diff --git a/.github/workflows/dynamic_arch.yml b/.github/workflows/dynamic_arch.yml
new file mode 100644
index 000000000..b6a4090bd
--- /dev/null
+++ b/.github/workflows/dynamic_arch.yml
@@ -0,0 +1,81 @@
+name: continuous build
+
+on: [push, pull_request]
+
+jobs:
+  build:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+        build: [cmake, make]
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v2
+
+      - name: Compilation cache
+        uses: actions/cache@v2
+        with:
+          path: ~/.ccache
+          # We include the commit sha in the cache key, as new cache entries are
+          # only created if there is no existing entry for the key yet.
+          key: ${{ runner.os }}-ccache-${{ github.sha }}
+          # Restore any ccache cache entry, if none for
+          # ${{ runner.os }}-ccache-${{ github.sha }} exists
+          restore-keys: |
+            ${{ runner.os }}-ccache
+
+      - name: Print system information
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            cat /proc/cpuinfo
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            sysctl -a | grep machdep.cpu
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+
+      - name: Install Dependencies
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            sudo apt-get install -y gfortran cmake ccache
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            brew install coreutils cmake ccache
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+          ccache -M 300M  # Limit the ccache size; Github's overall cache limit is 5GB
+
+      - name: Build
+        if: matrix.build == 'make'
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            export PATH="/usr/lib/ccache:${PATH}"
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            export PATH="$(brew --prefix)/opt/ccache/libexec:${PATH}"
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+
+          make -j$(nproc) DYNAMIC_ARCH=1 USE_OPENMP=0
+
+      - name: CMake build
+        if: matrix.build == 'cmake'
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            export PATH="/usr/lib/ccache:${PATH}"
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            export PATH="$(brew --prefix)/opt/ccache/libexec:${PATH}"
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+
+          mkdir build
+          cd build
+          cmake -DDYNAMIC_ARCH=1 -DNOFORTRAN=0 -DBUILD_WITHOUT_LAPACK=0  -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_BUILD_TYPE=Release ..
+          make -j$(nproc)
diff --git a/.github/workflows/nightly-Homebrew-build.yml b/.github/workflows/nightly-Homebrew-build.yml
index f55e73d23..8d7cfea2d 100644
--- a/.github/workflows/nightly-Homebrew-build.yml
+++ b/.github/workflows/nightly-Homebrew-build.yml
@@ -21,6 +21,7 @@ jobs:
   build-OpenBLAS-with-Homebrew:
     runs-on: macos-latest
     env:
+      DEVELOPER_DIR: /Applications/Xcode_11.4.1.app/Contents/Developer
       HOMEBREW_DEVELOPER: "ON"
       HOMEBREW_DISPLAY_INSTALL_TIMES: "ON"
       HOMEBREW_NO_ANALYTICS: "ON"
diff --git a/.gitignore b/.gitignore
index 6803a919e..bca79f043 100644
--- a/.gitignore
+++ b/.gitignore
@@ -70,6 +70,7 @@ test/SBLAT2.SUMM
 test/SBLAT3.SUMM
 test/ZBLAT2.SUMM
 test/ZBLAT3.SUMM
+test/SHBLAT3.SUMM
 test/cblat1
 test/cblat2
 test/cblat3
@@ -79,6 +80,7 @@ test/dblat3
 test/sblat1
 test/sblat2
 test/sblat3
+test/test_shgemm
 test/zblat1
 test/zblat2
 test/zblat3
diff --git a/.travis.yml b/.travis.yml
index c875572b2..101147353 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -16,7 +16,6 @@ matrix:
       before_script: &common-before
         - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
       script:
-        - set -e
         - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
         - make -C test $COMMON_FLAGS $BTYPE
         - make -C ctest $COMMON_FLAGS $BTYPE
@@ -108,7 +107,6 @@ matrix:
         - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
       before_script: *common-before
       script:
-        - set -e
         # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
         - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
               CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
@@ -151,7 +149,6 @@ matrix:
       before_script:
         - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
       script:
-        - set -e
         - mkdir build
         - CONFIG=Release
         - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 951271717..bb5322a1d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -6,7 +6,8 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 9.dev)
+set(OpenBLAS_PATCH_VERSION 10.dev)
+
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -23,6 +24,7 @@ option(BUILD_WITHOUT_CBLAS "Do not build the C interface (CBLAS) to the BLAS fun
 option(DYNAMIC_ARCH "Include support for multiple CPU targets, with automatic selection at runtime (x86/x86_64, aarch64 or ppc only)" OFF)
 option(DYNAMIC_OLDER "Include specific support for older x86 cpu models (Penryn,Dunnington,Atom,Nano,Opteron) with DYNAMIC_ARCH" OFF)
 option(BUILD_RELAPACK "Build with ReLAPACK (recursive implementation of several LAPACK functions on top of standard LAPACK)" OFF)
+option(USE_LOCKING "Use locks even in single-threaded builds to make them callable from multiple threads" OFF)
 if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
 option(NO_AFFINITY "Disable support for CPU affinity masks to avoid binding processes from e.g. R or numpy/scipy to a single core" ON)
 else()
@@ -86,9 +88,13 @@ if (NOT NO_LAPACK)
   list(APPEND SUBDIRS lapack)
 endif ()
 
+if (NOT DEFINED BUILD_HALF)
+ set (BUILD_HALF false)
+endif ()
 # set which float types we want to build for
 if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_COMPLEX AND NOT DEFINED BUILD_COMPLEX16)
   # if none are defined, build for all
+#  set(BUILD_HALF true)
   set(BUILD_SINGLE true)
   set(BUILD_DOUBLE true)
   set(BUILD_COMPLEX true)
@@ -120,6 +126,11 @@ if (BUILD_COMPLEX16)
   list(APPEND FLOAT_TYPES "ZCOMPLEX") # defines COMPLEX and DOUBLE
 endif ()
 
+if (BUILD_HALF)
+  message(STATUS "Building Half Precision")
+  list(APPEND FLOAT_TYPES "HALF") # defines nothing
+endif ()
+
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
   message(FATAL_ERROR "Detecting CPU failed. Please set TARGET explicitly, e.g. make TARGET=your_cpu_target. Please read README for details.")
 endif ()
@@ -234,7 +245,7 @@ if (BUILD_SHARED_LIBS AND BUILD_RELAPACK)
   if (NOT MSVC)
     target_link_libraries(${OpenBLAS_LIBNAME} "-Wl,-allow-multiple-definition")
   else()
-    target_link_libraries(${OpenBLAS_LIBNAME} "/FORCE:MULTIPLE")
+   set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /FORCE:MULTIPLE")
   endif()
 endif()
 
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index df497c1d2..aba39e56f 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -180,3 +180,13 @@ In chronological order:
   * [2019-12-23] optimize AVX2 CGEMM and ZGEMM
   * [2019-12-30] AVX2 CGEMM3M & ZGEMM3M kernels
   * [2020-01-07] optimize AVX2 SGEMM and STRMM
+
+* Rajalakshmi Srinivasaraghavan <https://github.com/RajalakshmiSR>
+  * [2020-04-15] Half-precision GEMM for bfloat16
+
+* Marius Hillenbrand <https://github.com/mhillenibm>
+  * [2020-05-12] Revise dynamic architecture detection for IBM z
+  * [2020-05-12] Add new sgemm and strmm kernel for IBM z14
+
+* Danfeng Zhang <https://github.com/craft-zhang>
+  * [2020-05-20] Improve performance of SGEMM and STRMM on Arm Cortex-A53
\ No newline at end of file
diff --git a/Changelog.txt b/Changelog.txt
index 5f924629b..cbf0b50f5 100644
--- a/Changelog.txt
+++ b/Changelog.txt
@@ -1,4 +1,77 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.10
+ 14-Jun-2020
+
+common:
+	* Improved thread locking behaviour in blas_server and parallel getrf                         
+	* Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
+	  due to overlong lines)
+	* Imported bugfix 403 from LAPACK (compile option "recursive" required
+	  for correctness with Intel and PGI)
+	* Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
+	* Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
+	* Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
+	  could lead to crashes at large matrix sizes  
+	* Restored internal soname in dynamic libraries on FreeBSD and Dragonfly 
+	* Added API (openblas_setaffinity) to set the thread affinity on Linux
+	* Added initial infrastructure for half-precision floating point 
+	  (bfloat16) support with a generic implementation of SHGEMM     
+	* Added CMAKE build system support for building the cblas_Xgemm3m
+	  functions
+	* Fixed CMAKE support for building in a path with embedded spaces
+	* Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
+	* Fixed GCC version detection in the Makefiles
+	* Allowed overriding the names of AR, AS and LD in Makefile builds
+
+POWER:
+	* Fixed big-endian POWER8 ELFv2 builds on FreeBSD
+	* Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
+	* Fixed CMAKE build support for POWER9
+	* fixed a potential race condition in the thread buffer allocation
+	* Worked around LAPACK test failures on PPC G4
+
+MIPS:
+	* Fixed a potential race condition in the thread buffer allocation
+	* Added support for MIPS 24K/24KE family based on P5600 kernels
+
+MIPS64:
+	* fixed a potential race condition in the thread buffer allocation
+	* Added TARGET=GENERIC
+
+ARMV7:
+	* Fixed a race condition in the thread buffer allocation
+
+ARMV8:
+	* Fixed a race condition in the thread buffer allocation
+	* Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
+	* Improved performance of the ThunderX2 DAXPY kernel
+	* Added an optimized SGEMM kernel for Cortex A53    
+	* Fixed Makefile support for INTERFACE64 (8-byte integer)
+
+x86_64:
+	* Fixed a syntax error in the CMAKE setup for SkylakeX
+	* Improved performance of STRSM on Haswell, SkylakeX and Ryzen
+	* Improved SGEMM performance on SGEMM for workloads with ldc a
+	  multiple of 1024
+	* Improved DGEMM performance on Skylake X
+	* Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
+	  builds created on SkylakeX
+	* Removed data alignment requirement in the SSE2 copy kernels
+	  that could cause spurious crashes
+	* Added a workaround for an optimizer bug in AppleClang 11.0.3
+	* Fixed LAPACK test failures due to wrong options for Intel Fortran
+	* Fixed compilation and LAPACK test results with recent Flang
+	  and AMD AOCC
+	* Fixed DYNAMIC_ARCH builds with CMAKE on OS X
+	* Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
+	  cblas_?sum, cblas_?gemm3m in the shared library on OS 
+	* Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
+	  show the name of an older generation chip supported by the same kernels)
+
+IBM Z:
+	* Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
+
 ====================================================================
 Version 0.3.9
  1-Mar-2020
diff --git a/Makefile b/Makefile
index 18320e6a3..e113026dd 100644
--- a/Makefile
+++ b/Makefile
@@ -264,6 +264,7 @@ lapack_prebuild :
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	-@echo "FC          = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "FFLAGS_DRV  = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS_NOOPT       = -O0 $(LAPACK_NOOPT)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "PNOOPT      = $(LAPACK_FPFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
diff --git a/Makefile.power b/Makefile.power
index 24d8aa8a7..5c431860f 100644
--- a/Makefile.power
+++ b/Makefile.power
@@ -9,6 +9,16 @@ else
 USE_OPENMP = 1
 endif
 
+ifeq ($(CORE), POWER10)
+ifeq ($(USE_OPENMP), 1)
+COMMON_OPT += -Ofast -mcpu=future -mtune=future -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+FCOMMON_OPT += -O2 -frecursive -mcpu=future -mtune=future -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
+else
+COMMON_OPT += -Ofast -mcpu=future -mtune=future -mvsx -malign-power -fno-fast-math
+FCOMMON_OPT += -O2 -frecursive -mcpu=future -mtune=future -malign-power -fno-fast-math
+endif
+endif
+
 ifeq ($(CORE), POWER9)
 ifeq ($(USE_OPENMP), 1)
 COMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp
diff --git a/Makefile.prebuild b/Makefile.prebuild
index b00f13368..48fb5e991 100644
--- a/Makefile.prebuild
+++ b/Makefile.prebuild
@@ -17,7 +17,11 @@ ifdef CPUIDEMU
 EXFLAGS = -DCPUIDEMU -DVENDOR=99
 endif
 
-ifeq ($(TARGET), 1004K)
+ifeq ($(TARGET), MIPS24K)
+TARGET_FLAGS = -mips32r2
+endif
+
+ifeq ($(TARGET), MIPS1004K)
 TARGET_FLAGS = -mips32r2
 endif
 
diff --git a/Makefile.rule b/Makefile.rule
index 724a60ec4..2c12177ee 100644
--- a/Makefile.rule
+++ b/Makefile.rule
@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.9.dev
+VERSION = 0.3.10.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -273,6 +273,9 @@ COMMON_PROF = -pg
 #
 # CPP_THREAD_SAFETY_TEST = 1
 
+
+# If you want to enable the experimental BFLOAT16 support
+# BUILD_HALF = 1
 #
 #  End of user configuration
 #
diff --git a/Makefile.system b/Makefile.system
index 2998c0e6a..2a0694d01 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -21,6 +21,8 @@ ifeq ($(ARCH), amd64)
 override ARCH=x86_64
 else ifeq ($(ARCH), powerpc64)
 override ARCH=power
+else ifeq ($(ARCH), powerpc)
+override ARCH=power
 else ifeq ($(ARCH), i386)
 override ARCH=x86
 else ifeq ($(ARCH), aarch64)
@@ -261,10 +263,10 @@ endif
 
 ARFLAGS	=
 CPP	= $(COMPILER) -E
-AR	= $(CROSS_SUFFIX)ar
-AS	= $(CROSS_SUFFIX)as
-LD	= $(CROSS_SUFFIX)ld
-RANLIB	= $(CROSS_SUFFIX)ranlib
+AR	?= $(CROSS_SUFFIX)ar
+AS	?= $(CROSS_SUFFIX)as
+LD	?= $(CROSS_SUFFIX)ld
+RANLIB	?= $(CROSS_SUFFIX)ranlib
 NM	= $(CROSS_SUFFIX)nm
 DLLWRAP = $(CROSS_SUFFIX)dllwrap
 OBJCOPY = $(CROSS_SUFFIX)objcopy
@@ -277,6 +279,17 @@ NO_LAPACK = 1
 override FEXTRALIB = 
 endif
 
+ifeq ($(C_COMPILER), GCC)
+GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
+GCCVERSIONEQ5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` = 5)
+GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
+GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
+GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
+GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 2)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+endif
+
 #
 #  OS dependent settings
 #
@@ -323,13 +336,7 @@ ifeq ($(C_COMPILER), CLANG)
 CCOMMON_OPT	+= -DMS_ABI
 endif
 
-ifeq ($(C_COMPILER), GCC)
 #Version tests for supporting specific features (MS_ABI, POWER9 intrinsics)
-GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
-GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
-GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
-GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
-GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
 ifeq ($(GCCVERSIONGT4), 1)
 # GCC Major version > 4
 # It is compatible with MSVC ABI.
@@ -343,7 +350,6 @@ ifeq ($(GCCMINORVERSIONGTEQ7), 1)
 CCOMMON_OPT	+= -DMS_ABI
 endif
 endif
-endif
 
 # Ensure the correct stack alignment on Win32
 # http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
@@ -563,8 +569,34 @@ DYNAMIC_CORE += EMAG8180
 endif
 
 ifeq ($(ARCH), zarch)
-DYNAMIC_CORE = Z13
+DYNAMIC_CORE = ZARCH_GENERIC
+
+# Z13 is supported since gcc-5.2, gcc-6, and in RHEL 7.3 and newer
+ifeq ($(GCCVERSIONGT5), 1)
+	ZARCH_SUPPORT_Z13 := 1
+else ifeq ($(GCCVERSIONEQ5), 1)
+ifeq ($(GCCMINORVERSIONGTEQ2), 1)
+	ZARCH_SUPPORT_Z13 := 1
+endif
+endif
+
+ifeq ($(wildcard /etc/redhat-release), /etc/redhat-release)
+ifeq ($(shell source /etc/os-release ; expr $$VERSION_ID \>= "7.3"), 1)
+	ZARCH_SUPPORT_Z13 := 1
+endif
+endif
+
+ifeq ($(ZARCH_SUPPORT_Z13), 1)
+DYNAMIC_CORE += Z13
+else
+$(info OpenBLAS: Not building Z13 kernels because gcc is older than 5.2 or 6.x)
+endif
+
+ifeq ($(GCCVERSIONGTEQ7), 1)
 DYNAMIC_CORE += Z14
+else
+$(info OpenBLAS: Not building Z14 kernels because gcc is older than 7.x)
+endif
 endif
 
 ifeq ($(ARCH), power)
@@ -572,14 +604,20 @@ DYNAMIC_CORE = POWER6
 DYNAMIC_CORE += POWER8
 ifneq ($(C_COMPILER), GCC)
 DYNAMIC_CORE += POWER9
+DYNAMIC_CORE += POWER10
 endif
 ifeq ($(C_COMPILER), GCC)
-GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
 ifeq ($(GCCVERSIONGT5), 1)
 DYNAMIC_CORE += POWER9
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
 endif
+GCCVERSIONGTEQ11 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 11)
+ifeq ($(GCCVERSIONGTEQ11), 1)
+DYNAMIC_CORE += POWER10
+else
+$(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)
+endif
 endif
 endif
 
@@ -690,7 +728,12 @@ CCOMMON_OPT += -march=mips64
 FCOMMON_OPT += -march=mips64
 endif
 
-ifeq ($(CORE), 1004K)
+ifeq ($(CORE), MIPS24K)
+CCOMMON_OPT += -mips32r2 -mtune=24kc  $(MSA_FLAGS)
+FCOMMON_OPT += -mips32r2 -mtune=24kc  $(MSA_FLAGS)
+endif
+
+ifeq ($(CORE), MIPS1004K)
 CCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
 FCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
 endif
@@ -755,6 +798,15 @@ endif
 
 ifeq ($(F_COMPILER), FLANG)
 CCOMMON_OPT += -DF_INTERFACE_FLANG
+FCOMMON_OPT += -Mrecursive -Kieee
+ifeq ($(OSNAME), Linux)
+ifeq ($(ARCH), x86_64)
+FLANG_VENDOR := $(shell expr `$(FC) --version|cut -f 1 -d "."|head -1`)
+ifeq ($(FLANG_VENDOR),AOCC)
+FCOMMON_OPT += -fno-unroll-loops
+endif
+endif
+endif
 ifdef BINARY64
 ifdef INTERFACE64
 ifneq ($(INTERFACE64), 0)
@@ -850,7 +902,7 @@ ifneq ($(INTERFACE64), 0)
 FCOMMON_OPT += -i8
 endif
 endif
-FCOMMON_OPT += -recursive
+FCOMMON_OPT += -recursive -fp-model strict -assume protect-parens
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -fopenmp
 endif
@@ -1119,6 +1171,10 @@ ifeq ($(USE_TLS), 1)
 CCOMMON_OPT += -DUSE_TLS
 endif
 
+ifeq ($(BUILD_HALF), 1)
+CCOMMON_OPT += -DBUILD_HALF
+endif
+
 CCOMMON_OPT += -DVERSION=\"$(VERSION)\"
 
 ifndef SYMBOLPREFIX
@@ -1145,6 +1201,7 @@ KERNELDIR	= $(TOPDIR)/kernel/$(ARCH)
 
 include $(TOPDIR)/Makefile.$(ARCH)
 
+CCOMMON_OPT     += -UASMNAME -UASMFNAME -UNAME -UCNAME -UCHAR_NAME -UCHAR_CNAME
 CCOMMON_OPT	+= -DASMNAME=$(FU)$(*F) -DASMFNAME=$(FU)$(*F)$(BU) -DNAME=$(*F)$(BU) -DCNAME=$(*F) -DCHAR_NAME=\"$(*F)$(BU)\" -DCHAR_CNAME=\"$(*F)\"
 
 ifeq ($(CORE), PPC440)
@@ -1237,7 +1294,6 @@ endif
 
 override CFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
 override PFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
-
 override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)
 override FPFLAGS    += $(FCOMMON_OPT) $(COMMON_PROF)
 #MAKEOVERRIDES =
@@ -1344,6 +1400,7 @@ export ARCH
 export CORE
 export LIBCORE
 export __BYTE_ORDER__
+export ELF_VERSION
 export PGCPATH
 export CONFIG
 export CC
@@ -1389,7 +1446,10 @@ export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE
 export NO_AVX512
+export BUILD_HALF
 
+export SHGEMM_UNROLL_M
+export SHGEMM_UNROLL_N
 export SGEMM_UNROLL_M
 export SGEMM_UNROLL_N
 export DGEMM_UNROLL_M
diff --git a/Makefile.tail b/Makefile.tail
index 2adede1a5..39902982b 100644
--- a/Makefile.tail
+++ b/Makefile.tail
@@ -1,3 +1,4 @@
+SHBLASOBJS_P = $(SHBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
 SBLASOBJS_P = $(SBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
 DBLASOBJS_P = $(DBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
 QBLASOBJS_P = $(QBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
@@ -9,8 +10,8 @@ COMMONOBJS_P = $(COMMONOBJS:.$(SUFFIX)=.$(PSUFFIX))
 
 HPLOBJS_P   = $(HPLOBJS:.$(SUFFIX)=.$(PSUFFIX))
 
-BLASOBJS    = $(SBLASOBJS)   $(DBLASOBJS)   $(CBLASOBJS)   $(ZBLASOBJS)
-BLASOBJS_P  = $(SBLASOBJS_P) $(DBLASOBJS_P) $(CBLASOBJS_P) $(ZBLASOBJS_P)
+BLASOBJS    = $(SHBLASOBJS)  $(SBLASOBJS)   $(DBLASOBJS)   $(CBLASOBJS)   $(ZBLASOBJS)
+BLASOBJS_P  = $(SHBLASOBJS_P) $(SBLASOBJS_P) $(DBLASOBJS_P) $(CBLASOBJS_P) $(ZBLASOBJS_P)
 
 ifdef EXPRECISION
 BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)
@@ -22,6 +23,7 @@ BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)
 BLASOBJS_P += $(QBLASOBJS_P) $(XBLASOBJS_P)
 endif
 
+$(SHBLASOBJS) $(SHBLASOBJS_P) : override CFLAGS += -DHALF -UDOUBLE  -UCOMPLEX
 $(SBLASOBJS) $(SBLASOBJS_P) : override CFLAGS += -UDOUBLE  -UCOMPLEX
 $(DBLASOBJS) $(DBLASOBJS_P) : override CFLAGS += -DDOUBLE  -UCOMPLEX
 $(QBLASOBJS) $(QBLASOBJS_P) : override CFLAGS += -DXDOUBLE -UCOMPLEX
@@ -29,6 +31,7 @@ $(CBLASOBJS) $(CBLASOBJS_P) : override CFLAGS += -UDOUBLE  -DCOMPLEX
 $(ZBLASOBJS) $(ZBLASOBJS_P) : override CFLAGS += -DDOUBLE  -DCOMPLEX
 $(XBLASOBJS) $(XBLASOBJS_P) : override CFLAGS += -DXDOUBLE -DCOMPLEX
 
+$(SHBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(SBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(DBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(QBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
diff --git a/Makefile.zarch b/Makefile.zarch
index 47ea1eb71..be1e34f6d 100644
--- a/Makefile.zarch
+++ b/Makefile.zarch
@@ -5,6 +5,6 @@ FCOMMON_OPT += -march=z13 -mzvector
 endif
 
 ifeq ($(CORE), Z14)
-CCOMMON_OPT += -march=z14 -mzvector
+CCOMMON_OPT += -march=z14 -mzvector -O3
 FCOMMON_OPT += -march=z14 -mzvector
 endif
diff --git a/README.md b/README.md
index 61393bd8f..6dc3c7b42 100644
--- a/README.md
+++ b/README.md
@@ -122,6 +122,11 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 - **AMD STEAMROLLER**: Uses Bulldozer codes with some optimizations.
 - **AMD ZEN**: Uses Haswell codes with some optimizations.
 
+#### MIPS32
+
+- **MIPS 1004K**: uses P5600 codes
+- **MIPS 24K**: uses P5600 codes
+
 #### MIPS64
 
 - **ICT Loongson 3A**: Optimized Level-3 BLAS and the part of Level-1,2.
diff --git a/TargetList.txt b/TargetList.txt
index f4a40ed02..4e54e3077 100644
--- a/TargetList.txt
+++ b/TargetList.txt
@@ -49,6 +49,7 @@ POWER6
 POWER7
 POWER8
 POWER9
+POWER10
 PPCG4
 PPC970
 PPC970MP
@@ -58,7 +59,8 @@ CELL
 
 3.MIPS CPU:
 P5600
-1004K
+MIPS1004K
+MIPS24K
 
 4.MIPS64 CPU:
 SICORTEX
diff --git a/azure-pipelines.yml b/azure-pipelines.yml
index 9b4c85367..639cb3558 100644
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@@ -49,3 +49,23 @@ jobs:
       # we need a privileged docker run for sde process attachment
       docker run --privileged intel_sde
     displayName: 'Run AVX512 SkylakeX docker build / test'
+
+- job: Windows_cl
+  pool:
+     vmImage: 'windows-latest'
+  steps:   
+  - task: CMake@1
+    inputs:
+      workingDirectory: 'build' # Optional
+      cmakeArgs: '-G "Visual Studio 16 2019" ..'
+  - task: CMake@1
+    inputs:
+      cmakeArgs: '--build . --config Release'
+      workingDirectory: 'build'
+  - script: |
+      cd build
+      cd utest
+      dir
+      openblas_utest.exe
+  
+      
diff --git a/benchmark/Makefile b/benchmark/Makefile
index 90d903ad7..2f70ceaf3 100644
--- a/benchmark/Makefile
+++ b/benchmark/Makefile
@@ -49,6 +49,12 @@ else
 GOTO_LAPACK_TARGETS=
 endif
 
+ifeq ($(BUILD_HALF),1)
+GOTO_HALF_TARGETS=shgemm.goto
+else
+GOTO_HALF_TARGETS=
+endif
+
 ifeq ($(OSNAME), WINNT)
 
 goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
@@ -91,7 +97,7 @@ goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
        sgetri.goto dgetri.goto cgetri.goto zgetri.goto \
        spotrf.goto dpotrf.goto cpotrf.goto zpotrf.goto \
        ssymm.goto dsymm.goto csymm.goto zsymm.goto \
-       saxpby.goto daxpby.goto caxpby.goto zaxpby.goto
+       saxpby.goto daxpby.goto caxpby.goto zaxpby.goto $(GOTO_HALF_TARGETS)
 
 acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
        scholesky.acml dcholesky.acml ccholesky.acml zcholesky.acml \
@@ -264,7 +270,7 @@ goto :: sgemm.goto dgemm.goto cgemm.goto zgemm.goto \
        samin.goto damin.goto camin.goto zamin.goto \
        smin.goto dmin.goto \
        saxpby.goto daxpby.goto caxpby.goto zaxpby.goto \
-       snrm2.goto dnrm2.goto scnrm2.goto dznrm2.goto $(GOTO_LAPACK_TARGETS)
+       snrm2.goto dnrm2.goto scnrm2.goto dznrm2.goto $(GOTO_LAPACK_TARGETS) $(GOTO_HALF_TARGETS)
 
 acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
        scholesky.acml dcholesky.acml ccholesky.acml zcholesky.acml \
@@ -614,6 +620,11 @@ zcholesky.essl : zcholesky.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
 ##################################### Sgemm ####################################################
+ifeq ($(BUILD_HALF),1)
+shgemm.goto : shgemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+endif
+
 sgemm.goto : sgemm.$(SUFFIX) ../$(LIBNAME)
 	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
@@ -1814,7 +1825,7 @@ zsymv.veclib : zsymv.$(SUFFIX)
 
 ##################################### Sgeev ####################################################
 sgeev.goto : sgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 sgeev.acml : sgeev.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1830,7 +1841,7 @@ sgeev.veclib : sgeev.$(SUFFIX)
 
 ##################################### Dgeev ####################################################
 dgeev.goto : dgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 dgeev.acml : dgeev.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1847,7 +1858,7 @@ dgeev.veclib : dgeev.$(SUFFIX)
 ##################################### Cgeev ####################################################
 
 cgeev.goto : cgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 cgeev.acml : cgeev.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1864,7 +1875,7 @@ cgeev.veclib : cgeev.$(SUFFIX)
 ##################################### Zgeev ####################################################
 
 zgeev.goto : zgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 zgeev.acml : zgeev.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1880,7 +1891,7 @@ zgeev.veclib : zgeev.$(SUFFIX)
 
 ##################################### Sgetri ####################################################
 sgetri.goto : sgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 sgetri.acml : sgetri.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1896,7 +1907,7 @@ sgetri.veclib : sgetri.$(SUFFIX)
 
 ##################################### Dgetri ####################################################
 dgetri.goto : dgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 dgetri.acml : dgetri.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1913,7 +1924,7 @@ dgetri.veclib : dgetri.$(SUFFIX)
 ##################################### Cgetri ####################################################
 
 cgetri.goto : cgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 cgetri.acml : cgetri.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -1930,7 +1941,7 @@ cgetri.veclib : cgetri.$(SUFFIX)
 ##################################### Zgetri ####################################################
 
 zgetri.goto : zgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
 
 zgetri.acml : zgetri.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
@@ -2916,6 +2927,11 @@ ccholesky.$(SUFFIX) : cholesky.c
 zcholesky.$(SUFFIX) : cholesky.c
 	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
 
+ifeq ($(BUILD_HALF),1)
+shgemm.$(SUFFIX) : gemm.c
+	$(CC) $(CFLAGS) -c -DHALF -UCOMPLEX -UDOUBLE -o $(@F) $^
+endif
+
 sgemm.$(SUFFIX) : gemm.c
 	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
 
diff --git a/benchmark/gemm.c b/benchmark/gemm.c
index dd016a7c3..d2235330b 100644
--- a/benchmark/gemm.c
+++ b/benchmark/gemm.c
@@ -39,6 +39,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifdef DOUBLE
 #define GEMM   BLASFUNC(dgemm)
+#elif defined(HALF)
+#define GEMM   BLASFUNC(shgemm)
 #else
 #define GEMM   BLASFUNC(sgemm)
 #endif
@@ -120,7 +122,8 @@ static void *huge_malloc(BLASLONG size){
 
 int main(int argc, char *argv[]){
 
-  FLOAT *a, *b, *c;
+  IFLOAT *a, *b;
+  FLOAT *c;
   FLOAT alpha[] = {1.0, 0.0};
   FLOAT beta [] = {0.0, 0.0};
   char transa = 'N';
@@ -184,10 +187,10 @@ int main(int argc, char *argv[]){
     k = to;
   }
 
-  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * m * k * COMPSIZE)) == NULL) {
+  if (( a = (IFLOAT *)malloc(sizeof(IFLOAT) * m * k * COMPSIZE)) == NULL) {
     fprintf(stderr,"Out of Memory!!\n");exit(1);
   }
-  if (( b = (FLOAT *)malloc(sizeof(FLOAT) * k * n * COMPSIZE)) == NULL) {
+  if (( b = (IFLOAT *)malloc(sizeof(IFLOAT) * k * n * COMPSIZE)) == NULL) {
     fprintf(stderr,"Out of Memory!!\n");exit(1);
   }
   if (( c = (FLOAT *)malloc(sizeof(FLOAT) * m * n * COMPSIZE)) == NULL) {
@@ -199,10 +202,10 @@ int main(int argc, char *argv[]){
 #endif
 
   for (i = 0; i < m * k * COMPSIZE; i++) {
-    a[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+    a[i] = ((IFLOAT) rand() / (IFLOAT) RAND_MAX) - 0.5;
   }
   for (i = 0; i < k * n * COMPSIZE; i++) {
-    b[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+    b[i] = ((IFLOAT) rand() / (IFLOAT) RAND_MAX) - 0.5;
   }
   for (i = 0; i < m * n * COMPSIZE; i++) {
     c[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
diff --git a/benchmark/zdot.c b/benchmark/zdot.c
index ed9d4d2e8..136135c9c 100644
--- a/benchmark/zdot.c
+++ b/benchmark/zdot.c
@@ -170,9 +170,11 @@ int main(int argc, char *argv[]){
 			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
    	}
     	gettimeofday( &start, (struct timezone *)0);
-
+#ifdef RETURN_BY_STACK
+    	DOT (&result , &m, x, &inc_x, y, &inc_y );
+#else
     	result = DOT (&m, x, &inc_x, y, &inc_y );
-
+#endif
     	gettimeofday( &stop, (struct timezone *)0);
 
     	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
diff --git a/c_check b/c_check
index c7899c84f..8234c2081 100644
--- a/c_check
+++ b/c_check
@@ -310,6 +310,7 @@ $linker_a = "";
 	    && ($flags !~ /advapi32/)
 	    && ($flags !~ /shell32/)
 	    && ($flags !~ /omp/)
+	    && ($flags !~ /[0-9]+/)
 	    ) {
 	    $linker_l .= $flags . " "
 	}
diff --git a/cmake/arch.cmake b/cmake/arch.cmake
index 9d51f777c..d56ba99cb 100644
--- a/cmake/arch.cmake
+++ b/cmake/arch.cmake
@@ -49,7 +49,7 @@ if (DYNAMIC_ARCH)
   endif ()
   
   if (POWER)
-    set(DYNAMIC_CORE POWER6 POWER8 POWER9)
+	  set(DYNAMIC_CORE POWER6 POWER8 POWER9 POWER10)
   endif ()
   
   if (X86)
diff --git a/cmake/fc.cmake b/cmake/fc.cmake
index cc330ae2c..fc1f9bb22 100644
--- a/cmake/fc.cmake
+++ b/cmake/fc.cmake
@@ -16,6 +16,7 @@ if (${F_COMPILER} STREQUAL "FLANG")
   if (USE_OPENMP)
     set(FCOMMON_OPT "${FCOMMON_OPT} -fopenmp")
   endif ()
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Mrecursive -Kieee")
 endif ()
 
 if (${F_COMPILER} STREQUAL "G77")
diff --git a/cmake/kernel.cmake b/cmake/kernel.cmake
index 9b238f004..4b505a102 100644
--- a/cmake/kernel.cmake
+++ b/cmake/kernel.cmake
@@ -113,11 +113,31 @@ macro(SetDefaultL1)
   set(ZSUMKERNEL zsum.S)
   set(QSUMKERNEL sum.S)
   set(XSUMKERNEL zsum.S)
+if (BUILD_HALF)
+  set(SHAMINKERNEL ../arm/amin.c)
+  set(SHAMAXKERNEL ../arm/amax.c)
+  set(SHMAXKERNEL ../arm/max.c)
+  set(SHMINKERNEL ../arm/min.c)
+  set(ISHAMAXKERNEL ../arm/iamax.c)
+  set(ISHAMINKERNEL ../arm/iamin.c)
+  set(ISHMAXKERNEL ../arm/imax.c)
+  set(ISHMINKERNEL ../arm/imin.c)
+  set(SHASUMKERNEL ../arm/asum.c)
+  set(SHAXPYKERNEL ../arm/axpy.c)
+  set(SHAXPBYKERNEL ../arm/axpby.c)
+  set(SHCOPYKERNEL ../arm/copy.c)
+  set(SHDOTKERNEL ../arm/dot.c)
+  set(SHROTKERNEL ../arm/rot.c)
+  set(SHSCALKERNEL ../arm/scal.c)
+  set(SHNRM2KERNEL ../arm/nrm2.c)
+  set(SHSUMKERNEL ../arm/sum.c)
+  set(SHSWAPKERNEL ../arm/swap.c)
+endif ()
 endmacro ()
 
 macro(SetDefaultL2)
-  set(SGEMVNKERNEL gemv_n.S)
-  set(SGEMVTKERNEL gemv_t.S)
+  set(SGEMVNKERNEL ../arm/gemv_n.c)
+  set(SGEMVTKERNEL ../arm/gemv_t.c)
   set(DGEMVNKERNEL gemv_n.S)
   set(DGEMVTKERNEL gemv_t.S)
   set(CGEMVNKERNEL zgemv_n.S)
@@ -161,6 +181,11 @@ macro(SetDefaultL2)
   set(XHEMV_L_KERNEL ../generic/zhemv_k.c)
   set(XHEMV_V_KERNEL ../generic/zhemv_k.c)
   set(XHEMV_M_KERNEL ../generic/zhemv_k.c)
+if (BUILD_HALF)
+  set(SHGEMVNKERNEL ../arm/gemv_n.c)
+  set(SHGEMVTKERNEL ../arm/gemv_t.c)
+  set(SHGERKERNEL ../generic/ger.c)
+endif ()
 endmacro ()
 
 macro(SetDefaultL3)
@@ -168,4 +193,18 @@ macro(SetDefaultL3)
   set(DGEADD_KERNEL ../generic/geadd.c)
   set(CGEADD_KERNEL ../generic/zgeadd.c)
   set(ZGEADD_KERNEL ../generic/zgeadd.c)
+if (BUILD_HALF)
+  set(SHGEADD_KERNEL ../generic/geadd.c)
+  set(SHGEMMKERNEL ../generic/gemmkernel_2x2.c)
+  set(SHGEMM_BETA  ../generic/gemm_beta.c)
+  set(SHGEMMINCOPY ../generic/gemm_ncopy_2.c)
+  set(SHGEMMITCOPY ../generic/gemm_tcopy_2.c)
+  set(SHGEMMONCOPY ../generic/gemm_ncopy_2.c)
+  set(SHGEMMOTCOPY ../generic/gemm_tcopy_2.c)
+  set(SHGEMMINCOPYOBJ shgemm_incopy.o)
+  set(SHGEMMITCOPYOBJ shgemm_itcopy.o)
+  set(SHGEMMONCOPYOBJ shgemm_oncopy.o)
+  set(SHGEMMOTCOPYOBJ shgemm_otcopy.o)
+endif ()
+
 endmacro ()
diff --git a/cmake/os.cmake b/cmake/os.cmake
index 2d25e7aaa..c644bc3f7 100644
--- a/cmake/os.cmake
+++ b/cmake/os.cmake
@@ -8,7 +8,7 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
   set(NO_EXPRECISION 1)
 endif ()
 
-if (${CMAKE_SYSTEM_NAME} MATCHES "FreeBSD|OpenBSD|NetBSD|DragonFly")
+if (${CMAKE_SYSTEM_NAME} MATCHES "FreeBSD|OpenBSD|NetBSD|DragonFly|Darwin")
   set(EXTRALIB "${EXTRALIB} -lm")
   set(NO_EXPRECISION 1)
 endif ()
diff --git a/cmake/prebuild.cmake b/cmake/prebuild.cmake
index 44e1473d1..30256870c 100644
--- a/cmake/prebuild.cmake
+++ b/cmake/prebuild.cmake
@@ -16,6 +16,8 @@
 # HAVE_SSE2
 # HAVE_SSE3
 # MAKE
+# SHGEMM_UNROLL_M
+# SHGEMM_UNROLL_N
 # SGEMM_UNROLL_M
 # SGEMM_UNROLL_N
 # DGEMM_UNROLL_M
@@ -418,7 +420,7 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
     set(ZGEMM_UNROLL_M 8)
     set(ZGEMM_UNROLL_N 2)
     set(SYMV_P 8)
-  elseif ("${TCORE}" STREQUAL "POWER9")
+  elseif ("${TCORE}" STREQUAL "POWER9" OR "${TCORE}" STREQUAL "POWER10")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE 32768\n"
       "#define L1_DATA_LINESIZE 128\n"
@@ -437,6 +439,8 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
     set(ZGEMM_UNROLL_N 2)
     set(SYMV_P 8)
   endif()
+  set(SHGEMM_UNROLL_M 8)
+  set(SHGEMM_UNROLL_N 4)
 
   # Or should this actually be NUM_CORES?
   if (${NUM_THREADS} GREATER 0)
@@ -488,7 +492,7 @@ else(NOT CMAKE_CROSSCOMPILING)
   if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
     try_compile(GETARCH_RESULT ${GETARCH_DIR}
       SOURCES ${GETARCH_SRC}
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${GETARCH_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I"${GETARCH_DIR}" -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
       OUTPUT_VARIABLE GETARCH_LOG
       COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
     )
@@ -516,7 +520,7 @@ execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH_BIN}" 1 OUTPUT_VARIABLE
   if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
     try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
       SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${GETARCH2_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I"${GETARCH2_DIR}" -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
       OUTPUT_VARIABLE GETARCH2_LOG
       COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
     )
diff --git a/cmake/system.cmake b/cmake/system.cmake
index ce980a7b9..d8dcc3cf3 100644
--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@@ -297,6 +297,16 @@ if (USE_SIMPLE_THREADED_LEVEL3)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_SIMPLE_THREADED_LEVEL3")
 endif ()
 
+if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+if (DEFINED MAX_STACK_ALLOC)
+if (NOT ${MAX_STACK_ALLOC} EQUAL 0)
+set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_STACK_ALLOC=${MAX_STACK_ALLOC}")
+endif ()
+else ()
+set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_STACK_ALLOC=2048")
+endif ()
+endif ()
+
 if (DEFINED LIBNAMESUFFIX)
   set(LIBPREFIX "libopenblas_${LIBNAMESUFFIX}")
 else ()
@@ -407,6 +417,14 @@ if (${CMAKE_C_COMPILER} STREQUAL "LSB" OR ${CMAKE_SYSTEM_NAME} STREQUAL "Windows
   set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DLAPACK_COMPLEX_STRUCTURE")
 endif ()
 
+if ("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+if ("${F_COMPILER}" STREQUAL "FLANG")
+if (${CMAKE_Fortran_COMPILER_VERSION} VERSION_LESS_EQUAL 3)
+  set(CMAKE_Fortran_FLAGS_RELEASE "${CMAKE_Fortran_FLAGS_RELEASE} -fno-unroll-loops")
+endif ()
+endif ()
+endif ()
+
 if (NOT DEFINED SUFFIX)
   set(SUFFIX o)
 endif ()
@@ -530,6 +548,8 @@ endif ()
 #export FUNCTION_PROFILE
 #export TARGET_CORE
 #
+#export SHGEMM_UNROLL_M
+#export SHGEMM_UNROLL_N
 #export SGEMM_UNROLL_M
 #export SGEMM_UNROLL_N
 #export DGEMM_UNROLL_M
diff --git a/cmake/utils.cmake b/cmake/utils.cmake
index fd93f8a70..1c21e776e 100644
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@@ -15,12 +15,36 @@ endfunction ()
 # Reads a Makefile into CMake vars.
 macro(ParseMakefileVars MAKEFILE_IN)
   message(STATUS "Reading vars from ${MAKEFILE_IN}...")
+        set (IfElse 0)
+        set (ElseSeen 0)
   file(STRINGS ${MAKEFILE_IN} makefile_contents)
   foreach (makefile_line ${makefile_contents})
+#message(STATUS "parsing ${makefile_line}")
+    if (${IfElse} GREATER 0)
+      string(REGEX MATCH "endif[ \t]*" line_match "${makefile_line}")
+      if (NOT "${line_match}" STREQUAL "")
+#           message(STATUS "ENDIF ${makefile_line}")
+        set (IfElse 0)
+        set (ElseSeen 0)
+        continue ()
+      endif ()
+      string(REGEX MATCH "else[ \t]*" line_match "${makefile_line}")
+        if (NOT "${line_match}" STREQUAL "") 
+#           message(STATUS "ELSE ${makefile_line}")
+           set (ElseSeen 1)        
+           continue ()  
+        endif()
+      if ( (${IfElse} EQUAL 2 AND ${ElseSeen} EQUAL 0) OR ( ${IfElse} EQUAL 1 AND ${ElseSeen} EQUAL 1))
+#           message(STATUS "skipping ${makefile_line}")
+         continue ()
+      endif ()    
+    endif ()    
     string(REGEX MATCH "([0-9_a-zA-Z]+)[ \t]*=[ \t]*(.+)$" line_match "${makefile_line}")
     if (NOT "${line_match}" STREQUAL "")
+#message(STATUS "match on ${line_match}")
       set(var_name ${CMAKE_MATCH_1})
-      set(var_value ${CMAKE_MATCH_2})
+#      set(var_value ${CMAKE_MATCH_2})
+      string(STRIP ${CMAKE_MATCH_2} var_value)
       # check for Makefile variables in the string, e.g. $(TSUFFIX)
       string(REGEX MATCHALL "\\$\\(([0-9_a-zA-Z]+)\\)" make_var_matches ${var_value})
       foreach (make_var ${make_var_matches})
@@ -33,7 +57,31 @@ macro(ParseMakefileVars MAKEFILE_IN)
     else ()
       string(REGEX MATCH "include \\$\\(KERNELDIR\\)/(.+)$" line_match "${makefile_line}")
       if (NOT "${line_match}" STREQUAL "")
+#message(STATUS "match on include ${line_match}")
         ParseMakefileVars(${KERNELDIR}/${CMAKE_MATCH_1})
+      else ()
+#        message(STATUS "unmatched line ${line_match}")
+        string(REGEX MATCH "ifeq \\(\\$\\(([_A-Z]+)\\),[ \t]*([0-9_A-Z]+)\\)" line_match "${makefile_line}")
+        if (NOT "${line_match}" STREQUAL "")
+#          message(STATUS "IFEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_2}")
+          if (DEFINED ${${CMAKE_MATCH_1}} AND ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2})
+#            message (STATUS "condition is true")
+            set (IfElse 1)
+          else ()
+            set (IfElse 2)
+          endif ()
+        else ()
+          string(REGEX MATCH "ifneq \\(\\$\\(([_A-Z]+)\\),[ \t]*([0-9_A-Z]+)\\)" line_match "${makefile_line}")
+          if (NOT "${line_match}" STREQUAL "")
+#            message(STATUS "IFNEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_2}")
+            if (NOT ( ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2}))
+#              message (STATUS "condition is true")
+              set (IfElse 1)
+            else ()
+              set (IfElse 2)
+            endif ()
+          endif ()
+        endif ()
       endif ()
     endif ()
   endforeach ()
@@ -163,6 +211,7 @@ function(GenerateNamedObjects sources_in)
   if (complex_only)
     list(REMOVE_ITEM float_list "SINGLE")
     list(REMOVE_ITEM float_list "DOUBLE")
+    list(REMOVE_ITEM float_list "HALF")
   elseif (real_only)
     list(REMOVE_ITEM float_list "COMPLEX")
     list(REMOVE_ITEM float_list "ZCOMPLEX")
@@ -176,6 +225,9 @@ function(GenerateNamedObjects sources_in)
       if (NOT no_float_type)
         string(SUBSTRING ${float_type} 0 1 float_char)
         string(TOLOWER ${float_char} float_char)
+	if (${float_type} STREQUAL "HALF")
+		set (float_char "sh")
+	endif ()
       endif ()
 
       if (NOT name_in)
@@ -210,6 +262,9 @@ function(GenerateNamedObjects sources_in)
       if (${float_type} STREQUAL "DOUBLE" OR ${float_type} STREQUAL "ZCOMPLEX")
         list(APPEND obj_defines "DOUBLE")
       endif ()
+      if (${float_type} STREQUAL "HALF")
+        list(APPEND obj_defines "HALF")
+      endif ()
       if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
         list(APPEND obj_defines "COMPLEX")
         if (mangle_complex_sources)
diff --git a/common.h b/common.h
index 762968e6f..00b34a3f7 100644
--- a/common.h
+++ b/common.h
@@ -257,6 +257,11 @@ typedef long BLASLONG;
 typedef unsigned long BLASULONG;
 #endif
 
+#ifndef BFLOAT16
+typedef unsigned short bfloat16;
+#define HALFCONVERSION 1
+#endif
+
 #ifdef USE64BITINT
 typedef BLASLONG blasint;
 #if defined(OS_WINDOWS) && defined(__64BIT__)
@@ -297,6 +302,13 @@ typedef int blasint;
 #define SIZE	8
 #define  BASE_SHIFT 3
 #define ZBASE_SHIFT 4
+#elif defined(HALF)
+#define IFLOAT	bfloat16
+#define XFLOAT IFLOAT
+#define FLOAT	float
+#define SIZE   2
+#define BASE_SHIFT 1
+#define ZBASE_SHIFT 2
 #else
 #define FLOAT	float
 #define SIZE    4
@@ -308,6 +320,10 @@ typedef int blasint;
 #define XFLOAT	FLOAT
 #endif
 
+#ifndef IFLOAT
+#define IFLOAT	FLOAT
+#endif
+
 #ifndef COMPLEX
 #define COMPSIZE  1
 #else
@@ -344,13 +360,8 @@ typedef int blasint;
 #endif
 #endif
 
-#ifdef POWER8
-#ifndef YIELDING
-#define YIELDING        __asm__ __volatile__ ("nop;nop;nop;nop;nop;nop;nop;nop;\n");
-#endif
-#endif
 
-#ifdef POWER9
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #ifndef YIELDING
 #define YIELDING        __asm__ __volatile__ ("nop;nop;nop;nop;nop;nop;nop;nop;\n");
 #endif
diff --git a/common_interface.h b/common_interface.h
index c350ac8ec..78f5be6b0 100644
--- a/common_interface.h
+++ b/common_interface.h
@@ -469,6 +469,8 @@ void BLASFUNC(xhbmv)(char *, blasint *, blasint *, xdouble *, xdouble *, blasint
 
 /* Level 3 routines */
 
+void BLASFUNC(shgemm)(char *, char *, blasint *, blasint *, blasint *, float *,
+	   bfloat16 *, blasint *, bfloat16 *, blasint *, float *, float *, blasint *);
 void BLASFUNC(sgemm)(char *, char *, blasint *, blasint *, blasint *, float *,
 	   float  *, blasint *, float  *, blasint *, float  *, float  *, blasint *);
 void BLASFUNC(dgemm)(char *, char *, blasint *, blasint *, blasint *, double *,
diff --git a/common_level3.h b/common_level3.h
index 6fa902be8..4e44a5e73 100644
--- a/common_level3.h
+++ b/common_level3.h
@@ -55,6 +55,8 @@ extern void sgemm_kernel_direct(BLASLONG M, BLASLONG N, BLASLONG K,
 extern int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K);
 
 
+int shgemm_beta(BLASLONG, BLASLONG, BLASLONG, float,
+	       bfloat16 *, BLASLONG, bfloat16 *, BLASLONG, float *, BLASLONG);
 int sgemm_beta(BLASLONG, BLASLONG, BLASLONG, float,
 	       float  *, BLASLONG, float   *, BLASLONG, float  *, BLASLONG);
 int dgemm_beta(BLASLONG, BLASLONG, BLASLONG, double,
@@ -76,6 +78,10 @@ int xgemm_beta(BLASLONG, BLASLONG, BLASLONG, xdouble *,
 	       xdouble *, BLASLONG, xdouble  *, BLASLONG, xdouble *, BLASLONG);
 #endif
 
+int shgemm_incopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
+int shgemm_itcopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
+int shgemm_oncopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
+int shgemm_otcopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
 int sgemm_incopy(BLASLONG m, BLASLONG n, float *a, BLASLONG lda, float *b);
 int sgemm_itcopy(BLASLONG m, BLASLONG n, float *a, BLASLONG lda, float *b);
 int sgemm_oncopy(BLASLONG m, BLASLONG n, float *a, BLASLONG lda, float *b);
@@ -499,6 +505,7 @@ int xher2k_kernel_UC(BLASLONG m, BLASLONG n, BLASLONG k, xdouble alpha_r, xdoubl
 int xher2k_kernel_LN(BLASLONG m, BLASLONG n, BLASLONG k, xdouble alpha_r, xdouble alpha_i, xdouble *a, xdouble *b, xdouble *c, BLASLONG ldc, BLASLONG offset, int flag);
 int xher2k_kernel_LC(BLASLONG m, BLASLONG n, BLASLONG k, xdouble alpha_r, xdouble alpha_i, xdouble *a, xdouble *b, xdouble *c, BLASLONG ldc, BLASLONG offset, int flag);
 
+int shgemm_kernel(BLASLONG, BLASLONG, BLASLONG, float,  bfloat16 *, bfloat16 *, float *, BLASLONG);
 int sgemm_kernel(BLASLONG, BLASLONG, BLASLONG, float,  float  *, float  *, float  *, BLASLONG);
 int dgemm_kernel(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG);
 
@@ -527,6 +534,11 @@ int cgemm3m_kernel(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float
 int zgemm3m_kernel(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
 int xgemm3m_kernel(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
 
+int shgemm_nn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_nt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_tn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_tt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+
 int sgemm_nn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_nt(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_tn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
@@ -619,6 +631,11 @@ int xgemm_cr(blas_arg_t *, BLASLONG *, BLASLONG *, xdouble *, xdouble *, BLASLON
 int xgemm_cc(blas_arg_t *, BLASLONG *, BLASLONG *, xdouble *, xdouble *, BLASLONG);
 #endif
 
+int shgemm_thread_nn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_thread_nt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_thread_tn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_thread_tt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+
 int sgemm_thread_nn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_thread_nt(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_thread_tn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
diff --git a/common_macro.h b/common_macro.h
index 13bb85794..8fe1f156f 100644
--- a/common_macro.h
+++ b/common_macro.h
@@ -39,6 +39,7 @@
 #ifndef COMMON_MACRO
 #define COMMON_MACRO
 
+#include "common_sh.h"
 #include "common_s.h"
 #include "common_d.h"
 #include "common_q.h"
@@ -642,6 +643,288 @@
 #define IMATCOPY_K_RT		DIMATCOPY_K_RT
 
 #define GEADD_K                 DGEADD_K
+
+#elif defined(HALF)
+
+#define	AMAX_K			SAMAX_K
+#define	AMIN_K			SAMIN_K
+#define	MAX_K			SMAX_K
+#define	MIN_K			SMIN_K
+#define	IAMAX_K			ISAMAX_K
+#define	IAMIN_K			ISAMIN_K
+#define	IMAX_K			ISMAX_K
+#define	IMIN_K			ISMIN_K
+#define	ASUM_K			SASUM_K
+#define	DOTU_K			SDOTU_K
+#define	DOTC_K			SDOTC_K
+#define	AXPYU_K			SAXPYU_K
+#define	AXPYC_K			SAXPYC_K
+#define AXPBY_K     SAXPBY_K
+#define SCAL_K			SSCAL_K
+#define GEMV_N			SGEMV_N
+#define GEMV_T			SGEMV_T
+#define SYMV_U			SSYMV_U
+#define SYMV_L			SSYMV_L
+#define	GERU_K			SGERU_K
+#define	GERC_K			SGERC_K
+#define	GERV_K			SGERV_K
+#define	GERD_K			SGERD_K
+#define	SUM_K			SSUM_K
+#define	SWAP_K			SSWAP_K
+#define	ROT_K			SROT_K
+#define COPY_K    SCOPY_K
+#define NRM2_K    SNRM2_K
+#define SYMV_THREAD_U		SSYMV_THREAD_U
+#define SYMV_THREAD_L		SSYMV_THREAD_L
+#define GEMM_BETA               SHGEMM_BETA
+#define	GEMM_KERNEL_N		SHGEMM_KERNEL
+#define	GEMM_KERNEL_L		SHGEMM_KERNEL
+#define	GEMM_KERNEL_R		SHGEMM_KERNEL
+#define	GEMM_KERNEL_B		SHGEMM_KERNEL
+
+#define	GEMM_NN			SHGEMM_NN
+#define	GEMM_CN			SHGEMM_TN
+#define	GEMM_TN			SHGEMM_TN
+#define	GEMM_NC			SHGEMM_NT
+#define	GEMM_NT			SHGEMM_NT
+#define	GEMM_CC			SHGEMM_TT
+#define	GEMM_CT			SHGEMM_TT
+#define	GEMM_TC			SHGEMM_TT
+#define	GEMM_TT			SHGEMM_TT
+#define	GEMM_NR			SHGEMM_NN
+#define	GEMM_TR			SHGEMM_TN
+#define	GEMM_CR			SHGEMM_TN
+#define	GEMM_RN			SHGEMM_NN
+#define	GEMM_RT			SHGEMM_NT
+#define	GEMM_RC			SHGEMM_NT
+#define	GEMM_RR			SHGEMM_NN
+#define	GEMM_ONCOPY		SHGEMM_ONCOPY
+#define	GEMM_OTCOPY		SHGEMM_OTCOPY
+#define	GEMM_INCOPY		SHGEMM_INCOPY
+#define	GEMM_ITCOPY		SHGEMM_ITCOPY
+#define SYMM_THREAD_LU          SSYMM_THREAD_LU
+#define SYMM_THREAD_LL          SSYMM_THREAD_LL
+#define SYMM_THREAD_RU          SSYMM_THREAD_RU
+#define SYMM_THREAD_RL          SSYMM_THREAD_RL
+#define SYMM_LU                 SSYMM_LU
+#define SYMM_LL                 SSYMM_LL
+#define SYMM_RU                 SSYMM_RU
+#define SYMM_RL                 SSYMM_RL
+
+
+#define HEMM_THREAD_LU          SHEMM_THREAD_LU
+#define HEMM_THREAD_LL          SHEMM_THREAD_LL
+#define HEMM_THREAD_RU          SHEMM_THREAD_RU
+#define HEMM_THREAD_RL          SHEMM_THREAD_RL
+
+#define	GEMM_THREAD_NN		SHGEMM_THREAD_NN
+#define	GEMM_THREAD_CN		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_TN		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_NC		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_NT		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_CC		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_CT		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_TC		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_TT		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_NR		SHGEMM_THREAD_NN
+#define	GEMM_THREAD_TR		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_CR		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_RN		SHGEMM_THREAD_NN
+#define	GEMM_THREAD_RT		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_RC		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_RR		SHGEMM_THREAD_NN
+
+#ifdef UNIT
+
+#define	TRMM_OUNCOPY		STRMM_OUNUCOPY
+#define	TRMM_OUTCOPY		STRMM_OUTUCOPY
+#define	TRMM_OLNCOPY		STRMM_OLNUCOPY
+#define	TRMM_OLTCOPY		STRMM_OLTUCOPY
+#define	TRSM_OUNCOPY		STRSM_OUNUCOPY
+#define	TRSM_OUTCOPY		STRSM_OUTUCOPY
+#define	TRSM_OLNCOPY		STRSM_OLNUCOPY
+#define	TRSM_OLTCOPY		STRSM_OLTUCOPY
+
+#define	TRMM_IUNCOPY		STRMM_IUNUCOPY
+#define	TRMM_IUTCOPY		STRMM_IUTUCOPY
+#define	TRMM_ILNCOPY		STRMM_ILNUCOPY
+#define	TRMM_ILTCOPY		STRMM_ILTUCOPY
+#define	TRSM_IUNCOPY		STRSM_IUNUCOPY
+#define	TRSM_IUTCOPY		STRSM_IUTUCOPY
+#define	TRSM_ILNCOPY		STRSM_ILNUCOPY
+#define	TRSM_ILTCOPY		STRSM_ILTUCOPY
+
+#else
+
+#define	TRMM_OUNCOPY		STRMM_OUNNCOPY
+#define	TRMM_OUTCOPY		STRMM_OUTNCOPY
+#define	TRMM_OLNCOPY		STRMM_OLNNCOPY
+#define	TRMM_OLTCOPY		STRMM_OLTNCOPY
+#define	TRSM_OUNCOPY		STRSM_OUNNCOPY
+#define	TRSM_OUTCOPY		STRSM_OUTNCOPY
+#define	TRSM_OLNCOPY		STRSM_OLNNCOPY
+#define	TRSM_OLTCOPY		STRSM_OLTNCOPY
+
+#define	TRMM_IUNCOPY		STRMM_IUNNCOPY
+#define	TRMM_IUTCOPY		STRMM_IUTNCOPY
+#define	TRMM_ILNCOPY		STRMM_ILNNCOPY
+#define	TRMM_ILTCOPY		STRMM_ILTNCOPY
+#define	TRSM_IUNCOPY		STRSM_IUNNCOPY
+#define	TRSM_IUTCOPY		STRSM_IUTNCOPY
+#define	TRSM_ILNCOPY		STRSM_ILNNCOPY
+#define	TRSM_ILTCOPY		STRSM_ILTNCOPY
+
+#define	TRMM_KERNEL_LN		STRMM_KERNEL_LN
+#define	TRMM_KERNEL_LT		STRMM_KERNEL_LT
+#define	TRMM_KERNEL_LR		STRMM_KERNEL_LN
+#define	TRMM_KERNEL_LC		STRMM_KERNEL_LT
+#define	TRMM_KERNEL_RN		STRMM_KERNEL_RN
+#define	TRMM_KERNEL_RT		STRMM_KERNEL_RT
+#define	TRMM_KERNEL_RR		STRMM_KERNEL_RN
+#define	TRMM_KERNEL_RC		STRMM_KERNEL_RT
+
+#define	TRSM_KERNEL_LN		STRSM_KERNEL_LN
+#define	TRSM_KERNEL_LT		STRSM_KERNEL_LT
+#define	TRSM_KERNEL_LR		STRSM_KERNEL_LN
+#define	TRSM_KERNEL_LC		STRSM_KERNEL_LT
+#define	TRSM_KERNEL_RN		STRSM_KERNEL_RN
+#define	TRSM_KERNEL_RT		STRSM_KERNEL_RT
+#define	TRSM_KERNEL_RR		STRSM_KERNEL_RN
+#define	TRSM_KERNEL_RC		STRSM_KERNEL_RT
+
+#define SYMM_IUTCOPY		SSYMM_IUTCOPY
+#define SYMM_ILTCOPY		SSYMM_ILTCOPY
+#define SYMM_OUTCOPY		SSYMM_OUTCOPY
+#define SYMM_OLTCOPY		SSYMM_OLTCOPY
+#define	TRMM_LNUU		STRMM_LNUU
+#define	TRMM_LNUN		STRMM_LNUN
+#define	TRMM_LNLU		STRMM_LNLU
+#define	TRMM_LNLN		STRMM_LNLN
+#define	TRMM_LTUU		STRMM_LTUU
+#define	TRMM_LTUN		STRMM_LTUN
+#define	TRMM_LTLU		STRMM_LTLU
+#define	TRMM_LTLN		STRMM_LTLN
+#define	TRMM_LRUU		STRMM_LNUU
+#define	TRMM_LRUN		STRMM_LNUN
+#define	TRMM_LRLU		STRMM_LNLU
+#define	TRMM_LRLN		STRMM_LNLN
+#define	TRMM_LCUU		STRMM_LTUU
+#define	TRMM_LCUN		STRMM_LTUN
+#define	TRMM_LCLU		STRMM_LTLU
+#define	TRMM_LCLN		STRMM_LTLN
+#define	TRMM_RNUU		STRMM_RNUU
+#define	TRMM_RNUN		STRMM_RNUN
+#define	TRMM_RNLU		STRMM_RNLU
+#define	TRMM_RNLN		STRMM_RNLN
+#define	TRMM_RTUU		STRMM_RTUU
+#define	TRMM_RTUN		STRMM_RTUN
+#define	TRMM_RTLU		STRMM_RTLU
+#define	TRMM_RTLN		STRMM_RTLN
+#define	TRMM_RRUU		STRMM_RNUU
+#define	TRMM_RRUN		STRMM_RNUN
+#define	TRMM_RRLU		STRMM_RNLU
+#define	TRMM_RRLN		STRMM_RNLN
+#define	TRMM_RCUU		STRMM_RTUU
+#define	TRMM_RCUN		STRMM_RTUN
+#define	TRMM_RCLU		STRMM_RTLU
+#define	TRMM_RCLN		STRMM_RTLN
+
+#define	TRSM_LNUU		STRSM_LNUU
+#define	TRSM_LNUN		STRSM_LNUN
+#define	TRSM_LNLU		STRSM_LNLU
+#define	TRSM_LNLN		STRSM_LNLN
+#define	TRSM_LTUU		STRSM_LTUU
+#define	TRSM_LTUN		STRSM_LTUN
+#define	TRSM_LTLU		STRSM_LTLU
+#define	TRSM_LTLN		STRSM_LTLN
+#define	TRSM_LRUU		STRSM_LNUU
+#define	TRSM_LRUN		STRSM_LNUN
+#define	TRSM_LRLU		STRSM_LNLU
+#define	TRSM_LRLN		STRSM_LNLN
+#define	TRSM_LCUU		STRSM_LTUU
+#define	TRSM_LCUN		STRSM_LTUN
+#define	TRSM_LCLU		STRSM_LTLU
+#define	TRSM_LCLN		STRSM_LTLN
+#define	TRSM_RNUU		STRSM_RNUU
+#define	TRSM_RNUN		STRSM_RNUN
+#define	TRSM_RNLU		STRSM_RNLU
+#define	TRSM_RNLN		STRSM_RNLN
+#define	TRSM_RTUU		STRSM_RTUU
+#define	TRSM_RTUN		STRSM_RTUN
+#define	TRSM_RTLU		STRSM_RTLU
+#define	TRSM_RTLN		STRSM_RTLN
+#define	TRSM_RRUU		STRSM_RNUU
+#define	TRSM_RRUN		STRSM_RNUN
+#define	TRSM_RRLU		STRSM_RNLU
+#define	TRSM_RRLN		STRSM_RNLN
+#define	TRSM_RCUU		STRSM_RTUU
+#define	TRSM_RCUN		STRSM_RTUN
+#define	TRSM_RCLU		STRSM_RTLU
+#define	TRSM_RCLN		STRSM_RTLN
+#define	SYRK_UN			SSYRK_UN
+#define	SYRK_UT			SSYRK_UT
+#define	SYRK_LN			SSYRK_LN
+#define	SYRK_LT			SSYRK_LT
+#define	SYRK_UR			SSYRK_UN
+#define	SYRK_UC			SSYRK_UT
+#define	SYRK_LR			SSYRK_LN
+#define	SYRK_LC			SSYRK_LT
+
+#define	SYRK_KERNEL_U		SSYRK_KERNEL_U
+#define	SYRK_KERNEL_L		SSYRK_KERNEL_L
+
+#define	HERK_UN			SSYRK_UN
+#define	HERK_LN			SSYRK_LN
+#define	HERK_UC			SSYRK_UT
+#define	HERK_LC			SSYRK_LT
+
+#define	HER2K_UN		SSYR2K_UN
+#define	HER2K_LN		SSYR2K_LN
+#define	HER2K_UC		SSYR2K_UT
+#define	HER2K_LC		SSYR2K_LT
+
+#define	SYR2K_UN		SSYR2K_UN
+#define	SYR2K_UT		SSYR2K_UT
+#define	SYR2K_LN		SSYR2K_LN
+#define	SYR2K_LT		SSYR2K_LT
+#define	SYR2K_UR		SSYR2K_UN
+#define	SYR2K_UC		SSYR2K_UT
+#define	SYR2K_LR		SSYR2K_LN
+#define	SYR2K_LC		SSYR2K_LT
+
+#define	SYR2K_KERNEL_U		SSYR2K_KERNEL_U
+#define	SYR2K_KERNEL_L		SSYR2K_KERNEL_L
+#define	SYRK_THREAD_UN		SSYRK_THREAD_UN
+#define	SYRK_THREAD_UT		SSYRK_THREAD_UT
+#define	SYRK_THREAD_LN		SSYRK_THREAD_LN
+#define	SYRK_THREAD_LT		SSYRK_THREAD_LT
+#define	SYRK_THREAD_UR		SSYRK_THREAD_UR
+#define	SYRK_THREAD_UC		SSYRK_THREAD_UC
+#define	SYRK_THREAD_LR		SSYRK_THREAD_LN
+#define	SYRK_THREAD_LC		SSYRK_THREAD_LT
+
+#define	HERK_THREAD_UN		SSYRK_THREAD_UN
+#define	HERK_THREAD_UT		SSYRK_THREAD_UT
+#define	HERK_THREAD_LN		SSYRK_THREAD_LN
+#define	HERK_THREAD_LT		SSYRK_THREAD_LT
+#define	HERK_THREAD_UR		SSYRK_THREAD_UR
+#define	HERK_THREAD_UC		SSYRK_THREAD_UC
+#define	HERK_THREAD_LR		SSYRK_THREAD_LN
+#define	HERK_THREAD_LC		SSYRK_THREAD_LT
+
+#define OMATCOPY_K_CN		SOMATCOPY_K_CN
+#define OMATCOPY_K_RN		SOMATCOPY_K_RN
+#define OMATCOPY_K_CT		SOMATCOPY_K_CT
+#define OMATCOPY_K_RT		SOMATCOPY_K_RT
+#define IMATCOPY_K_CN		SIMATCOPY_K_CN
+#define IMATCOPY_K_RN		SIMATCOPY_K_RN
+#define IMATCOPY_K_CT		SIMATCOPY_K_CT
+#define IMATCOPY_K_RT		SIMATCOPY_K_RT
+
+#define GEADD_K 		SGEADD_K
+
+#endif
+
 #else
 
 #define	AMAX_K			SAMAX_K
@@ -673,14 +956,14 @@
 #define	GEMV_S			SGEMV_S
 #define	GEMV_D			SGEMV_D
 
+
+#define	SYMV_U			SSYMV_U
+#define	SYMV_L			SSYMV_L
 #define	GERU_K			SGERU_K
 #define	GERC_K			SGERC_K
 #define	GERV_K			SGERV_K
 #define	GERD_K			SGERD_K
 
-#define	SYMV_U			SSYMV_U
-#define	SYMV_L			SSYMV_L
-
 #define	SYMV_THREAD_U		SSYMV_THREAD_U
 #define	SYMV_THREAD_L		SSYMV_THREAD_L
 
@@ -2202,6 +2485,9 @@
 #if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)
 extern BLASLONG gemm_offset_a;
 extern BLASLONG gemm_offset_b;
+extern BLASLONG shgemm_p;
+extern BLASLONG shgemm_q;
+extern BLASLONG shgemm_r;
 extern BLASLONG sgemm_p;
 extern BLASLONG sgemm_q;
 extern BLASLONG sgemm_r;
diff --git a/common_mips.h b/common_mips.h
index 2cc923043..7dc3ba246 100644
--- a/common_mips.h
+++ b/common_mips.h
@@ -43,6 +43,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifndef ASSEMBLER
 
+#if !defined(MIPS24K)
 static inline unsigned int rpcc(void){
   unsigned long ret;
 
@@ -53,6 +54,7 @@ static inline unsigned int rpcc(void){
   return ret;
 }
 #define RPCC_DEFINED
+#endif
 
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
@@ -92,7 +94,7 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
-#define BUFFER_SIZE     (16 << 20)
+#define BUFFER_SIZE     (16 << 21)
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
diff --git a/common_mips64.h b/common_mips64.h
index af638d60c..a06edfe08 100644
--- a/common_mips64.h
+++ b/common_mips64.h
@@ -227,7 +227,7 @@ REALNAME: ;\
 
 #define SEEK_ADDRESS
 
-#define BUFFER_SIZE     ( 32 << 20)
+#define BUFFER_SIZE     ( 32 << 21)
 
 #if defined(LOONGSON3A)
 #define PAGESIZE	(16UL << 10)
diff --git a/common_param.h b/common_param.h
index 574d5e176..c92609a76 100644
--- a/common_param.h
+++ b/common_param.h
@@ -47,6 +47,100 @@ typedef struct {
   int dtb_entries;
   int offsetA, offsetB, align;
 
+#ifdef BUILD_HALF
+  int shgemm_p, shgemm_q, shgemm_r;
+  int shgemm_unroll_m, shgemm_unroll_n, shgemm_unroll_mn;
+
+  float  (*shamax_k) (BLASLONG, float *, BLASLONG);
+  float  (*shamin_k) (BLASLONG, float *, BLASLONG);
+  float  (*shmax_k)  (BLASLONG, float *, BLASLONG);
+  float  (*shmin_k)  (BLASLONG, float *, BLASLONG);
+BLASLONG (*ishamax_k)(BLASLONG, float *, BLASLONG);
+BLASLONG (*ishamin_k)(BLASLONG, float *, BLASLONG);
+BLASLONG (*ishmax_k) (BLASLONG, float *, BLASLONG);
+BLASLONG (*ishmin_k) (BLASLONG, float *, BLASLONG);
+
+  float  (*shnrm2_k) (BLASLONG, float *, BLASLONG);
+  float  (*shasum_k) (BLASLONG, float *, BLASLONG);
+  float  (*shsum_k)  (BLASLONG, float *, BLASLONG);
+  int    (*shcopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  float  (*shdot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  double (*dshdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+
+  int    (*shrot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
+
+  int    (*shaxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  int    (*shscal_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  int    (*shswap_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+
+  int    (*shgemv_n) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+  int    (*shgemv_t) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+  int    (*shger_k)  (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+
+  int    (*shsymv_L) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
+  int    (*shsymv_U) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
+
+  int    (*shgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, float, bfloat16 *, bfloat16 *, float *, BLASLONG);
+  int    (*shgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, bfloat16 *, BLASLONG, bfloat16 *, BLASLONG, float *, BLASLONG);
+
+  int    (*shgemm_incopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+  int    (*shgemm_itcopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+  int    (*shgemm_oncopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+  int    (*shgemm_otcopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+
+  int    (*shtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+
+  int    (*shtrsm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+
+  int    (*shtrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+
+  int    (*shtrmm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+
+  int    (*shsymm_iutcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shsymm_iltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shsymm_outcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shsymm_oltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+
+  int	 (*shneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
+  int    (*shlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
+
+#endif
   int sgemm_p, sgemm_q, sgemm_r;
   int sgemm_unroll_m, sgemm_unroll_n, sgemm_unroll_mn;
 
@@ -84,6 +178,7 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int    (*sgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG);
   int    (*sgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float  *, BLASLONG);
 
+
   int    (*sgemm_incopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemm_itcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
@@ -907,6 +1002,15 @@ extern gotoblas_t *gotoblas;
 
 #define HAVE_EX_L2	gotoblas -> exclusive_cache
 
+#ifdef BUILD_HALF
+#define	SHGEMM_P		gotoblas -> shgemm_p
+#define	SHGEMM_Q		gotoblas -> shgemm_q
+#define	SHGEMM_R		gotoblas -> shgemm_r
+#define	SHGEMM_UNROLL_M	gotoblas -> shgemm_unroll_m
+#define	SHGEMM_UNROLL_N	gotoblas -> shgemm_unroll_n
+#define	SHGEMM_UNROLL_MN	gotoblas -> shgemm_unroll_mn
+#endif
+
 #define	SGEMM_P		gotoblas -> sgemm_p
 #define	SGEMM_Q		gotoblas -> sgemm_q
 #define	SGEMM_R		gotoblas -> sgemm_r
@@ -984,6 +1088,19 @@ extern gotoblas_t *gotoblas;
 #define HAVE_EX_L2	0
 #endif
 
+#ifdef BUILD_HALF
+#define	SHGEMM_P		SHGEMM_DEFAULT_P
+#define	SHGEMM_Q		SHGEMM_DEFAULT_Q
+#define	SHGEMM_R		SHGEMM_DEFAULT_R
+#define SHGEMM_UNROLL_M	SHGEMM_DEFAULT_UNROLL_M
+#define SHGEMM_UNROLL_N	SHGEMM_DEFAULT_UNROLL_N
+#ifdef  SHGEMM_DEFAULT_UNROLL_MN
+#define SHGEMM_UNROLL_MN	SHGEMM_DEFAULT_UNROLL_MN
+#else
+#define SHGEMM_UNROLL_MN	MAX((SHGEMM_UNROLL_M), (SHGEMM_UNROLL_N))
+#endif
+#endif
+
 #define	SGEMM_P		SGEMM_DEFAULT_P
 #define	SGEMM_Q		SGEMM_DEFAULT_Q
 #define	SGEMM_R		SGEMM_DEFAULT_R
@@ -1119,6 +1236,18 @@ extern gotoblas_t *gotoblas;
 #define GEMM_DEFAULT_R		DGEMM_DEFAULT_R
 #define GEMM_DEFAULT_UNROLL_M	DGEMM_DEFAULT_UNROLL_M
 #define GEMM_DEFAULT_UNROLL_N	DGEMM_DEFAULT_UNROLL_N
+#elif defined(HALF)
+#define GEMM_P			SHGEMM_P
+#define GEMM_Q			SHGEMM_Q
+#define GEMM_R			SHGEMM_R
+#define GEMM_UNROLL_M		SHGEMM_UNROLL_M
+#define GEMM_UNROLL_N		SHGEMM_UNROLL_N
+#define GEMM_UNROLL_MN		SHGEMM_UNROLL_MN
+#define GEMM_DEFAULT_P		SHGEMM_DEFAULT_P
+#define GEMM_DEFAULT_Q		SHGEMM_DEFAULT_Q
+#define GEMM_DEFAULT_R		SHGEMM_DEFAULT_R
+#define GEMM_DEFAULT_UNROLL_M	SHGEMM_DEFAULT_UNROLL_M
+#define GEMM_DEFAULT_UNROLL_N	SHGEMM_DEFAULT_UNROLL_N
 #else
 #define GEMM_P			SGEMM_P
 #define GEMM_Q			SGEMM_Q
@@ -1204,28 +1333,32 @@ extern gotoblas_t *gotoblas;
 #define GEMM_THREAD gemm_thread_n
 #endif
 
+#ifndef SHGEMM_DEFAULT_R
+#define SHGEMM_DEFAULT_R (((BUFFER_SIZE - ((SHGEMM_DEFAULT_P * SHGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SHGEMM_DEFAULT_Q *  4) - 15) & ~15UL)
+#endif
+
 #ifndef SGEMM_DEFAULT_R
-#define SGEMM_DEFAULT_R (((BUFFER_SIZE - ((SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SGEMM_DEFAULT_Q *  4) - 15) & ~15)
+#define SGEMM_DEFAULT_R (((BUFFER_SIZE - ((SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SGEMM_DEFAULT_Q *  4) - 15) & ~15UL)
 #endif
 
 #ifndef DGEMM_DEFAULT_R
-#define DGEMM_DEFAULT_R (((BUFFER_SIZE - ((DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (DGEMM_DEFAULT_Q *  8) - 15) & ~15)
+#define DGEMM_DEFAULT_R (((BUFFER_SIZE - ((DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (DGEMM_DEFAULT_Q *  8) - 15) & ~15UL)
 #endif
 
 #ifndef QGEMM_DEFAULT_R
-#define QGEMM_DEFAULT_R (((BUFFER_SIZE - ((QGEMM_DEFAULT_P * QGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (QGEMM_DEFAULT_Q * 16) - 15) & ~15)
+#define QGEMM_DEFAULT_R (((BUFFER_SIZE - ((QGEMM_DEFAULT_P * QGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (QGEMM_DEFAULT_Q * 16) - 15) & ~15UL)
 #endif
 
 #ifndef CGEMM_DEFAULT_R
-#define CGEMM_DEFAULT_R (((BUFFER_SIZE - ((CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (CGEMM_DEFAULT_Q *  8) - 15) & ~15)
+#define CGEMM_DEFAULT_R (((BUFFER_SIZE - ((CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (CGEMM_DEFAULT_Q *  8) - 15) & ~15UL)
 #endif
 
 #ifndef ZGEMM_DEFAULT_R
-#define ZGEMM_DEFAULT_R (((BUFFER_SIZE - ((ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (ZGEMM_DEFAULT_Q * 16) - 15) & ~15)
+#define ZGEMM_DEFAULT_R (((BUFFER_SIZE - ((ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (ZGEMM_DEFAULT_Q * 16) - 15) & ~15UL)
 #endif
 
 #ifndef XGEMM_DEFAULT_R
-#define XGEMM_DEFAULT_R (((BUFFER_SIZE - ((XGEMM_DEFAULT_P * XGEMM_DEFAULT_Q * 32 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (XGEMM_DEFAULT_Q * 32) - 15) & ~15)
+#define XGEMM_DEFAULT_R (((BUFFER_SIZE - ((XGEMM_DEFAULT_P * XGEMM_DEFAULT_Q * 32 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (XGEMM_DEFAULT_Q * 32) - 15) & ~15UL)
 #endif
 
 #ifndef SNUMOPT
diff --git a/common_power.h b/common_power.h
index e29d0f382..aa19794b5 100644
--- a/common_power.h
+++ b/common_power.h
@@ -68,7 +68,7 @@
 #endif
 
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #define MB		__asm__ __volatile__ ("eieio":::"memory")
 #define WMB		__asm__ __volatile__ ("eieio":::"memory")
 #define RMB		__asm__ __volatile__ ("eieio":::"memory")
@@ -272,7 +272,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define HAVE_PREFETCH
 #endif
 
-#if defined(POWER3) || defined(POWER6) || defined(PPCG4) || defined(CELL) || defined(POWER8) || defined(POWER9) || defined(PPC970)
+#if defined(POWER3) || defined(POWER6) || defined(PPCG4) || defined(CELL) || defined(POWER8) || defined(POWER9) || defined(POWER10) || defined(PPC970)
 #define DCBT_ARG	0
 #else
 #define DCBT_ARG	8
@@ -294,7 +294,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define L1_PREFETCH	dcbtst
 #endif
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #define L1_DUALFETCH
 #define L1_PREFETCHSIZE (16 + 128 * 100)
 #define L1_PREFETCH	dcbtst
@@ -843,7 +843,7 @@ Lmcount$lazy_ptr:
 #define BUFFER_SIZE     (  2 << 20)
 #elif defined(PPC440FP2)
 #define BUFFER_SIZE     ( 16 << 20)
-#elif defined(POWER8) || defined(POWER9)
+#elif defined(POWER8) || defined(POWER9) || defined(POWER10)
 #define BUFFER_SIZE     ( 64 << 20)
 #else
 #define BUFFER_SIZE     ( 16 << 20)
diff --git a/common_sh.h b/common_sh.h
new file mode 100644
index 000000000..7a0045762
--- /dev/null
+++ b/common_sh.h
@@ -0,0 +1,65 @@
+#ifndef COMMON_SH_H
+#define COMMON_SH_H
+
+#ifndef DYNAMIC_ARCH
+
+#define	SHGEMM_ONCOPY		shgemm_oncopy
+#define	SHGEMM_OTCOPY		shgemm_otcopy
+
+#if SHGEMM_DEFAULT_UNROLL_M == SHGEMM_DEFAULT_UNROLL_N
+#define	SHGEMM_INCOPY		shgemm_oncopy
+#define	SHGEMM_ITCOPY		shgemm_otcopy
+#else
+#define	SHGEMM_INCOPY		shgemm_incopy
+#define	SHGEMM_ITCOPY		shgemm_itcopy
+#endif
+#define	SHGEMM_BETA		shgemm_beta
+#define SHGEMM_KERNEL            shgemm_kernel
+
+#else
+
+#define	SHGEMM_ONCOPY		gotoblas -> shgemm_oncopy
+#define	SHGEMM_OTCOPY		gotoblas -> shgemm_otcopy
+#define	SHGEMM_INCOPY		gotoblas -> shgemm_incopy
+#define	SHGEMM_ITCOPY		gotoblas -> shgemm_itcopy
+#define	SHGEMM_BETA		gotoblas -> shgemm_beta
+#define	SHGEMM_KERNEL		gotoblas -> shgemm_kernel
+
+#endif
+
+#define	SHGEMM_NN		shgemm_nn
+#define	SHGEMM_CN		shgemm_tn
+#define	SHGEMM_TN		shgemm_tn
+#define	SHGEMM_NC		shgemm_nt
+#define	SHGEMM_NT		shgemm_nt
+#define	SHGEMM_CC		shgemm_tt
+#define	SHGEMM_CT		shgemm_tt
+#define	SHGEMM_TC		shgemm_tt
+#define	SHGEMM_TT		shgemm_tt
+#define	SHGEMM_NR		shgemm_nn
+#define	SHGEMM_TR		shgemm_tn
+#define	SHGEMM_CR		shgemm_tn
+#define	SHGEMM_RN		shgemm_nn
+#define	SHGEMM_RT		shgemm_nt
+#define	SHGEMM_RC		shgemm_nt
+#define	SHGEMM_RR		shgemm_nn
+
+#define	SHGEMM_THREAD_NN		shgemm_thread_nn
+#define	SHGEMM_THREAD_CN		shgemm_thread_tn
+#define	SHGEMM_THREAD_TN		shgemm_thread_tn
+#define	SHGEMM_THREAD_NC		shgemm_thread_nt
+#define	SHGEMM_THREAD_NT		shgemm_thread_nt
+#define	SHGEMM_THREAD_CC		shgemm_thread_tt
+#define	SHGEMM_THREAD_CT		shgemm_thread_tt
+#define	SHGEMM_THREAD_TC		shgemm_thread_tt
+#define	SHGEMM_THREAD_TT		shgemm_thread_tt
+#define	SHGEMM_THREAD_NR		shgemm_thread_nn
+#define	SHGEMM_THREAD_TR		shgemm_thread_tn
+#define	SHGEMM_THREAD_CR		shgemm_thread_tn
+#define	SHGEMM_THREAD_RN		shgemm_thread_nn
+#define	SHGEMM_THREAD_RT		shgemm_thread_nt
+#define	SHGEMM_THREAD_RC		shgemm_thread_nt
+#define	SHGEMM_THREAD_RR		shgemm_thread_nn
+
+#endif
+
diff --git a/common_x86_64.h b/common_x86_64.h
index 0247674cd..15d0c30aa 100644
--- a/common_x86_64.h
+++ b/common_x86_64.h
@@ -80,7 +80,7 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #endif
 
   do {
-    while (*address) {YIELDING;};
+    while (*address) {YIELDING;}
 
 #ifndef C_MSVC
     __asm__ __volatile__(
@@ -199,9 +199,9 @@ static __inline BLASLONG blas_quickdivide(BLASLONG x, BLASLONG y){
 #else
 extern unsigned int blas_quick_divide_table[];
 
-static __inline int blas_quickdivide(unsigned int x, unsigned int y){
+static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
 
-  unsigned int result;
+  volatile unsigned int result;
 
   if (y <= 1) return x;
 
@@ -215,7 +215,6 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
   y = blas_quick_divide_table[y];
 
   __asm__ __volatile__  ("mull %0" :"=d" (result), "+a"(x) : "0" (y));
-
   return result;
 }
 #endif
diff --git a/cpp_thread_test/cpp_thread_safety_common.h b/cpp_thread_test/cpp_thread_safety_common.h
index 60ab5bb2f..8005369a8 100644
--- a/cpp_thread_test/cpp_thread_safety_common.h
+++ b/cpp_thread_test/cpp_thread_safety_common.h
@@ -5,6 +5,14 @@ inline void pauser(){
     std::getline(std::cin, dummy);
 }
 
+void FailIfThreadsAreZero(uint32_t numConcurrentThreads) {
+	if(numConcurrentThreads == 0) {
+		std::cout<<"ERROR: Invalid parameter 0 for number of concurrent calls into OpenBLAS!"<<std::endl;
+		std::cout<<"CBLAS DGEMV thread safety test FAILED!"<<std::endl;
+		exit(-1);
+	}
+}
+
 void FillMatrices(std::vector<std::vector<double>>& matBlock, std::mt19937_64& PRNG, std::uniform_real_distribution<double>& rngdist, const blasint randomMatSize, const uint32_t numConcurrentThreads, const uint32_t numMat){
 	for(uint32_t i=0; i<numMat; i++){
 		for(uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize*randomMatSize); j++){
diff --git a/cpp_thread_test/dgemm_thread_safety.cpp b/cpp_thread_test/dgemm_thread_safety.cpp
index 1c5287524..104c64f2a 100644
--- a/cpp_thread_test/dgemm_thread_safety.cpp
+++ b/cpp_thread_test/dgemm_thread_safety.cpp
@@ -46,6 +46,8 @@ int main(int argc, char* argv[]){
 	std::cout<<"Number of concurrent calls into OpenBLAS : "<<numConcurrentThreads<<'\n';
 	std::cout<<"Number of testing rounds : "<<numTestRounds<<'\n';
 	std::cout<<"This test will need "<<(static_cast<uint64_t>(randomMatSize*randomMatSize)*numConcurrentThreads*3*8)/static_cast<double>(1024*1024)<<" MiB of RAM\n"<<std::endl;
+
+	FailIfThreadsAreZero(numConcurrentThreads);
 	
 	std::cout<<"Initializing random number generator..."<<std::flush;
 	std::mt19937_64 PRNG = InitPRNG();
diff --git a/cpp_thread_test/dgemv_thread_safety.cpp b/cpp_thread_test/dgemv_thread_safety.cpp
index 5411fec29..20ea38138 100644
--- a/cpp_thread_test/dgemv_thread_safety.cpp
+++ b/cpp_thread_test/dgemv_thread_safety.cpp
@@ -18,7 +18,7 @@ int main(int argc, char* argv[]){
 	uint32_t maxHwThreads = omp_get_max_threads();
 	
 	if (maxHwThreads < 52)
-		numConcurrentThreads = maxHwThreads -4;
+		numConcurrentThreads = maxHwThreads;
 	
 	if (argc > 4){
 		std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl;
@@ -47,6 +47,8 @@ int main(int argc, char* argv[]){
 	std::cout<<"Number of concurrent calls into OpenBLAS : "<<numConcurrentThreads<<'\n';
 	std::cout<<"Number of testing rounds : "<<numTestRounds<<'\n';
 	std::cout<<"This test will need "<<((static_cast<uint64_t>(randomMatSize*randomMatSize)*numConcurrentThreads*8)+(static_cast<uint64_t>(randomMatSize)*numConcurrentThreads*8*2))/static_cast<double>(1024*1024)<<" MiB of RAM\n"<<std::endl;
+
+	FailIfThreadsAreZero(numConcurrentThreads);
 	
 	std::cout<<"Initializing random number generator..."<<std::flush;
 	std::mt19937_64 PRNG = InitPRNG();
diff --git a/cpuid_mips.c b/cpuid_mips.c
index 6f2932c94..3a2e12393 100644
--- a/cpuid_mips.c
+++ b/cpuid_mips.c
@@ -73,11 +73,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_UNKNOWN     0
 #define CPU_P5600       1
 #define CPU_1004K	2
+#define CPU_24K		3
 
 static char *cpuname[] = {
   "UNKNOWN",
   "P5600",
-  "1004K"
+  "MIPS1004K",
+  "MIPS24K"
 };
 
 int detect(void){
@@ -105,6 +107,8 @@ int detect(void){
     return CPU_P5600;
   } else if (strstr(p, "1004K")) {
     return CPU_1004K;
+  } else if (strstr(p, " 24K")) {
+    return CPU_24K;
   } else  
     return CPU_UNKNOWN;
   }
@@ -121,7 +125,7 @@ void get_architecture(void){
 }
 
 void get_subarchitecture(void){
-  if(detect()==CPU_P5600|| detect()==CPU_1004K){
+  if(detect()==CPU_P5600|| detect()==CPU_1004K|| detect()==CPU_24K){
     printf("P5600");
   }else{
     printf("UNKNOWN");
@@ -146,7 +150,15 @@ void get_cpuconfig(void){
     printf("#define MIPS1004K\n");
     printf("#define L1_DATA_SIZE 32768\n");
     printf("#define L1_DATA_LINESIZE 32\n");
-    printf("#define L2_SIZE 26144\n");
+    printf("#define L2_SIZE 262144\n");
+    printf("#define DTB_DEFAULT_ENTRIES 8\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 4\n");
+  } else if (detect()==CPU_24K) {
+    printf("#define MIPS24K\n");
+    printf("#define L1_DATA_SIZE 32768\n");
+    printf("#define L1_DATA_LINESIZE 32\n");
+    printf("#define L2_SIZE 32768\n");
     printf("#define DTB_DEFAULT_ENTRIES 8\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 4\n");
@@ -159,7 +171,9 @@ void get_libname(void){
   if(detect()==CPU_P5600) {
     printf("p5600\n");
   } else if (detect()==CPU_1004K) {
-    printf("1004K\n");
+    printf("mips1004K\n");
+  } else if (detect()==CPU_24K) {
+    printf("mips24K\n");
   }else{
     printf("mips\n");
   }
diff --git a/cpuid_power.c b/cpuid_power.c
index d5ba6fb2c..b36aa4945 100644
--- a/cpuid_power.c
+++ b/cpuid_power.c
@@ -57,6 +57,7 @@
 #define CPUTYPE_PPCG4	   7
 #define CPUTYPE_POWER8     8
 #define CPUTYPE_POWER9     9
+#define CPUTYPE_POWER10    10
 
 char *cpuname[] = {
   "UNKNOWN",
@@ -68,7 +69,8 @@ char *cpuname[] = {
   "CELL",
   "PPCG4",
   "POWER8",
-  "POWER9"
+  "POWER9",
+  "POWER10"
 };
 
 char *lowercpuname[] = {
@@ -81,7 +83,8 @@ char *lowercpuname[] = {
   "cell",
   "ppcg4",
   "power8",
-  "power9"	
+  "power9",
+  "power10"
 };
 
 char *corename[] = {
@@ -94,7 +97,8 @@ char *corename[] = {
   "CELL",
   "PPCG4",
   "POWER8",
-  "POWER9"   	
+  "POWER9",
+  "POWER10"
 };
 
 int detect(void){
@@ -125,6 +129,7 @@ int detect(void){
   if (!strncasecmp(p, "POWER7", 6)) return CPUTYPE_POWER6;
   if (!strncasecmp(p, "POWER8", 6)) return CPUTYPE_POWER8;
   if (!strncasecmp(p, "POWER9", 6)) return CPUTYPE_POWER9;
+  if (!strncasecmp(p, "POWER10", 7)) return CPUTYPE_POWER10;
   if (!strncasecmp(p, "Cell",   4)) return CPUTYPE_CELL;
   if (!strncasecmp(p, "7447",   4)) return CPUTYPE_PPCG4;
 
@@ -157,6 +162,7 @@ int detect(void){
   if (!strncasecmp(p, "POWER7", 6)) return CPUTYPE_POWER6;
   if (!strncasecmp(p, "POWER8", 6)) return CPUTYPE_POWER8;
   if (!strncasecmp(p, "POWER9", 6)) return CPUTYPE_POWER9;
+  if (!strncasecmp(p, "POWER10", 7)) return CPUTYPE_POWER10;
   if (!strncasecmp(p, "Cell",   4)) return CPUTYPE_CELL;
   if (!strncasecmp(p, "7447",   4)) return CPUTYPE_PPCG4;
   return CPUTYPE_POWER5;
@@ -179,6 +185,9 @@ int detect(void){
 int id;
 __asm __volatile("mfpvr %0" : "=r"(id));
 switch ( id >> 16 ) {
+  case 0x80: // POWER10
+    return CPUTYPE_POWER10;
+    break;
   case 0x4e: // POWER9
     return CPUTYPE_POWER9;
     break;
diff --git a/cpuid_x86.c b/cpuid_x86.c
index e29adecae..356800b78 100644
--- a/cpuid_x86.c
+++ b/cpuid_x86.c
@@ -1406,6 +1406,17 @@ int get_cpuname(void){
 	    return CPUTYPE_SANDYBRIDGE;
           else
 	    return CPUTYPE_NEHALEM;
+    }
+      case 10: //family 6 exmodel 10
+        switch (model) {
+    case 5: // Comet Lake H and S
+    case 6: // Comet Lake U
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+        return CPUTYPE_SANDYBRIDGE;
+          else
+        return CPUTYPE_NEHALEM;
 	}
 	break;    
       }
@@ -1955,6 +1966,19 @@ int get_coretype(void){
 	    return CORE_NEHALEM;
         }
         break;
+      case 10:
+        switch (model) {
+	  case 5: // Comet Lake H and S
+    	  case 6: // Comet Lake U
+            if(support_avx())
+  #ifndef NO_AVX2
+              return CORE_HASWELL;
+  #else
+              return CORE_SANDYBRIDGE;
+  #endif
+            else
+              return CORE_NEHALEM;
+        }
       case 5:
         switch (model) {
 	case 6:
diff --git a/ctest/CMakeLists.txt b/ctest/CMakeLists.txt
index 14c9d1944..8d301c239 100644
--- a/ctest/CMakeLists.txt
+++ b/ctest/CMakeLists.txt
@@ -12,6 +12,9 @@ FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh
 foreach(float_type ${FLOAT_TYPES})
   string(SUBSTRING ${float_type} 0 1 float_char_upper)
   string(TOLOWER ${float_char_upper} float_char)
+  if (${float_char} STREQUAL "h")
+   continue()
+  endif()
   #level1
   add_executable(x${float_char}cblat1
     c_${float_char}blat1.f
diff --git a/driver/level3/Makefile b/driver/level3/Makefile
index e320092e3..09a62d9bf 100644
--- a/driver/level3/Makefile
+++ b/driver/level3/Makefile
@@ -19,6 +19,10 @@ ifeq ($(ARCH), MIPS)
 USE_GEMM3M = 1
 endif
 
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS       += shgemm_nn.$(SUFFIX) shgemm_nt.$(SUFFIX) shgemm_tn.$(SUFFIX) shgemm_tt.$(SUFFIX)
+endif
+
 SBLASOBJS	+= \
 	sgemm_nn.$(SUFFIX) sgemm_nt.$(SUFFIX) sgemm_tn.$(SUFFIX) sgemm_tt.$(SUFFIX) \
 	strmm_LNUU.$(SUFFIX) strmm_LNUN.$(SUFFIX) strmm_LNLU.$(SUFFIX) strmm_LNLN.$(SUFFIX) \
@@ -203,7 +207,9 @@ COMMONOBJS  += gemm_thread_m.$(SUFFIX) gemm_thread_n.$(SUFFIX) gemm_thread_mn.$(
 COMMONOBJS  += syrk_thread.$(SUFFIX)
 
 ifndef USE_SIMPLE_THREADED_LEVEL3
-
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS    += shgemm_thread_nn.$(SUFFIX) shgemm_thread_nt.$(SUFFIX) shgemm_thread_tn.$(SUFFIX) shgemm_thread_tt.$(SUFFIX)
+endif
 SBLASOBJS    += sgemm_thread_nn.$(SUFFIX) sgemm_thread_nt.$(SUFFIX) sgemm_thread_tn.$(SUFFIX) sgemm_thread_tt.$(SUFFIX)
 DBLASOBJS    += dgemm_thread_nn.$(SUFFIX) dgemm_thread_nt.$(SUFFIX) dgemm_thread_tn.$(SUFFIX) dgemm_thread_tt.$(SUFFIX)
 QBLASOBJS    += qgemm_thread_nn.$(SUFFIX) qgemm_thread_nt.$(SUFFIX) qgemm_thread_tn.$(SUFFIX) qgemm_thread_tt.$(SUFFIX)
@@ -283,6 +289,18 @@ endif
 
 all ::
 
+shgemm_nn.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_nt.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_tn.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_tt.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
+
 sgemm_nn.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
 
@@ -478,6 +496,17 @@ gemm_thread_variable.$(SUFFIX) : gemm_thread_variable.c ../../common.h
 beta_thread.$(SUFFIX) : beta_thread.c ../../common.h
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
+shgemm_thread_nn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_thread_nt.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_thread_tn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_thread_tt.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
 
 sgemm_thread_nn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
@@ -2652,6 +2681,18 @@ xtrsm_RCLU.$(SUFFIX) : trsm_R.c
 xtrsm_RCLN.$(SUFFIX) : trsm_R.c
 	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DTRANSA -UUPPER -UUNIT -DCONJ $< -o $(@F)
 
+shgemm_nn.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_nt.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_tn.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_tt.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
+
 sgemm_nn.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
 
@@ -2848,6 +2889,18 @@ beta_thread.$(PSUFFIX) : beta_thread.c ../../common.h
 	$(CC) -c $(PFLAGS) $< -o $(@F)
 
 
+shgemm_thread_nn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_thread_nt.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_thread_tn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_thread_tt.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
+
 sgemm_thread_nn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
 
diff --git a/driver/level3/level3.c b/driver/level3/level3.c
index 9aa67286f..c6bbb9ca9 100644
--- a/driver/level3/level3.c
+++ b/driver/level3/level3.c
@@ -62,18 +62,18 @@
 #ifndef ICOPY_OPERATION
 #if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
     defined(RN) || defined(RT) || defined(RC) || defined(RR)
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
 #ifndef OCOPY_OPERATION
 #if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
     defined(NR) || defined(TR) || defined(CR) || defined(RR)
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
@@ -173,7 +173,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 		  XFLOAT *sa, XFLOAT *sb, BLASLONG dummy){
   BLASLONG k, lda, ldb, ldc;
   FLOAT *alpha, *beta;
-  FLOAT *a, *b, *c;
+  IFLOAT *a, *b;
+  FLOAT *c;
   BLASLONG m_from, m_to, n_from, n_to;
 
   BLASLONG ls, is, js;
@@ -198,8 +199,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   k = K;
 
-  a = (FLOAT *)A;
-  b = (FLOAT *)B;
+  a = (IFLOAT *)A;
+  b = (IFLOAT *)B;
   c = (FLOAT *)C;
 
   lda = LDA;
diff --git a/driver/level3/level3_thread.c b/driver/level3/level3_thread.c
index ca0085e71..5a8d497d2 100644
--- a/driver/level3/level3_thread.c
+++ b/driver/level3/level3_thread.c
@@ -117,18 +117,18 @@ typedef struct {
 #ifndef ICOPY_OPERATION
 #if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
   defined(RN) || defined(RT) || defined(RC) || defined(RR)
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
 #ifndef OCOPY_OPERATION
 #if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
   defined(NR) || defined(TR) || defined(CR) || defined(RR)
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
@@ -219,15 +219,16 @@ typedef struct {
 #define STOP_RPCC(COUNTER)
 #endif
 
-static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
+static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IFLOAT *sb, BLASLONG mypos){
 
-  FLOAT *buffer[DIVIDE_RATE];
+  IFLOAT *buffer[DIVIDE_RATE];
 
   BLASLONG k, lda, ldb, ldc;
   BLASLONG m_from, m_to, n_from, n_to;
 
   FLOAT *alpha, *beta;
-  FLOAT *a, *b, *c;
+  IFLOAT *a, *b;
+  FLOAT *c;
   job_t *job = (job_t *)args -> common;
 
   BLASLONG nthreads_m;
@@ -255,8 +256,8 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   k = K;
 
-  a = (FLOAT *)A;
-  b = (FLOAT *)B;
+  a = (IFLOAT *)A;
+  b = (IFLOAT *)B;
   c = (FLOAT *)C;
 
   lda = LDA;
@@ -425,7 +426,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
           /* Apply kernel with local region of A and part of other region of B */
 	  START_RPCC();
 	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - js,  div_n), min_l, alpha,
-			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
+			   sa, (IFLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
 			   c, ldc, m_from, js);
           STOP_RPCC(kernel);
 
@@ -469,7 +470,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
           /* Apply kernel with local region of A and part of region of B */
 	  START_RPCC();
 	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - js, div_n), min_l, alpha,
-			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
+			   sa, (IFLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
 			   c, ldc, is, js);
           STOP_RPCC(kernel);
           
@@ -532,7 +533,7 @@ static int round_up(int remainder, int width, int multiple)
 
 
 static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
-		       *range_n, FLOAT *sa, FLOAT *sb,
+		       *range_n, IFLOAT *sa, IFLOAT *sb,
                        BLASLONG nthreads_m, BLASLONG nthreads_n) {
 
 #ifndef USE_OPENMP
@@ -728,7 +729,7 @@ EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
   return 0;
 }
 
-int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
+int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IFLOAT *sb, BLASLONG mypos){
 
   BLASLONG m = args -> m;
   BLASLONG n = args -> n;
diff --git a/driver/others/blas_server.c b/driver/others/blas_server.c
index f13b83dd4..04b614a6e 100644
--- a/driver/others/blas_server.c
+++ b/driver/others/blas_server.c
@@ -272,7 +272,7 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
       }
 }
 
-#if defined(OS_LINUX) && !defined(NO_AFFINITY)
+#if defined(OS_LINUX) && !defined(NO_AFFINITY) 
 int gotoblas_set_affinity(int);
 int gotoblas_set_affinity2(int);
 int get_node(void);
@@ -281,6 +281,8 @@ int get_node(void);
 static int increased_threads = 0;
 
 #ifdef OS_LINUX
+extern int openblas_get_num_threads(void);  
+
 int openblas_setaffinity(int thread_idx, size_t cpusetsize, cpu_set_t* cpu_set) {
   const int active_threads = openblas_get_num_threads();
 
@@ -602,7 +604,7 @@ int blas_thread_init(void){
       if(ret!=0){
 	struct rlimit rlim;
         const char *msg = strerror(ret);
-        fprintf(STDERR, "OpenBLAS blas_thread_init: pthread_create failed for thread %ld of %ld: %s\n", i+1,blas_num_threads,msg);
+        fprintf(STDERR, "OpenBLAS blas_thread_init: pthread_create failed for thread %ld of %d: %s\n", i+1,blas_num_threads,msg);
 #ifdef RLIMIT_NPROC
         if(0 == getrlimit(RLIMIT_NPROC, &rlim)) {
           fprintf(STDERR, "OpenBLAS blas_thread_init: RLIMIT_NPROC "
diff --git a/driver/others/dynamic.c b/driver/others/dynamic.c
index 2e87e186a..c03b0b21d 100644
--- a/driver/others/dynamic.c
+++ b/driver/others/dynamic.c
@@ -332,7 +332,7 @@ int support_avx512(){
   if((ebx & (1<<7)) == 0){
       ret=0;  //OS does not even support AVX2
   }
-  if((ebx & (1<<31)) != 0){
+  if((ebx & (1u<<31)) != 0){
     xgetbv(0, &eax, &edx);
     if((eax & 0xe0) == 0xe0)
       ret=1;  //OS supports AVX512VL
@@ -618,6 +618,18 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
+      case 10:
+    if (model == 5 || model == 6) {
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+    }
 	return NULL;
       }
       case 0xf:
@@ -632,7 +644,7 @@ static gotoblas_t *get_coretype(void){
         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
         if ( (eax & 0xffff)  >= 0x01) {
             cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
-            if ((edx & (1 << 30)) == 0 || (edx & (1 << 31)) == 0)
+            if ((edx & (1 << 30)) == 0 || (edx & (1u << 31)) == 0)
               return NULL;
           }
         else
@@ -764,18 +776,53 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_NORTHWOOD)    return corename[ 3];
   if (gotoblas == &gotoblas_PRESCOTT)     return corename[ 4];
   if (gotoblas == &gotoblas_BANIAS)       return corename[ 5];
-  if (gotoblas == &gotoblas_ATOM)         return corename[ 6];
+  if (gotoblas == &gotoblas_ATOM)
+#ifdef DYNAMIC_OLDER
+           return corename[ 6];
+#else
+           return corename[10];
+#endif
   if (gotoblas == &gotoblas_CORE2)        return corename[ 7];
-  if (gotoblas == &gotoblas_PENRYN)       return corename[ 8];
-  if (gotoblas == &gotoblas_DUNNINGTON)   return corename[ 9];
+  if (gotoblas == &gotoblas_PENRYN)
+#ifdef DYNAMIC_OLDER
+           return corename[ 8];
+#else
+           return corename[7];
+#endif
+  if (gotoblas == &gotoblas_DUNNINGTON)
+#ifdef DYNAMIC_OLDER
+           return corename[ 9];
+#else
+           return corename[7];
+#endif
   if (gotoblas == &gotoblas_NEHALEM)      return corename[10];
   if (gotoblas == &gotoblas_ATHLON)       return corename[11];
-  if (gotoblas == &gotoblas_OPTERON_SSE3) return corename[12];
-  if (gotoblas == &gotoblas_OPTERON)      return corename[13];
+  if (gotoblas == &gotoblas_OPTERON_SSE3)
+#ifdef DYNAMIC_OLDER
+           return corename[12];
+#else
+           return corename[7];
+#endif
+  if (gotoblas == &gotoblas_OPTERON)
+#ifdef DYNAMIC_OLDER
+           return corename[13];
+#else
+           return corename[7];
+#endif
   if (gotoblas == &gotoblas_BARCELONA)    return corename[14];
-  if (gotoblas == &gotoblas_NANO)         return corename[15];
+  if (gotoblas == &gotoblas_NANO)
+#ifdef DYNAMIC_OLDER
+           return corename[15];
+#else
+           return corename[10];
+#endif
   if (gotoblas == &gotoblas_SANDYBRIDGE)  return corename[16];
-  if (gotoblas == &gotoblas_BOBCAT)       return corename[17];
+  if (gotoblas == &gotoblas_BOBCAT)
+#ifdef DYNAMIC_OLDER
+           return corename[17];
+#else
+           return corename[7];
+#endif
   if (gotoblas == &gotoblas_BULLDOZER)    return corename[18];
   if (gotoblas == &gotoblas_PILEDRIVER)   return corename[19];
   if (gotoblas == &gotoblas_HASWELL)      return corename[20];
@@ -787,6 +834,7 @@ char *gotoblas_corename(void) {
 }
 
 
+
 static gotoblas_t *force_coretype(char *coretype){
 
 	int i ;
diff --git a/driver/others/dynamic_power.c b/driver/others/dynamic_power.c
index 8c831b998..811a5fae3 100644
--- a/driver/others/dynamic_power.c
+++ b/driver/others/dynamic_power.c
@@ -6,6 +6,9 @@ extern gotoblas_t gotoblas_POWER8;
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 extern gotoblas_t gotoblas_POWER9;
 #endif
+#if (!defined __GNUC__) || ( __GNUC__ >= 11)
+extern gotoblas_t gotoblas_POWER10;
+#endif
 
 extern void openblas_warning(int verbose, const char *msg);
 
@@ -13,7 +16,8 @@ static char *corename[] = {
 	"unknown",
 	"POWER6",
 	"POWER8",
-	"POWER9"
+	"POWER9",
+	"POWER10"
 };
 
 #define NUM_CORETYPES 4
@@ -23,6 +27,9 @@ char *gotoblas_corename(void) {
 	if (gotoblas == &gotoblas_POWER8)	return corename[2];
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	if (gotoblas == &gotoblas_POWER9)	return corename[3];
+#endif
+#if (!defined __GNUC__) || ( __GNUC__ >= 11)
+	if (gotoblas == &gotoblas_POWER10)	return corename[4];
 #endif
 	return corename[0];
 }
@@ -36,6 +43,10 @@ static gotoblas_t *get_coretype(void) {
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	if (__builtin_cpu_is("power9"))
 		return &gotoblas_POWER9;
+#endif
+#if (!defined __GNUC__) || ( __GNUC__ >= 11)
+	if (__builtin_cpu_is("isa_3_1") && __builtin_cpu_supports ("mma"))
+		return &gotoblas_POWER10;
 #endif
 	return NULL;
 }
@@ -61,6 +72,9 @@ static gotoblas_t *force_coretype(char * coretype) {
 	case  2: return (&gotoblas_POWER8);
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	case  3: return (&gotoblas_POWER9);
+#endif
+#if (!defined __GNUC__) || ( __GNUC__ >= 11)
+	case  4: return (&gotoblas_POWER10);
 #endif
 	default: return NULL;
 	}
diff --git a/driver/others/dynamic_zarch.c b/driver/others/dynamic_zarch.c
index 90d3051b1..403b34111 100644
--- a/driver/others/dynamic_zarch.c
+++ b/driver/others/dynamic_zarch.c
@@ -1,12 +1,58 @@
-
 #include "common.h"
+#include <stdbool.h>
 
+// Gate kernels for z13 and z14 on gcc version
+#if (__GNUC__ == 5 && __GNUC_MINOR__ >= 2) || __GNUC__ >= 6 ||           \
+    /* RHEL 7 since 7.3: */                                              \
+    (__GNUC__ == 4 && __GNUC_MINOR__ == 8 && __GNUC_PATCHLEVEL__ == 5 && \
+     __GNUC_RH_RELEASE__ >= 11)
+#define HAVE_Z13_SUPPORT
+#endif
+
+#if __GNUC__ >= 7
+#define HAVE_Z14_SUPPORT
+#endif
+
+// Guard the use of getauxval() on glibc version >= 2.16
+#ifdef __GLIBC__
+#include <features.h>
+#if __GLIBC_PREREQ(2, 16)
+#include <sys/auxv.h>
+#define HAVE_GETAUXVAL 1
+
+static unsigned long get_hwcap(void)
+{
+	unsigned long hwcap = getauxval(AT_HWCAP);
+	char *maskenv;
+
+	// honor requests for not using specific CPU features in LD_HWCAP_MASK
+	maskenv = getenv("LD_HWCAP_MASK");
+	if (maskenv)
+		hwcap &= strtoul(maskenv, NULL, 0);
+
+	return hwcap;
+	// note that a missing auxval is interpreted as no capabilities
+	// available, which is safe.
+}
+
+#else // __GLIBC_PREREQ(2, 16)
+#warn "Cannot detect SIMD support in Z13 or newer architectures since glibc is older than 2.16"
+
+static unsigned long get_hwcap(void) {
+	// treat missing support for getauxval() as no capabilities available,
+	// which is safe.
+	return 0;
+}
+#endif // __GLIBC_PREREQ(2, 16)
+#endif // __GLIBC
+
+extern gotoblas_t gotoblas_ZARCH_GENERIC;
+#ifdef HAVE_Z13_SUPPORT
 extern gotoblas_t gotoblas_Z13;
+#endif
+#ifdef HAVE_Z14_SUPPORT
 extern gotoblas_t gotoblas_Z14;
-//extern gotoblas_t gotoblas_Z15;
-//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
-//extern gotoblas_t gotoblas_Z14;
-//#endif
+#endif
 
 #define NUM_CORETYPES 4
 
@@ -16,47 +62,50 @@ static char* corename[] = {
 	"unknown",
 	"Z13",
 	"Z14",
-//	"Z15",
 	"ZARCH_GENERIC",
 };
 
 char* gotoblas_corename(void) {
+#ifdef HAVE_Z13_SUPPORT
 	if (gotoblas == &gotoblas_Z13)	return corename[1];
+#endif
+#ifdef HAVE_Z14_SUPPORT
 	if (gotoblas == &gotoblas_Z14)	return corename[2];
-//	if (gotoblas == &gotoblas_Z15)	return corename[3];
-//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
-//	if (gotoblas == &gotoblas_POWER9)	return corename[3];
-//#endif
-	return corename[0]; // try generic?
+#endif
+	if (gotoblas == &gotoblas_ZARCH_GENERIC) return corename[3];
+
+	return corename[0];
 }
 
-// __builtin_cpu_is is not supported by zarch
+/**
+ * Detect the fitting set of kernels by retrieving the CPU features supported by
+ * OS from the auxiliary value AT_HWCAP and choosing the set of kernels
+ * ("coretype") that exploits most of the features and can be compiled with the
+ * available gcc version.
+ * Note that we cannot use vector registers on a z13 or newer unless supported
+ * by the OS kernel (which needs to handle them properly during context switch).
+ */
 static gotoblas_t* get_coretype(void) {
-	FILE* infile;
-	char buffer[512], * p;
 
-	p = (char*)NULL;
-	infile = fopen("/proc/sysinfo", "r");
-	while (fgets(buffer, sizeof(buffer), infile)) {
-		if (!strncmp("Type", buffer, 4)) {
-			p = strchr(buffer, ':') + 2;
-#if 0
-			fprintf(stderr, "%s\n", p);
+	unsigned long hwcap __attribute__((unused)) = get_hwcap();
+
+	// z14 and z15 systems: exploit Vector Facility (SIMD) and
+	// Vector-Enhancements Facility 1 (float SIMD instructions), if present.
+#ifdef HAVE_Z14_SUPPORT
+	if ((hwcap & HWCAP_S390_VX) && (hwcap & HWCAP_S390_VXE))
+		return &gotoblas_Z14;
 #endif
-			break;
-		}
-	}
 
-	fclose(infile);
+	// z13: Vector Facility (SIMD for double)
+#ifdef HAVE_Z13_SUPPORT
+	if (hwcap & HWCAP_S390_VX)
+		return &gotoblas_Z13;
+#endif
 
-	if (strstr(p, "2964")) return &gotoblas_Z13;
-	if (strstr(p, "2965")) return &gotoblas_Z13;
-	if (strstr(p, "3906")) return &gotoblas_Z14;
-	if (strstr(p, "3907")) return &gotoblas_Z14;
-	if (strstr(p, "8561")) return &gotoblas_Z14;        // fallback z15 to z14
-	if (strstr(p, "8562")) return &gotoblas_Z14;        // fallback z15 to z14
-
-	return NULL; // should be ZARCH_GENERIC
+	// fallback in case of missing compiler support, systems before z13, or
+	// when the OS does not advertise support for the Vector Facility (e.g.,
+	// missing support in the OS kernel)
+	return &gotoblas_ZARCH_GENERIC;
 }
 
 static gotoblas_t* force_coretype(char* coretype) {
@@ -76,12 +125,13 @@ static gotoblas_t* force_coretype(char* coretype) {
 
 	switch (found)
 	{
+#ifdef HAVE_Z13_SUPPORT
 	case  1: return (&gotoblas_Z13);
+#endif
+#ifdef HAVE_Z14_SUPPORT
 	case  2: return (&gotoblas_Z14);
-//	case  3: return (&gotoblas_Z15);
-//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
-//	case  3: return (&gotoblas_POWER9);
-//#endif
+#endif
+	case  3: return (&gotoblas_ZARCH_GENERIC);
 	default: return NULL;
 	}
 	snprintf(message, 128, "Core not found: %s\n", coretype);
@@ -109,9 +159,9 @@ void gotoblas_dynamic_init(void) {
 
 	if (gotoblas == NULL)
 	{
-		snprintf(coremsg, 128, "Falling back to Z14 core\n");
+		snprintf(coremsg, 128, "Failed to detect system, falling back to generic z support.\n");
 		openblas_warning(1, coremsg);
-		gotoblas = &gotoblas_Z14;
+		gotoblas = &gotoblas_ZARCH_GENERIC;
 	}
 
 	if (gotoblas && gotoblas->init) {
diff --git a/driver/others/memory.c b/driver/others/memory.c
index 5abcbf3a4..a5595aed4 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -2070,7 +2070,7 @@ if (!release->address) return;
   if (munmap(release -> address, BUFFER_SIZE)) {
       int errsv=errno;
        perror("OpenBLAS : munmap failed:");
-       printf("error code=%d,\trelease->address=%lx\n",errsv,release->address);
+       printf("error code=%d,\trelease->address=%p\n",errsv,release->address);
   }
 }
 
diff --git a/driver/others/parameter.c b/driver/others/parameter.c
index 8bf7da78b..b1f3befae 100644
--- a/driver/others/parameter.c
+++ b/driver/others/parameter.c
@@ -62,6 +62,11 @@ BLASLONG gemm_offset_b = DEFAULT_GEMM_OFFSET_B;
 BLASLONG gemm_offset_b = GEMM_OFFSET_B;
 #endif
 
+#if SHGEMM_P == shgemm_p
+BLASLONG shgemm_p = DEFAULT_GEMM_P;
+#else
+BLASLONG shgemm_p = SHGEMM_P;
+#endif
 #if SGEMM_P == sgemm_p
 BLASLONG sgemm_p = DEFAULT_GEMM_P;
 #else
@@ -83,6 +88,11 @@ BLASLONG zgemm_p = DEFAULT_GEMM_P;
 BLASLONG zgemm_p = ZGEMM_P;
 #endif
 
+#if SHGEMM_Q == shgemm_q
+BLASLONG shgemm_q = DEFAULT_GEMM_Q;
+#else
+BLASLONG shgemm_q = SHGEMM_Q;
+#endif
 #if SGEMM_Q == sgemm_q
 BLASLONG sgemm_q = DEFAULT_GEMM_Q;
 #else
@@ -104,6 +114,11 @@ BLASLONG zgemm_q = DEFAULT_GEMM_Q;
 BLASLONG zgemm_q = ZGEMM_Q;
 #endif
 
+#if SHGEMM_R == shgemm_r
+BLASLONG shgemm_r = DEFAULT_GEMM_R;
+#else
+BLASLONG shgemm_r = SHGEMM_R;
+#endif
 #if SGEMM_R == sgemm_r
 BLASLONG sgemm_r = DEFAULT_GEMM_R;
 #else
@@ -597,6 +612,7 @@ void blas_set_parameter(void){
 
   size = BITMASK(cpuid3, 16, 0xff);
 
+  shgemm_p = 192 * (size + 1);
   sgemm_p = 192 * (size + 1);
   dgemm_p =  96 * (size + 1);
   cgemm_p =  96 * (size + 1);
@@ -610,6 +626,7 @@ void blas_set_parameter(void){
   xgemm_p =  16 * (size + 1);
 #endif
 
+  shgemm_r = (((BUFFER_SIZE - ((SHGEMM_P * SHGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SHGEMM_Q *  4)) - 15) & ~15;
   sgemm_r = (((BUFFER_SIZE - ((SGEMM_P * SGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SGEMM_Q *  4)) - 15) & ~15;
   dgemm_r = (((BUFFER_SIZE - ((DGEMM_P * DGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (DGEMM_Q *  8)) - 15) & ~15;
   cgemm_r = (((BUFFER_SIZE - ((CGEMM_P * CGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (CGEMM_Q *  8)) - 15) & ~15;
diff --git a/exports/Makefile b/exports/Makefile
index 60291b1ff..01a313b35 100644
--- a/exports/Makefile
+++ b/exports/Makefile
@@ -30,6 +30,10 @@ ifndef BUILD_LAPACK_DEPRECATED
 BUILD_LAPACK_DEPRECATED = 0
 endif
 
+ifndef BUILD_HALF
+BUILD_HALF = 0
+endif
+
 ifeq ($(OSNAME), WINNT)
 ifeq ($(F_COMPILER), GFORTRAN)
 ifndef ONLY_CBLAS
@@ -151,8 +155,12 @@ ifeq ($(F_COMPILER), INTEL)
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
 	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
+else ifeq ($(F_COMPILER), FLANG)
+	$(FC) $(FFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
+	-Wl,--whole-archive $< -Wl,--no-whole-archive \
+	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 else
-
 ifneq ($(C_COMPILER), LSB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
@@ -234,23 +242,23 @@ static : ../$(LIBNAME)
 	rm -f goto.$(SUFFIX)
 
 osx.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol osx $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol osx $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 aix.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol aix $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol aix $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 objcopy.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol objcopy $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol objcopy $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 objconv.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol objconv $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol objconv $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 test : linktest.c
 	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) -lm && echo OK.
 	rm -f linktest
 
 linktest.c : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol linktest  $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > linktest.c
+	perl ./gensymbol linktest  $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > linktest.c
 
 clean ::
 	@rm -f *.def *.dylib __.SYMDEF* *.renamed
diff --git a/exports/gensymbol b/exports/gensymbol
index d2894e6c8..73b4be248 100644
--- a/exports/gensymbol
+++ b/exports/gensymbol
@@ -40,17 +40,13 @@
     ztbsv,ztpmv,ztpsv,ztrmm,ztrmv,ztrsm,ztrsv,
     xerbla,
     saxpby,daxpby,caxpby,zaxpby,
+    somatcopy, domatcopy, comatcopy, zomatcopy,
+    simatcopy, dimatcopy, cimatcopy, zimatcopy,
     sgeadd,dgeadd,cgeadd,zgeadd,
-    somatcopy,
-    simatcopy,
-    domatcopy,
-    dimatcopy,
-    comatcopy,
-    cimatcopy,
-    zomatcopy,
-    zimatcopy,
+    ssum, dsum, scsum, dzsum
 );
 
+@halfblasobjs = (shgemm);
 @cblasobjs = (
     cblas_caxpy, cblas_ccopy, cblas_cdotc, cblas_cdotu, cblas_cgbmv, cblas_cgemm, cblas_cgemv,
     cblas_cgerc, cblas_cgeru, cblas_chbmv, cblas_chemm, cblas_chemv, cblas_cher2, cblas_cher2k,
@@ -80,9 +76,16 @@
     cblas_saxpby,cblas_daxpby,cblas_caxpby,cblas_zaxpby,
     cblas_somatcopy, cblas_domatcopy, cblas_comatcopy, cblas_zomatcopy,
     cblas_simatcopy, cblas_dimatcopy, cblas_cimatcopy, cblas_zimatcopy,
-    cblas_sgeadd, cblas_dgeadd,cblas_cgeadd, cblas_zgeadd
+    cblas_sgeadd, cblas_dgeadd,cblas_cgeadd, cblas_zgeadd,
+    cblas_isamin, cblas_idamin, cblas_icamin, cblas_izamin,
+    cblas_ismin, cblas_idmin, cblas_icmin, cblas_izmin,
+    cblas_ismax, cblas_idmax, cblas_icmax, cblas_izmax,
+    cblas_ssum, cblas_dsum, cblas_scsum, cblas_dzsum,
+    cblas_xerbla
 );
 
+@halfcblasobjs = (cblas_shgemm);
+
 @exblasobjs = (
     qamax,qamin,qasum,qaxpy,qcabs1,qcopy,qdot,qgbmv,qgemm,
     qgemv,qger,qmax,qmin,
@@ -3454,6 +3457,10 @@ use File::Spec;
 use File::Basename;
 my $dirname = File::Spec->catfile(dirname(dirname(File::Spec->rel2abs(__FILE__))), "lapack-netlib");
 
+if ($ARGV[12] == 1) {
+	@blasobjs = (@blasobjs, @halfblasobjs);
+	@cblasobjs = (@cblasobjs, @halfcblasobjs);
+}
 if ($ARGV[8] == 1) {
     #ONLY_CBLAS=1
     @underscore_objs = (@misc_underscore_objs);
@@ -3494,9 +3501,12 @@ if ($ARGV[1] eq "x86")    { @underscore_objs = (@underscore_objs, @gemm3mobjs);
 if ($ARGV[1] eq "ia64")   { @underscore_objs = (@underscore_objs, @gemm3mobjs); };
 if ($ARGV[1] eq "MIPS")   { @underscore_objs = (@underscore_objs, @gemm3mobjs); };
 
-
 if ($ARGV[4] == 0) {
     @no_underscore_objs = (@cblasobjs, @misc_no_underscore_objs);
+    if ($ARGV[1] eq "x86_64") { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
+    if ($ARGV[1] eq "x86")    { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
+    if ($ARGV[1] eq "ia64")   { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
+    if ($ARGV[1] eq "MIPS")   { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
 }else{
     #NO_CBLAS=1
     @no_underscore_objs = (@misc_no_underscore_objs);
diff --git a/f_check b/f_check
index fac8fc707..17d863224 100644
--- a/f_check
+++ b/f_check
@@ -334,7 +334,8 @@ if ($link ne "") {
 	    && ($flags !~ /kernel32/)
 	    && ($flags !~ /advapi32/)
 	    && ($flags !~ /shell32/)
-	    && ($flags !~ /omp/)
+	    && ($flags !~ /omp/ || ($vendor !~ /PGI/ && $flags =~ /omp/))
+	    && ($flags !~ /[0-9]+/)
 		&& ($flags !~ /^\-l$/)
 	    ) {
 	    $linker_l .= $flags . " ";
diff --git a/getarch.c b/getarch.c
index 145753bcc..164947f3e 100644
--- a/getarch.c
+++ b/getarch.c
@@ -650,6 +650,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "POWER9"
 #endif
 
+#if defined(FORCE_POWER10)
+#define FORCE
+#define ARCHITECTURE    "POWER"
+#define SUBARCHITECTURE "POWER10"
+#define SUBDIRNAME      "power"
+#define ARCHCONFIG   "-DPOWER10 " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=128 " \
+		     "-DL2_SIZE=4194304 -DL2_LINESIZE=128 " \
+		     "-DDTB_DEFAULT_ENTRIES=128 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=8 "
+#define LIBNAME   "power10"
+#define CORENAME  "POWER10"
+#endif
+
 #ifdef FORCE_PPCG4
 #define FORCE
 #define ARCHITECTURE    "POWER"
@@ -812,6 +825,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_MIPS1004K
+#define FORCE
+#define ARCHITECTURE    "MIPS"
+#define SUBARCHITECTURE "MIPS1004K"
+#define SUBDIRNAME      "mips"
+#define ARCHCONFIG   "-DMIPS1004K " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=262144 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=8 "
+#define LIBNAME   "mips1004K"
+#define CORENAME  "MIPS1004K"
+#else
+#endif
+
+#ifdef FORCE_MIPS24K
+#define FORCE
+#define ARCHITECTURE    "MIPS"
+#define SUBARCHITECTURE "MIPS24K"
+#define SUBDIRNAME      "mips"
+#define ARCHCONFIG   "-DMIPS24K " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=32768 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=8 "
+#define LIBNAME   "mips24K"
+#define CORENAME  "MIPS24K"
+#else
+#endif
+
 #ifdef FORCE_I6500
 #define FORCE
 #define ARCHITECTURE    "MIPS"
@@ -1334,10 +1375,12 @@ int main(int argc, char *argv[]){
 
 #if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 printf("__BYTE_ORDER__=__ORDER_BIG_ENDIAN__\n");
-#endif
-#if defined(__BIG_ENDIAN__) && __BIG_ENDIAN__ > 0
+#elif defined(__BIG_ENDIAN__) && __BIG_ENDIAN__ > 0
 printf("__BYTE_ORDER__=__ORDER_BIG_ENDIAN__\n");
 #endif
+#if defined(_CALL_ELF) && (_CALL_ELF == 2)
+printf("ELF_VERSION=2\n");
+#endif
 
 #ifdef MAKE_NB_JOBS
   #if MAKE_NB_JOBS > 0
diff --git a/getarch_2nd.c b/getarch_2nd.c
index cf9c578cb..a1d0ccac8 100644
--- a/getarch_2nd.c
+++ b/getarch_2nd.c
@@ -9,6 +9,8 @@
 int main(int argc, char **argv) {
 
   if ( (argc <= 1) || ((argc >= 2) && (*argv[1] == '0'))) {
+    printf("SHGEMM_UNROLL_M=%d\n", SHGEMM_DEFAULT_UNROLL_M);
+    printf("SHGEMM_UNROLL_N=%d\n", SHGEMM_DEFAULT_UNROLL_N);
     printf("SGEMM_UNROLL_M=%d\n", SGEMM_DEFAULT_UNROLL_M);
     printf("SGEMM_UNROLL_N=%d\n", SGEMM_DEFAULT_UNROLL_N);
     printf("DGEMM_UNROLL_M=%d\n", DGEMM_DEFAULT_UNROLL_M);
diff --git a/interface/CMakeLists.txt b/interface/CMakeLists.txt
index 5ea39f864..7a8fc6698 100644
--- a/interface/CMakeLists.txt
+++ b/interface/CMakeLists.txt
@@ -115,7 +115,7 @@ foreach (float_type ${FLOAT_TYPES})
     GenerateNamedObjects("syr2k.c" "HEMM" "her2k" ${CBLAS_FLAG} "" "" false ${float_type})
 
     if (USE_GEMM3M)
-      GenerateNamedObjects("gemm.c" "GEMM3M" "gemm3m" false "" "" false ${float_type})
+      GenerateNamedObjects("gemm.c" "GEMM3M" "gemm3m" ${CBLAS_FLAG} "" "" false ${float_type})
     endif()
   endif ()
   if (${float_type} STREQUAL "COMPLEX")
diff --git a/interface/Makefile b/interface/Makefile
index 3f0dcca28..44a9fdcf0 100644
--- a/interface/Makefile
+++ b/interface/Makefile
@@ -46,6 +46,9 @@ SBLAS3OBJS    = \
 		somatcopy.$(SUFFIX) simatcopy.$(SUFFIX)\
 		sgeadd.$(SUFFIX)
 
+ifeq ($(BUILD_HALF),1)
+SHBLAS3OBJS    = shgemm.$(SUFFIX)
+endif
 
 DBLAS1OBJS    = \
 		daxpy.$(SUFFIX) dswap.$(SUFFIX) \
@@ -277,6 +280,10 @@ CSBLAS3OBJS   = \
 	cblas_ssyrk.$(SUFFIX) cblas_ssyr2k.$(SUFFIX) cblas_somatcopy.$(SUFFIX)  cblas_simatcopy.$(SUFFIX)\
 	cblas_sgeadd.$(SUFFIX)
 
+ifeq ($(BUILD_HALF),1)
+CSHBLAS3OBJS = cblas_shgemm.$(SUFFIX)
+endif
+
 CDBLAS1OBJS   = \
 	cblas_idamax.$(SUFFIX) cblas_idamin.$(SUFFIX) cblas_dasum.$(SUFFIX) cblas_daxpy.$(SUFFIX) \
 	cblas_dcopy.$(SUFFIX) cblas_ddot.$(SUFFIX) \
@@ -367,6 +374,7 @@ override CFLAGS += -I.
 SBLAS1OBJS   += $(CSBLAS1OBJS)
 SBLAS2OBJS   += $(CSBLAS2OBJS)
 SBLAS3OBJS   += $(CSBLAS3OBJS)
+SHBLAS3OBJS  += $(CSHBLAS3OBJS)
 DBLAS1OBJS   += $(CDBLAS1OBJS)
 DBLAS2OBJS   += $(CDBLAS2OBJS)
 DBLAS3OBJS   += $(CDBLAS3OBJS)
@@ -380,6 +388,7 @@ ZBLAS3OBJS   += $(CZBLAS3OBJS)
 endif
 
 SBLASOBJS    = $(SBLAS1OBJS) $(SBLAS2OBJS) $(SBLAS3OBJS)
+SHBLASOBJS   = $(SHBLAS3OBJS)
 DBLASOBJS    = $(DBLAS1OBJS) $(DBLAS2OBJS) $(DBLAS3OBJS)
 QBLASOBJS    = $(QBLAS1OBJS) $(QBLAS2OBJS) $(QBLAS3OBJS)
 CBLASOBJS    = $(CBLAS1OBJS) $(CBLAS2OBJS) $(CBLAS3OBJS)
@@ -454,7 +463,7 @@ ZBLASOBJS += $(ZLAPACKOBJS)
 
 endif
 
-FUNCOBJS    = $(SBLASOBJS) $(DBLASOBJS) $(CBLASOBJS) $(ZBLASOBJS)
+FUNCOBJS    = $(SHBLASOBJS) $(SBLASOBJS) $(DBLASOBJS) $(CBLASOBJS) $(ZBLASOBJS)
 
 ifdef EXPRECISION
 FUNCOBJS   += $(QBLASOBJS) $(XBLASOBJS)
@@ -488,10 +497,10 @@ level1 : $(SBLAS1OBJS) $(DBLAS1OBJS) $(QBLAS1OBJS) $(CBLAS1OBJS) $(ZBLAS1OBJS) $
 level2 : $(SBLAS2OBJS) $(DBLAS2OBJS) $(QBLAS2OBJS) $(CBLAS2OBJS) $(ZBLAS2OBJS) $(XBLAS2OBJS)
 	$(AR) $(ARFLAGS) -ru $(TOPDIR)/$(LIBNAME) $^
 
-level3 : $(SBLAS3OBJS) $(DBLAS3OBJS) $(QBLAS3OBJS) $(CBLAS3OBJS) $(ZBLAS3OBJS) $(XBLAS3OBJS)
+level3 : $(SHBLAS3OBJS) $(SBLAS3OBJS) $(DBLAS3OBJS) $(QBLAS3OBJS) $(CBLAS3OBJS) $(ZBLAS3OBJS) $(XBLAS3OBJS)
 	$(AR) $(ARFLAGS) -ru $(TOPDIR)/$(LIBNAME) $^
 
-$(CSBLASOBJS) $(CSBLASOBJS_P) $(CDBLASOBJS) $(CDBLASOBJS_P) $(CQBLASOBJS) $(CQBLASOBJS_P) \
+$(CSHBLASOBJS) $(CSHBLASOBJS_P) $(CSBLASOBJS) $(CSBLASOBJS_P) $(CDBLASOBJS) $(CDBLASOBJS_P) $(CQBLASOBJS) $(CQBLASOBJS_P) \
 $(CCBLASOBJS) $(CCBLASOBJS_P) $(CZBLASOBJS) $(CZBLASOBJS_P) $(CXBLASOBJS) $(CXBLASOBJS_P) : override CFLAGS += -DCBLAS
 
 srot.$(SUFFIX) srot.$(PSUFFIX) : rot.c
@@ -1209,6 +1218,11 @@ zhpr2.$(SUFFIX) zhpr2.$(PSUFFIX) : zhpr2.c
 xhpr2.$(SUFFIX) xhpr2.$(PSUFFIX) : zhpr2.c
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
+ifeq ($(BUILD_HALF),1)
+shgemm.$(SUFFIX) shgemm.$(PSUFFIX) : gemm.c ../param.h
+	$(CC) -c $(CFLAGS) $< -o $(@F)
+endif
+
 sgemm.$(SUFFIX) sgemm.$(PSUFFIX) : gemm.c ../param.h
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
@@ -1770,6 +1784,11 @@ cblas_zhemv.$(SUFFIX) cblas_zhemv.$(PSUFFIX) : zhemv.c
 cblas_sgemm.$(SUFFIX) cblas_sgemm.$(PSUFFIX) : gemm.c ../param.h
 	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
 
+ifeq ($(BUILD_HALF),1)
+cblas_shgemm.$(SUFFIX) cblas_shgemm.$(PSUFFIX) : gemm.c ../param.h
+	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
+endif
+
 cblas_dgemm.$(SUFFIX) cblas_dgemm.$(PSUFFIX) : gemm.c ../param.h
 	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
 
diff --git a/interface/gemm.c b/interface/gemm.c
index 0b18d9a8c..99388e7d9 100644
--- a/interface/gemm.c
+++ b/interface/gemm.c
@@ -77,7 +77,7 @@
 #define GEMM_MULTITHREAD_THRESHOLD 4
 #endif
 
-static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
+static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, IFLOAT *, IFLOAT *, BLASLONG) = {
 #ifndef GEMM3M
   GEMM_NN, GEMM_TN, GEMM_RN, GEMM_CN,
   GEMM_NT, GEMM_TT, GEMM_RT, GEMM_CT,
@@ -108,8 +108,8 @@ static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLA
 void NAME(char *TRANSA, char *TRANSB,
 	  blasint *M, blasint *N, blasint *K,
 	  FLOAT *alpha,
-	  FLOAT *a, blasint *ldA,
-	  FLOAT *b, blasint *ldB,
+	  IFLOAT *a, blasint *ldA,
+	  IFLOAT *b, blasint *ldB,
 	  FLOAT *beta,
 	  FLOAT *c, blasint *ldC){
 
@@ -119,8 +119,8 @@ void NAME(char *TRANSA, char *TRANSB,
   blasint info;
 
   char transA, transB;
-  FLOAT *buffer;
-  FLOAT *sa, *sb;
+  IFLOAT *buffer;
+  IFLOAT *sa, *sb;
 
 #ifdef SMP
   double MNK;
diff --git a/kernel/CMakeLists.txt b/kernel/CMakeLists.txt
index 35e0fff25..d1349c5f8 100644
--- a/kernel/CMakeLists.txt
+++ b/kernel/CMakeLists.txt
@@ -41,6 +41,9 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     foreach (float_type ${FLOAT_TYPES})
       # a bit of metaprogramming here to pull out the appropriate KERNEL var
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       GenerateNamedObjects("${KERNELDIR}/${${float_char}AMAXKERNEL}" "USE_ABS" "amax_k" false "" "" false ${float_type})
       GenerateNamedObjects("${KERNELDIR}/${${float_char}AMINKERNEL}" "USE_ABS;USE_MIN" "amin_k" false "" "" false ${float_type})
       if (DEFINED ${float_char}MAXKERNEL)
@@ -93,6 +96,9 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     GenerateNamedObjects("generic/ger.c" "" "ger_k" false "" "" "" 3)
     foreach (float_type ${FLOAT_TYPES})
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GERUKERNEL}" "" "geru_k" false "" "" false ${float_type})
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GERCKERNEL}" "CONJ" "gerc_k" false "" "" false ${float_type})
@@ -124,17 +130,27 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     if (ARM OR ARM64 OR (TARGET_CORE MATCHES LONGSOON3B) OR (TARGET_CORE MATCHES GENERIC) OR (TARGET_CORE MATCHES HASWELL) OR (TARGET_CORE MATCHES ZEN) OR (TARGET_CORE MATCHES SKYLAKEX) )
       set(USE_TRMM true)
     endif ()
-    if (ZARCH OR (TARGET_CORE MATCHES POWER8) OR (TARGET_CORE MATCHES POWER9)) 
+    if (ZARCH OR (TARGET_CORE MATCHES POWER8) OR (TARGET_CORE MATCHES POWER9) OR (TARGET_CORE MATCHES POWER10))
       set(USE_TRMM true)
     endif ()
 
-    foreach (float_type SINGLE DOUBLE)
+    foreach (float_type SINGLE DOUBLE HALF)
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+        if (NOT ${BUILD_HALF})
+	  continue ()
+        else ()
+	  set (float_char "SH")
+      endif ()
+      endif ()
       GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "" "gemm_kernel" false "" "" false ${float_type})
     endforeach()
 
     foreach (float_type ${FLOAT_TYPES})
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       if (${float_char}GEMMINCOPY)
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMINCOPY}" "${float_type}" "${${float_char}GEMMINCOPYOBJ}" false "" "" true ${float_type})
       endif ()
@@ -470,9 +486,13 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
       GenerateNamedObjects("${KERNELDIR}/${${float_char}GEADD_KERNEL}" "" "geadd_k" false "" "" false ${float_type})
     endforeach ()
 
+
     # Makefile.LA
     if(NOT NO_LAPACK)
       foreach (float_type ${FLOAT_TYPES})
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
         if (NOT DEFINED ${float_char}NEG_TCOPY)
           if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C" OR ${float_char} STREQUAL "X")
             set(${float_char}NEG_TCOPY ../generic/zneg_tcopy.c)
@@ -516,6 +536,9 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     foreach (float_type ${FLOAT_TYPES})
       # a bit of metaprogramming here to pull out the appropriate KERNEL var
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       GenerateNamedObjects("generic/neg_tcopy_${${float_char}GEMM_UNROLL_M}.c" "" "neg_tcopy" false "" ${TSUFFIX}  false ${float_type})
       GenerateNamedObjects("generic/laswp_ncopy_${${float_char}GEMM_UNROLL_N}.c" "" "laswp_ncopy" false "" ${TSUFFIX}  false ${float_type})
     endforeach ()
diff --git a/kernel/Makefile.L3 b/kernel/Makefile.L3
index 6d96abb2e..86772cb22 100644
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@@ -51,6 +51,10 @@ ifeq ($(CORE), POWER9)
 USE_TRMM = 1
 endif
 
+ifeq ($(CORE), POWER10)
+USE_TRMM = 1
+endif
+
 ifeq ($(ARCH), zarch)
 USE_TRMM = 1
 endif
@@ -59,6 +63,25 @@ ifeq ($(CORE), Z14)
 USE_TRMM = 1
 endif
 
+ifeq ($(BUILD_HALF), 1)
+ifndef SHGEMMKERNEL
+SHGEMM_BETA = ../generic/gemm_beta.c
+SHGEMMKERNEL    = ../generic/gemmkernel_2x2.c
+SHGEMMINCOPY    = ../generic/gemm_ncopy_2.c
+SHGEMMITCOPY    = ../generic/gemm_tcopy_2.c
+SHGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+SHGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+SHGEMMINCOPYOBJ =  shgemm_incopy$(TSUFFIX).$(SUFFIX)
+SHGEMMITCOPYOBJ =  shgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SHGEMMONCOPYOBJ =  shgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SHGEMMOTCOPYOBJ =  shgemm_otcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+SHKERNELOBJS	+= \
+	shgemm_kernel$(TSUFFIX).$(SUFFIX) \
+	$(SHGEMMINCOPYOBJ) $(SHGEMMITCOPYOBJ) \
+	$(SHGEMMONCOPYOBJ) $(SHGEMMOTCOPYOBJ)
+endif
 
 SKERNELOBJS	+= \
 	sgemm_kernel$(TSUFFIX).$(SUFFIX) \
@@ -93,6 +116,9 @@ XKERNELOBJS	+= \
 	$(XGEMMINCOPYOBJ) $(XGEMMITCOPYOBJ) \
 	$(XGEMMONCOPYOBJ) $(XGEMMOTCOPYOBJ)
 
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS      += $(SHKERNELOBJS)
+endif
 SBLASOBJS	+= $(SKERNELOBJS)
 DBLASOBJS	+= $(DKERNELOBJS)
 QBLASOBJS	+= $(QKERNELOBJS)
@@ -100,6 +126,10 @@ CBLASOBJS	+= $(CKERNELOBJS)
 ZBLASOBJS	+= $(ZKERNELOBJS)
 XBLASOBJS	+= $(XKERNELOBJS)
 
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS += shgemm_beta$(TSUFFIX).$(SUFFIX)
+endif
+
 SBLASOBJS	+= \
 	sgemm_beta$(TSUFFIX).$(SUFFIX) \
 	strmm_kernel_LN$(TSUFFIX).$(SUFFIX) strmm_kernel_LT$(TSUFFIX).$(SUFFIX) \
@@ -389,6 +419,12 @@ ZBLASOBJS += \
 	zimatcopy_k_ctc$(TSUFFIX).$(SUFFIX) zimatcopy_k_rtc$(TSUFFIX).$(SUFFIX) \
 	zgeadd_k$(TSUFFIX).$(SUFFIX) 
 
+ifeq ($(BUILD_HALF), 1)
+SHGEMMINCOPYOBJ_P = $(SHGEMMINCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+SHGEMMITCOPYOBJ_P = $(SHGEMMITCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+SHGEMMONCOPYOBJ_P = $(SHGEMMONCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+SHGEMMOTCOPYOBJ_P = $(SHGEMMOTCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+endif
 
 SGEMMINCOPYOBJ_P = $(SGEMMINCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 SGEMMITCOPYOBJ_P = $(SGEMMITCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
@@ -415,6 +451,11 @@ XGEMMITCOPYOBJ_P = $(XGEMMITCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 XGEMMONCOPYOBJ_P = $(XGEMMONCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 XGEMMOTCOPYOBJ_P = $(XGEMMOTCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 
+ifeq ($(BUILD_HALF),1)
+$(KDIR)shgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SHGEMM_BETA)
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
 $(KDIR)sgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_BETA)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
@@ -433,12 +474,47 @@ $(KDIR)zgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_BETA)
 $(KDIR)xgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMM_BETA)
 	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX $< -o $@
 
+
+ifeq ($(BUILD_HALF), 1)
+
+$(KDIR)$(SHGEMMONCOPYOBJ) : $(KERNELDIR)/$(SHGEMMONCOPY)
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)$(SHGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SHGEMMOTCOPY)
+
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DHALF -UDOUBLE -UCOMPLEX $< -o - > shgemmotcopy.s
+	m4 shgemmotcopy.s > shgemmotcopy_nomacros.s
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemmotcopy_nomacros.s -o $@
+	rm shgemmotcopy.s shgemmotcopy_nomacros.s
+else
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
+ifneq ($(SHGEMM_UNROLL_M), $(SHGEMM_UNROLL_N))
+
+$(KDIR)$(SHGEMMINCOPYOBJ) : $(KERNELDIR)/$(SHGEMMINCOPY)
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)$(SHGEMMITCOPYOBJ) : $(KERNELDIR)/$(SHGEMMITCOPY)
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DHALF -UDOUBLE -UCOMPLEX $< -o - > shgemmitcopy.s
+	m4 shgemmitcopy.s > shgemmitcopy_nomacros.s
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemmitcopy_nomacros.s -o $@
+	rm shgemmitcopy.s shgemmitcopy_nomacros.s
+else
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
+endif
+endif
+
 $(KDIR)$(SGEMMONCOPYOBJ) : $(KERNELDIR)/$(SGEMMONCOPY)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SGEMMOTCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -UCOMPLEX $< -o sgemmotcopy.s	
+	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX $< -o - > sgemmotcopy.s	
 	m4 sgemmotcopy.s > sgemmotcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemmotcopy_nomacros.s -o $@
 	rm sgemmotcopy.s sgemmotcopy_nomacros.s
@@ -454,7 +530,7 @@ $(KDIR)$(SGEMMINCOPYOBJ) : $(KERNELDIR)/$(SGEMMINCOPY)
 
 $(KDIR)$(SGEMMITCOPYOBJ) : $(KERNELDIR)/$(SGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -UCOMPLEX $< -o sgemmitcopy.s	
+	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX $< -o - > sgemmitcopy.s	
 	m4 sgemmitcopy.s > sgemmitcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemmitcopy_nomacros.s -o $@
 	rm sgemmitcopy.s sgemmitcopy_nomacros.s
@@ -466,7 +542,7 @@ endif
 
 $(KDIR)$(DGEMMONCOPYOBJ) : $(KERNELDIR)/$(DGEMMONCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_ncopy.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_ncopy.s
 	m4 dgemm_ncopy.s > dgemm_ncopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_ncopy_nomacros.s -o $@
 	rm dgemm_ncopy.s dgemm_ncopy_nomacros.s
@@ -484,7 +560,7 @@ $(KDIR)$(DGEMMINCOPYOBJ) : $(KERNELDIR)/$(DGEMMINCOPY)
 
 $(KDIR)$(DGEMMITCOPYOBJ) : $(KERNELDIR)/$(DGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_itcopy.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_itcopy.s
 	m4 dgemm_itcopy.s > dgemm_itcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_itcopy_nomacros.s -o $@
 	rm dgemm_itcopy.s dgemm_itcopy_nomacros.s
@@ -527,7 +603,7 @@ $(KDIR)$(CGEMMINCOPYOBJ) : $(KERNELDIR)/$(CGEMMINCOPY)
 
 $(KDIR)$(CGEMMITCOPYOBJ) : $(KERNELDIR)/$(CGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -UDOUBLE -UCOMPLEX -E $< -o cgemm_itcopy.s
+	$(CC) $(CFLAGS) -UDOUBLE -UCOMPLEX -S $< -o - > cgemm_itcopy.s
 	m4 cgemm_itcopy.s > cgemm_itcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX cgemm_itcopy_nomacros.s -o $@
 	rm cgemm_itcopy.s cgemm_itcopy_nomacros.s
@@ -550,7 +626,7 @@ $(KDIR)$(ZGEMMINCOPYOBJ) : $(KERNELDIR)/$(ZGEMMINCOPY)
 
 $(KDIR)$(ZGEMMITCOPYOBJ) : $(KERNELDIR)/$(ZGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o zgemm_itcopy.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > zgemm_itcopy.s
 	m4 zgemm_itcopy.s > zgemm_itcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX zgemm_itcopy_nomacros.s -o $@
 	rm zgemm_itcopy.s zgemm_itcopy_nomacros.s
@@ -582,7 +658,7 @@ endif
 
 $(KDIR)sgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL) $(SGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -UCOMPLEX  $< -o sgemm_kernel$(TSUFFIX).s
+	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX  $< -o - > sgemm_kernel$(TSUFFIX).s
 	m4 sgemm_kernel$(TSUFFIX).s > sgemm_kernel$(TSUFFIX)_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm sgemm_kernel$(TSUFFIX).s sgemm_kernel$(TSUFFIX)_nomacros.s
@@ -590,9 +666,22 @@ else
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
+ifeq ($(BUILD_HALF), 1)
+
+$(KDIR)shgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SHGEMMKERNEL) $(SHGEMMDEPEND)
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DHALF -UDOUBLE -UCOMPLEX  $< -o - > shgemm_kernel$(TSUFFIX).s
+	m4 shgemm_kernel$(TSUFFIX).s > shgemm_kernel$(TSUFFIX)_nomacros.s
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemm_kernel$(TSUFFIX)_nomacros.s -o $@
+	rm shgemm_kernel$(TSUFFIX).s shgemm_kernel$(TSUFFIX)_nomacros.s
+else
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+endif
+
 $(KDIR)dgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_kernel$(TSUFFIX).s
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_kernel$(TSUFFIX).s
 	m4 dgemm_kernel$(TSUFFIX).s > dgemm_kernel$(TSUFFIX)_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm dgemm_kernel$(TSUFFIX).s dgemm_kernel$(TSUFFIX)_nomacros.s
@@ -605,7 +694,7 @@ $(KDIR)qgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL) $(QGEMMDEP
 
 $(KDIR)cgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DNN $< -o cgemm_kernel_n.s
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DNN $< -o - > cgemm_kernel_n.s
 	m4 cgemm_kernel_n.s > cgemm_kernel_n_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN cgemm_kernel_n_nomacros.s -o $@
 	rm cgemm_kernel_n.s cgemm_kernel_n_nomacros.s
@@ -615,7 +704,7 @@ endif
 
 $(KDIR)cgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DCN $< -o cgemm_kernel_l.s
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DCN $< -o - > cgemm_kernel_l.s
 	m4 cgemm_kernel_l.s > cgemm_kernel_l_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCN cgemm_kernel_l_nomacros.s -o $@
 	rm cgemm_kernel_l.s cgemm_kernel_l_nomacros.s
@@ -625,7 +714,7 @@ endif
 
 $(KDIR)cgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DNC  $< -o cgemm_kernel_r.s
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DNC  $< -o - > cgemm_kernel_r.s
 	m4 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
 	rm cgemm_kernel_r.s cgemm_kernel_r_nomacros.s
@@ -635,7 +724,7 @@ endif
 
 $(KDIR)cgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DCC $< -o cgemm_kernel_b.s
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DCC $< -o - > cgemm_kernel_b.s
 	m4 cgemm_kernel_b.s > cgemm_kernel_b_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCC cgemm_kernel_b_nomacros.s -o $@
 	rm cgemm_kernel_b.s cgemm_kernel_b_nomacros.s
@@ -645,7 +734,7 @@ endif
 
 $(KDIR)zgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DNN $< -o zgemm_kernel_n.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DNN $< -o - > zgemm_kernel_n.s
 	m4 zgemm_kernel_n.s > zgemm_kernel_n_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN zgemm_kernel_n_nomacros.s -o $@
 	rm zgemm_kernel_n.s zgemm_kernel_n_nomacros.s
@@ -655,7 +744,7 @@ endif
 
 $(KDIR)zgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DCN $< -o zgemm_kernel_l.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DCN $< -o - > zgemm_kernel_l.s
 	m4 zgemm_kernel_l.s > zgemm_kernel_l_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN zgemm_kernel_l_nomacros.s -o $@
 	rm zgemm_kernel_l.s zgemm_kernel_l_nomacros.s
@@ -665,7 +754,7 @@ endif
 
 $(KDIR)zgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DNC $< -o zgemm_kernel_r.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DNC $< -o - > zgemm_kernel_r.s
 	m4 zgemm_kernel_r.s > zgemm_kernel_r_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC zgemm_kernel_r_nomacros.s -o $@
 	rm zgemm_kernel_r.s zgemm_kernel_r_nomacros.s
@@ -675,7 +764,7 @@ endif
 
 $(KDIR)zgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DCC $< -o zgemm_kernel_b.s
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DCC $< -o - > zgemm_kernel_b.s
 	m4 zgemm_kernel_b.s > zgemm_kernel_b_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC zgemm_kernel_b_nomacros.s -o $@
 	rm zgemm_kernel_b.s zgemm_kernel_b_nomacros.s
@@ -699,7 +788,7 @@ $(KDIR)xgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMD
 ifdef USE_TRMM
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o strmmkernel_ln.s	
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > strmmkernel_ln.s	
 	m4 strmmkernel_ln.s > strmmkernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA strmmkernel_ln_nomacros.s -o $@
 	rm strmmkernel_ln.s strmmkernel_ln_nomacros.s
@@ -709,7 +798,7 @@ endif
 
 $(KDIR)strmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o strmmkernel_lt.s	
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > strmmkernel_lt.s	
 	m4 strmmkernel_lt.s > strmmkernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA strmmkernel_lt_nomacros.s -o $@
 	rm strmmkernel_lt.s strmmkernel_lt_nomacros.s
@@ -719,7 +808,7 @@ endif
 
 $(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o strmmkernel_rn.s	
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > strmmkernel_rn.s	
 	m4 strmmkernel_rn.s > strmmkernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA strmmkernel_rn_nomacros.s -o $@
 	rm strmmkernel_rn.s strmmkernel_rn_nomacros.s
@@ -729,7 +818,7 @@ endif
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o strmm_kernel_rt.s	
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
@@ -739,7 +828,7 @@ endif
 
 $(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o dtrmm_kernel_ln.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > dtrmm_kernel_ln.s
 	m4 dtrmm_kernel_ln.s > dtrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA dtrmm_kernel_ln_nomacros.s -o $@
 	rm dtrmm_kernel_ln.s dtrmm_kernel_ln_nomacros.s
@@ -749,7 +838,7 @@ endif
 
 $(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o dtrmm_kernel_lt.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > dtrmm_kernel_lt.s
 	m4 dtrmm_kernel_lt.s > dtrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA dtrmm_kernel_lt_nomacros.s -o $@
 	rm dtrmm_kernel_lt.s dtrmm_kernel_lt_nomacros.s
@@ -759,7 +848,7 @@ endif
 
 $(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o dtrmm_kernel_rn.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > dtrmm_kernel_rn.s
 	m4 dtrmm_kernel_rn.s > dtrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA dtrmm_kernel_rn_nomacros.s -o $@
 	rm dtrmm_kernel_rn.s dtrmm_kernel_rn_nomacros.s
@@ -769,7 +858,7 @@ endif
 
 $(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o dtrmm_kernel_rt.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > dtrmm_kernel_rt.s
 	m4 dtrmm_kernel_rt.s > dtrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA dtrmm_kernel_rt_nomacros.s -o $@
 	rm dtrmm_kernel_rt.s dtrmm_kernel_rt_nomacros.s
@@ -791,7 +880,7 @@ $(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
 
 $(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN  $< -o ctrmm_kernel_ln.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN  $< -o - > ctrmm_kernel_ln.s
 	m4 ctrmm_kernel_ln.s > ctrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_ln_nomacros.s -o $@
 	rm ctrmm_kernel_ln.s ctrmm_kernel_ln_nomacros.s
@@ -801,7 +890,7 @@ endif
 
 $(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o ctrmm_kernel_lt.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_lt.s
 	m4 ctrmm_kernel_lt.s > ctrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_lt_nomacros.s -o $@
 	rm ctrmm_kernel_lt.s ctrmm_kernel_lt_nomacros.s
@@ -811,7 +900,7 @@ endif
 
 $(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o ctrmm_kernel_lr.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lr.s
 	m4 ctrmm_kernel_lr.s > ctrmm_kernel_lr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN  ctrmm_kernel_lr_nomacros.s -o $@
 	rm ctrmm_kernel_lr.s ctrmm_kernel_lr_nomacros.s
@@ -821,7 +910,7 @@ endif
 
 $(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o ctrmm_kernel_lc.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lc.s
 	m4 ctrmm_kernel_lc.s > ctrmm_kernel_lc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ctrmm_kernel_lc_nomacros.s -o $@
 	rm ctrmm_kernel_lc_nomacros.s ctrmm_kernel_lc.s
@@ -831,7 +920,7 @@ endif
 
 $(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o ctrmm_kernel_rn.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rn.s
 	m4 ctrmm_kernel_rn.s > ctrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_rn_nomacros.s -o $@
 	rm ctrmm_kernel_rn.s ctrmm_kernel_rn_nomacros.s
@@ -841,7 +930,7 @@ endif
 
 $(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o ctrmm_kernel_rt.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rt.s
 	m4 ctrmm_kernel_rt.s > ctrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_rt_nomacros.s -o $@
 	rm ctrmm_kernel_rt.s ctrmm_kernel_rt_nomacros.s
@@ -851,7 +940,7 @@ endif
 
 $(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o ctrmm_kernel_rr.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_rr.s
 	m4 ctrmm_kernel_rr.s > ctrmm_kernel_rr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ctrmm_kernel_rr_nomacros.s -o $@
 	rm ctrmm_kernel_rr.s ctrmm_kernel_rr_nomacros.s
@@ -861,7 +950,7 @@ endif
 
 $(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o ctrmm_kernel_RC.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_RC.s
 	m4 ctrmm_kernel_RC.s > ctrmm_kernel_RC_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ctrmm_kernel_RC_nomacros.s -o $@
 	rm ctrmm_kernel_RC.s ctrmm_kernel_RC_nomacros.s
@@ -871,7 +960,7 @@ endif
 
 $(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o ztrmm_kernel_ln.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_ln.s
 	m4 ztrmm_kernel_ln.s > ztrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_ln_nomacros.s -o $@
 	rm ztrmm_kernel_ln.s ztrmm_kernel_ln_nomacros.s
@@ -881,7 +970,7 @@ endif
 
 $(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o ztrmm_kernel_lt.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_lt.s
 	m4 ztrmm_kernel_lt.s > ztrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_lt_nomacros.s -o $@
 	rm ztrmm_kernel_lt.s ztrmm_kernel_lt_nomacros.s
@@ -891,7 +980,7 @@ endif
 
 $(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o ztrmm_kernel_lr.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lr.s
 	m4 ztrmm_kernel_lr.s > ztrmm_kernel_lr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN ztrmm_kernel_lr_nomacros.s -o $@
 	rm ztrmm_kernel_lr.s ztrmm_kernel_lr_nomacros.s
@@ -901,7 +990,7 @@ endif
 
 $(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o ztrmm_kernel_lc.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lc.s
 	m4 ztrmm_kernel_lc.s >ztrmm_kernel_lc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ztrmm_kernel_lc_nomacros.s -o $@
 	rm ztrmm_kernel_lc.s ztrmm_kernel_lc_nomacros.s 
@@ -911,7 +1000,7 @@ endif
 
 $(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o ztrmm_kernel_rn.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rn.s
 	m4 ztrmm_kernel_rn.s > ztrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_rn_nomacros.s -o $@
 	rm ztrmm_kernel_rn.s ztrmm_kernel_rn_nomacros.s
@@ -921,7 +1010,7 @@ endif
 
 $(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o ztrmm_kernel_rt.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rt.s
 	m4 ztrmm_kernel_rt.s > ztrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_rt_nomacros.s -o $@
 	rm ztrmm_kernel_rt.s ztrmm_kernel_rt_nomacros.s
@@ -931,7 +1020,7 @@ endif
 
 $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o ztrmm_kernel_rr.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rr.s
 	m4 ztrmm_kernel_rr.s > ztrmm_kernel_rr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ztrmm_kernel_rr_nomacros.s -o $@
 	rm ztrmm_kernel_rr.s ztrmm_kernel_rr_nomacros.s
@@ -941,7 +1030,7 @@ endif
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o ztrmm_kernel_rc.s
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rc.s
 	m4 ztrmm_kernel_rc.s > ztrmm_kernel_rc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ztrmm_kernel_rc_nomacros.s -o $@
 	rm ztrmm_kernel_rc.s ztrmm_kernel_rc_nomacros.s
@@ -961,7 +1050,7 @@ $(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o strmm_kernel_rt.s	
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
@@ -1095,7 +1184,7 @@ $(KDIR)dtrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LN) $(DT
 
 $(KDIR)dtrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LT) $(DTRSMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o dtrsm_kernel_lt.s
+	$(CC) $(CFLAGS) -S -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o - > dtrsm_kernel_lt.s
 	m4 dtrsm_kernel_lt.s > dtrsm_kernel_lt_nomacros.s
 	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ dtrsm_kernel_lt_nomacros.s -o $@
 	rm dtrsm_kernel_lt.s dtrsm_kernel_lt_nomacros.s
@@ -2206,6 +2295,11 @@ $(KDIR)xtrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_
 $(KDIR)sgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMM_BETA)
 	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
+ifeq ($(BUILD_HALF),1)
+$(KDIR)shgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SHGEMM_BETA)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
 $(KDIR)dgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMM_BETA)
 	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
 
@@ -2221,6 +2315,24 @@ $(KDIR)zgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMM_BETA)
 $(KDIR)xgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMM_BETA)
 	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX $< -o $@
 
+
+ifeq ($(BUILD_HALF), 1)
+$(SHGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMONCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(SHGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMOTCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+ifneq ($(SHGEMM_UNROLL_M), $(SHGEMM_UNROLL_N))
+$(SHGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMINCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(SHGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMITCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+endif
+endif
+
 $(SGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(SGEMMONCOPY)
 	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
@@ -2325,6 +2437,12 @@ endif
 
 endif
 
+
+ifeq ($(BUILD_HALF), 1)
+$(KDIR)shgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SHGEMMKERNEL) $(SHGEMMDEPEND)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
 $(KDIR)sgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL) $(SGEMMDEPEND)
 	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
@@ -2342,7 +2460,7 @@ $(KDIR)cgemm_kernel_l$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMM
 
 $(KDIR)cgemm_kernel_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(PFLAGS) -E -UDOUBLE -DCOMPLEX -DNC $< -o cgemm_kernel_r.s
+	$(CC) $(PFLAGS) -S -UDOUBLE -DCOMPLEX -DNC $< -o - > cgemm_kernel_r.s
 	m4 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
 	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
 	rm cgemm_kernel_r.s cgemm_kernel_r_nomacros.s 
@@ -2388,7 +2506,7 @@ $(KDIR)strmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o strmm_kernel_rt.s	
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmmkernel_rn.s > strmm_kernel_rt_nomacros.s
 	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
diff --git a/kernel/arm64/KERNEL.CORTEXA53 b/kernel/arm64/KERNEL.CORTEXA53
index c1d33fa3e..eba38a92e 100644
--- a/kernel/arm64/KERNEL.CORTEXA53
+++ b/kernel/arm64/KERNEL.CORTEXA53
@@ -1,3 +1,187 @@
-include $(KERNELDIR)/KERNEL.ARMV8
+SAMINKERNEL  = ../arm/amin.c
+DAMINKERNEL  = ../arm/amin.c
+CAMINKERNEL  = ../arm/zamin.c
+ZAMINKERNEL  = ../arm/zamin.c
+
+SMAXKERNEL   = ../arm/max.c
+DMAXKERNEL   = ../arm/max.c
+
+SMINKERNEL   = ../arm/min.c
+DMINKERNEL   = ../arm/min.c
+
+ISAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = ../arm/iamin.c
+ICAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = ../arm/izamin.c
+
+ISMAXKERNEL  = ../arm/imax.c
+IDMAXKERNEL  = ../arm/imax.c
+
+ISMINKERNEL  = ../arm/imin.c
+IDMINKERNEL  = ../arm/imin.c
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+SAMAXKERNEL  = amax.S
+DAMAXKERNEL  = amax.S
+CAMAXKERNEL  = zamax.S
+ZAMAXKERNEL  = zamax.S
+
+SAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = axpy.S
+CAXPYKERNEL  = zaxpy.S
+ZAXPYKERNEL  = zaxpy.S
+
+SROTKERNEL   = rot.S
+DROTKERNEL   = rot.S
+CROTKERNEL   = zrot.S
+ZROTKERNEL   = zrot.S
+
+SSCALKERNEL  = scal.S
+DSCALKERNEL  = scal.S
+CSCALKERNEL  = zscal.S
+ZSCALKERNEL  = zscal.S
+
+SGEMVNKERNEL = gemv_n.S
+DGEMVNKERNEL = gemv_n.S
+CGEMVNKERNEL = zgemv_n.S
+ZGEMVNKERNEL = zgemv_n.S
+
+SGEMVTKERNEL = gemv_t.S
+DGEMVTKERNEL = gemv_t.S
+CGEMVTKERNEL = zgemv_t.S
+ZGEMVTKERNEL = zgemv_t.S
 
 
+SASUMKERNEL    = asum.S
+DASUMKERNEL    = asum.S
+CASUMKERNEL    = casum.S
+ZASUMKERNEL    = zasum.S
+
+SCOPYKERNEL    = copy.S
+DCOPYKERNEL    = copy.S
+CCOPYKERNEL    = copy.S
+ZCOPYKERNEL    = copy.S
+
+SSWAPKERNEL    = swap.S
+DSWAPKERNEL    = swap.S
+CSWAPKERNEL    = swap.S
+ZSWAPKERNEL    = swap.S
+
+ISAMAXKERNEL   = iamax.S
+IDAMAXKERNEL   = iamax.S
+ICAMAXKERNEL   = izamax.S
+IZAMAXKERNEL   = izamax.S
+
+SNRM2KERNEL    = nrm2.S
+DNRM2KERNEL    = nrm2.S
+CNRM2KERNEL    = znrm2.S
+ZNRM2KERNEL    = znrm2.S
+
+DDOTKERNEL     = dot.S
+SDOTKERNEL     = dot.S
+CDOTKERNEL     = zdot.S
+ZDOTKERNEL     = zdot.S
+DSDOTKERNEL    = dot.S
+
+DGEMM_BETA     = dgemm_beta.S
+SGEMM_BETA     = sgemm_beta.S
+
+ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 8x8)
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+else
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+endif
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+ifeq ($(SGEMM_UNROLL_M), 16)
+SGEMMITCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
+endif
+ifeq ($(SGEMM_UNROLL_M), 4)
+SGEMMINCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+endif
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+SGEMMOTCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+
+ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
+
+ifeq ($(DGEMM_UNROLL_M), 8)
+DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
+DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
+else
+DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
+endif
+
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifeq ($(DGEMM_UNROLL_N), 4)
+DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
+DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
+else
+DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
+endif
+
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    =  cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
diff --git a/kernel/arm64/daxpy_thunderx2t99.S b/kernel/arm64/daxpy_thunderx2t99.S
index b8d0af5c2..baf39150f 100644
--- a/kernel/arm64/daxpy_thunderx2t99.S
+++ b/kernel/arm64/daxpy_thunderx2t99.S
@@ -98,11 +98,58 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	X, X, #128
 .endm
 
+/*
+ * No need to do software prefetches if the vector fits
+ * into L1 cache
+ */
+.macro KERNEL_F16_L1CACHE
+	ldp	q4, q5, [X]
+	ldp	q16, q17, [Y]
+
+	ldp	q6, q7, [X, #32]
+	ldp	q18, q19, [Y, #32]
+
+	fmla	v16.2d, v4.2d, v0.d[0]
+	fmla	v17.2d, v5.2d, v0.d[0]
+
+	stp	q16, q17, [Y]
+
+	ldp	q20, q21, [X, #64]
+	ldp	q24, q25, [Y, #64]
+
+	fmla	v18.2d, v6.2d, v0.d[0]
+	fmla	v19.2d, v7.2d, v0.d[0]
+
+	stp	q18, q19, [Y, #32]
+
+	ldp	q22, q23, [X, #96]
+	ldp	q26, q27, [Y, #96]
+
+	fmla	v24.2d, v20.2d, v0.d[0]
+	fmla	v25.2d, v21.2d, v0.d[0]
+
+	stp	q24, q25, [Y, #64]
+
+	fmla	v26.2d, v22.2d, v0.d[0]
+	fmla	v27.2d, v23.2d, v0.d[0]
+
+	stp	q26, q27, [Y, #96]
+
+	add	Y, Y, #128
+	add	X, X, #128
+.endm
+
 .macro KERNEL_F32
 	KERNEL_F16 
 	KERNEL_F16 
 .endm
 
+
+.macro KERNEL_F32_L1CACHE
+	KERNEL_F16_L1CACHE
+	KERNEL_F16_L1CACHE
+.endm
+
 .macro INIT_S
 	lsl	INC_X, INC_X, #3
 	lsl	INC_Y, INC_Y, #3
@@ -138,6 +185,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	I, xzr
 	beq	.Ldaxpy_kernel_F1
 
+	cmp	N, #2048
+	ble	.Ldaxpy_kernel_F32_L1CACHE
+
 	.align 5
 .Ldaxpy_kernel_F32:
 
@@ -145,6 +195,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	subs	I, I, #1
 	bne	.Ldaxpy_kernel_F32
+	b	.Ldaxpy_kernel_F1
+
+	.align 5
+.Ldaxpy_kernel_F32_L1CACHE:
+
+	KERNEL_F32_L1CACHE
+
+	subs	I, I, #1
+	bne	.Ldaxpy_kernel_F32_L1CACHE
 
 .Ldaxpy_kernel_F1:
 
diff --git a/kernel/arm64/sgemm_kernel_8x8_cortexa53.S b/kernel/arm64/sgemm_kernel_8x8_cortexa53.S
new file mode 100644
index 000000000..628a928ca
--- /dev/null
+++ b/kernel/arm64/sgemm_kernel_8x8_cortexa53.S
@@ -0,0 +1,2299 @@
+/*******************************************************************************
+Copyright (c) 2015, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+/*                   X0          X1          X2          s0        X3        x4       x5           x6  */
+/*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc) */
+
+#define origM		x0
+#define origN		x1
+#define origK		x2
+#define origPA		x3
+#define origPB		x4
+#define pC		x5
+#define LDC		x6
+#define offset		x7
+#define counterL	x8
+#define counterI	x9
+#define counterJ	x10
+#define pB		x11
+#define pCRow0		x12
+#define pCRow1		x13
+#define pCRow2		x14
+#define pA		x15
+#define temp		x16
+
+#define alpha0		s10
+#define alphaV0		v10.s[0]
+#define alpha1		s11
+#define alphaV1		v11.s[0]
+#define alpha2		s14
+#define alphaV2		v14.s[0]
+#define alpha3		s15
+#define alphaV3		v15.s[0]
+
+#define A_PRE_SIZE	640
+#define B_PRE_SIZE	224
+#define C_PRE_SIZE	96
+
+// 00 origM
+// 01 origN
+// 02 origK
+// 03 origPA
+// 04 origPB
+// 05 pC
+// 06 origLDC -> LDC
+// 07 offset
+// 08 counterL
+// 09 counterI
+// 10 counterJ
+// 11 pB
+// 12 pCRow0
+// 13 pCRow1
+// 14 pCRow2
+// 15 pA
+// 16 temp
+// 17
+// 18 must save
+// 19 must save
+// 20 must save pA0_2, pA0_3
+// 21 must save pA0_6, pA0_7
+// 22 must save pA1_2, pA1_3
+// 23 must save pA1_6, pA1_7
+// 24 must save pB0_2, pB0_3
+// 25 must save pB0_6, pB0_7
+// 26 must save pB1_2, pB1_3
+// 27 must save pB1_6, pB1_7
+// 28 must save
+// 29 frame
+// 30 link
+// 31 sp
+
+//v00 ALPHA -> pA0_0, pA0_1, pA0_2, pA0_3
+//v01 pA0_4, pA0_5, pA0_6, pA0_7
+//v02 pA1_0, pA1_1, pA1_2, pA1_3
+//v03 pA1_4, pA1_5, pA1_6, pA1_7
+//v04 pB0_0, pB0_1, pB0_2, pB0_3
+//v05 pB0_4, pB0_5, pB0_6, pB0_7
+//v06 pB1_0, pB1_1, pB1_2, pB1_3
+//v07 pB1_4, pB1_5, pB1_6, pB1_7
+//v08 must save
+//v09 must save
+//v10 must save ALPHA0
+//v11 must save ALPHA1
+//v12 must save
+//v13 must save
+//v14 must save ALPHA2
+//v15 must save ALPHA3
+//v16 must save C00, C01, C02, C03
+//v17 must save C04, C05, C06, C07
+//v18 C08, C09, C10, C11
+//v19 C12, C13, C14, C15
+//v20 C16, C17, C18, C19
+//v21 C20, C21, C22, C23
+//v22 C24, C25, C26, C27
+//v23 C28, C29, C30, C31
+//v24 C32, C33, C34, C35
+//v25 C36, C37, C38, C39
+//v26 C40, C41, C42, C43
+//v27 C44, C45, C46, C47
+//v28 C48, C49, C50, C51
+//v29 C52, C53, C54, C55
+//v30 C56, C57, C58, C59
+//v31 C60, C61, C62, C63
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+.macro INIT8x8
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s18, s16
+	fmov		s19, s17
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s22, s17
+	fmov		s23, s18
+	fmov		s24, wzr
+	fmov		s25, s16
+	fmov		s26, s17
+	fmov		s27, s18
+	fmov		s28, wzr
+	fmov		s29, s16
+	fmov		s30, s17
+	fmov		s31, s18
+.endm
+
+.macro KERNEL8x8_I
+	ldp	q0, q1, [pA], #32
+	ldp	q4, q5, [pB], #32
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d3, [pA, #8]
+	ldr	d7, [pB, #8]
+	ldr	x22, [pA], #16
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v17.4s, v1.4s, v4.s[0]
+	ldr	x23, [pA], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v19.4s, v1.4s, v4.s[1]
+	fmul	v20.4s, v0.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmul	v21.4s, v1.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v23.4s, v1.4s, v4.s[3]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v25.4s, v1.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v27.4s, v1.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v29.4s, v1.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+	fmul	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #16
+	fmla	v17.4s, v1.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x23, [pA], #8
+	fmla	v19.4s, v1.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmla	v20.4s, v0.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x20, [pA], #16
+	fmla	v17.4s, v3.4s, v6.s[0]
+	ldr	x24, [pB], #16
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x21, [pA], #8
+	fmla	v19.4s, v3.4s, v6.s[1]
+	ldr	x25, [pB], #8
+	fmla	v20.4s, v2.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v3.d[1], x23
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v17.4s, v3.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_SUB
+	ldp	q0, q1, [pA], #32
+	ldp	q4, q5, [pB], #32
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro SAVE8x8
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+	add	pCRow1, pCRow0, LDC
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q2, q3, [pCRow1]
+	fmla	v2.4s, v18.4s, alphaV2
+	fmla	v3.4s, v19.4s, alphaV3
+	stp	q2, q3, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q4, q5, [pCRow2]
+	fmla	v4.4s, v20.4s, alphaV0
+	fmla	v5.4s, v21.4s, alphaV1
+	stp	q4, q5, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q6, q7, [pCRow1]
+	fmla	v6.4s, v22.4s, alphaV2
+	fmla	v7.4s, v23.4s, alphaV3
+	stp	q6, q7, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q0, q1, [pCRow2]
+	fmla	v0.4s, v24.4s, alphaV0
+	fmla	v1.4s, v25.4s, alphaV1
+	stp	q0, q1, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q2, q3, [pCRow1]
+	fmla	v2.4s, v26.4s, alphaV2
+	fmla	v3.4s, v27.4s, alphaV3
+	stp	q2, q3, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q4, q5, [pCRow2]
+	fmla	v4.4s, v28.4s, alphaV0
+	fmla	v5.4s, v29.4s, alphaV1
+	stp	q4, q5, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+
+	ldp	q6, q7, [pCRow1]
+	fmla	v6.4s, v30.4s, alphaV2
+	fmla	v7.4s, v31.4s, alphaV3
+	stp	q6, q7, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL4x8_I
+	ldr	q0, [pA], #16
+	ldp	q4, q5, [pB], #32
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d7, [pB, #8]
+	ldr	x22, [pA], #8
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	ldr	x22, [pA], #8
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	ldr	x20, [pA], #8
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x24, [pB], #16
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x25, [pB], #8
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_SUB
+	ldr	q0, [pA], #16
+	ldp	q4, q5, [pB], #32
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro SAVE4x8
+	add	pCRow1, pCRow0, LDC
+
+	ldr	q0, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	str	q0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	q2, [pCRow1]
+	fmla	v2.4s, v18.4s, alphaV2
+	str	q2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	q4, [pCRow2]
+	fmla	v4.4s, v20.4s, alphaV0
+	str	q4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	q6, [pCRow1]
+	fmla	v6.4s, v22.4s, alphaV2
+	str	q6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	q0, [pCRow2]
+	fmla	v0.4s, v24.4s, alphaV0
+	str	q0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	q2, [pCRow1]
+	fmla	v2.4s, v26.4s, alphaV2
+	str	q2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	q4, [pCRow2]
+	fmla	v4.4s, v28.4s, alphaV0
+	str	q4, [pCRow2]
+
+	ldr	q6, [pCRow1]
+	fmla	v6.4s, v30.4s, alphaV2
+	str	q6, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL2x8_SUB
+	ldr	d0, [pA], #8
+	ldp	q4, q5, [pB], #32
+
+	fmla	v16.2s, v0.2s, v4.s[0]
+	fmla	v18.2s, v0.2s, v4.s[1]
+	fmla	v20.2s, v0.2s, v4.s[2]
+	fmla	v22.2s, v0.2s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.2s, v0.2s, v5.s[0]
+	fmla	v26.2s, v0.2s, v5.s[1]
+	fmla	v28.2s, v0.2s, v5.s[2]
+	fmla	v30.2s, v0.2s, v5.s[3]
+.endm
+
+.macro SAVE2x8
+	add	pCRow1, pCRow0, LDC
+
+	ldr	d0, [pCRow0]
+	fmla	v0.2s, v16.2s, alphaV0
+	str	d0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	d2, [pCRow1]
+	fmla	v2.2s, v18.2s, alphaV2
+	str	d2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	d4, [pCRow2]
+	fmla	v4.2s, v20.2s, alphaV0
+	str	d4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	d6, [pCRow1]
+	fmla	v6.2s, v22.2s, alphaV2
+	str	d6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	d0, [pCRow2]
+	fmla	v0.2s, v24.2s, alphaV0
+	str	d0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	d2, [pCRow1]
+	fmla	v2.2s, v26.2s, alphaV2
+	str	d2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	d4, [pCRow2]
+	fmla	v4.2s, v28.2s, alphaV0
+	str	d4, [pCRow2]
+
+	ldr	d6, [pCRow1]
+	fmla	v6.2s, v30.2s, alphaV2
+	str	d6, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL1x8_SUB
+	ldp	q4, q5, [pB], #32
+	ldr	s0, [pA], #4
+
+	fmla	s16, s0, v4.s[0]
+	fmla	s18, s0, v4.s[1]
+	fmla	s20, s0, v4.s[2]
+	fmla	s22, s0, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	s24, s0, v5.s[0]
+	fmla	s26, s0, v5.s[1]
+	fmla	s28, s0, v5.s[2]
+	fmla	s30, s0, v5.s[3]
+.endm
+
+.macro SAVE1x8
+	add	pCRow1, pCRow0, LDC
+
+	ldr	s0, [pCRow0]
+	fmla	s0, s16, alphaV0
+	str	s0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	s2, [pCRow1]
+	fmla	s2, s18, alphaV2
+	str	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	s4, [pCRow2]
+	fmla	s4, s20, alphaV0
+	str	s4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	s6, [pCRow1]
+	fmla	s6, s22, alphaV2
+	str	s6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	s0, [pCRow2]
+	fmla	s0, s24, alphaV0
+	str	s0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	s2, [pCRow1]
+	fmla	s2, s26, alphaV2
+	str	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	s4, [pCRow2]
+	fmla	s4, s28, alphaV0
+	str	s4, [pCRow2]
+
+	ldr	s6, [pCRow1]
+	fmla	s6, s30, alphaV2
+	str	s6, [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x4
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s18, wzr
+	fmov		s19, s16
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s22, wzr
+	fmov		s23, s16
+.endm
+
+.macro KERNEL8x4_I
+	ldp	q0, q1, [pA], #32
+	ldr	q4, [pB], #16
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d3, [pA, #8]
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #16
+	fmul	v17.4s, v1.4s, v4.s[0]
+	ldr	x26, [pB], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x23, [pA], #8
+	fmul	v19.4s, v1.4s, v4.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v21.4s, v1.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v23.4s, v1.4s, v4.s[3]
+.endm
+
+.macro KERNEL8x4_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	ldr	x22, [pA], #16
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #8
+	fmla	v17.4s, v1.4s, v4.s[0]
+	ldr	x23, [pA], #8
+	fmla	v18.4s, v0.4s, v4.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+.endm
+
+.macro KERNEL8x4_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	ldr	x20, [pA], #16
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x24, [pB], #8
+	fmla	v17.4s, v3.4s, v6.s[0]
+	ldr	x21, [pA], #8
+	fmla	v18.4s, v2.4s, v6.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+.endm
+
+.macro KERNEL8x4_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v3.d[1], x23
+	fmla	v16.4s, v2.4s, v6.s[0]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v17.4s, v3.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+.endm
+
+.macro KERNEL8x4_SUB
+	ldp	q0, q1, [pA], #32
+	ldr	q4, [pB], #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+.endm
+
+.macro SAVE8x4
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+	add	pCRow1, pCRow0, LDC
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q4, q5, [pCRow1]
+	fmla	v4.4s, v18.4s, alphaV0
+	fmla	v5.4s, v19.4s, alphaV1
+	stp	q4, q5, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q0, q1, [pCRow2]
+	fmla	v0.4s, v20.4s, alphaV0
+	fmla	v1.4s, v21.4s, alphaV1
+	stp	q0, q1, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+
+	ldp	q4, q5, [pCRow1]
+	fmla	v4.4s, v22.4s, alphaV0
+	fmla	v5.4s, v23.4s, alphaV1
+	stp	q4, q5, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+
+.macro INIT4x4
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, wzr
+.endm
+
+.macro KERNEL4x4_I
+	ldr	q0, [pA], #16
+	ldr	q4, [pB], #16
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x26, [pB], #8
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+.endm
+
+.macro KERNEL4x4_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	x22, [pA], #8
+	ldr	x26, [pB], #8
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+.endm
+
+.macro KERNEL4x4_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	x20, [pA], #8
+	ldr	x24, [pB], #8
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+.endm
+
+.macro KERNEL4x4_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+.endm
+
+.macro KERNEL4x4_SUB
+	ldr	q0, [pA], #16
+	ldr	q4, [pB], #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+.endm
+
+.macro SAVE4x4
+	ldr	q0, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	str	q0, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+	ldr	q1, [pCRow1]
+	fmla	v1.4s, v18.4s, alphaV2
+	str	q1, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	ldr	q2, [pCRow2]
+	fmla	v2.4s, v20.4s, alphaV0
+	str	q2, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+	ldr	q3, [pCRow1]
+	fmla	v3.4s, v22.4s, alphaV2
+	str	q3, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x4
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+.endm
+
+.macro KERNEL2x4_SUB
+	ldr	d0, [pA], #8
+	ldr	q4, [pB], #16
+
+	fmla	v16.2s, v0.2s, v4.s[0]
+	fmla	v18.2s, v0.2s, v4.s[1]
+	fmla	v20.2s, v0.2s, v4.s[2]
+	fmla	v22.2s, v0.2s, v4.s[3]
+.endm
+
+.macro SAVE2x4
+	ldr	d8, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	str	d8, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+	ldr	d12, [pCRow1]
+	fmla	v12.2s, v18.2s, alphaV1
+	str	d12, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	ldr	d8, [pCRow2]
+	fmla	v8.2s, v20.2s, alphaV2
+	str	d8, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+	ldr	d12, [pCRow1]
+	fmla	v12.2s, v22.2s, alphaV3
+	str	d12, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x4
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL1x4_SUB
+	ldr	s0, [pA]
+	add	pA, pA, #4
+
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+	fmla	v20.2s, v9.2s, v0.s[0]
+.endm
+
+.macro SAVE1x4
+	add	pCRow1, pCRow0, LDC
+	ld1	{v8.s}[0], [pCRow0]
+	ld1	{v8.s}[1], [pCRow1]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	add	pCRow1, pCRow2, LDC
+	ld1	{v12.s}[0], [pCRow2]
+	ld1	{v12.s}[1], [pCRow1]
+	fmla	v12.2s, v20.2s, alphaV1
+	st1	{v12.s}[0], [pCRow2]
+	st1	{v12.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s18, s17
+	fmov	s19, s16
+.endm
+
+.macro KERNEL8x2_SUB
+	ldp	q0, q1, [pA], #32
+	ldr	d4, [pB], #8
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+.endm
+
+.macro SAVE8x2
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+	add	pCRow1, pCRow0, LDC
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q4, q5, [pCRow1]
+	fmla	v4.4s, v18.4s, alphaV0
+	fmla	v5.4s, v19.4s, alphaV1
+	stp	q4, q5, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s20, s17
+	fmov	s21, s16
+.endm
+
+.macro KERNEL4x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v21.2s, v1.2s, v8.s[1]
+.endm
+
+.macro SAVE4x2
+	ld1	{v8.2s, v9.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	fmla	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+	ld1	{v12.2s, v13.2s}, [pCRow1]
+	fmla	v12.2s, v20.2s, alphaV2
+	fmla	v13.2s, v21.2s, alphaV3
+	st1	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x2
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL2x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+.endm
+
+.macro SAVE2x2
+	ld1	{v8.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow1 , pCRow0, LDC
+	ld1	{v12.2s}, [pCRow1]
+	fmla	v12.2s, v20.2s, alphaV1
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x2
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL1x2_SUB
+	ld1	{v8.2s} , [pB]
+	add	pB , pB, #8
+
+	ldr	s0 , [pA]
+	add	pA, pA, #4
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+.endm
+
+.macro SAVE1x2
+	add	pCRow1 , pCRow0, LDC
+	ld1	{v8.s}[0], [pCRow0]
+	ld1	{v8.s}[1], [pCRow1]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x1
+	fmov	s16, wzr
+	fmov	s17, wzr
+.endm
+
+.macro KERNEL8x1_SUB
+	ldr	s4, [pB], #4
+	ldp	q0, q1, [pA], #32
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v17.4s, v1.4s, v4.s[0]
+.endm
+
+.macro SAVE8x1
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x1
+	fmov	s16, wzr
+	fmov	s17, s16
+.endm
+
+.macro KERNEL4x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA , pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro SAVE4x1
+	ld1	{v8.2s, v9.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	fmla	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x1
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL2x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s}, [pA]
+	add	pA , pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+.endm
+
+.macro SAVE2x1
+	ld1	{v8.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x1
+	fmov	s16, wzr
+.endm
+
+.macro KERNEL1x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ldr	s0, [pA]
+	add	pA , pA, #4
+
+	fmadd	s16, s0, s8, s16
+.endm
+
+.macro SAVE1x1
+	ldr	s8, [pCRow0]
+	fmla	s8, s16, alphaV0
+	str	s8, [pCRow0]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+.Lsgemm_kernel_begin:
+
+	.align 5
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+
+	fmov	alpha0, s0
+	fmov	alpha1, s0
+	fmov	alpha2, s0
+	fmov	alpha3, s0
+
+	lsl	LDC, LDC, #2			// ldc = ldc * 4
+
+	mov	pB, origPB
+
+	mov	counterJ, origN
+	asr	counterJ, counterJ, #3		// J = J / 8
+	cmp	counterJ, #0
+	ble	.Lsgemm_kernel_L4_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_BEGIN:
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC, pC, LDC, lsl #3
+
+	mov	pA, origPA			// pA = start of A array
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI, #0
+	ble	.Lsgemm_kernel_L8_M4_BEGIN
+
+.Lsgemm_kernel_L8_M8_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// L = K / 8
+	cmp	counterL , #2			// is there at least 16 to do?
+	blt	.Lsgemm_kernel_L8_M8_32
+
+	KERNEL8x8_I				// do one in the K
+	KERNEL8x8_M2				// do another in the K
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L8_M8_22a
+	.align 5
+
+.Lsgemm_kernel_L8_M8_22:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M8_22
+
+.Lsgemm_kernel_L8_M8_22a:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	 .Lsgemm_kernel_L8_M8_44
+
+.Lsgemm_kernel_L8_M8_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L8_M8_40
+
+	KERNEL8x8_I
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	.Lsgemm_kernel_L8_M8_44
+
+.Lsgemm_kernel_L8_M8_40:
+
+	INIT8x8
+
+.Lsgemm_kernel_L8_M8_44:
+
+	ands	counterL , origK, #7
+	ble	.Lsgemm_kernel_L8_M8_100
+
+.Lsgemm_kernel_L8_M8_46:
+
+	KERNEL8x8_SUB
+
+	subs	counterL, counterL, 1
+	bgt	.Lsgemm_kernel_L8_M8_46
+
+.Lsgemm_kernel_L8_M8_100:
+
+	SAVE8x8
+
+.Lsgemm_kernel_L8_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lsgemm_kernel_L8_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L8_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L8_M2_BEGIN
+
+.Lsgemm_kernel_L8_M4_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lsgemm_kernel_L8_M4_32
+
+	KERNEL4x8_I				// do one in the K
+	KERNEL4x8_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L8_M4_22a
+	.align 5
+
+.Lsgemm_kernel_L8_M4_22:
+
+	KERNEL4x8_M1
+	KERNEL4x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M4_22
+
+.Lsgemm_kernel_L8_M4_22a:
+
+	KERNEL4x8_M1
+	KERNEL4x8_E
+
+	b	 .Lsgemm_kernel_L8_M4_44
+
+.Lsgemm_kernel_L8_M4_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L8_M4_40
+
+	KERNEL4x8_I
+	KERNEL4x8_E
+
+	b	.Lsgemm_kernel_L8_M4_44
+
+.Lsgemm_kernel_L8_M4_40:
+
+	INIT4x8
+
+.Lsgemm_kernel_L8_M4_44:
+
+	ands	counterL , origK, #1
+	ble	.Lsgemm_kernel_L8_M4_100
+
+.Lsgemm_kernel_L8_M4_46:
+
+	KERNEL4x8_SUB
+
+.Lsgemm_kernel_L8_M4_100:
+
+	SAVE4x8
+
+.Lsgemm_kernel_L8_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L8_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L8_M1_BEGIN
+
+.Lsgemm_kernel_L8_M2_20:
+
+	INIT2x8
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L8_M2_40
+
+.Lsgemm_kernel_L8_M2_22:
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M2_22
+
+
+.Lsgemm_kernel_L8_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L8_M2_100
+
+.Lsgemm_kernel_L8_M2_42:
+
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M2_42
+
+.Lsgemm_kernel_L8_M2_100:
+
+	SAVE2x8
+
+.Lsgemm_kernel_L8_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L8_END
+
+.Lsgemm_kernel_L8_M1_20:
+
+	INIT1x8
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L8_M1_40
+
+.Lsgemm_kernel_L8_M1_22:
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M1_22
+
+
+.Lsgemm_kernel_L8_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L8_M1_100
+
+.Lsgemm_kernel_L8_M1_42:
+
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M1_42
+
+.Lsgemm_kernel_L8_M1_100:
+
+	SAVE1x8
+
+.Lsgemm_kernel_L8_END:
+	lsl	temp, origK, #5			// B = B + K * 4 * 8
+	add	origPB, origPB, temp
+
+	subs	counterJ, counterJ , #1		// j--
+	bgt	.Lsgemm_kernel_L8_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #7
+	ble	.Lsgemm_kernel_L999
+
+	tst	counterJ , #4
+	ble	.Lsgemm_kernel_L2_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #2
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI, #0
+	ble	.Lsgemm_kernel_L4_M4_BEGIN
+
+.Lsgemm_kernel_L4_M8_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lsgemm_kernel_L4_M8_32
+
+	KERNEL8x4_I				// do one in the K
+	KERNEL8x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L4_M8_22a
+	.align 5
+
+.Lsgemm_kernel_L4_M8_22:
+
+	KERNEL8x4_M1
+	KERNEL8x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M8_22
+
+.Lsgemm_kernel_L4_M8_22a:
+
+	KERNEL8x4_M1
+	KERNEL8x4_E
+
+	b	 .Lsgemm_kernel_L4_M8_44
+
+.Lsgemm_kernel_L4_M8_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L4_M8_40
+
+	KERNEL8x4_I
+	KERNEL8x4_E
+
+	b	.Lsgemm_kernel_L4_M8_44
+
+.Lsgemm_kernel_L4_M8_40:
+
+	INIT8x4
+
+.Lsgemm_kernel_L4_M8_44:
+
+	ands	counterL , origK, #1
+	ble	.Lsgemm_kernel_L4_M8_100
+
+.Lsgemm_kernel_L4_M8_46:
+
+	KERNEL8x4_SUB
+
+.Lsgemm_kernel_L4_M8_100:
+
+	SAVE8x4
+
+.Lsgemm_kernel_L4_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lsgemm_kernel_L4_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L4_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L4_M2_BEGIN
+
+.Lsgemm_kernel_L4_M4_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lsgemm_kernel_L4_M4_32
+
+	KERNEL4x4_I				// do one in the K
+	KERNEL4x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L4_M4_22a
+	.align 5
+
+.Lsgemm_kernel_L4_M4_22:
+
+	KERNEL4x4_M1
+	KERNEL4x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M4_22
+
+.Lsgemm_kernel_L4_M4_22a:
+
+	KERNEL4x4_M1
+	KERNEL4x4_E
+
+	b	 .Lsgemm_kernel_L4_M4_44
+
+.Lsgemm_kernel_L4_M4_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L4_M4_40
+
+	KERNEL4x4_I
+	KERNEL4x4_E
+
+	b	.Lsgemm_kernel_L4_M4_44
+
+.Lsgemm_kernel_L4_M4_40:
+
+	INIT4x4
+
+.Lsgemm_kernel_L4_M4_44:
+
+	ands	counterL , origK, #1
+	ble	.Lsgemm_kernel_L4_M4_100
+
+.Lsgemm_kernel_L4_M4_46:
+
+	KERNEL4x4_SUB
+
+.Lsgemm_kernel_L4_M4_100:
+
+	SAVE4x4
+
+.Lsgemm_kernel_L4_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L4_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L4_M1_BEGIN
+
+.Lsgemm_kernel_L4_M2_20:
+
+	INIT2x4
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L4_M2_40
+
+.Lsgemm_kernel_L4_M2_22:
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M2_22
+
+
+.Lsgemm_kernel_L4_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L4_M2_100
+
+.Lsgemm_kernel_L4_M2_42:
+
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M2_42
+
+.Lsgemm_kernel_L4_M2_100:
+
+	SAVE2x4
+
+.Lsgemm_kernel_L4_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L4_END
+
+.Lsgemm_kernel_L4_M1_20:
+
+	INIT1x4
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L4_M1_40
+
+.Lsgemm_kernel_L4_M1_22:
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M1_22
+
+
+.Lsgemm_kernel_L4_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L4_M1_100
+
+.Lsgemm_kernel_L4_M1_42:
+
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M1_42
+
+.Lsgemm_kernel_L4_M1_100:
+
+	SAVE1x4
+
+.Lsgemm_kernel_L4_END:
+	add	origPB, origPB, origK, lsl #4	// B = B + K * 4 * 4
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_BEGIN:   // less than 2 left in N direction
+
+	mov	counterJ , origN
+	tst	counterJ , #3
+	ble	.Lsgemm_kernel_L999
+
+	tst	counterJ , #2
+	ble	.Lsgemm_kernel_L1_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #1
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI,#0
+	ble	.Lsgemm_kernel_L2_M4_BEGIN
+
+.Lsgemm_kernel_L2_M8_20:
+
+	INIT8x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lsgemm_kernel_L2_M8_40
+	.align 5
+
+.Lsgemm_kernel_L2_M8_22:
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M8_22
+
+
+.Lsgemm_kernel_L2_M8_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M8_100
+
+.Lsgemm_kernel_L2_M8_42:
+
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M8_42
+
+.Lsgemm_kernel_L2_M8_100:
+
+	SAVE8x2
+
+.Lsgemm_kernel_L2_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lsgemm_kernel_L2_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L2_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L2_M2_BEGIN
+
+.Lsgemm_kernel_L2_M4_20:
+
+	INIT4x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lsgemm_kernel_L2_M4_40
+	.align 5
+
+.Lsgemm_kernel_L2_M4_22:
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M4_22
+
+
+.Lsgemm_kernel_L2_M4_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M4_100
+
+.Lsgemm_kernel_L2_M4_42:
+
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M4_42
+
+.Lsgemm_kernel_L2_M4_100:
+
+	SAVE4x2
+
+.Lsgemm_kernel_L2_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L2_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L2_M1_BEGIN
+
+.Lsgemm_kernel_L2_M2_20:
+
+	INIT2x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+        cmp	counterL,#0
+	ble	.Lsgemm_kernel_L2_M2_40
+
+.Lsgemm_kernel_L2_M2_22:
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M2_22
+
+
+.Lsgemm_kernel_L2_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M2_100
+
+.Lsgemm_kernel_L2_M2_42:
+
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M2_42
+
+.Lsgemm_kernel_L2_M2_100:
+
+	SAVE2x2
+
+.Lsgemm_kernel_L2_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L2_END
+
+.Lsgemm_kernel_L2_M1_20:
+
+	INIT1x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+        cmp     counterL, #0
+	ble	.Lsgemm_kernel_L2_M1_40
+
+.Lsgemm_kernel_L2_M1_22:
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M1_22
+
+
+.Lsgemm_kernel_L2_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M1_100
+
+.Lsgemm_kernel_L2_M1_42:
+
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M1_42
+
+.Lsgemm_kernel_L2_M1_100:
+
+	SAVE1x2
+
+.Lsgemm_kernel_L2_END:
+
+	add	origPB, origPB, origK, lsl #3	// B = B + K * 2 * 4
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #1
+	ble	.Lsgemm_kernel_L999 // done
+
+
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC , pC , LDC			// Update pC to point to next
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3
+	cmp	counterI, #0
+	ble	.Lsgemm_kernel_L1_M4_BEGIN
+
+.Lsgemm_kernel_L1_M8_20:
+
+	INIT8x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M8_40
+	.align 5
+
+.Lsgemm_kernel_L1_M8_22:
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M8_22
+
+
+.Lsgemm_kernel_L1_M8_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M8_100
+
+.Lsgemm_kernel_L1_M8_42:
+
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M8_42
+
+.Lsgemm_kernel_L1_M8_100:
+
+	SAVE8x1
+
+.Lsgemm_kernel_L1_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lsgemm_kernel_L1_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L1_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L1_M2_BEGIN
+
+.Lsgemm_kernel_L1_M4_20:
+
+	INIT4x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M4_40
+	.align 5
+
+.Lsgemm_kernel_L1_M4_22:
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M4_22
+
+
+.Lsgemm_kernel_L1_M4_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M4_100
+
+.Lsgemm_kernel_L1_M4_42:
+
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M4_42
+
+.Lsgemm_kernel_L1_M4_100:
+
+	SAVE4x1
+
+.Lsgemm_kernel_L1_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L1_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L1_M1_BEGIN
+
+.Lsgemm_kernel_L1_M2_20:
+
+	INIT2x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M2_40
+
+.Lsgemm_kernel_L1_M2_22:
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M2_22
+
+
+.Lsgemm_kernel_L1_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M2_100
+
+.Lsgemm_kernel_L1_M2_42:
+
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M2_42
+
+.Lsgemm_kernel_L1_M2_100:
+
+	SAVE2x1
+
+.Lsgemm_kernel_L1_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L1_END
+
+.Lsgemm_kernel_L1_M1_20:
+
+	INIT1x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M1_40
+
+.Lsgemm_kernel_L1_M1_22:
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M1_22
+
+
+.Lsgemm_kernel_L1_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M1_100
+
+.Lsgemm_kernel_L1_M1_42:
+
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M1_42
+
+.Lsgemm_kernel_L1_M1_100:
+
+	SAVE1x1
+
+.Lsgemm_kernel_L1_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L999:
+	mov	x0, #0				// set return value
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+	ret
+
+	EPILOGUE
+
diff --git a/kernel/arm64/sgemm_ncopy_8.S b/kernel/arm64/sgemm_ncopy_8.S
new file mode 100644
index 000000000..f99b1d992
--- /dev/null
+++ b/kernel/arm64/sgemm_ncopy_8.S
@@ -0,0 +1,562 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+#define	M	x0
+#define	N	x1
+#define	A00	x2
+#define	LDA	x3
+#define	B00	x4
+
+#define	A01	x5
+#define	A02	x6
+#define	A03	x7
+#define	A04	x8
+#define	A05	x9
+#define	A06	x10
+#define	A07	x11
+#define	A08	x12
+
+#define I	x13
+#define	J	x14
+#define	K	x15
+
+#define	TEMP1	x16
+#define	TEMP2	x17
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+
+.macro SAVE_REGS
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+.endm
+
+.macro RESTORE_REGS
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+.endm
+
+.macro COPY4x8
+	ldr	q0, [A01], #16
+	ldr	q1, [A02], #16
+	ins	v8.s[0], v0.s[0]
+	ins	v10.s[0], v0.s[1]
+	ins	v12.s[0], v0.s[2]
+	ins	v14.s[0], v0.s[3]
+	ins	v8.s[1], v1.s[0]
+	ins	v10.s[1], v1.s[1]
+	ins	v12.s[1], v1.s[2]
+	ins	v14.s[1], v1.s[3]
+
+	ldr	q2, [A03], #16
+	ldr	q3, [A04], #16
+	ins	v8.s[2], v2.s[0]
+	ins	v10.s[2], v2.s[1]
+	ins	v12.s[2], v2.s[2]
+	ins	v14.s[2], v2.s[3]
+	ins	v8.s[3], v3.s[0]
+	ins	v10.s[3], v3.s[1]
+	ins	v12.s[3], v3.s[2]
+	ins	v14.s[3], v3.s[3]
+
+	ldr	q4, [A05], #16
+	ldr	q5, [A06], #16
+	ins	v9.s[0], v4.s[0]
+	ins	v11.s[0], v4.s[1]
+	ins	v13.s[0], v4.s[2]
+	ins	v15.s[0], v4.s[3]
+	ins	v9.s[1], v5.s[0]
+	ins	v11.s[1], v5.s[1]
+	ins	v13.s[1], v5.s[2]
+	ins	v15.s[1], v5.s[3]
+
+	ldr	q6, [A07], #16
+	ldr	q7, [A08], #16
+	ins	v9.s[2], v6.s[0]
+	ins	v11.s[2], v6.s[1]
+	ins	v13.s[2], v6.s[2]
+	ins	v15.s[2], v6.s[3]
+	ins	v9.s[3], v7.s[0]
+	ins	v11.s[3], v7.s[1]
+	ins	v13.s[3], v7.s[2]
+	ins	v15.s[3], v7.s[3]
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [B00], #64
+	st1	{v12.4s, v13.4s, v14.4s, v15.4s}, [B00], #64
+.endm
+
+.macro COPY2x8
+	ldr	d0, [A01], #8
+	ldr	d1, [A02], #8
+	ins	v8.s[0], v0.s[0]
+	ins	v10.s[0], v0.s[1]
+	ins	v8.s[1], v1.s[0]
+	ins	v10.s[1], v1.s[1]
+
+	ldr	d2, [A03], #8
+	ldr	d3, [A04], #8
+	ins	v8.s[2], v2.s[0]
+	ins	v10.s[2], v2.s[1]
+	ins	v8.s[3], v3.s[0]
+	ins	v10.s[3], v3.s[1]
+
+	ldr	d4, [A05], #8
+	ldr	d5, [A06], #8
+	ins	v9.s[0], v4.s[0]
+	ins	v11.s[0], v4.s[1]
+	ins	v9.s[1], v5.s[0]
+	ins	v11.s[1], v5.s[1]
+
+	ldr	d6, [A07], #8
+	ldr	d7, [A08], #8
+	ins	v9.s[2], v6.s[0]
+	ins	v11.s[2], v6.s[1]
+	ins	v9.s[3], v7.s[0]
+	ins	v11.s[3], v7.s[1]
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [B00], #64
+.endm
+
+.macro COPY1x8
+	ldr	s0, [A01], #4
+	ldr	s1, [A02], #4
+	ins	v8.s[0], v0.s[0]
+	ins	v8.s[1], v1.s[0]
+
+	ldr	s2, [A03], #4
+	ldr	s3, [A04], #4
+	ins	v8.s[2], v2.s[0]
+	ins	v8.s[3], v3.s[0]
+
+	ldr	s4, [A05], #4
+	ldr	s5, [A06], #4
+	ins	v9.s[0], v4.s[0]
+	ins	v9.s[1], v5.s[0]
+
+	ldr	s6, [A07], #4
+	ldr	s7, [A08], #4
+	ins	v9.s[2], v6.s[0]
+	ins	v9.s[3], v7.s[0]
+
+	st1	{v8.4s, v9.4s}, [B00], #32
+.endm
+
+.macro COPY4x4
+	ldr	q0, [A01], #16
+	ldr	q1, [A02], #16
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v10.s[0], v0.s[2]
+	ins	v11.s[0], v0.s[3]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+	ins	v10.s[1], v1.s[2]
+	ins	v11.s[1], v1.s[3]
+
+	ldr	q2, [A03], #16
+	ldr	q3, [A04], #16
+	ins	v8.s[2], v2.s[0]
+	ins	v9.s[2], v2.s[1]
+	ins	v10.s[2], v2.s[2]
+	ins	v11.s[2], v2.s[3]
+	ins	v8.s[3], v3.s[0]
+	ins	v9.s[3], v3.s[1]
+	ins	v10.s[3], v3.s[2]
+	ins	v11.s[3], v3.s[3]
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [B00], #64
+.endm
+
+.macro COPY2x4
+	ldr	d0, [A01], #8
+	ldr	d1, [A02], #8
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+
+	ldr	d2, [A03], #8
+	ldr	d3, [A04], #8
+	ins	v8.s[2], v2.s[0]
+	ins	v9.s[2], v2.s[1]
+	ins	v8.s[3], v3.s[0]
+	ins	v9.s[3], v3.s[1]
+
+	st1	{v8.4s, v9.4s}, [B00], #32
+.endm
+
+.macro COPY1x4
+	ldr	s0, [A01], #4
+	ldr	s1, [A02], #4
+	ins	v8.s[0], v0.s[0]
+	ins	v8.s[1], v1.s[0]
+
+	ldr	s2, [A03], #4
+	ldr	s3, [A04], #4
+	ins	v8.s[2], v2.s[0]
+	ins	v8.s[3], v3.s[0]
+
+	st1	{v8.4s}, [B00], #16
+.endm
+
+.macro COPY4x2
+	ldr	q0, [A01], #16
+	ldr	q1, [A02], #16
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v10.s[0], v0.s[2]
+	ins	v11.s[0], v0.s[3]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+	ins	v10.s[1], v1.s[2]
+	ins	v11.s[1], v1.s[3]
+
+	st1	{v8.2s, v9.2s, v10.2s, v11.2s}, [B00], #32
+.endm
+
+.macro COPY2x2
+	ldr	d0, [A01], #8
+	ldr	d1, [A02], #8
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+
+	st1	{v8.2s, v9.2s}, [B00], #16
+.endm
+
+.macro COPY1x2
+	ldr	s0, [A01], #4
+	ldr	s1, [A02], #4
+	ins	v8.s[0], v0.s[0]
+	ins	v8.s[1], v1.s[0]
+
+	st1	{v8.2s}, [B00], #8
+.endm
+
+.macro COPY1x1
+	ldr	s0, [A01], #4
+	str	s0, [B00], #4
+.endm
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+
+	SAVE_REGS
+
+	lsl	LDA, LDA, #2					// LDA = LDA * SIZE
+
+.Lsgemm_ncopy_L8_BEGIN:
+
+	asr	J, N, #3					// J = N / 8
+	cmp 	J, #0
+	ble	.Lsgemm_ncopy_L4_BEGIN
+
+	.align	5
+.Lsgemm_ncopy_L8_M4_BEGIN:
+
+	mov	A01, A00
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A05, A04, LDA
+	add	A06, A05, LDA
+	add	A07, A06, LDA
+	add	A08, A07, LDA
+	add	A00, A08, LDA
+
+	asr	I, M, #2					// I = M / 4
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L8_M4_40
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A01
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_1:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_1
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A02
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_2:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_2
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A03
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_3:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_3
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A04
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_4:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_4
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A05
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_5:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_5
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A06
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_6:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_6
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A07
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_7:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_7
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A08
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_8:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_8
+
+	.align	5
+.Lsgemm_ncopy_L8_M4_20:
+
+	COPY4x8
+
+	subs	I, I, #1
+	bne	.Lsgemm_ncopy_L8_M4_20
+
+.Lsgemm_ncopy_L8_M4_40:
+
+	and	I, M, #2
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L8_M4_60
+
+	COPY2x8
+
+.Lsgemm_ncopy_L8_M4_60:
+
+	and	I, M, #1
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L8_M4_END
+
+	COPY1x8
+
+.Lsgemm_ncopy_L8_M4_END:
+
+	subs	J , J, #1						// j--
+	bne	.Lsgemm_ncopy_L8_M4_BEGIN
+
+/*********************************************************************************************/
+
+.Lsgemm_ncopy_L4_BEGIN:
+
+	tst	N, #7
+	ble	.Lsgemm_ncopy_L999
+
+	tst	N, #4
+	ble	.Lsgemm_ncopy_L2_BEGIN
+
+.Lsgemm_ncopy_L4_M4_BEGIN:
+	mov	A01, A00
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A00, A04, LDA
+
+	asr	I, M, #2					// I = M / 4
+	cmp 	I, #0
+	ble	.Lsgemm_ncopy_L4_M4_40
+
+	.align	5
+.Lsgemm_ncopy_L4_M4_20:
+
+	COPY4x4
+
+	subs	I, I, #1
+	bne	.Lsgemm_ncopy_L4_M4_20
+
+.Lsgemm_ncopy_L4_M4_40:
+
+	and	I, M, #2
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L4_M4_60
+
+	COPY2x4
+
+.Lsgemm_ncopy_L4_M4_60:
+
+	and	I, M, #1
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L4_M4_END
+
+	COPY1x4
+
+.Lsgemm_ncopy_L4_M4_END:
+
+
+/*********************************************************************************************/
+
+.Lsgemm_ncopy_L2_BEGIN:
+
+	tst	N, #2
+	ble	.Lsgemm_ncopy_L1_BEGIN
+
+.Lsgemm_ncopy_L2_M4_BEGIN:
+
+	mov	A01, A00
+	add	A02, A01, LDA
+	add	A00, A02, LDA
+
+	asr	I, M, #2					// I = M / 4
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L2_M4_40
+
+	.align	5
+.Lsgemm_ncopy_L2_M4_20:
+
+	COPY4x2
+
+	subs	I , I , #1
+	bne	.Lsgemm_ncopy_L2_M4_20
+
+
+.Lsgemm_ncopy_L2_M4_40:
+
+	and	I, M, #2
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L2_M4_60
+
+	COPY2x2
+
+.Lsgemm_ncopy_L2_M4_60:
+
+	and	I, M, #1
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L2_M4_END
+
+	COPY1x2
+
+.Lsgemm_ncopy_L2_M4_END:
+
+.Lsgemm_ncopy_L1_BEGIN:
+
+	tst	N, #1
+	ble	.Lsgemm_ncopy_L999
+
+.Lsgemm_ncopy_L1_M1_BEGIN:
+
+	mov	A01, A00
+
+	mov	I, M
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L1_M1_END
+
+	.align	5
+.Lsgemm_ncopy_L1_M1_20:
+
+	COPY1x1
+
+	subs	I, I, #1
+	bne	.Lsgemm_ncopy_L1_M1_20
+
+.Lsgemm_ncopy_L1_M1_END:
+
+.Lsgemm_ncopy_L999:
+
+	mov	x0, #0
+	RESTORE_REGS
+	ret
+
+	EPILOGUE
diff --git a/kernel/arm64/sgemm_tcopy_8.S b/kernel/arm64/sgemm_tcopy_8.S
new file mode 100644
index 000000000..7d81ba266
--- /dev/null
+++ b/kernel/arm64/sgemm_tcopy_8.S
@@ -0,0 +1,707 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+#define	M		x0
+#define	N		x1
+#define	A		x2
+#define	LDA		x3
+#define	B		x4
+
+#define M8		x5
+
+#define	A01		x6
+#define	A02		x7
+#define	A03		x8
+#define	A04		x9
+#define	A05		x10
+#define	A06		x11
+#define	A07		x12
+#define	A08		x13
+
+#define	B01		x14
+#define	B02		x15
+#define	B03		x16
+#define	B04		x17
+#define	B00		x22
+
+
+#define I		x18
+#define	J		x19
+
+#define TEMP1		x20
+
+#define A_PREFETCH	256
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+.macro SAVE_REGS
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+.endm
+
+.macro RESTORE_REGS
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x8
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldp	q0, q1, [A01]
+	ldp	q2, q3, [A02]
+	add	A01, A01, #32
+	add	A02, A02, #32
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B00]
+	add	TEMP1, B00, #64
+
+	ldp	q4, q5, [A03]
+	ldp	q6, q7, [A04]
+	add	A03, A03, #32
+	add	A04, A04, #32
+
+	st1	{v4.4s, v5.4s, v6.4s, v7.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	ldp	q8, q9, [A05]
+	ldp	q10, q11, [A06]
+	add	A05, A05, #32
+	add	A06, A06, #32
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	ldp	q12, q13, [A07]
+	ldp	q14, q15, [A08]
+	add	A07, A07, #32
+	add	A08, A08, #32
+
+	st1	{v12.4s, v13.4s, v14.4s, v15.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x8
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldr	q0, [A01]
+	ldr	q1, [A02]
+	ldr	q2, [A03]
+	ldr	q3, [A04]
+	add	A01, A01, #16
+	add	A02, A02, #16
+	add	A03, A03, #16
+	add	A04, A04, #16
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B01]
+	add	B01, B01, #64
+
+	ldr	q4, [A05]
+	ldr	q5, [A06]
+	ldr	q6, [A07]
+	ldr	q7, [A08]
+
+	add	A05, A05, #16
+	add	A06, A06, #16
+	add	A07, A07, #16
+	add	A08, A08, #16
+
+	st1	{v4.4s, v5.4s, v6.4s, v7.4s}, [B01]
+	add	B01, B01, #64
+.endm
+
+.macro COPY2x8
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	ldr	d1, [A02]
+	ldr	d2, [A03]
+	ldr	d3, [A04]
+
+	add	A01, A01, #8
+	add	A02, A02, #8
+	add	A03, A03, #8
+	add	A04, A04, #8
+
+	stp	d0, d1, [B02]
+	add	B02, B02, #16
+	stp	d2, d3, [B02]
+	add	B02, B02, #16
+
+	ldr	d4, [A05]
+	ldr	d5, [A06]
+	ldr	d6, [A07]
+	ldr	d7, [A08]
+
+	add	A05, A05, #8
+	add	A06, A06, #8
+	add	A07, A07, #8
+	add	A08, A08, #8
+
+	stp	d4, d5, [B02]
+	add	B02, B02, #16
+	stp	d6, d7, [B02]
+	add	B02, B02, #16
+
+.endm
+
+.macro COPY1x8
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	ldr	s1, [A02]
+	ldr	s2, [A03]
+	ldr	s3, [A04]
+
+	add	A01, A01, #4
+	add	A02, A02, #4
+	add	A03, A03, #4
+	add	A04, A04, #4
+
+	stp	s0, s1, [B03]
+	add	B03, B03, #8
+	stp	s2, s3, [B03]
+	add	B03, B03, #8
+
+	ldr	s4, [A05]
+	ldr	s5, [A06]
+	ldr	s6, [A07]
+	ldr	s7, [A08]
+
+	ldr	d4, [A05], #8
+	ldr	d5, [A06], #8
+	ldr	d6, [A07], #8
+	ldr	d7, [A08], #8
+
+	stp	s4, s5, [B03]
+	add	B03, B03, #8
+	stp	s6, s7, [B03]
+	add	B03, B03, #8
+
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x4
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldp	q0, q1, [A01]
+	ldp	q2, q3, [A02]
+	add	A01, A01, #32
+	add	A02, A02, #32
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B00]
+	add	TEMP1, B00, #64
+
+	ldp	q4, q5, [A03]
+	ldp	q6, q7, [A04]
+	add	A03, A03, #32
+	add	A04, A04, #32
+
+	st1	{v4.4s, v5.4s, v6.4s, v7.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x4
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldr	q0, [A01]
+	ldr	q1, [A02]
+	ldr	q2, [A03]
+	ldr	q3, [A04]
+	add	A01, A01, #16
+	add	A02, A02, #16
+	add	A03, A03, #16
+	add	A04, A04, #16
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B01]
+
+	add	B01, B01, #64
+.endm
+
+.macro COPY2x4
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	ldr	d1, [A02]
+	ldr	d2, [A03]
+	ldr	d3, [A04]
+
+	add	A01, A01, #8
+	add	A02, A02, #8
+	add	A03, A03, #8
+	add	A04, A04, #8
+
+	stp	d0, d1, [B02]
+	add	B02, B02, #16
+	stp	d2, d3, [B02]
+
+	add	B02, B02, #16
+.endm
+
+.macro COPY1x4
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	ldr	s1, [A02]
+	ldr	s2, [A03]
+	ldr	s3, [A04]
+
+	add	A01, A01, #4
+	add	A02, A02, #4
+	add	A03, A03, #4
+	add	A04, A04, #4
+
+	stp	s0, s1, [B03]
+	add	B03, B03, #8
+	stp	s2, s3, [B03]
+	add	B03, B03, #8
+
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x2
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ld1	{v0.4s, v1.4s}, [A01]
+	ld1	{v2.4s, v3.4s}, [A02]
+	add	A01, A01, #32
+	add	A02, A02, #32
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B00]
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x2
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ldr	q0, [A01]
+	ldr	q1, [A02]
+	add	A01, A01, #16
+	add	A02, A02, #16
+
+	stp	q0, q1, [B01]
+	add	B01, B01, #32
+.endm
+
+.macro COPY2x2
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	ldr	d1, [A02]
+
+	add	A01, A01, #8
+	add	A02, A02, #8
+
+	stp	d0, d1, [B02]
+	add	B02, B02, #16
+.endm
+
+.macro COPY1x2
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	ldr	s1, [A02]
+
+	add	A01, A01, #4
+	add	A02, A02, #4
+
+	stp	s0, s1, [B03]
+
+	add	B03, B03, #8
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x1
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldp	q0, q1, [A01]
+	add	A01, A01, #32
+	stp	q0, q1, [B00]
+
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x1
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldr q0, [A01]
+	add	A01, A01, #16
+	str q0, [B01]
+
+	add	B01, B01, #16
+.endm
+
+.macro COPY2x1
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	add	A01, A01, #8
+	str d0, [B02]
+
+	add	B02, B02, #8
+.endm
+
+.macro COPY1x1
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	add	A01, A01, #4
+	str	s0, [B03]
+
+	add	B03, B03, #4
+.endm
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+
+	SAVE_REGS
+
+	lsl	LDA, LDA, #2					// LDA = LDA * SIZE
+
+	lsl	TEMP1, M, #2					// TEMP1 = M * SIZE
+
+	and	B01 , N , #-8
+	and	B02 , N , #-4
+	and	B03 , N , #-2
+
+	mul	B01, B01, TEMP1
+	mul	B02, B02, TEMP1
+	mul	B03, B03, TEMP1
+
+	add	B01 , B01, B
+	add	B02 , B02, B
+	add	B03 , B03, B
+
+	lsl	M8, M, #5					// M8 = M * 8 * SIZE
+
+.Lsgemm_tcopy_L8_BEGIN:
+
+	asr 	J, M, #3					// J = M / 8
+	cmp 	J, #0
+	ble	.Lsgemm_tcopy_L4_BEGIN
+
+	.align	5
+.Lsgemm_tcopy_L8_M8_BEGIN:
+
+	mov	A01, A
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A05, A04, LDA
+	add	A06, A05, LDA
+	add	A07, A06, LDA
+	add	A08, A07, LDA
+	add	A, A08, LDA
+
+	mov	B00, B
+	add	B, B00, #256					// B = B + 8 * 8 * SIZE
+
+	asr	I, N, #3					// I = N / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L8_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L8_M8_20:
+
+	COPY8x8
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L8_M8_20
+
+.Lsgemm_tcopy_L8_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L8_M8_60
+
+	COPY4x8
+
+.Lsgemm_tcopy_L8_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L8_M8_80
+
+	COPY2x8
+
+.Lsgemm_tcopy_L8_M8_80:
+
+	tst	N, #1
+	ble	.Lsgemm_tcopy_L8_M8_END
+
+	COPY1x8
+
+.Lsgemm_tcopy_L8_M8_END:
+
+	subs	J, J, #1						// j--
+	bne	.Lsgemm_tcopy_L8_M8_BEGIN
+
+/*********************************************************************************************/
+
+.Lsgemm_tcopy_L4_BEGIN:
+
+	tst	M, #7
+	ble	.Lsgemm_tcopy_L999
+
+	tst	M, #4
+	ble	.Lsgemm_tcopy_L2_BEGIN
+
+.Lsgemm_tcopy_L4_M8_BEGIN:
+
+	mov	A01, A
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A, A04, LDA
+
+	mov	B00, B
+	add	B, B00, #128					// B = B + 4 * 8 * SIZE
+
+	asr	I, N, #3					// I = N / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L4_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L4_M8_20:
+
+	COPY8x4
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L4_M8_20
+
+.Lsgemm_tcopy_L4_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L4_M8_60
+
+	COPY4x4
+
+.Lsgemm_tcopy_L4_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L4_M8_80
+
+	COPY2x4
+
+.Lsgemm_tcopy_L4_M8_80:
+
+	tst	N , #1
+	ble	.Lsgemm_tcopy_L4_M8_END
+
+	COPY1x4
+
+
+.Lsgemm_tcopy_L4_M8_END:
+
+/*********************************************************************************************/
+
+.Lsgemm_tcopy_L2_BEGIN:
+
+	tst	M, #3
+	ble	.Lsgemm_tcopy_L999
+
+	tst	M, #2
+	ble	.Lsgemm_tcopy_L1_BEGIN
+
+.Lsgemm_tcopy_L2_M16_BEGIN:
+
+	mov	A01, A
+	add	A02, A01, LDA
+	add	A, A02, LDA
+
+	mov	B00, B
+	add	B, B00, #64					// B = B + 2 * 8 * SIZE
+
+	asr	I, N, #3					// I = N / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L2_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L2_M8_20:
+
+	COPY8x2
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L2_M8_20
+
+.Lsgemm_tcopy_L2_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L2_M8_60
+
+	COPY4x2
+
+.Lsgemm_tcopy_L2_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L2_M8_80
+
+	COPY2x2
+
+.Lsgemm_tcopy_L2_M8_80:
+
+	tst	N , #1
+	ble	.Lsgemm_tcopy_L2_M8_END
+
+	COPY1x2
+
+.Lsgemm_tcopy_L2_M8_END:
+
+/*********************************************************************************************/
+
+.Lsgemm_tcopy_L1_BEGIN:
+
+	tst	M, #1
+	ble	.Lsgemm_tcopy_L999
+
+
+.Lsgemm_tcopy_L1_M16_BEGIN:
+
+	mov	A01, A						// A01 = A
+	mov	B00, B
+
+	asr	I, N, #3					// I = M / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L1_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L1_M8_20:
+
+	COPY8x1
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L1_M8_20
+
+.Lsgemm_tcopy_L1_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L1_M8_60
+
+	COPY4x1
+
+.Lsgemm_tcopy_L1_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L1_M8_80
+
+	COPY2x1
+
+.Lsgemm_tcopy_L1_M8_80:
+
+	tst	N , #1
+	ble	.Lsgemm_tcopy_L1_M8_END
+
+	COPY1x1
+
+
+.Lsgemm_tcopy_L1_M8_END:
+
+.Lsgemm_tcopy_L999:
+
+	mov	x0, #0						// set return value
+	RESTORE_REGS
+	ret
+
+	EPILOGUE
diff --git a/kernel/arm64/strmm_kernel_8x8_cortexa53.S b/kernel/arm64/strmm_kernel_8x8_cortexa53.S
new file mode 100644
index 000000000..4b84623f3
--- /dev/null
+++ b/kernel/arm64/strmm_kernel_8x8_cortexa53.S
@@ -0,0 +1,2823 @@
+/*******************************************************************************
+Copyright (c) 2015, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+/*                   X0          X1          X2          s0        X3        x4       x5           x6               x7 */
+/*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc, BLASLONG offset) */
+
+#define origM		x0
+#define origN		x1
+#define origK		x2
+#define origPA		x3
+#define origPB		x4
+#define pC		x5
+#define LDC		x6
+#define offset		x7
+#define counterL	x8
+#define counterI	x9
+#define counterJ	x10
+#define pB		x11
+#define pCRow0		x12
+#define pCRow1		x13
+#define pCRow2		x14
+#define pA		x15
+#define temp		x16
+#define tempOffset	x17
+#define tempK		x18
+
+#define alpha0		s10
+#define alphaV0		v10.s[0]
+#define alpha1		s11
+#define alphaV1		v11.s[0]
+#define alpha2		s14
+#define alphaV2		v14.s[0]
+#define alpha3		s15
+#define alphaV3		v15.s[0]
+
+// 00 origM
+// 01 origN
+// 02 origK
+// 03 origPA
+// 04 origPB
+// 05 pC
+// 06 origLDC -> LDC
+// 07 offset
+// 08 counterL
+// 09 counterI
+// 10 counterJ
+// 11 pB
+// 12 pCRow0
+// 13 pCRow1
+// 14 pCRow2
+// 15 pA
+// 16 temp
+// 17 tempOffset
+// 18 must save tempK
+// 19 must save
+// 20 must save pA0_2, pA0_3
+// 21 must save pA0_6, pA0_7
+// 22 must save pA1_2, pA1_3
+// 23 must save pA1_6, pA1_7
+// 24 must save pB0_2, pB0_3
+// 25 must save pB0_6, pB0_7
+// 26 must save pB1_2, pB1_3
+// 27 must save pB1_6, pB1_7
+// 28 must save
+// 29 frame
+// 30 link
+// 31 sp
+
+//v00 ALPHA -> pA0_0, pA0_1, pA0_2, pA0_3
+//v01 pA0_4, pA0_5, pA0_6, pA0_7
+//v02 pA1_0, pA1_1, pA1_2, pA1_3
+//v03 pA1_4, pA1_5, pA1_6, pA1_7
+//v04 pB0_0, pB0_1, pB0_2, pB0_3
+//v05 pB0_4, pB0_5, pB0_6, pB0_7
+//v06 pB1_0, pB1_1, pB1_2, pB1_3
+//v07 pB1_4, pB1_5, pB1_6, pB1_7
+//v08 must save
+//v09 must save
+//v10 must save ALPHA0
+//v11 must save ALPHA1
+//v12 must save
+//v13 must save
+//v14 must save ALPHA2
+//v15 must save ALPHA3
+//v16 must save C00, C01, C02, C03
+//v17 must save C04, C05, C06, C07
+//v18 C08, C09, C10, C11
+//v19 C12, C13, C14, C15
+//v20 C16, C17, C18, C19
+//v21 C20, C21, C22, C23
+//v22 C24, C25, C26, C27
+//v23 C28, C29, C30, C31
+//v24 C32, C33, C34, C35
+//v25 C36, C37, C38, C39
+//v26 C40, C41, C42, C43
+//v27 C44, C45, C46, C47
+//v28 C48, C49, C50, C51
+//v29 C52, C53, C54, C55
+//v30 C56, C57, C58, C59
+//v31 C60, C61, C62, C63
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+.macro INIT8x8
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s18, s16
+	fmov		s19, s17
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s22, s17
+	fmov		s23, s18
+	fmov		s24, wzr
+	fmov		s25, s16
+	fmov		s26, s17
+	fmov		s27, s18
+	fmov		s28, wzr
+	fmov		s29, s16
+	fmov		s30, s17
+	fmov		s31, s18
+.endm
+
+.macro KERNEL8x8_I
+	ld1	{v0.4s, v1.4s}, [pA], #32
+	ld1	{v4.4s, v5.4s}, [pB], #32
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d3, [pA, #8]
+	ldr	d7, [pB, #8]
+
+	ldr	x22, [pA], #16
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v17.4s, v1.4s, v4.s[0]
+	ldr	x23, [pA], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v19.4s, v1.4s, v4.s[1]
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v21.4s, v1.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v23.4s, v1.4s, v4.s[3]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v25.4s, v1.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v27.4s, v1.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v29.4s, v1.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+	fmul	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #16
+	fmla	v17.4s, v1.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x23, [pA], #8
+	fmla	v19.4s, v1.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x20, [pA], #16
+	fmla	v17.4s, v3.4s, v6.s[0]
+	ldr	x24, [pB], #16
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x21, [pA], #8
+	fmla	v19.4s, v3.4s, v6.s[1]
+	ldr	x25, [pB], #8
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v3.d[1], x23
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v17.4s, v3.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+	ld1	{v1.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro SAVE8x8
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v2.4s, v18.4s, alphaV2
+	fmul	v3.4s, v19.4s, alphaV3
+	st1 	{v2.4s, v3.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v4.4s, v20.4s, alphaV0
+	fmul	v5.4s, v21.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v6.4s, v22.4s, alphaV2
+	fmul	v7.4s, v23.4s, alphaV3
+	st1 	{v6.4s, v7.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v0.4s, v24.4s, alphaV0
+	fmul	v1.4s, v25.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v2.4s, v26.4s, alphaV2
+	fmul	v3.4s, v27.4s, alphaV3
+	st1 	{v2.4s, v3.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v4.4s, v28.4s, alphaV0
+	fmul	v5.4s, v29.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow2]
+
+	fmul	v6.4s, v30.4s, alphaV2
+	fmul	v7.4s, v31.4s, alphaV3
+	st1 	{v6.4s, v7.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+
+.macro INIT4x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL4x8_I
+	ld1	{v0.4s}, [pA], #16
+	ld1	{v4.4s, v5.4s}, [pB], #32
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d7, [pB, #8]
+	ldr	x21, [pA], #8
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x21, [pA], #8
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x26, [pB], #16
+	fmla	v20.4s, v0.4s, v4.s[2]
+	ldr	x27, [pB], #8
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x21
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x20, [pA], #8
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x24, [pB], #16
+	fmla	v20.4s, v2.4s, v6.s[2]
+	ldr	x25, [pB], #8
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_E
+	fmov	v2.d[1], x21
+	fmov	v6.d[1], x26
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro SAVE4x8
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.4s, v16.4s, alphaV0
+	st1 	{v0.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.4s, v18.4s, alphaV2
+	st1 	{v2.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.4s, v20.4s, alphaV0
+	st1 	{v4.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v6.4s, v22.4s, alphaV2
+	st1 	{v6.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v0.4s, v24.4s, alphaV0
+	st1 	{v0.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.4s, v26.4s, alphaV2
+	st1 	{v2.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.4s, v28.4s, alphaV0
+	st1 	{v4.4s}, [pCRow2]
+
+
+	fmul	v6.4s, v30.4s, alphaV2
+	st1 	{v6.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL2x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v4.s[0]
+	fmla	v18.2s, v0.2s, v4.s[1]
+	fmla	v20.2s, v0.2s, v4.s[2]
+	fmla	v22.2s, v0.2s, v4.s[3]
+	fmla	v24.2s, v0.2s, v5.s[0]
+	fmla	v26.2s, v0.2s, v5.s[1]
+	fmla	v28.2s, v0.2s, v5.s[2]
+	fmla	v30.2s, v0.2s, v5.s[3]
+.endm
+
+.macro SAVE2x8
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.2s, v16.2s, alphaV0
+	st1 	{v0.2s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.2s, v18.2s, alphaV2
+	st1 	{v2.2s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.2s, v20.2s, alphaV0
+	st1 	{v4.2s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v6.2s, v22.2s, alphaV2
+	st1 	{v6.2s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v0.2s, v24.2s, alphaV0
+	st1 	{v0.2s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.2s, v26.2s, alphaV2
+	st1 	{v2.2s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.2s, v28.2s, alphaV0
+	st1 	{v4.2s}, [pCRow2]
+
+
+	fmul	v6.2s, v30.2s, alphaV2
+	st1 	{v6.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL1x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ldr	s0, [pA]
+	add	pA, pA, #4
+
+	fmla	s16, s0, v4.s[0]
+	fmla	s18, s0, v4.s[1]
+	fmla	s20, s0, v4.s[2]
+	fmla	s22, s0, v4.s[3]
+	fmla	s24, s0, v5.s[0]
+	fmla	s26, s0, v5.s[1]
+	fmla	s28, s0, v5.s[2]
+	fmla	s30, s0, v5.s[3]
+.endm
+
+.macro SAVE1x8
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	s0, s16, alphaV0
+	str 	s0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	s2, s18, alphaV2
+	str 	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	s4, s20, alphaV0
+	str 	s4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	s6, s22, alphaV2
+	str 	s6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	s0, s24, alphaV0
+	str 	s0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	s2, s26, alphaV2
+	str 	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	s4, s28, alphaV0
+	str 	s4, [pCRow2]
+
+
+	fmul	s6, s30, alphaV2
+	str 	s6, [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x4
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s24, wzr
+	fmov		s25, s16
+	fmov		s28, wzr
+	fmov		s29, s16
+.endm
+
+.macro KERNEL8x4_I
+	ld1	{v8.4s}, [pB], #16
+	ld1	{v0.4s, v1.4s}, [pA], #32
+
+	ldr	d9, [pB], #8
+	ldr	d2, [pA], #8
+	ldr	d3, [pA, #8]
+	fmul	v16.4s, v0.4s, v8.s[0]
+	ldr	x25, [pB], #8
+	fmul	v17.4s, v1.4s, v8.s[0]
+	ldr	x22, [pA], #16
+	fmul	v20.4s, v0.4s, v8.s[1]
+	ldr	x23, [pA], #8
+	fmul	v21.4s, v1.4s, v8.s[1]
+	fmul	v24.4s, v0.4s, v8.s[2]
+	fmul	v25.4s, v1.4s, v8.s[2]
+	fmul	v28.4s, v0.4s, v8.s[3]
+	fmul	v29.4s, v1.4s, v8.s[3]
+.endm
+
+.macro KERNEL8x4_M1
+	ldr	d9, [pB], #8
+	fmov	v8.d[1], x24
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	fmla	v16.4s, v0.4s, v8.s[0]
+	ldr	x25, [pB], #8
+	fmla	v17.4s, v1.4s, v8.s[0]
+	ldr	x22, [pA], #16
+	fmla	v20.4s, v0.4s, v8.s[1]
+	ldr	x23, [pA], #8
+	fmla	v21.4s, v1.4s, v8.s[1]
+	fmla	v24.4s, v0.4s, v8.s[2]
+	fmla	v25.4s, v1.4s, v8.s[2]
+	fmla	v28.4s, v0.4s, v8.s[3]
+	fmla	v29.4s, v1.4s, v8.s[3]
+.endm
+
+.macro KERNEL8x4_M2
+	ldr	d8, [pB], #8
+	fmov	v9.d[1], x25
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	fmla	v16.4s, v2.4s, v9.s[0]
+	ldr	x24, [pB], #8
+	fmla	v17.4s, v3.4s, v9.s[0]
+	ldr	x20, [pA], #16
+	fmla	v20.4s, v2.4s, v9.s[1]
+	ldr	x21, [pA], #8
+	fmla	v21.4s, v3.4s, v9.s[1]
+	fmla	v24.4s, v2.4s, v9.s[2]
+	fmla	v25.4s, v3.4s, v9.s[2]
+	fmla	v28.4s, v2.4s, v9.s[3]
+	fmla	v29.4s, v3.4s, v9.s[3]
+.endm
+
+.macro KERNEL8x4_E
+	fmov	v9.d[1], x25
+	fmov	v2.d[1], x22
+	fmov	v3.d[1], x23
+	fmla	v16.4s, v2.4s, v9.s[0]
+	fmla	v17.4s, v3.4s, v9.s[0]
+	fmla	v20.4s, v2.4s, v9.s[1]
+	fmla	v21.4s, v3.4s, v9.s[1]
+	fmla	v24.4s, v2.4s, v9.s[2]
+	fmla	v25.4s, v3.4s, v9.s[2]
+	fmla	v28.4s, v2.4s, v9.s[3]
+	fmla	v29.4s, v3.4s, v9.s[3]
+.endm
+
+.macro KERNEL8x4_SUB
+	ld1	{v8.4s}, [pB], #16
+	ld1	{v0.4s, v1.4s}, [pA], #32
+	fmla	v16.4s, v0.4s, v8.s[0]
+	fmla	v17.4s, v1.4s, v8.s[0]
+	fmla	v20.4s, v0.4s, v8.s[1]
+	fmla	v21.4s, v1.4s, v8.s[1]
+	fmla	v24.4s, v0.4s, v8.s[2]
+	fmla	v25.4s, v1.4s, v8.s[2]
+	fmla	v28.4s, v0.4s, v8.s[3]
+	fmla	v29.4s, v1.4s, v8.s[3]
+.endm
+
+.macro SAVE8x4
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v4.4s, v20.4s, alphaV0
+	fmul	v5.4s, v21.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v0.4s, v24.4s, alphaV0
+	fmul	v1.4s, v25.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow2]
+
+
+	fmul	v4.4s, v28.4s, alphaV0
+	fmul	v5.4s, v29.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+
+.macro INIT4x4
+	fmov		s16, wzr
+	fmov		s17, s16
+	fmov		s20, s17
+	fmov		s21, s16
+	fmov		s24, s17
+	fmov		s25, s16
+	fmov		s28, s17
+	fmov		s29, s16
+.endm
+
+.macro KERNEL4x4_I
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmul	v16.2s, v0.2s, v8.s[0]
+	fmul	v29.2s, v1.2s, v9.s[1]
+
+	fmul	v20.2s, v0.2s, v8.s[1]
+	fmul	v25.2s, v1.2s, v9.s[0]
+
+	fmul	v24.2s, v0.2s, v9.s[0]
+	fmul	v21.2s, v1.2s, v8.s[1]
+
+	fmul	v28.2s, v0.2s, v9.s[1]
+	fmul	v17.2s, v1.2s, v8.s[0]
+
+	ld1	{v12.2s, v13.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v4.2s, v5.2s}, [pA]
+	add	pA, pA, #16
+.endm
+
+.macro KERNEL4x4_M1
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v29.2s, v1.2s, v9.s[1]
+
+	ld1	{v12.2s, v13.2s}, [pB]		// For next round
+	add	pB, pB, #16
+
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v25.2s, v1.2s, v9.s[0]
+
+	ld1	{v4.2s, v5.2s}, [pA]		// For next round
+	add	pA, pA, #16
+
+	fmla	v24.2s, v0.2s, v9.s[0]
+	fmla	v21.2s, v1.2s, v8.s[1]
+
+	prfm	PLDL1KEEP, [pB, #512]
+
+	fmla	v28.2s, v0.2s, v9.s[1]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro KERNEL4x4_M2
+	fmla	v16.2s, v4.2s, v12.s[0]
+	fmla	v29.2s, v5.2s, v13.s[1]
+
+	ld1	{v8.2s, v9.2s}, [pB]		// For next round
+	add	pB, pB, #16
+
+	fmla	v20.2s, v4.2s, v12.s[1]
+	fmla	v25.2s, v5.2s, v13.s[0]
+
+	ld1	{v0.2s, v1.2s}, [pA]		// For next round
+	add	pA, pA, #16
+
+	fmla	v24.2s, v4.2s, v13.s[0]
+	fmla	v21.2s, v5.2s, v12.s[1]
+
+	prfm	PLDL1KEEP, [pA, #512]
+
+	fmla	v28.2s, v4.2s, v13.s[1]
+	fmla	v17.2s, v5.2s, v12.s[0]
+.endm
+
+.macro KERNEL4x4_E
+	fmla	v16.2s, v4.2s, v12.s[0]
+	fmla	v29.2s, v5.2s, v13.s[1]
+
+	fmla	v20.2s, v4.2s, v12.s[1]
+	fmla	v25.2s, v5.2s, v13.s[0]
+
+	fmla	v24.2s, v4.2s, v13.s[0]
+	fmla	v21.2s, v5.2s, v12.s[1]
+
+	fmla	v28.2s, v4.2s, v13.s[1]
+	fmla	v17.2s, v5.2s, v12.s[0]
+.endm
+
+.macro KERNEL4x4_SUB
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v29.2s, v1.2s, v9.s[1]
+
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v25.2s, v1.2s, v9.s[0]
+
+	fmla	v24.2s, v0.2s, v9.s[0]
+	fmla	v21.2s, v1.2s, v8.s[1]
+
+	fmla	v28.2s, v0.2s, v9.s[1]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro SAVE4x4
+
+	fmul	v8.2s, v16.2s, alphaV0
+	fmul	v9.2s, v17.2s, alphaV1
+	st1 	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV2
+	fmul	v13.2s, v21.2s, alphaV3
+	st1 	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v8.2s, v24.2s, alphaV0
+	fmul	v9.2s, v25.2s, alphaV1
+	st1 	{v8.2s, v9.2s}, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v12.2s, v28.2s, alphaV2
+	fmul	v13.2s, v29.2s, alphaV3
+	st1 	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x4
+	fmov		s16, wzr
+	fmov		s20, s16
+	fmov		s24, s20
+	fmov		s28, s16
+.endm
+
+.macro KERNEL2x4_SUB
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v24.2s, v0.2s, v9.s[0]
+	fmla	v28.2s, v0.2s, v9.s[1]
+.endm
+
+.macro SAVE2x4
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV1
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v8.2s, v24.2s, alphaV2
+	st1	{v8.2s}, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v12.2s, v28.2s, alphaV3
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x4
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL1x4_SUB
+	ldr	s0, [pA]
+	add	pA, pA, #4
+
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+	fmla	v20.2s, v9.2s, v0.s[0]
+.endm
+
+.macro SAVE1x4
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v12.2s, v20.2s, alphaV1
+	st1	{v12.s}[0], [pCRow2]
+	st1	{v12.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s20, s17
+	fmov	s21, s16
+.endm
+
+.macro KERNEL8x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+	ld1	{v1.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v8.s[0]
+	fmla	v17.4s, v1.4s, v8.s[0]
+
+	fmla	v20.4s, v0.4s, v8.s[1]
+	fmla	v21.4s, v1.4s, v8.s[1]
+.endm
+
+.macro SAVE8x2
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v4.4s, v20.4s, alphaV0
+	fmul	v5.4s, v21.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s20, s17
+	fmov	s21, s16
+.endm
+
+.macro KERNEL4x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v21.2s, v1.2s, v8.s[1]
+.endm
+
+.macro SAVE4x2
+
+	fmul	v8.2s, v16.2s, alphaV0
+	fmul	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV2
+	fmul	v13.2s, v21.2s, alphaV3
+	st1	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x2
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL2x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+.endm
+
+.macro SAVE2x2
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow1 , pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV1
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x2
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL1x2_SUB
+	ld1	{v8.2s} , [pB]
+	add	pB , pB, #8
+
+	ldr	s0 , [pA]
+	add	pA, pA, #4
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+.endm
+
+.macro SAVE1x2
+	add	pCRow1 , pCRow0, LDC
+
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x1
+	fmov	s16, wzr
+	fmov	s17, wzr
+.endm
+
+.macro KERNEL8x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+	ld1	{v1.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v8.s[0]
+	fmla	v17.4s, v1.4s, v8.s[0]
+.endm
+
+.macro SAVE8x1
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x1
+	fmov	s16, wzr
+	fmov	s17, s16
+.endm
+
+.macro KERNEL4x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA , pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro SAVE4x1
+
+	fmul	v8.2s, v16.2s, alphaV0
+	fmul	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x1
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL2x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s}, [pA]
+	add	pA , pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+.endm
+
+.macro SAVE2x1
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x1
+	fmov	s16, wzr
+.endm
+
+.macro KERNEL1x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ldr	s0, [pA]
+	add	pA , pA, #4
+
+	fmadd 	s16, s0, s8, s16  
+.endm
+
+.macro SAVE1x1
+
+	fmul	s8, s16, alpha0
+	str 	s8, [pCRow0]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+.Lstrmm_kernel_begin:
+
+	.align 5
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+
+	fmov	alpha0, s0
+	fmov	alpha1, s0
+	fmov	alpha2, s0
+	fmov	alpha3, s0
+
+	lsl	LDC, LDC, #2			// ldc = ldc * 4
+
+#if !defined(LEFT)
+	neg	tempOffset, offset
+#endif
+	mov	pB, origPB
+
+	mov	counterJ, origN
+	asr 	counterJ, counterJ, #3		// J = J / 8
+	cmp 	counterJ, #0
+	ble	.Lstrmm_kernel_L4_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_BEGIN:
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC, pC, LDC, lsl #3
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+
+	mov	pA, origPA			// pA = start of A array
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3		// counterI = counterI / 8
+	cmp 	counterI, #0
+	ble	.Lstrmm_kernel_L8_M4_BEGIN
+
+.Lstrmm_kernel_L8_M8_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #3		// L = K / 8
+	cmp	counterL , #2			// is there at least 16 to do?
+	blt	.Lstrmm_kernel_L8_M8_32
+
+	KERNEL8x8_I				// do one in the K
+	KERNEL8x8_M2				// do another in the K
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L8_M8_22a
+	.align 5
+
+.Lstrmm_kernel_L8_M8_22:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M8_22
+
+.Lstrmm_kernel_L8_M8_22a:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	 .Lstrmm_kernel_L8_M8_44
+
+.Lstrmm_kernel_L8_M8_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L8_M8_40
+
+	KERNEL8x8_I
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	.Lstrmm_kernel_L8_M8_44
+
+.Lstrmm_kernel_L8_M8_40:
+
+	INIT8x8
+
+.Lstrmm_kernel_L8_M8_44:
+
+	ands	counterL , tempK, #7
+	ble	.Lstrmm_kernel_L8_M8_100
+
+.Lstrmm_kernel_L8_M8_46:
+
+	KERNEL8x8_SUB
+
+	subs	counterL, counterL, 1
+	bgt	.Lstrmm_kernel_L8_M8_46
+
+.Lstrmm_kernel_L8_M8_100:
+
+	SAVE8x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+
+.Lstrmm_kernel_L8_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lstrmm_kernel_L8_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L8_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L8_M2_BEGIN
+
+.Lstrmm_kernel_L8_M4_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #4
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #5
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lstrmm_kernel_L8_M4_32
+
+	KERNEL4x8_I				// do one in the K
+	KERNEL4x8_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L8_M4_22a
+	.align 5
+
+.Lstrmm_kernel_L8_M4_22:
+
+	KERNEL4x8_M1
+	KERNEL4x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M4_22
+
+.Lstrmm_kernel_L8_M4_22a:
+
+	KERNEL4x8_M1
+	KERNEL4x8_E
+
+	b	 .Lstrmm_kernel_L8_M4_44
+
+.Lstrmm_kernel_L8_M4_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L8_M4_40
+
+	KERNEL4x8_I
+	KERNEL4x8_E
+
+	b	.Lstrmm_kernel_L8_M4_44
+
+.Lstrmm_kernel_L8_M4_40:
+
+	INIT4x8
+
+.Lstrmm_kernel_L8_M4_44:
+
+	ands	counterL , tempK, #1
+	ble	.Lstrmm_kernel_L8_M4_100
+
+.Lstrmm_kernel_L8_M4_46:
+
+	KERNEL4x8_SUB
+
+.Lstrmm_kernel_L8_M4_100:
+
+	SAVE4x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	lsl	temp, tempK, #5
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+
+.Lstrmm_kernel_L8_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L8_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L8_M1_BEGIN
+
+.Lstrmm_kernel_L8_M2_20:
+
+	INIT2x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #5
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L8_M2_40
+
+.Lstrmm_kernel_L8_M2_22:
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M2_22
+
+
+.Lstrmm_kernel_L8_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L8_M2_100
+
+.Lstrmm_kernel_L8_M2_42:
+
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M2_42
+
+.Lstrmm_kernel_L8_M2_100:
+
+	SAVE2x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #5
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+
+.Lstrmm_kernel_L8_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L8_END
+
+.Lstrmm_kernel_L8_M1_20:
+
+	INIT1x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #5
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L8_M1_40
+
+.Lstrmm_kernel_L8_M1_22:
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M1_22
+
+
+.Lstrmm_kernel_L8_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L8_M1_100
+
+.Lstrmm_kernel_L8_M1_42:
+
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M1_42
+
+.Lstrmm_kernel_L8_M1_100:
+
+	SAVE1x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #1
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #2
+	add	pA, pA, temp
+	lsl	temp, tempK, #5
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #1
+#endif
+
+.Lstrmm_kernel_L8_END:
+	lsl	temp, origK, #5			// B = B + K * 4 * 8
+	add	origPB, origPB, temp
+
+#if !defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+
+	subs	counterJ, counterJ , #1		// j--
+	bgt	.Lstrmm_kernel_L8_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #7
+	ble	.Lstrmm_kernel_L999
+
+	tst	counterJ , #4
+	ble	.Lstrmm_kernel_L2_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #2
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3		// counterI = counterI / 8
+	cmp 	counterI, #0
+	ble	.Lstrmm_kernel_L4_M4_BEGIN
+
+.Lstrmm_kernel_L4_M8_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #4
+#endif
+
+	asr 	counterL , tempK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lstrmm_kernel_L4_M8_32
+
+	KERNEL8x4_I				// do one in the K
+	KERNEL8x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L4_M8_22a
+	.align 5
+
+.Lstrmm_kernel_L4_M8_22:
+
+	KERNEL8x4_M1
+	KERNEL8x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M8_22
+
+.Lstrmm_kernel_L4_M8_22a:
+
+	KERNEL8x4_M1
+	KERNEL8x4_E
+
+	b	 .Lstrmm_kernel_L4_M8_44
+
+.Lstrmm_kernel_L4_M8_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L4_M8_40
+
+	KERNEL8x4_I
+	KERNEL8x4_E
+
+	b	.Lstrmm_kernel_L4_M8_44
+
+.Lstrmm_kernel_L4_M8_40:
+
+	INIT8x4
+
+.Lstrmm_kernel_L4_M8_44:
+
+	ands	counterL , tempK, #1
+	ble	.Lstrmm_kernel_L4_M8_100
+
+.Lstrmm_kernel_L4_M8_46:
+
+	KERNEL8x4_SUB
+
+.Lstrmm_kernel_L4_M8_100:
+
+	SAVE8x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	lsl	temp, tempK, #4
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+.Lstrmm_kernel_L4_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lstrmm_kernel_L4_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L4_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L4_M2_BEGIN
+
+.Lstrmm_kernel_L4_M4_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #4
+#endif
+	asr 	counterL , tempK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lstrmm_kernel_L4_M4_32
+
+	KERNEL4x4_I				// do one in the K
+	KERNEL4x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L4_M4_22a
+	.align 5
+
+.Lstrmm_kernel_L4_M4_22:
+
+	KERNEL4x4_M1
+	KERNEL4x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M4_22
+
+.Lstrmm_kernel_L4_M4_22a:
+
+	KERNEL4x4_M1
+	KERNEL4x4_E
+
+	b	 .Lstrmm_kernel_L4_M4_44
+
+.Lstrmm_kernel_L4_M4_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L4_M4_40
+
+	KERNEL4x4_I
+	KERNEL4x4_E
+
+	b	.Lstrmm_kernel_L4_M4_44
+
+.Lstrmm_kernel_L4_M4_40:
+
+	INIT4x4
+
+.Lstrmm_kernel_L4_M4_44:
+
+	ands	counterL , tempK, #1
+	ble	.Lstrmm_kernel_L4_M4_100
+
+.Lstrmm_kernel_L4_M4_46:
+
+	KERNEL4x4_SUB
+
+.Lstrmm_kernel_L4_M4_100:
+
+	SAVE4x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+.Lstrmm_kernel_L4_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L4_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L4_M1_BEGIN
+
+.Lstrmm_kernel_L4_M2_20:
+
+	INIT2x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #4
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L4_M2_40
+
+.Lstrmm_kernel_L4_M2_22:
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M2_22
+
+
+.Lstrmm_kernel_L4_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L4_M2_100
+
+.Lstrmm_kernel_L4_M2_42:
+
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M2_42
+
+.Lstrmm_kernel_L4_M2_100:
+
+	SAVE2x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #4
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+.Lstrmm_kernel_L4_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L4_END
+
+.Lstrmm_kernel_L4_M1_20:
+
+	INIT1x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #4
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L4_M1_40
+
+.Lstrmm_kernel_L4_M1_22:
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M1_22
+
+
+.Lstrmm_kernel_L4_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L4_M1_100
+
+.Lstrmm_kernel_L4_M1_42:
+
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M1_42
+
+.Lstrmm_kernel_L4_M1_100:
+
+	SAVE1x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #1
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #2
+	add	pA, pA, temp
+	lsl	temp, tempK, #4
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #1
+#endif
+.Lstrmm_kernel_L4_END:
+	add	origPB, origPB, origK, lsl #4	// B = B + K * 4 * 4
+#if !defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_BEGIN:   // less than 2 left in N direction
+
+	mov	counterJ , origN
+	tst	counterJ , #3
+	ble	.Lstrmm_kernel_L999
+
+	tst	counterJ , #2
+	ble	.Lstrmm_kernel_L1_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #1
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI,#0
+	ble	.Lstrmm_kernel_L2_M4_BEGIN
+
+.Lstrmm_kernel_L2_M8_20:
+
+	INIT8x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lstrmm_kernel_L2_M8_40
+	.align 5
+
+.Lstrmm_kernel_L2_M8_22:
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M8_22
+
+
+.Lstrmm_kernel_L2_M8_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M8_100
+
+.Lstrmm_kernel_L2_M8_42:
+
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M8_42
+
+.Lstrmm_kernel_L2_M8_100:
+
+	SAVE8x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+.Lstrmm_kernel_L2_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lstrmm_kernel_L2_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L2_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L2_M2_BEGIN
+
+.Lstrmm_kernel_L2_M4_20:
+
+	INIT4x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #4
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lstrmm_kernel_L2_M4_40
+	.align 5
+
+.Lstrmm_kernel_L2_M4_22:
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M4_22
+
+
+.Lstrmm_kernel_L2_M4_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M4_100
+
+.Lstrmm_kernel_L2_M4_42:
+
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M4_42
+
+.Lstrmm_kernel_L2_M4_100:
+
+	SAVE4x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+.Lstrmm_kernel_L2_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L2_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L2_M1_BEGIN
+
+.Lstrmm_kernel_L2_M2_20:
+
+	INIT2x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+        cmp	counterL,#0
+	ble	.Lstrmm_kernel_L2_M2_40
+
+.Lstrmm_kernel_L2_M2_22:
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M2_22
+
+
+.Lstrmm_kernel_L2_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M2_100
+
+.Lstrmm_kernel_L2_M2_42:
+
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M2_42
+
+.Lstrmm_kernel_L2_M2_100:
+
+	SAVE2x2
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+
+.Lstrmm_kernel_L2_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L2_END
+
+.Lstrmm_kernel_L2_M1_20:
+
+	INIT1x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+        cmp     counterL, #0
+	ble	.Lstrmm_kernel_L2_M1_40
+
+.Lstrmm_kernel_L2_M1_22:
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M1_22
+
+
+.Lstrmm_kernel_L2_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M1_100
+
+.Lstrmm_kernel_L2_M1_42:
+
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M1_42
+
+.Lstrmm_kernel_L2_M1_100:
+
+	SAVE1x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #1
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #2
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #1
+#endif
+.Lstrmm_kernel_L2_END:
+#if !defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+	add	origPB, origPB, origK, lsl #3	// B = B + K * 2 * 4
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #1
+	ble	.Lstrmm_kernel_L999 // done
+
+
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC , pC , LDC			// Update pC to point to next
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3
+	cmp	counterI, #0
+	ble	.Lstrmm_kernel_L1_M4_BEGIN
+
+.Lstrmm_kernel_L1_M8_20:
+
+	INIT8x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M8_40
+	.align 5
+
+.Lstrmm_kernel_L1_M8_22:
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M8_22
+
+
+.Lstrmm_kernel_L1_M8_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M8_100
+
+.Lstrmm_kernel_L1_M8_42:
+
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M8_42
+
+.Lstrmm_kernel_L1_M8_100:
+
+	SAVE8x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #1
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	lsl	temp, tempK, #2
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+.Lstrmm_kernel_L1_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lstrmm_kernel_L1_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L1_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L1_M2_BEGIN
+
+.Lstrmm_kernel_L1_M4_20:
+
+	INIT4x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #4
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M4_40
+	.align 5
+
+.Lstrmm_kernel_L1_M4_22:
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M4_22
+
+
+.Lstrmm_kernel_L1_M4_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M4_100
+
+.Lstrmm_kernel_L1_M4_42:
+
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M4_42
+
+.Lstrmm_kernel_L1_M4_100:
+
+	SAVE4x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #1
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	lsl	temp, tempK, #2
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+.Lstrmm_kernel_L1_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L1_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L1_M1_BEGIN
+
+.Lstrmm_kernel_L1_M2_20:
+
+	INIT2x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M2_40
+
+.Lstrmm_kernel_L1_M2_22:
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M2_22
+
+
+.Lstrmm_kernel_L1_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M2_100
+
+.Lstrmm_kernel_L1_M2_42:
+
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M2_42
+
+.Lstrmm_kernel_L1_M2_100:
+
+	SAVE2x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #1
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #2
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+.Lstrmm_kernel_L1_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L1_END
+
+.Lstrmm_kernel_L1_M1_20:
+
+	INIT1x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M1_40
+
+.Lstrmm_kernel_L1_M1_22:
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M1_22
+
+
+.Lstrmm_kernel_L1_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M1_100
+
+.Lstrmm_kernel_L1_M1_42:
+
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M1_42
+
+.Lstrmm_kernel_L1_M1_100:
+
+	SAVE1x1
+
+.Lstrmm_kernel_L1_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L999:
+	mov	x0, #0				// set return value
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+	ret
+
+	EPILOGUE
+
diff --git a/kernel/common_param.h b/kernel/common_param.h
deleted file mode 100644
index eab14b0a6..000000000
--- a/kernel/common_param.h
+++ /dev/null
@@ -1,1274 +0,0 @@
-/*********************************************************************/
-/* Copyright 2009, 2010 The University of Texas at Austin.           */
-/* All rights reserved.                                              */
-/*                                                                   */
-/* Redistribution and use in source and binary forms, with or        */
-/* without modification, are permitted provided that the following   */
-/* conditions are met:                                               */
-/*                                                                   */
-/*   1. Redistributions of source code must retain the above         */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer.                                                  */
-/*                                                                   */
-/*   2. Redistributions in binary form must reproduce the above      */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer in the documentation and/or other materials       */
-/*      provided with the distribution.                              */
-/*                                                                   */
-/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
-/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
-/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
-/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
-/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
-/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
-/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
-/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
-/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
-/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
-/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
-/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
-/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
-/*    POSSIBILITY OF SUCH DAMAGE.                                    */
-/*                                                                   */
-/* The views and conclusions contained in the software and           */
-/* documentation are those of the authors and should not be          */
-/* interpreted as representing official policies, either expressed   */
-/* or implied, of The University of Texas at Austin.                 */
-/*********************************************************************/
-
-#ifndef COMMON_PARAM_H
-#define COMMON_PARAM_H
-
-#ifndef ASSEMBLER
-
-#ifdef DYNAMIC_ARCH
-
-typedef struct {
-  int dtb_entries;
-  int offsetA, offsetB, align;
-
-  int sgemm_p, sgemm_q, sgemm_r;
-  int sgemm_unroll_m, sgemm_unroll_n, sgemm_unroll_mn;
-
-  int exclusive_cache;
-
-  float  (*samax_k) (BLASLONG, float *, BLASLONG);
-  float  (*samin_k) (BLASLONG, float *, BLASLONG);
-  float  (*smax_k)  (BLASLONG, float *, BLASLONG);
-  float  (*smin_k)  (BLASLONG, float *, BLASLONG);
-BLASLONG (*isamax_k)(BLASLONG, float *, BLASLONG);
-BLASLONG (*isamin_k)(BLASLONG, float *, BLASLONG);
-BLASLONG (*ismax_k) (BLASLONG, float *, BLASLONG);
-BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
-
-  float  (*snrm2_k) (BLASLONG, float *, BLASLONG);
-  float  (*sasum_k) (BLASLONG, float *, BLASLONG);
-  float  (*ssum_k)  (BLASLONG, float *, BLASLONG);
-  int    (*scopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  float  (*sdot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  double (*dsdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-
-  int    (*srot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
-
-  int    (*saxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  int    (*sscal_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  int    (*sswap_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-
-  int    (*sgemv_n) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*sgemv_t) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*sger_k)  (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-
-  int    (*ssymv_L) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-  int    (*ssymv_U) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-
-  int    (*sgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG);
-  int    (*sgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float  *, BLASLONG);
-
-  int    (*sgemm_incopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*sgemm_itcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*sgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*sgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-
-  int    (*strsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*strsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*strsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*strsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-
-  int    (*strsm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*strsm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-
-  int    (*strmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*strmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*strmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*strmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
-
-  int    (*strmm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*strmm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int    (*ssymm_iutcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ssymm_iltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ssymm_outcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ssymm_oltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int	 (*sneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*slaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
-
-  int dgemm_p, dgemm_q, dgemm_r;
-  int dgemm_unroll_m, dgemm_unroll_n, dgemm_unroll_mn;
-
-  double (*damax_k) (BLASLONG, double *, BLASLONG);
-  double (*damin_k) (BLASLONG, double *, BLASLONG);
-  double (*dmax_k)  (BLASLONG, double *, BLASLONG);
-  double (*dmin_k)  (BLASLONG, double *, BLASLONG);
-BLASLONG (*idamax_k)(BLASLONG, double *, BLASLONG);
-BLASLONG (*idamin_k)(BLASLONG, double *, BLASLONG);
-BLASLONG (*idmax_k) (BLASLONG, double *, BLASLONG);
-BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
-
-  double (*dnrm2_k) (BLASLONG, double *, BLASLONG);
-  double (*dasum_k) (BLASLONG, double *, BLASLONG);
-  double (*dsum_k)  (BLASLONG, double *, BLASLONG);
-  int    (*dcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  double (*ddot_k)  (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*drot_k)  (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
-
-  int    (*daxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*dscal_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*dswap_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-
-  int    (*dgemv_n) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*dgemv_t) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*dger_k)  (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-
-  int    (*dsymv_L) (BLASLONG, BLASLONG, double,  double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-  int    (*dsymv_U) (BLASLONG, BLASLONG, double,  double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-
-  int    (*dgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG);
-  int    (*dgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double  *, BLASLONG);
-
-  int    (*dgemm_incopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*dgemm_itcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*dgemm_oncopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*dgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-
-  int    (*dtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*dtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*dtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*dtrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-
-  int    (*dtrsm_iunucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_iunncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_iutucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_iutncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_ilnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_ilnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_iltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_iltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_ounucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_ounncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_outucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_outncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_olnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_olnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_oltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*dtrsm_oltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-
-  int    (*dtrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*dtrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*dtrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*dtrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
-
-  int    (*dtrmm_iunucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_iunncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_iutucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_iutncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_ilnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_ilnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_iltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_iltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_ounucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_ounncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_outucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_outncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_olnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_olnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_oltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dtrmm_oltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int    (*dsymm_iutcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dsymm_iltcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dsymm_outcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*dsymm_oltcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int	 (*dneg_tcopy)   (BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*dlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, double *, BLASLONG, blasint *, double *);
-
-#ifdef EXPRECISION
-
-  int qgemm_p, qgemm_q, qgemm_r;
-  int qgemm_unroll_m, qgemm_unroll_n, qgemm_unroll_mn;
-
- xdouble (*qamax_k) (BLASLONG, xdouble *, BLASLONG);
- xdouble (*qamin_k) (BLASLONG, xdouble *, BLASLONG);
- xdouble (*qmax_k)  (BLASLONG, xdouble *, BLASLONG);
- xdouble (*qmin_k)  (BLASLONG, xdouble *, BLASLONG);
-BLASLONG (*iqamax_k)(BLASLONG, xdouble *, BLASLONG);
-BLASLONG (*iqamin_k)(BLASLONG, xdouble *, BLASLONG);
-BLASLONG (*iqmax_k) (BLASLONG, xdouble *, BLASLONG);
-BLASLONG (*iqmin_k) (BLASLONG, xdouble *, BLASLONG);
-
- xdouble (*qnrm2_k) (BLASLONG, xdouble *, BLASLONG);
- xdouble (*qasum_k) (BLASLONG, xdouble *, BLASLONG);
- xdouble (*qsum_k)  (BLASLONG, xdouble *, BLASLONG);
-  int    (*qcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
- xdouble (*qdot_k)  (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*qrot_k)  (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
-
-  int    (*qaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*qscal_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*qswap_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-
-  int    (*qgemv_n) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*qgemv_t) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*qger_k)  (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-
-  int    (*qsymv_L) (BLASLONG, BLASLONG, xdouble,  xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-  int    (*qsymv_U) (BLASLONG, BLASLONG, xdouble,  xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-
-  int    (*qgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
-  int    (*qgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble  *, BLASLONG);
-
-  int    (*qgemm_incopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*qgemm_itcopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*qgemm_oncopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*qgemm_otcopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-
-  int    (*qtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*qtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*qtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*qtrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-
-  int    (*qtrsm_iunucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_iunncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_iutucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_iutncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_ilnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_ilnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_iltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_iltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_ounucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_ounncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_outucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_outncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_olnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_olnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_oltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrsm_oltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*qtrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*qtrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*qtrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*qtrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-
-  int    (*qtrmm_iunucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_iunncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_iutucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_iutncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_ilnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_ilnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_iltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_iltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_ounucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_ounncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_outucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_outncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_olnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_olnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_oltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qtrmm_oltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*qsymm_iutcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qsymm_iltcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qsymm_outcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*qsymm_oltcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int	 (*qneg_tcopy)   (BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*qlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, xdouble *, BLASLONG, blasint *, xdouble *);
-
-#endif
-
-  int cgemm_p, cgemm_q, cgemm_r;
-  int cgemm_unroll_m, cgemm_unroll_n, cgemm_unroll_mn;
-
-  float (*camax_k) (BLASLONG, float *, BLASLONG);
-  float (*camin_k) (BLASLONG, float *, BLASLONG);
-BLASLONG (*icamax_k)(BLASLONG, float *, BLASLONG);
-BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
-
-  float (*cnrm2_k) (BLASLONG, float *, BLASLONG);
-  float (*casum_k) (BLASLONG, float *, BLASLONG);
-  float (*csum_k)  (BLASLONG, float *, BLASLONG);
-  int    (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  openblas_complex_float (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  openblas_complex_float (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  int    (*csrot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
-
-  int    (*caxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  int    (*caxpyc_k)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  int    (*cscal_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  int    (*cswap_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
-
-  int    (*cgemv_n) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_t) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_r) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_c) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_o) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_u) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_s) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemv_d) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgeru_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgerc_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgerv_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgerd_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
-
-  int    (*csymv_L) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-  int    (*csymv_U) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-  int    (*chemv_L) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-  int    (*chemv_U) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-  int    (*chemv_M) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-  int    (*chemv_V) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
-
-  int    (*cgemm_kernel_n )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
-  int    (*cgemm_kernel_l )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
-  int    (*cgemm_kernel_r )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
-  int    (*cgemm_kernel_b )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
-  int    (*cgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float  *, BLASLONG);
-
-  int    (*cgemm_incopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm_itcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-
-  int    (*ctrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_LC)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_RR)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrsm_kernel_RC)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-
-  int    (*ctrsm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-  int    (*ctrsm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-
-  int    (*ctrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_RR)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_RC)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-  int    (*ctrmm_kernel_LC)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
-
-  int    (*ctrmm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*ctrmm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int    (*csymm_iutcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm_iltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm_outcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm_oltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int    (*chemm_iutcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm_iltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm_outcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm_oltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int cgemm3m_p, cgemm3m_q, cgemm3m_r;
-  int cgemm3m_unroll_m, cgemm3m_unroll_n, cgemm3m_unroll_mn;
-
-  int    (*cgemm3m_kernel)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
-
-  int    (*cgemm3m_incopyb)(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm3m_incopyr)(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm3m_incopyi)(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm3m_itcopyb)(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm3m_itcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*cgemm3m_itcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-
-  int    (*cgemm3m_oncopyb)(BLASLONG, BLASLONG, float *, BLASLONG, float, float, float *);
-  int    (*cgemm3m_oncopyr)(BLASLONG, BLASLONG, float *, BLASLONG, float, float, float *);
-  int    (*cgemm3m_oncopyi)(BLASLONG, BLASLONG, float *, BLASLONG, float, float, float *);
-  int    (*cgemm3m_otcopyb)(BLASLONG, BLASLONG, float *, BLASLONG, float, float, float *);
-  int    (*cgemm3m_otcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, float, float, float *);
-  int    (*cgemm3m_otcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, float, float, float *);
-
-  int    (*csymm3m_iucopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm3m_ilcopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm3m_iucopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm3m_ilcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm3m_iucopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*csymm3m_ilcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int    (*csymm3m_oucopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*csymm3m_olcopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*csymm3m_oucopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*csymm3m_olcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*csymm3m_oucopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*csymm3m_olcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-
-  int    (*chemm3m_iucopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm3m_ilcopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm3m_iucopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm3m_ilcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm3m_iucopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-  int    (*chemm3m_ilcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
-  int    (*chemm3m_oucopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*chemm3m_olcopyb)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*chemm3m_oucopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*chemm3m_olcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*chemm3m_oucopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-  int    (*chemm3m_olcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-
-  int	 (*cneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
-  int    (*claswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
-
-  int zgemm_p, zgemm_q, zgemm_r;
-  int zgemm_unroll_m, zgemm_unroll_n, zgemm_unroll_mn;
-
-  double (*zamax_k) (BLASLONG, double *, BLASLONG);
-  double (*zamin_k) (BLASLONG, double *, BLASLONG);
-BLASLONG (*izamax_k)(BLASLONG, double *, BLASLONG);
-BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
-
-  double (*znrm2_k) (BLASLONG, double *, BLASLONG);
-  double (*zasum_k) (BLASLONG, double *, BLASLONG);
-  double (*zsum_k)  (BLASLONG, double *, BLASLONG);
-  int    (*zcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  openblas_complex_double (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  openblas_complex_double (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*zdrot_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
-
-  int    (*zaxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*zaxpyc_k)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*zscal_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  int    (*zswap_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-
-  int    (*zgemv_n) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_t) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_r) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_c) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_o) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_u) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_s) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemv_d) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgeru_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgerc_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgerv_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgerd_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
-
-  int    (*zsymv_L) (BLASLONG, BLASLONG, double,  double, double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-  int    (*zsymv_U) (BLASLONG, BLASLONG, double,  double, double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-  int    (*zhemv_L) (BLASLONG, BLASLONG, double,  double, double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-  int    (*zhemv_U) (BLASLONG, BLASLONG, double,  double, double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-  int    (*zhemv_M) (BLASLONG, BLASLONG, double,  double, double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-  int    (*zhemv_V) (BLASLONG, BLASLONG, double,  double, double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-
-  int    (*zgemm_kernel_n )(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
-  int    (*zgemm_kernel_l )(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
-  int    (*zgemm_kernel_r )(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
-  int    (*zgemm_kernel_b )(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
-  int    (*zgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double  *, BLASLONG);
-
-  int    (*zgemm_incopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm_itcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm_oncopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-
-  int    (*ztrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_LC)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_RR)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrsm_kernel_RC)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-
-  int    (*ztrsm_iunucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_iunncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_iutucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_iutncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_ilnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_ilnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_iltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_iltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_ounucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_ounncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_outucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_outncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_olnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_olnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_oltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-  int    (*ztrsm_oltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, double *);
-
-  int    (*ztrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_RR)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_RC)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-  int    (*ztrmm_kernel_LC)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
-
-  int    (*ztrmm_iunucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_iunncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_iutucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_iutncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_ilnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_ilnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_iltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_iltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_ounucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_ounncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_outucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_outncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_olnucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_olnncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_oltucopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*ztrmm_oltncopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int    (*zsymm_iutcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm_iltcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm_outcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm_oltcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int    (*zhemm_iutcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm_iltcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm_outcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm_oltcopy)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int zgemm3m_p, zgemm3m_q, zgemm3m_r;
-  int zgemm3m_unroll_m, zgemm3m_unroll_n, zgemm3m_unroll_mn;
-
-  int    (*zgemm3m_kernel)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
-
-  int    (*zgemm3m_incopyb)(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm3m_incopyr)(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm3m_incopyi)(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm3m_itcopyb)(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm3m_itcopyr)(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zgemm3m_itcopyi)(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-
-  int    (*zgemm3m_oncopyb)(BLASLONG, BLASLONG, double *, BLASLONG, double, double, double *);
-  int    (*zgemm3m_oncopyr)(BLASLONG, BLASLONG, double *, BLASLONG, double, double, double *);
-  int    (*zgemm3m_oncopyi)(BLASLONG, BLASLONG, double *, BLASLONG, double, double, double *);
-  int    (*zgemm3m_otcopyb)(BLASLONG, BLASLONG, double *, BLASLONG, double, double, double *);
-  int    (*zgemm3m_otcopyr)(BLASLONG, BLASLONG, double *, BLASLONG, double, double, double *);
-  int    (*zgemm3m_otcopyi)(BLASLONG, BLASLONG, double *, BLASLONG, double, double, double *);
-
-  int    (*zsymm3m_iucopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm3m_ilcopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm3m_iucopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm3m_ilcopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm3m_iucopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zsymm3m_ilcopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int    (*zsymm3m_oucopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zsymm3m_olcopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zsymm3m_oucopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zsymm3m_olcopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zsymm3m_oucopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zsymm3m_olcopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-
-  int    (*zhemm3m_iucopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm3m_ilcopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm3m_iucopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm3m_ilcopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm3m_iucopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-  int    (*zhemm3m_ilcopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double *);
-
-  int    (*zhemm3m_oucopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zhemm3m_olcopyb)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zhemm3m_oucopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zhemm3m_olcopyr)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zhemm3m_oucopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-  int    (*zhemm3m_olcopyi)(BLASLONG, BLASLONG, double *, BLASLONG, BLASLONG, BLASLONG, double, double, double *);
-
-  int	 (*zneg_tcopy)   (BLASLONG, BLASLONG, double *, BLASLONG, double *);
-  int    (*zlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, double *, BLASLONG, blasint *, double *);
-
-#ifdef EXPRECISION
-
-  int xgemm_p, xgemm_q, xgemm_r;
-  int xgemm_unroll_m, xgemm_unroll_n, xgemm_unroll_mn;
-
-  xdouble (*xamax_k) (BLASLONG, xdouble *, BLASLONG);
-  xdouble (*xamin_k) (BLASLONG, xdouble *, BLASLONG);
-BLASLONG (*ixamax_k)(BLASLONG, xdouble *, BLASLONG);
-BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
-
-  xdouble (*xnrm2_k) (BLASLONG, xdouble *, BLASLONG);
-  xdouble (*xasum_k) (BLASLONG, xdouble *, BLASLONG);
-  xdouble (*xsum_k) (BLASLONG, xdouble *, BLASLONG);
-  int    (*xcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  openblas_complex_xdouble (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  openblas_complex_xdouble (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*xqrot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
-
-  int    (*xaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*xaxpyc_k)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*xscal_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  int    (*xswap_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-
-  int    (*xgemv_n) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_t) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_r) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_c) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_o) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_u) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_s) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemv_d) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgeru_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgerc_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgerv_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgerd_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *);
-
-  int    (*xsymv_L) (BLASLONG, BLASLONG, xdouble,  xdouble, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-  int    (*xsymv_U) (BLASLONG, BLASLONG, xdouble,  xdouble, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-  int    (*xhemv_L) (BLASLONG, BLASLONG, xdouble,  xdouble, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-  int    (*xhemv_U) (BLASLONG, BLASLONG, xdouble,  xdouble, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-  int    (*xhemv_M) (BLASLONG, BLASLONG, xdouble,  xdouble, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-  int    (*xhemv_V) (BLASLONG, BLASLONG, xdouble,  xdouble, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble  *, BLASLONG, xdouble *);
-
-  int    (*xgemm_kernel_n )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
-  int    (*xgemm_kernel_l )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
-  int    (*xgemm_kernel_r )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
-  int    (*xgemm_kernel_b )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
-  int    (*xgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble  *, BLASLONG);
-
-  int    (*xgemm_incopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm_itcopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm_oncopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm_otcopy   )(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-
-  int    (*xtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_LC)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_RR)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrsm_kernel_RC)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-
-  int    (*xtrsm_iunucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_iunncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_iutucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_iutncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_ilnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_ilnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_iltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_iltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_ounucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_ounncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_outucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_outncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_olnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_olnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_oltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrsm_oltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*xtrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_RR)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_RC)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-  int    (*xtrmm_kernel_LC)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG, BLASLONG);
-
-  int    (*xtrmm_iunucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_iunncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_iutucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_iutncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_ilnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_ilnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_iltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_iltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_ounucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_ounncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_outucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_outncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_olnucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_olnncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_oltucopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xtrmm_oltncopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*xsymm_iutcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm_iltcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm_outcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm_oltcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*xhemm_iutcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm_iltcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm_outcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm_oltcopy)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int xgemm3m_p, xgemm3m_q, xgemm3m_r;
-  int xgemm3m_unroll_m, xgemm3m_unroll_n, xgemm3m_unroll_mn;
-
-  int    (*xgemm3m_kernel)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
-
-  int    (*xgemm3m_incopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm3m_incopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm3m_incopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm3m_itcopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm3m_itcopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xgemm3m_itcopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-
-  int    (*xgemm3m_oncopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xgemm3m_oncopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xgemm3m_oncopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xgemm3m_otcopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xgemm3m_otcopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xgemm3m_otcopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble, xdouble *);
-
-  int    (*xsymm3m_iucopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm3m_ilcopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm3m_iucopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm3m_ilcopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm3m_iucopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xsymm3m_ilcopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*xsymm3m_oucopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xsymm3m_olcopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xsymm3m_oucopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xsymm3m_olcopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xsymm3m_oucopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xsymm3m_olcopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-
-  int    (*xhemm3m_iucopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm3m_ilcopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm3m_iucopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm3m_ilcopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm3m_iucopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-  int    (*xhemm3m_ilcopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble *);
-
-  int    (*xhemm3m_oucopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xhemm3m_olcopyb)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xhemm3m_oucopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xhemm3m_olcopyr)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xhemm3m_oucopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-  int    (*xhemm3m_olcopyi)(BLASLONG, BLASLONG, xdouble *, BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *);
-
-  int	 (*xneg_tcopy)   (BLASLONG, BLASLONG, xdouble *, BLASLONG, xdouble *);
-  int    (*xlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, xdouble *, BLASLONG, blasint *, xdouble *);
-
-#endif
-
-
-  void (*init)(void);
-
-  int snum_opt, dnum_opt, qnum_opt;
-
-  int    (*saxpby_k)     (BLASLONG, float, float*, BLASLONG,float, float*, BLASLONG);
-  int    (*daxpby_k)     (BLASLONG, double, double*, BLASLONG,double, double*, BLASLONG);
-  int    (*caxpby_k)     (BLASLONG, float, float,  float*, BLASLONG,float,float, float*, BLASLONG);
-  int    (*zaxpby_k)     (BLASLONG, double, double,  double*, BLASLONG,double,double, double*, BLASLONG);
-
-  int    (*somatcopy_k_cn)	(BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*somatcopy_k_ct)	(BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*somatcopy_k_rn)	(BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*somatcopy_k_rt)	(BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
-
-  int    (*domatcopy_k_cn)	(BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*domatcopy_k_ct)	(BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*domatcopy_k_rn)	(BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*domatcopy_k_rt)	(BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
-
-  int    (*comatcopy_k_cn)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*comatcopy_k_ct)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*comatcopy_k_rn)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*comatcopy_k_rt)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-
-  int    (*comatcopy_k_cnc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*comatcopy_k_ctc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*comatcopy_k_rnc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-  int    (*comatcopy_k_rtc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
-
-  int    (*zomatcopy_k_cn)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*zomatcopy_k_ct)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*zomatcopy_k_rn)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*zomatcopy_k_rt)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-
-  int    (*zomatcopy_k_cnc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*zomatcopy_k_ctc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*zomatcopy_k_rnc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-  int    (*zomatcopy_k_rtc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
-
-  int    (*simatcopy_k_cn)	(BLASLONG, BLASLONG, float, float*, BLASLONG);
-  int    (*simatcopy_k_ct)	(BLASLONG, BLASLONG, float, float*, BLASLONG);
-  int    (*simatcopy_k_rn)	(BLASLONG, BLASLONG, float, float*, BLASLONG);
-  int    (*simatcopy_k_rt)	(BLASLONG, BLASLONG, float, float*, BLASLONG);
-
-  int    (*dimatcopy_k_cn)	(BLASLONG, BLASLONG, double, double*, BLASLONG);
-  int    (*dimatcopy_k_ct)	(BLASLONG, BLASLONG, double, double*, BLASLONG);
-  int    (*dimatcopy_k_rn)	(BLASLONG, BLASLONG, double, double*, BLASLONG);
-  int    (*dimatcopy_k_rt)	(BLASLONG, BLASLONG, double, double*, BLASLONG);
-
-  int    (*cimatcopy_k_cn)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-  int    (*cimatcopy_k_ct)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-  int    (*cimatcopy_k_rn)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-  int    (*cimatcopy_k_rt)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-
-  int    (*cimatcopy_k_cnc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-  int    (*cimatcopy_k_ctc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-  int    (*cimatcopy_k_rnc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-  int    (*cimatcopy_k_rtc)	(BLASLONG, BLASLONG, float, float, float*, BLASLONG);
-
-  int    (*zimatcopy_k_cn)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-  int    (*zimatcopy_k_ct)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-  int    (*zimatcopy_k_rn)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-  int    (*zimatcopy_k_rt)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-
-  int    (*zimatcopy_k_cnc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-  int    (*zimatcopy_k_ctc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-  int    (*zimatcopy_k_rnc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-  int    (*zimatcopy_k_rtc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG);
-
-  int    (*sgeadd_k) (BLASLONG, BLASLONG, float, float *, BLASLONG, float, float *, BLASLONG); 
-  int    (*dgeadd_k) (BLASLONG, BLASLONG, double, double *, BLASLONG, double, double *, BLASLONG); 
-  int    (*cgeadd_k) (BLASLONG, BLASLONG, float, float,  float *,  BLASLONG, float, float, float *, BLASLONG); 
-  int    (*zgeadd_k) (BLASLONG, BLASLONG, double, double, double *, BLASLONG, double, double, double *, BLASLONG); 
-
-} gotoblas_t;
-
-extern gotoblas_t *gotoblas;
-
-#define DTB_ENTRIES  gotoblas -> dtb_entries
-#define GEMM_OFFSET_A	gotoblas -> offsetA
-#define GEMM_OFFSET_B	gotoblas -> offsetB
-#define GEMM_ALIGN	gotoblas -> align
-
-#define HAVE_EX_L2	gotoblas -> exclusive_cache
-
-#define	SGEMM_P		gotoblas -> sgemm_p
-#define	SGEMM_Q		gotoblas -> sgemm_q
-#define	SGEMM_R		gotoblas -> sgemm_r
-#define	SGEMM_UNROLL_M	gotoblas -> sgemm_unroll_m
-#define	SGEMM_UNROLL_N	gotoblas -> sgemm_unroll_n
-#define SGEMM_UNROLL_MN	gotoblas -> sgemm_unroll_mn
-
-#define	DGEMM_P		gotoblas -> dgemm_p
-#define	DGEMM_Q		gotoblas -> dgemm_q
-#define	DGEMM_R		gotoblas -> dgemm_r
-#define	DGEMM_UNROLL_M	gotoblas -> dgemm_unroll_m
-#define	DGEMM_UNROLL_N	gotoblas -> dgemm_unroll_n
-#define DGEMM_UNROLL_MN	gotoblas -> dgemm_unroll_mn
-
-#define	QGEMM_P		gotoblas -> qgemm_p
-#define	QGEMM_Q		gotoblas -> qgemm_q
-#define	QGEMM_R		gotoblas -> qgemm_r
-#define	QGEMM_UNROLL_M	gotoblas -> qgemm_unroll_m
-#define	QGEMM_UNROLL_N	gotoblas -> qgemm_unroll_n
-#define QGEMM_UNROLL_MN	gotoblas -> qgemm_unroll_mn
-
-#define	CGEMM_P		gotoblas -> cgemm_p
-#define	CGEMM_Q		gotoblas -> cgemm_q
-#define	CGEMM_R		gotoblas -> cgemm_r
-#define	CGEMM_UNROLL_M	gotoblas -> cgemm_unroll_m
-#define	CGEMM_UNROLL_N	gotoblas -> cgemm_unroll_n
-#define CGEMM_UNROLL_MN	gotoblas -> cgemm_unroll_mn
-
-#define	ZGEMM_P		gotoblas -> zgemm_p
-#define	ZGEMM_Q		gotoblas -> zgemm_q
-#define	ZGEMM_R		gotoblas -> zgemm_r
-#define	ZGEMM_UNROLL_M	gotoblas -> zgemm_unroll_m
-#define	ZGEMM_UNROLL_N	gotoblas -> zgemm_unroll_n
-#define ZGEMM_UNROLL_MN	gotoblas -> zgemm_unroll_mn
-
-#define	XGEMM_P		gotoblas -> xgemm_p
-#define	XGEMM_Q		gotoblas -> xgemm_q
-#define	XGEMM_R		gotoblas -> xgemm_r
-#define	XGEMM_UNROLL_M	gotoblas -> xgemm_unroll_m
-#define	XGEMM_UNROLL_N	gotoblas -> xgemm_unroll_n
-#define XGEMM_UNROLL_MN	gotoblas -> xgemm_unroll_mn
-
-#define	CGEMM3M_P		gotoblas -> cgemm3m_p
-#define	CGEMM3M_Q		gotoblas -> cgemm3m_q
-#define	CGEMM3M_R		gotoblas -> cgemm3m_r
-#define	CGEMM3M_UNROLL_M	gotoblas -> cgemm3m_unroll_m
-#define	CGEMM3M_UNROLL_N	gotoblas -> cgemm3m_unroll_n
-#define CGEMM3M_UNROLL_MN	gotoblas -> cgemm3m_unroll_mn
-
-#define	ZGEMM3M_P		gotoblas -> zgemm3m_p
-#define	ZGEMM3M_Q		gotoblas -> zgemm3m_q
-#define	ZGEMM3M_R		gotoblas -> zgemm3m_r
-#define	ZGEMM3M_UNROLL_M	gotoblas -> zgemm3m_unroll_m
-#define	ZGEMM3M_UNROLL_N	gotoblas -> zgemm3m_unroll_n
-#define ZGEMM3M_UNROLL_MN	gotoblas -> zgemm3m_unroll_mn
-
-#define	XGEMM3M_P		gotoblas -> xgemm3m_p
-#define	XGEMM3M_Q		gotoblas -> xgemm3m_q
-#define	XGEMM3M_R		gotoblas -> xgemm3m_r
-#define	XGEMM3M_UNROLL_M	gotoblas -> xgemm3m_unroll_m
-#define	XGEMM3M_UNROLL_N	gotoblas -> xgemm3m_unroll_n
-#define XGEMM3M_UNROLL_MN	gotoblas -> xgemm3m_unroll_mn
-
-#else
-
-#define DTB_ENTRIES  DTB_DEFAULT_ENTRIES
-
-#define GEMM_OFFSET_A	GEMM_DEFAULT_OFFSET_A
-#define GEMM_OFFSET_B	GEMM_DEFAULT_OFFSET_B
-#define GEMM_ALIGN	GEMM_DEFAULT_ALIGN
-
-#ifdef HAVE_EXCLUSIVE_CACHE
-#define HAVE_EX_L2	1
-#else
-#define HAVE_EX_L2	0
-#endif
-
-#define	SGEMM_P		SGEMM_DEFAULT_P
-#define	SGEMM_Q		SGEMM_DEFAULT_Q
-#define	SGEMM_R		SGEMM_DEFAULT_R
-#define SGEMM_UNROLL_M	SGEMM_DEFAULT_UNROLL_M
-#define SGEMM_UNROLL_N	SGEMM_DEFAULT_UNROLL_N
-#ifdef  SGEMM_DEFAULT_UNROLL_MN
-#define SGEMM_UNROLL_MN	SGEMM_DEFAULT_UNROLL_MN
-#else
-#define SGEMM_UNROLL_MN	MAX((SGEMM_UNROLL_M), (SGEMM_UNROLL_N))
-#endif
-
-#define	DGEMM_P		DGEMM_DEFAULT_P
-#define	DGEMM_Q		DGEMM_DEFAULT_Q
-#define	DGEMM_R		DGEMM_DEFAULT_R
-#define DGEMM_UNROLL_M	DGEMM_DEFAULT_UNROLL_M
-#define DGEMM_UNROLL_N	DGEMM_DEFAULT_UNROLL_N
-#ifdef  DGEMM_DEFAULT_UNROLL_MN
-#define DGEMM_UNROLL_MN	DGEMM_DEFAULT_UNROLL_MN
-#else
-#define DGEMM_UNROLL_MN	MAX((DGEMM_UNROLL_M), (DGEMM_UNROLL_N))
-#endif
-
-#define	QGEMM_P		QGEMM_DEFAULT_P
-#define	QGEMM_Q		QGEMM_DEFAULT_Q
-#define	QGEMM_R		QGEMM_DEFAULT_R
-#define QGEMM_UNROLL_M	QGEMM_DEFAULT_UNROLL_M
-#define QGEMM_UNROLL_N	QGEMM_DEFAULT_UNROLL_N
-#define QGEMM_UNROLL_MN	MAX((QGEMM_UNROLL_M), (QGEMM_UNROLL_N))
-
-#define	CGEMM_P		CGEMM_DEFAULT_P
-#define	CGEMM_Q		CGEMM_DEFAULT_Q
-#define	CGEMM_R		CGEMM_DEFAULT_R
-#define CGEMM_UNROLL_M	CGEMM_DEFAULT_UNROLL_M
-#define CGEMM_UNROLL_N	CGEMM_DEFAULT_UNROLL_N
-#ifdef  CGEMM_DEFAULT_UNROLL_MN
-#define CGEMM_UNROLL_MN	CGEMM_DEFAULT_UNROLL_MN
-#else
-#define CGEMM_UNROLL_MN	MAX((CGEMM_UNROLL_M), (CGEMM_UNROLL_N))
-#endif
-
-#define	ZGEMM_P		ZGEMM_DEFAULT_P
-#define	ZGEMM_Q		ZGEMM_DEFAULT_Q
-#define	ZGEMM_R		ZGEMM_DEFAULT_R
-#define ZGEMM_UNROLL_M	ZGEMM_DEFAULT_UNROLL_M
-#define ZGEMM_UNROLL_N	ZGEMM_DEFAULT_UNROLL_N
-#ifdef  ZGEMM_DEFAULT_UNROLL_MN
-#define ZGEMM_UNROLL_MN	ZGEMM_DEFAULT_UNROLL_MN
-#else
-#define ZGEMM_UNROLL_MN	MAX((ZGEMM_UNROLL_M), (ZGEMM_UNROLL_N))
-#endif
-
-#define	XGEMM_P		XGEMM_DEFAULT_P
-#define	XGEMM_Q		XGEMM_DEFAULT_Q
-#define	XGEMM_R		XGEMM_DEFAULT_R
-#define XGEMM_UNROLL_M	XGEMM_DEFAULT_UNROLL_M
-#define XGEMM_UNROLL_N	XGEMM_DEFAULT_UNROLL_N
-#define XGEMM_UNROLL_MN	MAX((XGEMM_UNROLL_M), (XGEMM_UNROLL_N))
-
-#ifdef CGEMM3M_DEFAULT_UNROLL_N
-
-#define	CGEMM3M_P		CGEMM3M_DEFAULT_P
-#define	CGEMM3M_Q		CGEMM3M_DEFAULT_Q
-#define	CGEMM3M_R		CGEMM3M_DEFAULT_R
-#define CGEMM3M_UNROLL_M	CGEMM3M_DEFAULT_UNROLL_M
-#define CGEMM3M_UNROLL_N	CGEMM3M_DEFAULT_UNROLL_N
-#define CGEMM3M_UNROLL_MN	MAX((CGEMM3M_UNROLL_M), (CGEMM3M_UNROLL_N))
-
-#else
-
-#define	CGEMM3M_P		SGEMM_DEFAULT_P
-#define	CGEMM3M_Q		SGEMM_DEFAULT_Q
-#define	CGEMM3M_R		SGEMM_DEFAULT_R
-#define CGEMM3M_UNROLL_M	SGEMM_DEFAULT_UNROLL_M
-#define CGEMM3M_UNROLL_N	SGEMM_DEFAULT_UNROLL_N
-#define CGEMM3M_UNROLL_MN	MAX((CGEMM_UNROLL_M), (CGEMM_UNROLL_N))
-
-#endif
-
-
-#ifdef ZGEMM3M_DEFAULT_UNROLL_N
-
-#define	ZGEMM3M_P		ZGEMM3M_DEFAULT_P
-#define	ZGEMM3M_Q		ZGEMM3M_DEFAULT_Q
-#define	ZGEMM3M_R		ZGEMM3M_DEFAULT_R
-#define ZGEMM3M_UNROLL_M	ZGEMM3M_DEFAULT_UNROLL_M
-#define ZGEMM3M_UNROLL_N	ZGEMM3M_DEFAULT_UNROLL_N
-#define ZGEMM3M_UNROLL_MN	MAX((ZGEMM_UNROLL_M), (ZGEMM_UNROLL_N))
-
-#else
-
-#define	ZGEMM3M_P		DGEMM_DEFAULT_P
-#define	ZGEMM3M_Q		DGEMM_DEFAULT_Q
-#define	ZGEMM3M_R		DGEMM_DEFAULT_R
-#define ZGEMM3M_UNROLL_M	DGEMM_DEFAULT_UNROLL_M
-#define ZGEMM3M_UNROLL_N	DGEMM_DEFAULT_UNROLL_N
-#define ZGEMM3M_UNROLL_MN	MAX((ZGEMM_UNROLL_M), (ZGEMM_UNROLL_N))
-
-#endif
-
-#define	XGEMM3M_P		QGEMM_DEFAULT_P
-#define	XGEMM3M_Q		QGEMM_DEFAULT_Q
-#define	XGEMM3M_R		QGEMM_DEFAULT_R
-#define XGEMM3M_UNROLL_M	QGEMM_DEFAULT_UNROLL_M
-#define XGEMM3M_UNROLL_N	QGEMM_DEFAULT_UNROLL_N
-#define XGEMM3M_UNROLL_MN	MAX((QGEMM_UNROLL_M), (QGEMM_UNROLL_N))
-
-
-#endif
-#endif
-
-#ifndef COMPLEX
-#if   defined(XDOUBLE)
-#define GEMM_P			QGEMM_P
-#define GEMM_Q			QGEMM_Q
-#define GEMM_R			QGEMM_R
-#define GEMM_UNROLL_M		QGEMM_UNROLL_M
-#define GEMM_UNROLL_N		QGEMM_UNROLL_N
-#define GEMM_UNROLL_MN		QGEMM_UNROLL_MN
-#define GEMM_DEFAULT_P		QGEMM_DEFAULT_P
-#define GEMM_DEFAULT_Q		QGEMM_DEFAULT_Q
-#define GEMM_DEFAULT_R		QGEMM_DEFAULT_R
-#define GEMM_DEFAULT_UNROLL_M	QGEMM_DEFAULT_UNROLL_M
-#define GEMM_DEFAULT_UNROLL_N	QGEMM_DEFAULT_UNROLL_N
-#elif defined(DOUBLE)
-#define GEMM_P			DGEMM_P
-#define GEMM_Q			DGEMM_Q
-#define GEMM_R			DGEMM_R
-#define GEMM_UNROLL_M		DGEMM_UNROLL_M
-#define GEMM_UNROLL_N		DGEMM_UNROLL_N
-#define GEMM_UNROLL_MN		DGEMM_UNROLL_MN
-#define GEMM_DEFAULT_P		DGEMM_DEFAULT_P
-#define GEMM_DEFAULT_Q		DGEMM_DEFAULT_Q
-#define GEMM_DEFAULT_R		DGEMM_DEFAULT_R
-#define GEMM_DEFAULT_UNROLL_M	DGEMM_DEFAULT_UNROLL_M
-#define GEMM_DEFAULT_UNROLL_N	DGEMM_DEFAULT_UNROLL_N
-#else
-#define GEMM_P			SGEMM_P
-#define GEMM_Q			SGEMM_Q
-#define GEMM_R			SGEMM_R
-#define GEMM_UNROLL_M		SGEMM_UNROLL_M
-#define GEMM_UNROLL_N		SGEMM_UNROLL_N
-#define GEMM_UNROLL_MN		SGEMM_UNROLL_MN
-#define GEMM_DEFAULT_P		SGEMM_DEFAULT_P
-#define GEMM_DEFAULT_Q		SGEMM_DEFAULT_Q
-#define GEMM_DEFAULT_R		SGEMM_DEFAULT_R
-#define GEMM_DEFAULT_UNROLL_M	SGEMM_DEFAULT_UNROLL_M
-#define GEMM_DEFAULT_UNROLL_N	SGEMM_DEFAULT_UNROLL_N
-#endif
-#else
-#if   defined(XDOUBLE)
-#define GEMM_P			XGEMM_P
-#define GEMM_Q			XGEMM_Q
-#define GEMM_R			XGEMM_R
-#define GEMM_UNROLL_M		XGEMM_UNROLL_M
-#define GEMM_UNROLL_N		XGEMM_UNROLL_N
-#define GEMM_UNROLL_MN		XGEMM_UNROLL_MN
-#define GEMM_DEFAULT_P		XGEMM_DEFAULT_P
-#define GEMM_DEFAULT_Q		XGEMM_DEFAULT_Q
-#define GEMM_DEFAULT_R		XGEMM_DEFAULT_R
-#define GEMM_DEFAULT_UNROLL_M	XGEMM_DEFAULT_UNROLL_M
-#define GEMM_DEFAULT_UNROLL_N	XGEMM_DEFAULT_UNROLL_N
-#elif defined(DOUBLE)
-#define GEMM_P			ZGEMM_P
-#define GEMM_Q			ZGEMM_Q
-#define GEMM_R			ZGEMM_R
-#define GEMM_UNROLL_M		ZGEMM_UNROLL_M
-#define GEMM_UNROLL_N		ZGEMM_UNROLL_N
-#define GEMM_UNROLL_MN		ZGEMM_UNROLL_MN
-#define GEMM_DEFAULT_P		ZGEMM_DEFAULT_P
-#define GEMM_DEFAULT_Q		ZGEMM_DEFAULT_Q
-#define GEMM_DEFAULT_R		ZGEMM_DEFAULT_R
-#define GEMM_DEFAULT_UNROLL_M	ZGEMM_DEFAULT_UNROLL_M
-#define GEMM_DEFAULT_UNROLL_N	ZGEMM_DEFAULT_UNROLL_N
-#else
-#define GEMM_P			CGEMM_P
-#define GEMM_Q			CGEMM_Q
-#define GEMM_R			CGEMM_R
-#define GEMM_UNROLL_M		CGEMM_UNROLL_M
-#define GEMM_UNROLL_N		CGEMM_UNROLL_N
-#define GEMM_UNROLL_MN		CGEMM_UNROLL_MN
-#define GEMM_DEFAULT_P		CGEMM_DEFAULT_P
-#define GEMM_DEFAULT_Q		CGEMM_DEFAULT_Q
-#define GEMM_DEFAULT_R		CGEMM_DEFAULT_R
-#define GEMM_DEFAULT_UNROLL_M	CGEMM_DEFAULT_UNROLL_M
-#define GEMM_DEFAULT_UNROLL_N	CGEMM_DEFAULT_UNROLL_N
-#endif
-#endif
-
-#ifdef XDOUBLE
-#define GEMM3M_UNROLL_M	XGEMM3M_UNROLL_M
-#define GEMM3M_UNROLL_N	XGEMM3M_UNROLL_N
-#elif defined(DOUBLE)
-#define GEMM3M_UNROLL_M	ZGEMM3M_UNROLL_M
-#define GEMM3M_UNROLL_N	ZGEMM3M_UNROLL_N
-#else
-#define GEMM3M_UNROLL_M	CGEMM3M_UNROLL_M
-#define GEMM3M_UNROLL_N	CGEMM3M_UNROLL_N
-#endif
-
-
-#ifndef QGEMM_DEFAULT_UNROLL_M
-#define QGEMM_DEFAULT_UNROLL_M 2
-#endif
-
-#ifndef QGEMM_DEFAULT_UNROLL_N
-#define QGEMM_DEFAULT_UNROLL_N 2
-#endif
-
-#ifndef XGEMM_DEFAULT_UNROLL_M
-#define XGEMM_DEFAULT_UNROLL_M 2
-#endif
-
-#ifndef XGEMM_DEFAULT_UNROLL_N
-#define XGEMM_DEFAULT_UNROLL_N 2
-#endif
-
-#ifndef GEMM_THREAD
-#define GEMM_THREAD gemm_thread_n
-#endif
-
-#ifndef SGEMM_DEFAULT_R
-#define SGEMM_DEFAULT_R (((BUFFER_SIZE - ((SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SGEMM_DEFAULT_Q *  4) - 15) & ~15UL)
-#endif
-
-#ifndef DGEMM_DEFAULT_R
-#define DGEMM_DEFAULT_R (((BUFFER_SIZE - ((DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (DGEMM_DEFAULT_Q *  8) - 15) & ~15UL)
-#endif
-
-#ifndef QGEMM_DEFAULT_R
-#define QGEMM_DEFAULT_R (((BUFFER_SIZE - ((QGEMM_DEFAULT_P * QGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (QGEMM_DEFAULT_Q * 16) - 15) & ~15UL)
-#endif
-
-#ifndef CGEMM_DEFAULT_R
-#define CGEMM_DEFAULT_R (((BUFFER_SIZE - ((CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (CGEMM_DEFAULT_Q *  8) - 15) & ~15UL)
-#endif
-
-#ifndef ZGEMM_DEFAULT_R
-#define ZGEMM_DEFAULT_R (((BUFFER_SIZE - ((ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (ZGEMM_DEFAULT_Q * 16) - 15) & ~15UL)
-#endif
-
-#ifndef XGEMM_DEFAULT_R
-#define XGEMM_DEFAULT_R (((BUFFER_SIZE - ((XGEMM_DEFAULT_P * XGEMM_DEFAULT_Q * 32 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (XGEMM_DEFAULT_Q * 32) - 15) & ~15UL)
-#endif
-
-#ifndef SNUMOPT
-#define SNUMOPT		2
-#endif
-
-#ifndef DNUMOPT
-#define DNUMOPT		2
-#endif
-
-#ifndef QNUMOPT
-#define QNUMOPT		1
-#endif
-
-#ifndef GEMM3M_P
-#ifdef XDOUBLE
-#define GEMM3M_P	XGEMM3M_P
-#elif defined(DOUBLE)
-#define GEMM3M_P	ZGEMM3M_P
-#else
-#define GEMM3M_P	CGEMM3M_P
-#endif
-#endif
-
-#ifndef GEMM3M_Q
-#ifdef XDOUBLE
-#define GEMM3M_Q	XGEMM3M_Q
-#elif defined(DOUBLE)
-#define GEMM3M_Q	ZGEMM3M_Q
-#else
-#define GEMM3M_Q	CGEMM3M_Q
-#endif
-#endif
-
-#ifndef GEMM3M_R
-#ifdef XDOUBLE
-#define GEMM3M_R	XGEMM3M_R
-#elif defined(DOUBLE)
-#define GEMM3M_R	ZGEMM3M_R
-#else
-#define GEMM3M_R	CGEMM3M_R
-#endif
-#endif
-
-
-#endif
diff --git a/kernel/generic/gemm_beta.c b/kernel/generic/gemm_beta.c
index fa9d7680d..ccb772cc7 100644
--- a/kernel/generic/gemm_beta.c
+++ b/kernel/generic/gemm_beta.c
@@ -39,7 +39,7 @@
 #include "common.h"
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT beta,
-	  FLOAT *dummy2, BLASLONG dummy3, FLOAT *dummy4, BLASLONG dummy5,
+	  IFLOAT *dummy2, BLASLONG dummy3, IFLOAT *dummy4, BLASLONG dummy5,
 	  FLOAT *c, BLASLONG ldc){
 
 
diff --git a/kernel/generic/gemm_ncopy_16.c b/kernel/generic/gemm_ncopy_16.c
index 5f91d0dbe..d3ab46472 100644
--- a/kernel/generic/gemm_ncopy_16.c
+++ b/kernel/generic/gemm_ncopy_16.c
@@ -39,24 +39,24 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
-  FLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
-  FLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12;
-  FLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
+  IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
+  IFLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12;
+  IFLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16;
 
-  FLOAT *boffset;
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT *boffset;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
 
   aoffset = a;
   boffset = b;
diff --git a/kernel/generic/gemm_ncopy_2.c b/kernel/generic/gemm_ncopy_2.c
index b728c713f..415860f81 100644
--- a/kernel/generic/gemm_ncopy_2.c
+++ b/kernel/generic/gemm_ncopy_2.c
@@ -39,10 +39,10 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
-  FLOAT *a_offset, *a_offset1, *a_offset2;
-  FLOAT *b_offset;
+  IFLOAT *a_offset, *a_offset1, *a_offset2;
+  IFLOAT *b_offset;
 
   a_offset = a;
   b_offset = b;
diff --git a/kernel/generic/gemm_ncopy_8.c b/kernel/generic/gemm_ncopy_8.c
index a49a778e6..aaf9c8917 100644
--- a/kernel/generic/gemm_ncopy_8.c
+++ b/kernel/generic/gemm_ncopy_8.c
@@ -39,30 +39,30 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
-  FLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
+  IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
 
-  FLOAT *boffset;
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
-  FLOAT ctemp33, ctemp34, ctemp35, ctemp36;
-  FLOAT ctemp37, ctemp38, ctemp39, ctemp40;
-  FLOAT ctemp41, ctemp42, ctemp43, ctemp44;
-  FLOAT ctemp45, ctemp46, ctemp47, ctemp48;
-  FLOAT ctemp49, ctemp50, ctemp51, ctemp52;
-  FLOAT ctemp53, ctemp54, ctemp55, ctemp56;
-  FLOAT ctemp57, ctemp58, ctemp59, ctemp60;
-  FLOAT ctemp61, ctemp62, ctemp63, ctemp64;
+  IFLOAT *boffset;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT ctemp33, ctemp34, ctemp35, ctemp36;
+  IFLOAT ctemp37, ctemp38, ctemp39, ctemp40;
+  IFLOAT ctemp41, ctemp42, ctemp43, ctemp44;
+  IFLOAT ctemp45, ctemp46, ctemp47, ctemp48;
+  IFLOAT ctemp49, ctemp50, ctemp51, ctemp52;
+  IFLOAT ctemp53, ctemp54, ctemp55, ctemp56;
+  IFLOAT ctemp57, ctemp58, ctemp59, ctemp60;
+  IFLOAT ctemp61, ctemp62, ctemp63, ctemp64;
 
 
   aoffset = a;
diff --git a/kernel/generic/gemm_tcopy_16.c b/kernel/generic/gemm_tcopy_16.c
index 56268ebf2..14252599a 100644
--- a/kernel/generic/gemm_tcopy_16.c
+++ b/kernel/generic/gemm_tcopy_16.c
@@ -39,22 +39,22 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
 
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2;
-  FLOAT *boffset;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2;
+  IFLOAT *boffset;
 
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
 
   aoffset   = a;
   boffset   = b;
diff --git a/kernel/generic/gemm_tcopy_2.c b/kernel/generic/gemm_tcopy_2.c
index 5695b13c2..b4aa4de57 100644
--- a/kernel/generic/gemm_tcopy_2.c
+++ b/kernel/generic/gemm_tcopy_2.c
@@ -39,11 +39,11 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
 
-  FLOAT *a_offset, *a_offset1, *a_offset2;
-  FLOAT *b_offset, *b_offset1, *b_offset2;
+  IFLOAT *a_offset, *a_offset1, *a_offset2;
+  IFLOAT *b_offset, *b_offset1, *b_offset2;
 
   a_offset = a;
   b_offset = b;
diff --git a/kernel/generic/gemm_tcopy_8.c b/kernel/generic/gemm_tcopy_8.c
index b28f3d219..3e8a839db 100644
--- a/kernel/generic/gemm_tcopy_8.c
+++ b/kernel/generic/gemm_tcopy_8.c
@@ -39,32 +39,32 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
 
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
-  FLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
+  IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
 
-  FLOAT *boffset,  *boffset1, *boffset2, *boffset3, *boffset4;
+  IFLOAT *boffset,  *boffset1, *boffset2, *boffset3, *boffset4;
 
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
-  FLOAT ctemp33, ctemp34, ctemp35, ctemp36;
-  FLOAT ctemp37, ctemp38, ctemp39, ctemp40;
-  FLOAT ctemp41, ctemp42, ctemp43, ctemp44;
-  FLOAT ctemp45, ctemp46, ctemp47, ctemp48;
-  FLOAT ctemp49, ctemp50, ctemp51, ctemp52;
-  FLOAT ctemp53, ctemp54, ctemp55, ctemp56;
-  FLOAT ctemp57, ctemp58, ctemp59, ctemp60;
-  FLOAT ctemp61, ctemp62, ctemp63, ctemp64;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT ctemp33, ctemp34, ctemp35, ctemp36;
+  IFLOAT ctemp37, ctemp38, ctemp39, ctemp40;
+  IFLOAT ctemp41, ctemp42, ctemp43, ctemp44;
+  IFLOAT ctemp45, ctemp46, ctemp47, ctemp48;
+  IFLOAT ctemp49, ctemp50, ctemp51, ctemp52;
+  IFLOAT ctemp53, ctemp54, ctemp55, ctemp56;
+  IFLOAT ctemp57, ctemp58, ctemp59, ctemp60;
+  IFLOAT ctemp61, ctemp62, ctemp63, ctemp64;
 
   aoffset   = a;
   boffset   = b;
diff --git a/kernel/generic/gemmkernel_2x2.c b/kernel/generic/gemmkernel_2x2.c
index 01f1c67b5..cc7bb8e48 100644
--- a/kernel/generic/gemmkernel_2x2.c
+++ b/kernel/generic/gemmkernel_2x2.c
@@ -1,13 +1,32 @@
 #include "common.h"
-int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#if defined(HALF) && defined(HALFCONVERSION)
+static float
+bfloat16tof32 (bfloat16 f16)
+{
+  float result = 0;
+  unsigned short* q = (unsigned short*)(&result);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  q[0] = f16;
+#else
+  q[1] = f16;
+#endif
+  return result;
+}
+#define BF16TOF32(x) (bfloat16tof32(x))
+#else
+#define BF16TOF32(x) x
+#endif
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,IFLOAT* ba,IFLOAT* bb,FLOAT* C,BLASLONG ldc
 #ifdef TRMMKERNEL
 		,BLASLONG offset
 #endif
 		)
 {
    BLASLONG i,j,k;
-   FLOAT *C0,*C1,*ptrba,*ptrbb;
-   FLOAT res0,res1,res2,res3,load0,load1,load2,load3,load4,load5,load6,load7;
+   FLOAT *C0,*C1;
+   IFLOAT *ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3;
+   IFLOAT load0,load1,load2,load3,load4,load5,load6,load7;
    for (j=0; j<bn/2; j+=1)
      {
         C0 = C;
@@ -24,36 +43,36 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[2*0+0];
                   load1 = ptrbb[2*0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*0+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   load3 = ptrbb[2*0+1];
-                  res2 = res2+load0*load3;
-                  res3 = res3+load2*load3;
+                  res2 = res2+BF16TOF32(load0)*BF16TOF32(load3);
+                  res3 = res3+BF16TOF32(load2)*BF16TOF32(load3);
                   load4 = ptrba[2*1+0];
                   load5 = ptrbb[2*1+0];
-                  res0 = res0+load4*load5;
+                  res0 = res0+BF16TOF32(load4)*BF16TOF32(load5);
                   load6 = ptrba[2*1+1];
-                  res1 = res1+load6*load5;
+                  res1 = res1+BF16TOF32(load6)*BF16TOF32(load5);
                   load7 = ptrbb[2*1+1];
-                  res2 = res2+load4*load7;
-                  res3 = res3+load6*load7;
+                  res2 = res2+BF16TOF32(load4)*BF16TOF32(load7);
+                  res3 = res3+BF16TOF32(load6)*BF16TOF32(load7);
                   load0 = ptrba[2*2+0];
                   load1 = ptrbb[2*2+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*2+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   load3 = ptrbb[2*2+1];
-                  res2 = res2+load0*load3;
-                  res3 = res3+load2*load3;
+                  res2 = res2+BF16TOF32(load0)*BF16TOF32(load3);
+                  res3 = res3+BF16TOF32(load2)*BF16TOF32(load3);
                   load4 = ptrba[2*3+0];
                   load5 = ptrbb[2*3+0];
-                  res0 = res0+load4*load5;
+                  res0 = res0+BF16TOF32(load4)*BF16TOF32(load5);
                   load6 = ptrba[2*3+1];
-                  res1 = res1+load6*load5;
+                  res1 = res1+BF16TOF32(load6)*BF16TOF32(load5);
                   load7 = ptrbb[2*3+1];
-                  res2 = res2+load4*load7;
-                  res3 = res3+load6*load7;
+                  res2 = res2+BF16TOF32(load4)*BF16TOF32(load7);
+                  res3 = res3+BF16TOF32(load6)*BF16TOF32(load7);
                   ptrba = ptrba+8;
                   ptrbb = ptrbb+8;
                }
@@ -61,12 +80,12 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[2*0+0];
                   load1 = ptrbb[2*0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*0+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   load3 = ptrbb[2*0+1];
-                  res2 = res2+load0*load3;
-                  res3 = res3+load2*load3;
+                  res2 = res2+BF16TOF32(load0)*BF16TOF32(load3);
+                  res3 = res3+BF16TOF32(load2)*BF16TOF32(load3);
                   ptrba = ptrba+2;
                   ptrbb = ptrbb+2;
                }
@@ -90,9 +109,9 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[0+0];
                   load1 = ptrbb[2*0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrbb[2*0+1];
-                  res1 = res1+load0*load2;
+                  res1 = res1+BF16TOF32(load0)*BF16TOF32(load2);
                   ptrba = ptrba+1;
                   ptrbb = ptrbb+2;
                }
@@ -121,9 +140,9 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[2*0+0];
                   load1 = ptrbb[0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*0+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   ptrba = ptrba+2;
                   ptrbb = ptrbb+1;
                }
@@ -141,7 +160,7 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[0+0];
                   load1 = ptrbb[0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   ptrba = ptrba+1;
                   ptrbb = ptrbb+1;
                }
diff --git a/kernel/mips/KERNEL.1004K b/kernel/mips/KERNEL.MIPS1004K
similarity index 100%
rename from kernel/mips/KERNEL.1004K
rename to kernel/mips/KERNEL.MIPS1004K
diff --git a/kernel/mips/KERNEL.MIPS24K b/kernel/mips/KERNEL.MIPS24K
new file mode 100644
index 000000000..67135356e
--- /dev/null
+++ b/kernel/mips/KERNEL.MIPS24K
@@ -0,0 +1 @@
+include $(KERNELDIR)/KERNEL.P5600
diff --git a/kernel/power/KERNEL.POWER10 b/kernel/power/KERNEL.POWER10
new file mode 100644
index 000000000..39f5e9414
--- /dev/null
+++ b/kernel/power/KERNEL.POWER10
@@ -0,0 +1,225 @@
+ifeq ($(__BYTE_ORDER__),__ORDER_BIG_ENDIAN__)
+include $(KERNELDIR)/KERNEL.POWER8
+else
+
+#SGEMM_BETA = ../generic/gemm_beta.c
+#DGEMM_BETA = ../generic/gemm_beta.c
+#CGEMM_BETA = ../generic/zgemm_beta.c
+#ZGEMM_BETA = ../generic/zgemm_beta.c
+
+SHGEMM_BETA = ../generic/gemm_beta.c
+SHGEMMKERNEL    = shgemm_kernel_power10.c
+SHGEMMINCOPY    = ../generic/gemm_ncopy_16.c
+SHGEMMITCOPY    = ../generic/gemm_tcopy_16.c
+SHGEMMONCOPY    = ../generic/gemm_ncopy_8.c
+SHGEMMOTCOPY    = ../generic/gemm_tcopy_8.c
+SHGEMMINCOPYOBJ =  shgemm_incopy$(TSUFFIX).$(SUFFIX)
+SHGEMMITCOPYOBJ =  shgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SHGEMMONCOPYOBJ =  shgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SHGEMMOTCOPYOBJ =  shgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+STRMMKERNEL	= sgemm_kernel_power10.c
+DTRMMKERNEL	= dgemm_kernel_power10.c
+CTRMMKERNEL	= cgemm_kernel_power10.S
+ZTRMMKERNEL	= zgemm_kernel_power10.S
+
+SGEMMKERNEL    =  sgemm_kernel_power10.c
+SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
+SGEMMITCOPY    = sgemm_tcopy_16_power8.S
+SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
+SGEMMOTCOPY    = sgemm_tcopy_8_power8.S
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_power10.c
+DGEMMINCOPY    = ../generic/gemm_ncopy_16.c
+DGEMMITCOPY    =  dgemm_tcopy_16_power8.S
+DGEMMONCOPY    =  dgemm_ncopy_4_power8.S
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    = cgemm_kernel_power10.S
+CGEMMINCOPY    = ../generic/zgemm_ncopy_8.c
+CGEMMITCOPY    = ../generic/zgemm_tcopy_8.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    = zgemm_kernel_power10.S
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMINCOPY    = ../generic/zgemm_ncopy_8.c
+ZGEMMITCOPY    = zgemm_tcopy_8_power8.S
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= dtrsm_kernel_LT_16x4_power8.S
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+#Todo: CGEMM3MKERNEL should be 4x4 blocksizes.
+#CGEMM3MKERNEL    =  zgemm3m_kernel_8x4_sse3.S
+#ZGEMM3MKERNEL    =  zgemm3m_kernel_4x4_sse3.S
+
+#Pure C for other kernels
+#SAMAXKERNEL  = ../arm/amax.c
+#DAMAXKERNEL  = ../arm/amax.c
+#CAMAXKERNEL  = ../arm/zamax.c
+#ZAMAXKERNEL  = ../arm/zamax.c
+#
+#SAMINKERNEL  = ../arm/amin.c
+#DAMINKERNEL  = ../arm/amin.c
+#CAMINKERNEL  = ../arm/zamin.c
+#ZAMINKERNEL  = ../arm/zamin.c
+#
+#SMAXKERNEL   = ../arm/max.c
+#DMAXKERNEL   = ../arm/max.c
+#
+#SMINKERNEL   = ../arm/min.c
+#DMINKERNEL   = ../arm/min.c
+#
+ifneq ($(GCCVERSIONGTEQ9),1)
+ISAMAXKERNEL = isamax_power9.S
+else
+ISAMAXKERNEL = isamax.c
+endif
+IDAMAXKERNEL = idamax.c
+ifneq ($(GCCVERSIONGTEQ9),1)
+ICAMAXKERNEL = icamax_power9.S
+else
+ICAMAXKERNEL = icamax.c
+endif
+IZAMAXKERNEL = izamax.c
+#
+ifneq ($(GCCVERSIONGTEQ9),1)
+ISAMINKERNEL = isamin_power9.S
+else
+ISAMINKERNEL = isamin.c
+endif
+IDAMINKERNEL = idamin.c
+ifneq ($(GCCVERSIONGTEQ9),1)
+ICAMINKERNEL = icamin_power9.S
+else
+ICAMINKERNEL = icamin.c
+endif
+IZAMINKERNEL = izamin.c
+#
+#ISMAXKERNEL  = ../arm/imax.c
+#IDMAXKERNEL  = ../arm/imax.c
+#
+#ISMINKERNEL  = ../arm/imin.c
+#IDMINKERNEL  = ../arm/imin.c
+#
+SASUMKERNEL  = sasum.c
+DASUMKERNEL  = dasum.c
+CASUMKERNEL  = casum.c
+ZASUMKERNEL  = zasum.c
+#
+SAXPYKERNEL  = saxpy.c
+DAXPYKERNEL  = daxpy.c
+ifneq ($(GCCVERSIONGTEQ9),1)
+CAXPYKERNEL  = caxpy_power9.S
+else
+CAXPYKERNEL  = caxpy.c
+endif
+ZAXPYKERNEL  = zaxpy.c
+#
+SCOPYKERNEL  = scopy.c
+DCOPYKERNEL  = dcopy.c
+CCOPYKERNEL  = ccopy.c
+ZCOPYKERNEL  = zcopy.c
+#
+SDOTKERNEL   =  sdot.c
+DDOTKERNEL   =  ddot.c
+DSDOTKERNEL  =  sdot.c
+ifneq ($(GCCVERSIONGTEQ9),1)
+CDOTKERNEL   =  cdot_power9.S
+else
+CDOTKERNEL   =  cdot.c
+endif
+ZDOTKERNEL   =  zdot.c
+#
+SNRM2KERNEL  = ../arm/nrm2.c
+DNRM2KERNEL  = ../arm/nrm2.c
+CNRM2KERNEL  = ../arm/znrm2.c
+ZNRM2KERNEL  = ../arm/znrm2.c
+#
+SROTKERNEL   = srot.c
+DROTKERNEL   = drot.c
+CROTKERNEL   = crot.c
+ZROTKERNEL   = zrot.c
+#
+SSCALKERNEL  = sscal.c
+DSCALKERNEL  = dscal.c
+CSCALKERNEL  = zscal.c
+ZSCALKERNEL  = zscal.c
+#
+SSWAPKERNEL  = sswap.c
+DSWAPKERNEL  = dswap.c
+CSWAPKERNEL  = cswap.c
+ZSWAPKERNEL  = zswap.c
+#
+
+SGEMVNKERNEL = sgemv_n.c
+DGEMVNKERNEL = dgemv_n.c
+CGEMVNKERNEL = cgemv_n.c
+ZGEMVNKERNEL = zgemv_n_4.c
+#
+SGEMVTKERNEL = sgemv_t.c
+DGEMVTKERNEL = dgemv_t.c
+CGEMVTKERNEL = cgemv_t.c
+ZGEMVTKERNEL = zgemv_t_4.c
+
+
+#SSYMV_U_KERNEL =  ../generic/symv_k.c
+#SSYMV_L_KERNEL =  ../generic/symv_k.c
+#DSYMV_U_KERNEL =  ../generic/symv_k.c
+#DSYMV_L_KERNEL =  ../generic/symv_k.c
+#QSYMV_U_KERNEL =  ../generic/symv_k.c
+#QSYMV_L_KERNEL =  ../generic/symv_k.c
+#CSYMV_U_KERNEL =  ../generic/zsymv_k.c
+#CSYMV_L_KERNEL =  ../generic/zsymv_k.c
+#ZSYMV_U_KERNEL =  ../generic/zsymv_k.c
+#ZSYMV_L_KERNEL =  ../generic/zsymv_k.c
+#XSYMV_U_KERNEL =  ../generic/zsymv_k.c
+#XSYMV_L_KERNEL =  ../generic/zsymv_k.c
+
+#ZHEMV_U_KERNEL =  ../generic/zhemv_k.c
+#ZHEMV_L_KERNEL =  ../generic/zhemv_k.c
+
+LSAME_KERNEL = ../generic/lsame.c
+SCABS_KERNEL	= ../generic/cabs.c
+DCABS_KERNEL	= ../generic/cabs.c
+QCABS_KERNEL	= ../generic/cabs.c
+
+#Dump kernel
+CGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
+ZGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
+
+endif
diff --git a/kernel/power/KERNEL.POWER8 b/kernel/power/KERNEL.POWER8
index c7867012b..7fba5b4d6 100644
--- a/kernel/power/KERNEL.POWER8
+++ b/kernel/power/KERNEL.POWER8
@@ -12,7 +12,7 @@ SGEMMKERNEL    =  sgemm_kernel_16x8_power8.S
 SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
-SGEMMOTCOPY    = sgemm_tcopy_8_power8.S 
+SGEMMOTCOPY    = sgemm_tcopy_8_power8.S
 SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
@@ -232,3 +232,11 @@ QCABS_KERNEL	= ../generic/cabs.c
 #Dump kernel
 CGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
 ZGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
+
+ifeq ($(__BYTE_ORDER__)$(ELF_VERSION),__ORDER_BIG_ENDIAN__2)
+IDAMAXKERNEL  = ../arm/iamax.c
+IDAMINKERNEL  = ../arm/iamin.c
+IZAMAXKERNEL  = ../arm/izamax.c
+IZAMINKERNEL  = ../arm/izamin.c
+endif
+
diff --git a/kernel/power/KERNEL.POWER9 b/kernel/power/KERNEL.POWER9
index aabb5d976..ab8fbfcd9 100644
--- a/kernel/power/KERNEL.POWER9
+++ b/kernel/power/KERNEL.POWER9
@@ -16,7 +16,7 @@ SGEMMKERNEL    =  sgemm_kernel_power9.S
 SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
-SGEMMOTCOPY    = sgemm_tcopy_8_power8.S 
+SGEMMOTCOPY    = sgemm_tcopy_8_power8.S
 SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
diff --git a/kernel/power/KERNEL.PPCG4 b/kernel/power/KERNEL.PPCG4
index f615754bb..54660b54d 100644
--- a/kernel/power/KERNEL.PPCG4
+++ b/kernel/power/KERNEL.PPCG4
@@ -20,8 +20,10 @@ ZAXPYKERNEL	= zaxpy_ppc440.S
 
 SDOTKERNEL	=  dot_ppc440.S
 DDOTKERNEL	=  dot_ppc440.S
-CDOTKERNEL	= zdot_ppc440.S
-ZDOTKERNEL	= zdot_ppc440.S
+#CDOTKERNEL	= zdot_ppc440.S
+#ZDOTKERNEL	= zdot_ppc440.S
+CDOTKERNEL	= ../arm/zdot.c
+ZDOTKERNEL	= ../arm/zdot.c
 
 ISAMAXKERNEL	=  iamax_ppc440.S
 IDAMAXKERNEL	=  iamax_ppc440.S
@@ -52,8 +54,11 @@ ZNRM2KERNEL	= znrm2_ppc440.S
 
 SROTKERNEL	=  rot_ppc440.S
 DROTKERNEL	=  rot_ppc440.S
-CROTKERNEL	= zrot_ppc440.S
-ZROTKERNEL	= zrot_ppc440.S
+#CROTKERNEL	= zrot_ppc440.S
+#ZROTKERNEL	= zrot_ppc440.S
+CROTKERNEL	= ../arm/zrot.c
+ZROTKERNEL	= ../arm/zrot.c
+
 
 SSCALKERNEL	=  scal_ppc440.S
 DSCALKERNEL	=  scal_ppc440.S
@@ -78,13 +83,18 @@ DGEMMINCOPYOBJ =
 DGEMMITCOPYOBJ =
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
-CGEMMKERNEL    =  zgemm_kernel_altivec_g4.S
-CGEMMINCOPY    =  ../generic/zgemm_ncopy_8.c
-CGEMMITCOPY    =  ../generic/zgemm_tcopy_8.c
+#CGEMMKERNEL    =  zgemm_kernel_altivec_g4.S
+#CGEMMINCOPY    =  ../generic/zgemm_ncopy_8.c
+#CGEMMITCOPY    =  ../generic/zgemm_tcopy_8.c
+CGEMMKERNEL    =  zgemm_kernel.S
+CGEMMINCOPY    =
+CGEMMONCOPY    =
 CGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
 CGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
-CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
-CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMINCOPYOBJ =  
+#cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  
+#cgemm_itcopy$(TSUFFIX).$(SUFFIX)
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
 CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 ZGEMMKERNEL    =  zgemm_kernel_g4.S
diff --git a/kernel/power/casum.c b/kernel/power/casum.c
index a9ece0768..3478a39ef 100644
--- a/kernel/power/casum.c
+++ b/kernel/power/casum.c
@@ -46,7 +46,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#if defined(POWER8)  || defined(POWER9)
+#if defined(POWER8)  || defined(POWER9) || defined(POWER10)
 #include "casum_microk_power8.c"
 #endif
 
diff --git a/kernel/power/ccopy.c b/kernel/power/ccopy.c
index 50df84cc5..cbe5b48d2 100644
--- a/kernel/power/ccopy.c
+++ b/kernel/power/ccopy.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "ccopy_microk_power8.c"
 #endif
 
diff --git a/kernel/power/cgemm_kernel_8x4_power8.S b/kernel/power/cgemm_kernel_8x4_power8.S
index 2bc99974f..6be8c128c 100644
--- a/kernel/power/cgemm_kernel_8x4_power8.S
+++ b/kernel/power/cgemm_kernel_8x4_power8.S
@@ -424,7 +424,7 @@ L999:
 	lwz	r16,  204(SP)
 	lwz	r15,  208(SP)
 	lwz	r14,  212(SP)
-        addi    r11, 224
+        addi    r11, SP, 224
 #endif
         lvx     v20, r11, r0
         addi    r11, r11, 16
@@ -459,4 +459,4 @@ L999:
 	blr
 
 	EPILOGUE
-#endif^
+#endif
diff --git a/kernel/power/cgemm_kernel_power10.S b/kernel/power/cgemm_kernel_power10.S
new file mode 100644
index 000000000..e04f948dd
--- /dev/null
+++ b/kernel/power/cgemm_kernel_power10.S
@@ -0,0 +1,286 @@
+/***************************************************************************
+Copyright (c) 2013-2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#include "def_vsx.h"
+
+ 
+#define LOAD	ld
+#define STACKSIZE  (512 )  
+#define FLINK_SAVE (STACKSIZE+16) /* 16($r12) */  
+#define	M	r3
+#define	N	r4
+#define	K	r5
+
+
+#define A	r8
+#define	B	r9
+#define	C	r10
+#define	LDC	r6
+#define OFFSET	r7
+
+
+#define alpha_r vs51
+#define alpha_i vs55
+#define save_permute_1 vs59
+#define permute_mask vs63
+#define o0	0
+ 
+
+#define T1	r11
+#define T2	r12
+#define T3	r14
+#define T4	r15
+#define T5	r16
+#define T6	r17
+#define L	r18
+#define T7	r19
+#define T8	r20
+#define TEMP_REG	r21
+#define	I	r22
+#define J	r23
+#define AO	r24
+#define	BO	r25
+#define	CO 	r26
+#define T9	r27
+#define	T10	r28
+#define	PRE	r29
+
+#define T12	r30
+#define T13	r31
+
+#include "cgemm_macros_power10.S"
+
+.equ    perm_const1, 0x0405060700010203
+.equ    perm_const2, 0x0c0d0e0f08090a0b
+.equ save_permute_12, 0x0c0d0e0f1c1d1e1f
+.equ save_permute_11, 0x0405060714151617
+
+
+
+#ifndef NEEDPARAM
+
+	PROLOGUE
+	PROFCODE
+
+
+	addi	SP, SP, -STACKSIZE
+	mflr r0
+
+
+	stfd	f14,    0(SP)
+	stfd	f15,    8(SP)
+	stfd	f16,   16(SP)
+	stfd	f17,   24(SP)
+
+	stfd	f18,   32(SP)
+	stfd	f19,   40(SP)
+	stfd	f20,   48(SP)
+	stfd	f21,   56(SP)
+
+	stfd	f22,   64(SP)
+	stfd	f23,   72(SP)
+	stfd	f24,   80(SP)
+	stfd	f25,   88(SP)
+
+	stfd	f26,   96(SP)
+	stfd	f27,  104(SP)
+	stfd	f28,  112(SP)
+	stfd	f29,  120(SP)
+
+	stfd	f30,  128(SP)
+	stfd	f31,  136(SP)
+
+
+	std	r31,  144(SP)
+	std	r30,  152(SP)
+	std	r29,  160(SP)
+	std	r28,  168(SP)
+	std	r27,  176(SP)
+	std	r26,  184(SP)
+	std	r25,  192(SP)
+	std	r24,  200(SP)
+	std	r23,  208(SP)
+	std	r22,  216(SP)
+	std	r21,  224(SP)
+	std	r20,  232(SP)
+	std	r19,  240(SP)
+	std	r18,  248(SP)
+	std	r17,  256(SP)
+	std	r16,  264(SP)
+	std	r15,  272(SP)
+	std	r14,  280(SP)
+ 
+ 
+  stxv    vs52,  288(SP)
+  stxv    vs53,  304(SP)
+  stxv    vs54,  320(SP)
+  stxv    vs55,  336(SP)
+  stxv    vs56,  352(SP)
+  stxv    vs57,  368(SP)
+  stxv    vs58,  384(SP)
+  stxv    vs59,  400(SP)
+  stxv    vs60,  416(SP)
+  stxv    vs61,  432(SP)
+  stxv    vs62,  448(SP)
+  stxv    vs63,  464(SP)
+  std     r0,   FLINK_SAVE(SP)
+ 
+
+
+	ld	LDC, FRAMESLOT(0) + STACKSIZE(SP)
+
+
+
+#ifdef TRMMKERNEL
+	ld	OFFSET,  FRAMESLOT(1) + STACKSIZE(SP)
+#endif
+   slwi    LDC, LDC, ZBASE_SHIFT
+
+ 
+ 
+	/*alpha is stored in f1. convert to single and splat*/
+    xscvdpspn alpha_r,vs1 
+    xscvdpspn alpha_i,vs2 
+	xxspltw   alpha_r,alpha_r,0 
+	xxspltw   alpha_i,alpha_i,0 
+/*load reverse permute mask for big endian
+  uint128 = 0xc0d0e0f08090a0b0405060700010203
+*/ 
+		
+	lis T2, perm_const2@highest
+	lis T1, perm_const1@highest
+	lis T3, save_permute_12@highest
+	lis T4, save_permute_11@highest
+
+	
+	ori T2, T2, perm_const2@higher
+	ori T1, T1, perm_const1@higher
+	ori T3, T3, save_permute_12@higher
+	ori T4, T4, save_permute_11@higher
+
+	
+	rldicr T2, T2, 32, 31
+	rldicr T1, T1, 32, 31
+	rldicr T3, T3, 32, 31
+	rldicr T4, T4, 32, 31 
+
+	oris T2, T2, perm_const2@h
+	oris T1, T1, perm_const1@h
+	oris T3, T3, save_permute_12@h
+	oris T4, T4, save_permute_11@h
+
+	
+	ori T2, T2, perm_const2@l  
+	ori T1, T1, perm_const1@l
+	ori T3, T3, save_permute_12@l  
+	ori T4, T4, save_permute_11@l
+
+	
+  li r0,0
+  li PRE,512
+
+#if defined(CC) || defined(CR) || defined(RC) || defined(RR) 
+/*negate for this case as we will use addition -1*(a+b) */
+  xvnegsp alpha_r,alpha_r
+  xvnegsp alpha_i,alpha_i
+#endif
+
+	mtvsrdd permute_mask,T2,T1
+	mtvsrdd save_permute_1,T3,T4 	
+
+     /*mask is reverse permute so we have to make it inner permute */
+ 	xxpermdi	permute_mask,	permute_mask,	permute_mask,2 
+
+#include "cgemm_logic_power10.S"
+
+.L999: 
+	lfd	f14,    0(SP)
+	lfd	f15,    8(SP)
+	lfd	f16,   16(SP)
+	lfd	f17,   24(SP)
+
+	lfd	f18,   32(SP)
+	lfd	f19,   40(SP)
+	lfd	f20,   48(SP)
+	lfd	f21,   56(SP)
+
+	lfd	f22,   64(SP)
+	lfd	f23,   72(SP)
+	lfd	f24,   80(SP)
+	lfd	f25,   88(SP)
+
+	lfd	f26,   96(SP)
+	lfd	f27,  104(SP)
+	lfd	f28,  112(SP)
+	lfd	f29,  120(SP)
+
+	lfd	f30,  128(SP)
+	lfd	f31,  136(SP)
+
+	ld	r31,  144(SP)
+	ld	r30,  152(SP)
+	ld	r29,  160(SP)
+	ld	r28,  168(SP)
+	ld	r27,  176(SP)
+	ld	r26,  184(SP)
+	ld	r25,  192(SP)
+	ld	r24,  200(SP)
+	ld	r23,  208(SP)
+	ld	r22,  216(SP)
+	ld	r21,  224(SP)
+	ld	r20,  232(SP)
+	ld	r19,  240(SP)
+	ld	r18,  248(SP)
+	ld	r17,  256(SP)
+	ld	r16,  264(SP)
+	ld	r15,  272(SP)
+	ld	r14,  280(SP)
+
+	ld    r0, 	 FLINK_SAVE(SP)	
+ 
+    lxv    vs52,  288(SP)
+    lxv    vs53,  304(SP)
+    lxv    vs54,  320(SP)
+    lxv    vs55,  336(SP)
+    lxv    vs56,  352(SP)
+    lxv    vs57,  368(SP)
+    lxv    vs58,  384(SP) 
+    lxv    vs59,  400(SP)
+	mtlr r0
+    lxv    vs60,  416(SP)
+    lxv    vs61,  432(SP) 
+    lxv    vs62,  448(SP)
+    lxv    vs63,  464(SP)
+
+	addi	SP, SP, STACKSIZE 
+	blr
+
+
+	EPILOGUE
+#endif
diff --git a/kernel/power/cgemm_logic_power10.S b/kernel/power/cgemm_logic_power10.S
new file mode 100644
index 000000000..3700ac87b
--- /dev/null
+++ b/kernel/power/cgemm_logic_power10.S
@@ -0,0 +1,2814 @@
+/***************************************************************************
+Copyright (c) 2013-2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define MY_ALIGN .align 3
+b CGEMM_L4
+/*                MINI SUBROUTINES                            */      
+/*                4x8 MAIN 128x+2 LOOP                     */      
+
+
+CGEMM_L4x8_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD4x8_2 
+    MY_ALIGN
+CGEMM_L4x8_LOOP:
+/*----------------------------------------*/   
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL4x8_L2 128,64,0,0 
+CGEMM_L4x8_K128:
+/*----------------------------------------*/   
+    KERNEL4x8_L2 128,64,1,0
+    dcbt    AO, T2  
+    KERNEL4x8_L2 128,64,2,0
+    KERNEL4x8_L2 128,64,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL4x8_L2 128,64,4,0
+    KERNEL4x8_L2 128,64,5,0
+    dcbt    AO, T4  
+    KERNEL4x8_L2 128,64,6,0
+    KERNEL4x8_L2 128,64,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL4x8_L2 128,64,8,0
+    KERNEL4x8_L2 128,64,9,0
+    KERNEL4x8_L2 128,64,10,0
+    KERNEL4x8_L2 128,64,11,0  
+    dcbt    BO, T4
+    KERNEL4x8_L2 128,64,12,0
+    KERNEL4x8_L2 128,64,13,0
+    KERNEL4x8_L2 128,64,14,0
+    KERNEL4x8_L2 128,64,15,0  
+    KERNEL4x8_L2 128,64,16,0
+    KERNEL4x8_L2 128,64,17,0 
+    KERNEL4x8_L2 128,64,18,0
+    KERNEL4x8_L2 128,64,19,0  
+    KERNEL4x8_L2 128,64,20,0
+    KERNEL4x8_L2 128,64,21,0 
+    KERNEL4x8_L2 128,64,22,0
+    KERNEL4x8_L2 128,64,23,0   
+    KERNEL4x8_L2 128,64,24,0
+    KERNEL4x8_L2 128,64,25,0
+    KERNEL4x8_L2 128,64,26,0
+    KERNEL4x8_L2 128,64,27,0  
+    KERNEL4x8_L2 128,64,28,0
+    KERNEL4x8_L2 128,64,29,0
+    KERNEL4x8_L2 128,64,30,0
+    KERNEL4x8_L2 128,64,31,0 
+    KERNEL4x8_L2 128,64,32,0
+    KERNEL4x8_L2 128,64,33,0
+    KERNEL4x8_L2 128,64,34,0
+    KERNEL4x8_L2 128,64,35,0 
+    KERNEL4x8_L2 128,64,36,0
+    KERNEL4x8_L2 128,64,37,0
+    KERNEL4x8_L2 128,64,38,0
+    KERNEL4x8_L2 128,64,39,0  
+    KERNEL4x8_L2 128,64,40,0
+    KERNEL4x8_L2 128,64,41,0
+    KERNEL4x8_L2 128,64,42,0
+    KERNEL4x8_L2 128,64,43,0  
+    KERNEL4x8_L2 128,64,44,0
+    KERNEL4x8_L2 128,64,45,0
+    KERNEL4x8_L2 128,64,46,0
+    KERNEL4x8_L2 128,64,47,0 
+    KERNEL4x8_L2 128,64,48,0
+    KERNEL4x8_L2 128,64,49,0 
+    KERNEL4x8_L2 128,64,50,0
+    KERNEL4x8_L2 128,64,51,0  
+    KERNEL4x8_L2 128,64,52,0
+    KERNEL4x8_L2 128,64,53,0 
+    KERNEL4x8_L2 128,64,54,0
+    KERNEL4x8_L2 128,64,55,0  
+    KERNEL4x8_L2 128,64,56,0
+    KERNEL4x8_L2 128,64,57,0
+    KERNEL4x8_L2 128,64,58,0
+    KERNEL4x8_L2 128,64,59,0  
+    KERNEL4x8_L2 128,64,60,0
+    KERNEL4x8_L2 128,64,61,0
+    KERNEL4x8_L2 128,64,62,0 
+    KERNEL4x8_L2 128,64,63,1  
+    bdnz    CGEMM_L4x8_LOOP
+    MY_ALIGN  
+CGEMM_L4x8_LOOP_END:
+/*----------------------------------------*/   
+    END4x8_2
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x8_L64_SUB:
+/*----------------------------------------*/   
+    LOAD4x8_2  
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL4x8_L2 128,64,0,0 
+    KERNEL4x8_L2 128,64,1,0
+    dcbt    AO, T2  
+    KERNEL4x8_L2 128,64,2,0
+    KERNEL4x8_L2 128,64,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL4x8_L2 128,64,4,0
+    KERNEL4x8_L2 128,64,5,0
+    dcbt    AO, T4  
+    KERNEL4x8_L2 128,64,6,0
+    KERNEL4x8_L2 128,64,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL4x8_L2 128,64,8,0
+    KERNEL4x8_L2 128,64,9,0
+    KERNEL4x8_L2 128,64,10,0
+    KERNEL4x8_L2 128,64,11,0  
+    dcbt    BO, T4
+    KERNEL4x8_L2 128,64,12,0
+    KERNEL4x8_L2 128,64,13,0
+    KERNEL4x8_L2 128,64,14,0
+    KERNEL4x8_L2 128,64,15,0  
+    KERNEL4x8_L2 128,64,16,0
+    KERNEL4x8_L2 128,64,17,0 
+    KERNEL4x8_L2 128,64,18,0
+    KERNEL4x8_L2 128,64,19,0  
+    KERNEL4x8_L2 128,64,20,0
+    KERNEL4x8_L2 128,64,21,0 
+    KERNEL4x8_L2 128,64,22,0
+    KERNEL4x8_L2 128,64,23,0   
+    KERNEL4x8_L2 128,64,24,0
+    KERNEL4x8_L2 128,64,25,0
+    KERNEL4x8_L2 128,64,26,0
+    KERNEL4x8_L2 128,64,27,0  
+    KERNEL4x8_L2 128,64,28,0
+    KERNEL4x8_L2 128,64,29,0
+    KERNEL4x8_L2 128,64,30,0
+    KERNEL4x8_E2 128,64,31,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x8_L32_SUB:
+/*----------------------------------------*/   
+    LOAD4x8_2  
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL4x8_L2 128,64,0,0 
+    KERNEL4x8_L2 128,64,1,0
+    dcbt    AO, T2  
+    KERNEL4x8_L2 128,64,2,0
+    KERNEL4x8_L2 128,64,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL4x8_L2 128,64,4,0
+    KERNEL4x8_L2 128,64,5,0
+    dcbt    AO, T4  
+    KERNEL4x8_L2 128,64,6,0
+    KERNEL4x8_L2 128,64,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL4x8_L2 128,64,8,0
+    KERNEL4x8_L2 128,64,9,0
+    KERNEL4x8_L2 128,64,10,0
+    KERNEL4x8_L2 128,64,11,0  
+    dcbt    BO, T4
+    KERNEL4x8_L2 128,64,12,0
+    KERNEL4x8_L2 128,64,13,0
+    KERNEL4x8_L2 128,64,14,0
+    KERNEL4x8_E2 128,64,15,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x8_L16_SUB:
+/*----------------------------------------*/   
+    LOAD4x8_2 
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL4x8_L2 128,64,0,0 
+    KERNEL4x8_L2 128,64,1,0
+    dcbt    AO, T2  
+    KERNEL4x8_L2 128,64,2,0
+    KERNEL4x8_L2 128,64,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL4x8_L2 128,64,4,0
+    KERNEL4x8_L2 128,64,5,0
+    dcbt    AO, T4  
+    KERNEL4x8_L2 128,64,6,0
+    KERNEL4x8_E2 128,64,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x4_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD4x4_2  
+    MY_ALIGN
+CGEMM_L4x4_LOOP:
+/*----------------------------------------*/   
+    KERNEL4x4_L2 64,64,0,0
+CGEMM_L4x4_K32:
+/*----------------------------------------*/   
+    KERNEL4x4_L2 64,64,1,0   
+    KERNEL4x4_L2 64,64,2,0
+    KERNEL4x4_L2 64,64,3,0  
+    KERNEL4x4_L2 64,64,4,0
+    KERNEL4x4_L2 64,64,5,0 
+    KERNEL4x4_L2 64,64,6,0
+    KERNEL4x4_L2 64,64,7,0
+    KERNEL4x4_L2 64,64,8,0
+    KERNEL4x4_L2 64,64,9,0   
+    KERNEL4x4_L2 64,64,10,0
+    KERNEL4x4_L2 64,64,11,0  
+    KERNEL4x4_L2 64,64,12,0
+    KERNEL4x4_L2 64,64,13,0 
+    KERNEL4x4_L2 64,64,14,0
+    KERNEL4x4_L2 64,64,15,1    
+    bdnz    CGEMM_L4x4_LOOP
+    MY_ALIGN  
+CGEMM_L4x4_LOOP_END:
+/*----------------------------------------*/   
+    END4x4_2 
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x4_L16_SUB:
+/*----------------------------------------*/   
+    LOAD4x4_2
+    KERNEL4x4_L2 64,64,0,0
+    KERNEL4x4_L2 64,64,1,0   
+    KERNEL4x4_L2 64,64,2,0
+    KERNEL4x4_L2 64,64,3,0  
+    KERNEL4x4_L2 64,64,4,0
+    KERNEL4x4_L2 64,64,5,0 
+    KERNEL4x4_L2 64,64,6,0
+    KERNEL4x4_E2 64,64,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x4_L8_SUB:
+/*----------------------------------------*/   
+    LOAD4x4_2
+    KERNEL4x4_L2 64,64,0,0
+    KERNEL4x4_L2 64,64,1,0   
+    KERNEL4x4_L2 64,64,2,0
+    KERNEL4x4_E2 64,64,3,1 
+    blr
+
+
+CGEMM_4x2_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD4x2_2  
+    MY_ALIGN 
+CGEMM_L4x2_LOOP:
+/*----------------------------------------*/   
+    KERNEL4x2_L2 32,64,0,0 
+CGEMM_L4x2_K32:
+/*----------------------------------------*/   
+    KERNEL4x2_L2 32,64,1,0  
+    KERNEL4x2_L2 32,64,2,0
+    KERNEL4x2_L2 32,64,3,0  
+    KERNEL4x2_L2 32,64,4,0
+    KERNEL4x2_L2 32,64,5,0 
+    KERNEL4x2_L2 32,64,6,0
+    KERNEL4x2_L2 32,64,7,0
+    KERNEL4x2_L2 32,64,8,0
+    KERNEL4x2_L2 32,64,9,0  
+    KERNEL4x2_L2 32,64,10,0
+    KERNEL4x2_L2 32,64,11,0  
+    KERNEL4x2_L2 32,64,12,0
+    KERNEL4x2_L2 32,64,13,0 
+    KERNEL4x2_L2 32,64,14,0
+    KERNEL4x2_L2 32,64,15,1   
+    bdnz    CGEMM_L4x2_LOOP
+    MY_ALIGN  
+
+
+CGEMM_L4x2_LOOP_END:
+/*----------------------------------------*/   
+    END4x2_2 
+    blr
+    MY_ALIGN
+CGEMM_4x2_L16_SUB:
+/*----------------------------------------*/   
+    LOAD4x2_2
+    KERNEL4x2_L2 32,64,0,0
+    KERNEL4x2_L2 32,64,1,0  
+    KERNEL4x2_L2 32,64,2,0
+    KERNEL4x2_L2 32,64,3,0  
+    KERNEL4x2_L2 32,64,4,0
+    KERNEL4x2_L2 32,64,5,0 
+    KERNEL4x2_L2 32,64,6,0
+    KERNEL4x2_E2 32,64,7,1
+    blr
+    MY_ALIGN
+CGEMM_4x2_L8_SUB:
+/*----------------------------------------*/   
+    LOAD4x2_2
+    KERNEL4x2_L2 32,64,0,0
+    KERNEL4x2_L2 32,64,1,0  
+    KERNEL4x2_L2 32,64,2,0
+    KERNEL4x2_E2 32,64,3,1  
+    blr
+
+
+CGEMM_4x1_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD4x1_2  
+    MY_ALIGN
+CGEMM_L4x1_LOOP:
+/*----------------------------------------*/   
+    KERNEL4x1_L2 16,64,0,0 
+CGEMM_L4x1_K32:
+/*----------------------------------------*/   
+    KERNEL4x1_L2 16,64,1,0  
+    KERNEL4x1_L2 16,64,2,0
+    KERNEL4x1_L2 16,64,3,0  
+    KERNEL4x1_L2 16,64,4,0
+    KERNEL4x1_L2 16,64,5,0 
+    KERNEL4x1_L2 16,64,6,0
+    KERNEL4x1_L2 16,64,7,0
+    KERNEL4x1_L2 16,64,8,0
+    KERNEL4x1_L2 16,64,9,0  
+    KERNEL4x1_L2 16,64,10,0
+    KERNEL4x1_L2 16,64,11,0  
+    KERNEL4x1_L2 16,64,12,0
+    KERNEL4x1_L2 16,64,13,0 
+    KERNEL4x1_L2 16,64,14,0
+    KERNEL4x1_L2 16,64,15,1   
+    bdnz    CGEMM_L4x1_LOOP
+    MY_ALIGN  
+CGEMM_L4x1_LOOP_END:
+/*----------------------------------------*/   
+    END4x1_2 
+    blr
+
+    MY_ALIGN
+CGEMM_4x1_L16_SUB:
+/*----------------------------------------*/   
+    LOAD4x1_2
+    KERNEL4x1_L2 16,64,0,0
+    KERNEL4x1_L2 16,64,1,0  
+    KERNEL4x1_L2 16,64,2,0
+    KERNEL4x1_L2 16,64,3,0  
+    KERNEL4x1_L2 16,64,4,0
+    KERNEL4x1_L2 16,64,5,0 
+    KERNEL4x1_L2 16,64,6,0
+    KERNEL4x1_E2 16,64,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_4x1_L8_SUB:
+/*----------------------------------------*/   
+    LOAD4x1_2
+    KERNEL4x1_L2 16,64,0,0
+    KERNEL4x1_L2 16,64,1,0  
+    KERNEL4x1_L2 16,64,2,0
+    KERNEL4x1_E2 16,64,3,1  
+    blr
+
+
+
+/*             MAIN LOOP BEGINS               */   
+    MY_ALIGN
+
+
+CGEMM_L4:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    neg TEMP_REG, OFFSET 
+#endif   
+    /* Pre set value in vs57 as 0xffff0000ffff0000 for masking */
+    vspltisb v24, -1
+    vspltisb v25, 0
+    xxsldwi vs57, vs56, vs57, 1
+    xxpermdi vs57, vs57, vs57, 3
+    srawi.    J,  N,  2
+    ble   CGEMM_L4_END
+
+
+CGEMM_L4_BEGIN:
+/*----------------------------------------*/   
+    mr    CO, C
+    slwi    T1, LDC , 2     
+    add     T2,C,LDC    
+    mr    AO, A  
+    add   C,  C,  T1
+#if defined(TRMMKERNEL) && defined(LEFT)   
+    mr TEMP_REG, OFFSET  /*off = offset;*/
+#endif     
+    srawi.    I,  M,  3
+    ble   CGEMM_L4x8_END
+    dcbt    CO,r0  /*just prefetch*/
+    dcbt    T2,r0    
+
+
+CGEMM_L4x8_BEGIN:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,8,4
+#else    
+    mr    BO, B  
+    dcbt    B,  r0  
+#endif     
+    dcbt    AO, r0
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,8,4
+    mr T1, T6
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512   
+    srawi.   T8, T1, 7 /**(T1-2) % 128x */
+#else   
+    mr T1, K
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512 
+    srawi.   T8, T1, 7 /**(K-2) % 128x */
+#endif   
+    ZERO4x8  
+    ble   CGEMM_L4x8_SUB0
+    bl CGEMM_L4x8_LMAIN_SUB
+    andi.   L,  T1, 127
+    ble   CGEMM_L4x8_SAVE
+    b   CGEMM_L4x8_SUB2
+
+
+CGEMM_L4x8_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 255
+    cmpwi   T6,129
+#else   
+    andi.   L,  K,  255
+    cmpwi   K,129
+#endif       
+    li T8,1
+    bne CMP4x8_128K
+    addi BO,BO,-32
+    addi AO,AO,-64 
+    LOAD4x8O 64,32 
+    END4x8_WITHOUT_ADD   
+    LOAD4x8_2O  128, 64 
+    mtctr   T8    
+    bl CGEMM_L4x8_K128   
+    b CGEMM_L4x8_SAVE  
+    CMP4x8_128K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,128
+#else    
+    cmpwi   K,128
+#endif        
+    bne CGEMM_L4x8_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-64
+    addi AO,AO,-128   
+    LOAD4x8_2O 128,64
+    bl CGEMM_L4x8_K128   
+    b CGEMM_L4x8_SAVE 
+    MY_ALIGN
+
+
+CGEMM_L4x8_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 64
+    ble CGEMM_L4x8_SUB2_32
+    bl  CGEMM_4x8_L64_SUB
+    MY_ALIGN
+
+
+CGEMM_L4x8_SUB2_32:
+/*----------------------------------------*/   
+    andi.      T1,L, 32
+    ble CGEMM_L4x8_SUB2_16    
+    bl  CGEMM_4x8_L32_SUB
+    MY_ALIGN 
+
+
+CGEMM_L4x8_SUB2_16:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L4x8_SUB2_8
+    bl  CGEMM_4x8_L16_SUB  
+    MY_ALIGN    
+
+
+CGEMM_L4x8_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L4x8_SUB2_4
+    LOAD4x8_2
+    KERNEL4x8_L2  128,64, 0,0
+    KERNEL4x8_L2  128,64, 1,0
+    KERNEL4x8_L2  128,64, 2,0
+    KERNEL4x8_E2  128,64, 3,1
+    MY_ALIGN   
+
+
+CGEMM_L4x8_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L4x8_SUB2_2
+    LOAD4x8_2
+    KERNEL4x8_L2  128,64, 0,0
+    KERNEL4x8_E2  128,64, 1,1
+    MY_ALIGN
+
+
+CGEMM_L4x8_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L4x8_SUB2_1
+    LOAD4x8_2 
+    KERNEL4x8_E2  128,64, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L4x8_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L4x8_SAVE 
+    KERNEL4x8
+
+    MY_ALIGN
+CGEMM_L4x8_SAVE:
+/*----------------------------------------*/   
+    addic.    I,  I,  -1
+    MY_ALIGN
+    SAVE4x8
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,8,4
+#endif     
+    bgt   CGEMM_L4x8_BEGIN
+    andi.   T2, M,  7
+    ble   CGEMM_L4x1_END
+    andi.   T1, M,  4
+    ble   CGEMM_L4x4_END
+    b   CGEMM_L4x4_BEGIN
+    MY_ALIGN 
+
+
+CGEMM_L4x8_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L4x4_BEGIN:
+/*----------------------------------------*/   
+    andi.   T2, M,  7
+    ble   CGEMM_L4x1_END
+    andi.   T1, M,  4
+    ble   CGEMM_L4x4_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,4,4
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,4,4
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO4x4
+    ble   CGEMM_L4x4_SUB0 
+    bl CGEMM_4x4_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L4x4_SAVE
+    b    CGEMM_L4x4_SUB2
+
+
+CGEMM_L4x4_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP4x4_32K
+    addi BO,BO,-32
+    addi AO,AO,-32  
+    LOAD4x4O 32,32 
+    END4x4_WITHOUT_ADD   
+    LOAD4x4_2O  64, 64 
+    mtctr   T8    
+    bl CGEMM_L4x4_K32   
+    b CGEMM_L4x4_SAVE  
+    CMP4x4_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L4x4_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-64
+    addi AO,AO,-64   
+    LOAD4x4_2O 64,64
+    bl CGEMM_L4x4_K32   
+    b CGEMM_L4x4_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L4x4_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L4x4_SUB2_8
+    bl  CGEMM_4x4_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L4x4_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L4x4_SUB2_4
+    bl CGEMM_4x4_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L4x4_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L4x4_SUB2_2
+    LOAD4x4_2
+    KERNEL4x4_L2  64,64, 0,0
+    KERNEL4x4_E2  64,64, 1,1
+    MY_ALIGN
+
+
+CGEMM_L4x4_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L4x4_SUB2_1
+    LOAD4x4_2
+    KERNEL4x4_E2  64,64, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L4x4_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L4x4_SAVE 
+    KERNEL4x4
+
+
+CGEMM_L4x4_SAVE:
+/*----------------------------------------*/   
+    SAVE4x4
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,4,4
+#endif     
+
+
+CGEMM_L4x4_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L4x2_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M,  2
+    ble   CGEMM_L4x2_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,2,4
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,2,4
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO4x2
+    ble   CGEMM_L4x2_SUB0 
+    bl CGEMM_4x2_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L4x2_SAVE
+    b   CGEMM_L4x2_SUB2
+
+
+CGEMM_L4x2_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP4x2_32K
+    addi BO,BO,-32
+    addi AO,AO,-16  
+    LOAD4x2O 16,32 
+    END4x2_WITHOUT_ADD   
+    LOAD4x2_2O  32, 64  
+    mtctr   T8    
+    bl CGEMM_L4x2_K32   
+    b CGEMM_L4x2_SAVE  
+    CMP4x2_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L4x2_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-64
+    addi AO,AO,-32   
+    LOAD4x2_2O 32,64
+    bl CGEMM_L4x2_K32   
+    b CGEMM_L4x2_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L4x2_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L4x2_SUB2_8
+    bl CGEMM_4x2_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L4x2_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L4x2_SUB2_4
+    bl CGEMM_4x2_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L4x2_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L4x2_SUB2_2
+    LOAD4x2_2
+    KERNEL4x2_L2  32,64, 0,0
+    KERNEL4x2_E2  32,64, 1,1
+    MY_ALIGN
+
+
+CGEMM_L4x2_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L4x2_SUB2_1
+    LOAD4x2_2
+    KERNEL4x2_E2  32,64, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L4x2_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L4x2_SAVE 
+    KERNEL4x2
+
+    MY_ALIGN
+CGEMM_L4x2_SAVE:
+/*----------------------------------------*/   
+    SAVE4x2
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,2,4
+#endif     
+
+
+CGEMM_L4x2_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L4x1_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M,  1
+    ble   CGEMM_L4x1_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,1,4
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,1,4
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO4x1
+    ble   CGEMM_L4x1_SUB0 
+    bl CGEMM_4x1_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L4x1_SAVE
+    b   CGEMM_L4x1_SUB2
+
+
+CGEMM_L4x1_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP4x1_32K
+    addi BO,BO,-32
+    addi AO,AO,-8  
+    LOAD4x1O 8,32 
+    END4x1_WITHOUT_ADD   
+    LOAD4x1_2O  16, 64  
+    mtctr   T8    
+    bl CGEMM_L4x1_K32   
+    b CGEMM_L4x1_SAVE  
+    CMP4x1_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L4x1_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-64
+    addi AO,AO,-16   
+    LOAD4x1_2O 16,64
+    bl CGEMM_L4x1_K32   
+    b CGEMM_L4x1_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L4x1_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L4x1_SUB2_8
+    bl CGEMM_4x1_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L4x1_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L4x1_SUB2_4
+    bl CGEMM_4x1_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L4x1_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L4x1_SUB2_2
+    LOAD4x1_2
+    KERNEL4x1_L2  16,64, 0,0
+    KERNEL4x1_E2  16,64, 1,1
+    MY_ALIGN
+
+
+CGEMM_L4x1_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L4x1_SUB2_1
+    LOAD4x1_2
+    KERNEL4x1_E2  16,64, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L4x1_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L4x1_SAVE 
+    KERNEL4x1
+
+    MY_ALIGN
+CGEMM_L4x1_SAVE:
+/*----------------------------------------*/  
+     
+    SAVE4x1
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,1,4
+#endif   
+
+
+CGEMM_L4x1_END:
+/*----------------------------------------*/   
+    slwi    T1, K,  5
+    addic.    J,  J,  -1
+    add   B,  B,  T1
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    addi TEMP_REG, TEMP_REG, 4
+#endif   
+    bgt   CGEMM_L4_BEGIN
+
+
+CGEMM_L4_END:
+
+b CGEMM_L2
+/*                MINI SUBROUTINES                            */      
+/*                2x8 MAIN 128x+2 LOOP                     */
+
+
+CGEMM_L2x8_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD2x8_2 
+    MY_ALIGN
+CGEMM_L2x8_LOOP:
+/*----------------------------------------*/   
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_L2 128,32,0,0 
+CGEMM_L2x8_K128:
+/*----------------------------------------*/   
+    KERNEL2x8_L2 128,32,1,0
+    dcbt    AO, T2  
+    KERNEL2x8_L2 128,32,2,0
+    KERNEL2x8_L2 128,32,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_L2 128,32,4,0
+    KERNEL2x8_L2 128,32,5,0
+    dcbt    AO, T4  
+    KERNEL2x8_L2 128,32,6,0
+    KERNEL2x8_L2 128,32,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL2x8_L2 128,32,8,0
+    KERNEL2x8_L2 128,32,9,0
+    KERNEL2x8_L2 128,32,10,0
+    KERNEL2x8_L2 128,32,11,0  
+    dcbt    BO, T4
+    KERNEL2x8_L2 128,32,12,0
+    KERNEL2x8_L2 128,32,13,0
+    KERNEL2x8_L2 128,32,14,0
+    KERNEL2x8_L2 128,32,15,0  
+    KERNEL2x8_L2 128,32,16,0
+    KERNEL2x8_L2 128,32,17,0 
+    KERNEL2x8_L2 128,32,18,0
+    KERNEL2x8_L2 128,32,19,0  
+    KERNEL2x8_L2 128,32,20,0
+    KERNEL2x8_L2 128,32,21,0 
+    KERNEL2x8_L2 128,32,22,0
+    KERNEL2x8_L2 128,32,23,0   
+    KERNEL2x8_L2 128,32,24,0
+    KERNEL2x8_L2 128,32,25,0
+    KERNEL2x8_L2 128,32,26,0
+    KERNEL2x8_L2 128,32,27,0  
+    KERNEL2x8_L2 128,32,28,0
+    KERNEL2x8_L2 128,32,29,0
+    KERNEL2x8_L2 128,32,30,0
+    KERNEL2x8_L2 128,32,31,0 
+    KERNEL2x8_L2 128,32,32,0
+    KERNEL2x8_L2 128,32,33,0
+    KERNEL2x8_L2 128,32,34,0
+    KERNEL2x8_L2 128,32,35,0 
+    KERNEL2x8_L2 128,32,36,0
+    KERNEL2x8_L2 128,32,37,0
+    KERNEL2x8_L2 128,32,38,0
+    KERNEL2x8_L2 128,32,39,0  
+    KERNEL2x8_L2 128,32,40,0
+    KERNEL2x8_L2 128,32,41,0
+    KERNEL2x8_L2 128,32,42,0
+    KERNEL2x8_L2 128,32,43,0  
+    KERNEL2x8_L2 128,32,44,0
+    KERNEL2x8_L2 128,32,45,0
+    KERNEL2x8_L2 128,32,46,0
+    KERNEL2x8_L2 128,32,47,0 
+    KERNEL2x8_L2 128,32,48,0
+    KERNEL2x8_L2 128,32,49,0 
+    KERNEL2x8_L2 128,32,50,0
+    KERNEL2x8_L2 128,32,51,0  
+    KERNEL2x8_L2 128,32,52,0
+    KERNEL2x8_L2 128,32,53,0 
+    KERNEL2x8_L2 128,32,54,0
+    KERNEL2x8_L2 128,32,55,0  
+    KERNEL2x8_L2 128,32,56,0
+    KERNEL2x8_L2 128,32,57,0
+    KERNEL2x8_L2 128,32,58,0
+    KERNEL2x8_L2 128,32,59,0  
+    KERNEL2x8_L2 128,32,60,0
+    KERNEL2x8_L2 128,32,61,0
+    KERNEL2x8_L2 128,32,62,0 
+    KERNEL2x8_L2 128,32,63,1  
+    bdnz    CGEMM_L2x8_LOOP
+    MY_ALIGN  
+CGEMM_L2x8_LOOP_END:
+/*----------------------------------------*/   
+    END2x8_2
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x8_L64_SUB:
+/*----------------------------------------*/   
+    LOAD2x8_2  
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_L2 128,32,0,0 
+    KERNEL2x8_L2 128,32,1,0
+    dcbt    AO, T2  
+    KERNEL2x8_L2 128,32,2,0
+    KERNEL2x8_L2 128,32,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_L2 128,32,4,0
+    KERNEL2x8_L2 128,32,5,0
+    dcbt    AO, T4  
+    KERNEL2x8_L2 128,32,6,0
+    KERNEL2x8_L2 128,32,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL2x8_L2 128,32,8,0
+    KERNEL2x8_L2 128,32,9,0
+    KERNEL2x8_L2 128,32,10,0
+    KERNEL2x8_L2 128,32,11,0  
+    dcbt    BO, T4
+    KERNEL2x8_L2 128,32,12,0
+    KERNEL2x8_L2 128,32,13,0
+    KERNEL2x8_L2 128,32,14,0
+    KERNEL2x8_L2 128,32,15,0  
+    KERNEL2x8_L2 128,32,16,0
+    KERNEL2x8_L2 128,32,17,0 
+    KERNEL2x8_L2 128,32,18,0
+    KERNEL2x8_L2 128,32,19,0  
+    KERNEL2x8_L2 128,32,20,0
+    KERNEL2x8_L2 128,32,21,0 
+    KERNEL2x8_L2 128,32,22,0
+    KERNEL2x8_L2 128,32,23,0   
+    KERNEL2x8_L2 128,32,24,0
+    KERNEL2x8_L2 128,32,25,0
+    KERNEL2x8_L2 128,32,26,0
+    KERNEL2x8_L2 128,32,27,0  
+    KERNEL2x8_L2 128,32,28,0
+    KERNEL2x8_L2 128,32,29,0
+    KERNEL2x8_L2 128,32,30,0
+    KERNEL2x8_E2 128,32,31,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x8_L32_SUB:
+/*----------------------------------------*/   
+    LOAD2x8_2  
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_L2 128,32,0,0 
+    KERNEL2x8_L2 128,32,1,0
+    dcbt    AO, T2  
+    KERNEL2x8_L2 128,32,2,0
+    KERNEL2x8_L2 128,32,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_L2 128,32,4,0
+    KERNEL2x8_L2 128,32,5,0
+    dcbt    AO, T4  
+    KERNEL2x8_L2 128,32,6,0
+    KERNEL2x8_L2 128,32,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL2x8_L2 128,32,8,0
+    KERNEL2x8_L2 128,32,9,0
+    KERNEL2x8_L2 128,32,10,0
+    KERNEL2x8_L2 128,32,11,0  
+    dcbt    BO, T4
+    KERNEL2x8_L2 128,32,12,0
+    KERNEL2x8_L2 128,32,13,0
+    KERNEL2x8_L2 128,32,14,0
+    KERNEL2x8_E2 128,32,15,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x8_L16_SUB:
+/*----------------------------------------*/   
+    LOAD2x8_2 
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_L2 128,32,0,0 
+    KERNEL2x8_L2 128,32,1,0
+    dcbt    AO, T2  
+    KERNEL2x8_L2 128,32,2,0
+    KERNEL2x8_L2 128,32,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_L2 128,32,4,0
+    KERNEL2x8_L2 128,32,5,0
+    dcbt    AO, T4  
+    KERNEL2x8_L2 128,32,6,0
+    KERNEL2x8_E2 128,32,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x4_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD2x4_2  
+    MY_ALIGN
+CGEMM_L2x4_LOOP:
+/*----------------------------------------*/   
+    KERNEL2x4_L2 64,32,0,0
+CGEMM_L2x4_K32:
+/*----------------------------------------*/   
+    KERNEL2x4_L2 64,32,1,0   
+    KERNEL2x4_L2 64,32,2,0
+    KERNEL2x4_L2 64,32,3,0  
+    KERNEL2x4_L2 64,32,4,0
+    KERNEL2x4_L2 64,32,5,0 
+    KERNEL2x4_L2 64,32,6,0
+    KERNEL2x4_L2 64,32,7,0
+    KERNEL2x4_L2 64,32,8,0
+    KERNEL2x4_L2 64,32,9,0   
+    KERNEL2x4_L2 64,32,10,0
+    KERNEL2x4_L2 64,32,11,0  
+    KERNEL2x4_L2 64,32,12,0
+    KERNEL2x4_L2 64,32,13,0 
+    KERNEL2x4_L2 64,32,14,0
+    KERNEL2x4_L2 64,32,15,1    
+    bdnz    CGEMM_L2x4_LOOP
+    MY_ALIGN  
+CGEMM_L2x4_LOOP_END:
+/*----------------------------------------*/   
+    END2x4_2 
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x4_L16_SUB:
+/*----------------------------------------*/   
+    LOAD2x4_2
+    KERNEL2x4_L2 64,32,0,0
+    KERNEL2x4_L2 64,32,1,0   
+    KERNEL2x4_L2 64,32,2,0
+    KERNEL2x4_L2 64,32,3,0  
+    KERNEL2x4_L2 64,32,4,0
+    KERNEL2x4_L2 64,32,5,0 
+    KERNEL2x4_L2 64,32,6,0
+    KERNEL2x4_E2 64,32,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x4_L8_SUB:
+/*----------------------------------------*/   
+    LOAD2x4_2
+    KERNEL2x4_L2 64,32,0,0
+    KERNEL2x4_L2 64,32,1,0   
+    KERNEL2x4_L2 64,32,2,0
+    KERNEL2x4_E2 64,32,3,1 
+    blr
+
+
+CGEMM_2x2_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD2x2_2  
+    MY_ALIGN 
+CGEMM_L2x2_LOOP:
+/*----------------------------------------*/   
+    KERNEL2x2_L2 32,32,0,0 
+CGEMM_L2x2_K32:
+/*----------------------------------------*/   
+    KERNEL2x2_L2 32,32,1,0  
+    KERNEL2x2_L2 32,32,2,0
+    KERNEL2x2_L2 32,32,3,0  
+    KERNEL2x2_L2 32,32,4,0
+    KERNEL2x2_L2 32,32,5,0 
+    KERNEL2x2_L2 32,32,6,0
+    KERNEL2x2_L2 32,32,7,0
+    KERNEL2x2_L2 32,32,8,0
+    KERNEL2x2_L2 32,32,9,0  
+    KERNEL2x2_L2 32,32,10,0
+    KERNEL2x2_L2 32,32,11,0  
+    KERNEL2x2_L2 32,32,12,0
+    KERNEL2x2_L2 32,32,13,0 
+    KERNEL2x2_L2 32,32,14,0
+    KERNEL2x2_L2 32,32,15,1   
+    bdnz    CGEMM_L2x2_LOOP
+    MY_ALIGN  
+
+
+CGEMM_L2x2_LOOP_END:
+/*----------------------------------------*/   
+    END2x2_2 
+    blr
+    MY_ALIGN
+CGEMM_2x2_L16_SUB:
+/*----------------------------------------*/   
+    LOAD2x2_2
+    KERNEL2x2_L2 32,32,0,0
+    KERNEL2x2_L2 32,32,1,0  
+    KERNEL2x2_L2 32,32,2,0
+    KERNEL2x2_L2 32,32,3,0  
+    KERNEL2x2_L2 32,32,4,0
+    KERNEL2x2_L2 32,32,5,0 
+    KERNEL2x2_L2 32,32,6,0
+    KERNEL2x2_E2 32,32,7,1
+    blr
+    MY_ALIGN
+CGEMM_2x2_L8_SUB:
+/*----------------------------------------*/   
+    LOAD2x2_2
+    KERNEL2x2_L2 32,32,0,0
+    KERNEL2x2_L2 32,32,1,0  
+    KERNEL2x2_L2 32,32,2,0
+    KERNEL2x2_E2 32,32,3,1  
+    blr
+
+
+CGEMM_2x1_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD2x1_2  
+    MY_ALIGN
+CGEMM_L2x1_LOOP:
+/*----------------------------------------*/   
+    KERNEL2x1_L2 16,32,0,0 
+CGEMM_L2x1_K32:
+/*----------------------------------------*/   
+    KERNEL2x1_L2 16,32,1,0  
+    KERNEL2x1_L2 16,32,2,0
+    KERNEL2x1_L2 16,32,3,0  
+    KERNEL2x1_L2 16,32,4,0
+    KERNEL2x1_L2 16,32,5,0 
+    KERNEL2x1_L2 16,32,6,0
+    KERNEL2x1_L2 16,32,7,0
+    KERNEL2x1_L2 16,32,8,0
+    KERNEL2x1_L2 16,32,9,0  
+    KERNEL2x1_L2 16,32,10,0
+    KERNEL2x1_L2 16,32,11,0  
+    KERNEL2x1_L2 16,32,12,0
+    KERNEL2x1_L2 16,32,13,0 
+    KERNEL2x1_L2 16,32,14,0
+    KERNEL2x1_L2 16,32,15,1   
+    bdnz    CGEMM_L2x1_LOOP
+    MY_ALIGN  
+CGEMM_L2x1_LOOP_END:
+/*----------------------------------------*/   
+    END2x1_2 
+    blr
+
+    MY_ALIGN
+CGEMM_2x1_L16_SUB:
+/*----------------------------------------*/   
+    LOAD2x1_2
+    KERNEL2x1_L2 16,32,0,0
+    KERNEL2x1_L2 16,32,1,0  
+    KERNEL2x1_L2 16,32,2,0
+    KERNEL2x1_L2 16,32,3,0  
+    KERNEL2x1_L2 16,32,4,0
+    KERNEL2x1_L2 16,32,5,0 
+    KERNEL2x1_L2 16,32,6,0
+    KERNEL2x1_E2 16,32,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_2x1_L8_SUB:
+/*----------------------------------------*/   
+    LOAD2x1_2
+    KERNEL2x1_L2 16,32,0,0
+    KERNEL2x1_L2 16,32,1,0  
+    KERNEL2x1_L2 16,32,2,0
+    KERNEL2x1_E2 16,32,3,1  
+    blr
+
+
+
+/*             MAIN LOOP BEGINS               */   
+    MY_ALIGN
+
+
+CGEMM_L2:
+/*----------------------------------------*/   
+
+    andi.    J,  N,  2
+    ble   CGEMM_L2_END
+
+
+CGEMM_L2_BEGIN:
+/*----------------------------------------*/   
+    mr    CO, C
+    slwi    T1, LDC , 1     
+    add     T2,C,LDC    
+    mr    AO, A  
+    add   C,  C,  T1
+#if defined(TRMMKERNEL) && defined(LEFT)   
+    mr TEMP_REG, OFFSET  /*off = offset;*/
+#endif     
+    srawi.    I,  M, 3
+    ble   CGEMM_L2x8_END
+    dcbt    CO,r0  /*just prefetch*/
+    dcbt    T2,r0    
+
+
+CGEMM_L2x8_BEGIN:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,8,2
+#else    
+    mr    BO, B  
+    dcbt    B,  r0  
+#endif     
+    dcbt    AO, r0
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,8,2
+    mr T1, T6
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512   
+    srawi.   T8, T1, 7 /**(T1-2) % 128x */
+#else   
+    mr T1, K
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512 
+    srawi.   T8, T1, 7 /**(K-2) % 128x */
+#endif   
+    ZERO2x8  
+    ble   CGEMM_L2x8_SUB0
+    bl CGEMM_L2x8_LMAIN_SUB
+    andi.   L,  T1, 127
+    ble   CGEMM_L2x8_SAVE
+    b   CGEMM_L2x8_SUB2
+
+
+CGEMM_L2x8_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 255
+    cmpwi   T6,129
+#else   
+    andi.   L,  K,  255
+    cmpwi   K,129
+#endif       
+    li T8,1
+    bne CMP2x8_128K
+    addi BO,BO,-16
+    addi AO,AO,-64 
+    LOAD2x8O 64,16 
+    END2x8_WITHOUT_ADD   
+    LOAD2x8_2O  128, 32 
+    mtctr   T8    
+    bl CGEMM_L2x8_K128   
+    b CGEMM_L2x8_SAVE  
+    CMP2x8_128K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,128
+#else    
+    cmpwi   K,128
+#endif        
+    bne CGEMM_L2x8_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-32
+    addi AO,AO,-128   
+    LOAD2x8_2O 128,32
+    bl CGEMM_L2x8_K128   
+    b CGEMM_L2x8_SAVE 
+    MY_ALIGN
+
+
+CGEMM_L2x8_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 64
+    ble CGEMM_L2x8_SUB2_32
+    bl  CGEMM_2x8_L64_SUB
+    MY_ALIGN
+
+
+CGEMM_L2x8_SUB2_32:
+/*----------------------------------------*/   
+    andi.      T1,L, 32
+    ble CGEMM_L2x8_SUB2_16    
+    bl  CGEMM_2x8_L32_SUB
+    MY_ALIGN 
+
+
+CGEMM_L2x8_SUB2_16:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L2x8_SUB2_8
+    bl  CGEMM_2x8_L16_SUB  
+    MY_ALIGN    
+
+
+CGEMM_L2x8_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L2x8_SUB2_4
+    LOAD2x8_2
+    KERNEL2x8_L2  128,32, 0,0
+    KERNEL2x8_L2  128,32, 1,0
+    KERNEL2x8_L2  128,32, 2,0
+    KERNEL2x8_E2  128,32, 3,1
+    MY_ALIGN   
+
+
+CGEMM_L2x8_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L2x8_SUB2_2
+    LOAD2x8_2
+    KERNEL2x8_L2  128,32, 0,0
+    KERNEL2x8_E2  128,32, 1,1
+    MY_ALIGN
+
+
+CGEMM_L2x8_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L2x8_SUB2_1
+    LOAD2x8_2 
+    KERNEL2x8_E2  128,32, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L2x8_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L2x8_SAVE 
+    KERNEL2x8
+
+    MY_ALIGN
+CGEMM_L2x8_SAVE:
+/*----------------------------------------*/   
+    addic.    I,  I,  -1
+    MY_ALIGN
+    SAVE2x8
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,8,2
+#endif     
+    bgt   CGEMM_L2x8_BEGIN
+    andi.   T2, M,  7
+    ble   CGEMM_L2x1_END
+    andi.   T1, M,  4
+    ble   CGEMM_L2x4_END
+    b   CGEMM_L2x4_BEGIN
+    MY_ALIGN 
+
+
+CGEMM_L2x8_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L2x4_BEGIN:
+/*----------------------------------------*/   
+    andi.   T2, M,  7
+    ble   CGEMM_L2x1_END
+    andi.   T1, M,  4
+    ble   CGEMM_L2x4_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,4,2
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,4,2
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO2x4
+    ble   CGEMM_L2x4_SUB0 
+    bl CGEMM_2x4_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L2x4_SAVE
+    b    CGEMM_L2x4_SUB2
+
+
+CGEMM_L2x4_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP2x4_32K
+    addi BO,BO,-16
+    addi AO,AO,-32  
+    LOAD2x4O 32,16 
+    END2x4_WITHOUT_ADD   
+    LOAD2x4_2O  64, 32 
+    mtctr   T8    
+    bl CGEMM_L2x4_K32   
+    b CGEMM_L2x4_SAVE  
+    CMP2x4_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L2x4_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-32
+    addi AO,AO,-64   
+    LOAD2x4_2O 64,32
+    bl CGEMM_L2x4_K32   
+    b CGEMM_L2x4_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L2x4_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L2x4_SUB2_8
+    bl  CGEMM_2x4_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L2x4_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L2x4_SUB2_4
+    bl CGEMM_2x4_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L2x4_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L2x4_SUB2_2
+    LOAD2x4_2
+    KERNEL2x4_L2  64,32, 0,0
+    KERNEL2x4_E2  64,32, 1,1
+    MY_ALIGN
+
+
+CGEMM_L2x4_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L2x4_SUB2_1
+    LOAD2x4_2
+    KERNEL2x4_E2  64,32, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L2x4_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L2x4_SAVE 
+    KERNEL2x4
+
+
+CGEMM_L2x4_SAVE:
+/*----------------------------------------*/   
+    SAVE2x4
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,4,2
+#endif     
+
+
+CGEMM_L2x4_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L2x2_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M,  2
+    ble   CGEMM_L2x2_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,2,2
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,2,2
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO2x2
+    ble   CGEMM_L2x2_SUB0 
+    bl CGEMM_2x2_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L2x2_SAVE
+    b   CGEMM_L2x2_SUB2
+
+
+CGEMM_L2x2_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP2x2_32K
+    addi BO,BO,-16
+    addi AO,AO,-16  
+    LOAD2x2O 16,16 
+    END2x2_WITHOUT_ADD   
+    LOAD2x2_2O  32, 32  
+    mtctr   T8    
+    bl CGEMM_L2x2_K32   
+    b CGEMM_L2x2_SAVE  
+    CMP2x2_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L2x2_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-32
+    addi AO,AO,-32   
+    LOAD2x2_2O 32,32
+    bl CGEMM_L2x2_K32   
+    b CGEMM_L2x2_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L2x2_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L2x2_SUB2_8
+    bl CGEMM_2x2_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L2x2_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L2x2_SUB2_4
+    bl CGEMM_2x2_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L2x2_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L2x2_SUB2_2
+    LOAD2x2_2
+    KERNEL2x2_L2  32,32, 0,0
+    KERNEL2x2_E2  32,32, 1,1
+    MY_ALIGN
+
+
+CGEMM_L2x2_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L2x2_SUB2_1
+    LOAD2x2_2
+    KERNEL2x2_E2  32,32, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L2x2_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L2x2_SAVE 
+    KERNEL2x2
+
+    MY_ALIGN
+CGEMM_L2x2_SAVE:
+/*----------------------------------------*/   
+    SAVE2x2
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,2,2
+#endif     
+
+
+CGEMM_L2x2_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L2x1_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M,  1
+    ble   CGEMM_L2x1_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,1,2
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,1,2
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO2x1
+    ble   CGEMM_L2x1_SUB0 
+    bl CGEMM_2x1_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L2x1_SAVE
+    b   CGEMM_L2x1_SUB2
+
+
+CGEMM_L2x1_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP2x1_32K
+    addi BO,BO,-16
+    addi AO,AO,-8  
+    LOAD2x1O 8,16 
+    END2x1_WITHOUT_ADD   
+    LOAD2x1_2O  16, 32  
+    mtctr   T8    
+    bl CGEMM_L2x1_K32   
+    b CGEMM_L2x1_SAVE  
+    CMP2x1_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L2x1_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-32
+    addi AO,AO,-16   
+    LOAD2x1_2O 16,32
+    bl CGEMM_L2x1_K32   
+    b CGEMM_L2x1_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L2x1_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L2x1_SUB2_8
+    bl CGEMM_2x1_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L2x1_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L2x1_SUB2_4
+    bl CGEMM_2x1_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L2x1_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L2x1_SUB2_2
+    LOAD2x1_2
+    KERNEL2x1_L2  16,32, 0,0
+    KERNEL2x1_E2  16,32, 1,1
+    MY_ALIGN
+
+
+CGEMM_L2x1_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L2x1_SUB2_1
+    LOAD2x1_2
+    KERNEL2x1_E2  16,32, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L2x1_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L2x1_SAVE 
+    KERNEL2x1
+
+    MY_ALIGN
+CGEMM_L2x1_SAVE:
+/*----------------------------------------*/  
+     
+    SAVE2x1
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,1,2
+#endif   
+
+
+CGEMM_L2x1_END:
+/*----------------------------------------*/   
+    slwi    T1, K,  4
+
+    add   B,  B,  T1
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    addi TEMP_REG, TEMP_REG, 2
+#endif   
+
+CGEMM_L2_END:
+
+
+b CGEMM_L1
+/*                MINI SUBROUTINES                            */      
+/*                1x8 MAIN 128x+2 LOOP                     */      
+
+
+CGEMM_L1x8_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD1x8_2 
+    MY_ALIGN
+CGEMM_L1x8_LOOP:
+/*----------------------------------------*/   
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_L2 128,16,0,0 
+CGEMM_L1x8_K128:
+/*----------------------------------------*/   
+    KERNEL1x8_L2 128,16,1,0
+    dcbt    AO, T2  
+    KERNEL1x8_L2 128,16,2,0
+    KERNEL1x8_L2 128,16,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_L2 128,16,4,0
+    KERNEL1x8_L2 128,16,5,0
+    dcbt    AO, T4  
+    KERNEL1x8_L2 128,16,6,0
+    KERNEL1x8_L2 128,16,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL1x8_L2 128,16,8,0
+    KERNEL1x8_L2 128,16,9,0
+    KERNEL1x8_L2 128,16,10,0
+    KERNEL1x8_L2 128,16,11,0  
+    dcbt    BO, T4
+    KERNEL1x8_L2 128,16,12,0
+    KERNEL1x8_L2 128,16,13,0
+    KERNEL1x8_L2 128,16,14,0
+    KERNEL1x8_L2 128,16,15,0  
+    KERNEL1x8_L2 128,16,16,0
+    KERNEL1x8_L2 128,16,17,0 
+    KERNEL1x8_L2 128,16,18,0
+    KERNEL1x8_L2 128,16,19,0  
+    KERNEL1x8_L2 128,16,20,0
+    KERNEL1x8_L2 128,16,21,0 
+    KERNEL1x8_L2 128,16,22,0
+    KERNEL1x8_L2 128,16,23,0   
+    KERNEL1x8_L2 128,16,24,0
+    KERNEL1x8_L2 128,16,25,0
+    KERNEL1x8_L2 128,16,26,0
+    KERNEL1x8_L2 128,16,27,0  
+    KERNEL1x8_L2 128,16,28,0
+    KERNEL1x8_L2 128,16,29,0
+    KERNEL1x8_L2 128,16,30,0
+    KERNEL1x8_L2 128,16,31,0 
+    KERNEL1x8_L2 128,16,32,0
+    KERNEL1x8_L2 128,16,33,0
+    KERNEL1x8_L2 128,16,34,0
+    KERNEL1x8_L2 128,16,35,0 
+    KERNEL1x8_L2 128,16,36,0
+    KERNEL1x8_L2 128,16,37,0
+    KERNEL1x8_L2 128,16,38,0
+    KERNEL1x8_L2 128,16,39,0  
+    KERNEL1x8_L2 128,16,40,0
+    KERNEL1x8_L2 128,16,41,0
+    KERNEL1x8_L2 128,16,42,0
+    KERNEL1x8_L2 128,16,43,0  
+    KERNEL1x8_L2 128,16,44,0
+    KERNEL1x8_L2 128,16,45,0
+    KERNEL1x8_L2 128,16,46,0
+    KERNEL1x8_L2 128,16,47,0 
+    KERNEL1x8_L2 128,16,48,0
+    KERNEL1x8_L2 128,16,49,0 
+    KERNEL1x8_L2 128,16,50,0
+    KERNEL1x8_L2 128,16,51,0  
+    KERNEL1x8_L2 128,16,52,0
+    KERNEL1x8_L2 128,16,53,0 
+    KERNEL1x8_L2 128,16,54,0
+    KERNEL1x8_L2 128,16,55,0  
+    KERNEL1x8_L2 128,16,56,0
+    KERNEL1x8_L2 128,16,57,0
+    KERNEL1x8_L2 128,16,58,0
+    KERNEL1x8_L2 128,16,59,0  
+    KERNEL1x8_L2 128,16,60,0
+    KERNEL1x8_L2 128,16,61,0
+    KERNEL1x8_L2 128,16,62,0 
+    KERNEL1x8_L2 128,16,63,1  
+    bdnz    CGEMM_L1x8_LOOP
+    MY_ALIGN  
+CGEMM_L1x8_LOOP_END:
+/*----------------------------------------*/   
+    END1x8_2
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x8_L64_SUB:
+/*----------------------------------------*/   
+    LOAD1x8_2  
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_L2 128,16,0,0 
+    KERNEL1x8_L2 128,16,1,0
+    dcbt    AO, T2  
+    KERNEL1x8_L2 128,16,2,0
+    KERNEL1x8_L2 128,16,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_L2 128,16,4,0
+    KERNEL1x8_L2 128,16,5,0
+    dcbt    AO, T4  
+    KERNEL1x8_L2 128,16,6,0
+    KERNEL1x8_L2 128,16,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL1x8_L2 128,16,8,0
+    KERNEL1x8_L2 128,16,9,0
+    KERNEL1x8_L2 128,16,10,0
+    KERNEL1x8_L2 128,16,11,0  
+    dcbt    BO, T4
+    KERNEL1x8_L2 128,16,12,0
+    KERNEL1x8_L2 128,16,13,0
+    KERNEL1x8_L2 128,16,14,0
+    KERNEL1x8_L2 128,16,15,0  
+    KERNEL1x8_L2 128,16,16,0
+    KERNEL1x8_L2 128,16,17,0 
+    KERNEL1x8_L2 128,16,18,0
+    KERNEL1x8_L2 128,16,19,0  
+    KERNEL1x8_L2 128,16,20,0
+    KERNEL1x8_L2 128,16,21,0 
+    KERNEL1x8_L2 128,16,22,0
+    KERNEL1x8_L2 128,16,23,0   
+    KERNEL1x8_L2 128,16,24,0
+    KERNEL1x8_L2 128,16,25,0
+    KERNEL1x8_L2 128,16,26,0
+    KERNEL1x8_L2 128,16,27,0  
+    KERNEL1x8_L2 128,16,28,0
+    KERNEL1x8_L2 128,16,29,0
+    KERNEL1x8_L2 128,16,30,0
+    KERNEL1x8_E2 128,16,31,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x8_L32_SUB:
+/*----------------------------------------*/   
+    LOAD1x8_2  
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_L2 128,16,0,0 
+    KERNEL1x8_L2 128,16,1,0
+    dcbt    AO, T2  
+    KERNEL1x8_L2 128,16,2,0
+    KERNEL1x8_L2 128,16,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_L2 128,16,4,0
+    KERNEL1x8_L2 128,16,5,0
+    dcbt    AO, T4  
+    KERNEL1x8_L2 128,16,6,0
+    KERNEL1x8_L2 128,16,7,0  
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL1x8_L2 128,16,8,0
+    KERNEL1x8_L2 128,16,9,0
+    KERNEL1x8_L2 128,16,10,0
+    KERNEL1x8_L2 128,16,11,0  
+    dcbt    BO, T4
+    KERNEL1x8_L2 128,16,12,0
+    KERNEL1x8_L2 128,16,13,0
+    KERNEL1x8_L2 128,16,14,0
+    KERNEL1x8_E2 128,16,15,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x8_L16_SUB:
+/*----------------------------------------*/   
+    LOAD1x8_2 
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_L2 128,16,0,0 
+    KERNEL1x8_L2 128,16,1,0
+    dcbt    AO, T2  
+    KERNEL1x8_L2 128,16,2,0
+    KERNEL1x8_L2 128,16,3,0 
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_L2 128,16,4,0
+    KERNEL1x8_L2 128,16,5,0
+    dcbt    AO, T4  
+    KERNEL1x8_L2 128,16,6,0
+    KERNEL1x8_E2 128,16,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x4_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD1x4_2  
+    MY_ALIGN
+CGEMM_L1x4_LOOP:
+/*----------------------------------------*/   
+    KERNEL1x4_L2 64,16,0,0
+CGEMM_L1x4_K32:
+/*----------------------------------------*/   
+    KERNEL1x4_L2 64,16,1,0   
+    KERNEL1x4_L2 64,16,2,0
+    KERNEL1x4_L2 64,16,3,0  
+    KERNEL1x4_L2 64,16,4,0
+    KERNEL1x4_L2 64,16,5,0 
+    KERNEL1x4_L2 64,16,6,0
+    KERNEL1x4_L2 64,16,7,0
+    KERNEL1x4_L2 64,16,8,0
+    KERNEL1x4_L2 64,16,9,0   
+    KERNEL1x4_L2 64,16,10,0
+    KERNEL1x4_L2 64,16,11,0  
+    KERNEL1x4_L2 64,16,12,0
+    KERNEL1x4_L2 64,16,13,0 
+    KERNEL1x4_L2 64,16,14,0
+    KERNEL1x4_L2 64,16,15,1    
+    bdnz    CGEMM_L1x4_LOOP
+    MY_ALIGN  
+CGEMM_L1x4_LOOP_END:
+/*----------------------------------------*/   
+    END1x4_2 
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x4_L16_SUB:
+/*----------------------------------------*/   
+    LOAD1x4_2
+    KERNEL1x4_L2 64,16,0,0
+    KERNEL1x4_L2 64,16,1,0   
+    KERNEL1x4_L2 64,16,2,0
+    KERNEL1x4_L2 64,16,3,0  
+    KERNEL1x4_L2 64,16,4,0
+    KERNEL1x4_L2 64,16,5,0 
+    KERNEL1x4_L2 64,16,6,0
+    KERNEL1x4_E2 64,16,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x4_L8_SUB:
+/*----------------------------------------*/   
+    LOAD1x4_2
+    KERNEL1x4_L2 64,16,0,0
+    KERNEL1x4_L2 64,16,1,0   
+    KERNEL1x4_L2 64,16,2,0
+    KERNEL1x4_E2 64,16,3,1 
+    blr
+
+
+CGEMM_1x2_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD1x2_2  
+    MY_ALIGN 
+CGEMM_L1x2_LOOP:
+/*----------------------------------------*/   
+    KERNEL1x2_L2 32,16,0,0 
+CGEMM_L1x2_K32:
+/*----------------------------------------*/   
+    KERNEL1x2_L2 32,16,1,0  
+    KERNEL1x2_L2 32,16,2,0
+    KERNEL1x2_L2 32,16,3,0  
+    KERNEL1x2_L2 32,16,4,0
+    KERNEL1x2_L2 32,16,5,0 
+    KERNEL1x2_L2 32,16,6,0
+    KERNEL1x2_L2 32,16,7,0
+    KERNEL1x2_L2 32,16,8,0
+    KERNEL1x2_L2 32,16,9,0  
+    KERNEL1x2_L2 32,16,10,0
+    KERNEL1x2_L2 32,16,11,0  
+    KERNEL1x2_L2 32,16,12,0
+    KERNEL1x2_L2 32,16,13,0 
+    KERNEL1x2_L2 32,16,14,0
+    KERNEL1x2_L2 32,16,15,1   
+    bdnz    CGEMM_L1x2_LOOP
+    MY_ALIGN  
+
+
+CGEMM_L1x2_LOOP_END:
+/*----------------------------------------*/   
+    END1x2_2 
+    blr
+    MY_ALIGN
+CGEMM_1x2_L16_SUB:
+/*----------------------------------------*/   
+    LOAD1x2_2
+    KERNEL1x2_L2 32,16,0,0
+    KERNEL1x2_L2 32,16,1,0  
+    KERNEL1x2_L2 32,16,2,0
+    KERNEL1x2_L2 32,16,3,0  
+    KERNEL1x2_L2 32,16,4,0
+    KERNEL1x2_L2 32,16,5,0 
+    KERNEL1x2_L2 32,16,6,0
+    KERNEL1x2_E2 32,16,7,1
+    blr
+    MY_ALIGN
+CGEMM_1x2_L8_SUB:
+/*----------------------------------------*/   
+    LOAD1x2_2
+    KERNEL1x2_L2 32,16,0,0
+    KERNEL1x2_L2 32,16,1,0  
+    KERNEL1x2_L2 32,16,2,0
+    KERNEL1x2_E2 32,16,3,1  
+    blr
+
+
+CGEMM_1x1_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD1x1_2  
+    MY_ALIGN
+CGEMM_L1x1_LOOP:
+/*----------------------------------------*/   
+    KERNEL1x1_L2 16,16,0,0 
+CGEMM_L1x1_K32:
+/*----------------------------------------*/   
+    KERNEL1x1_L2 16,16,1,0  
+    KERNEL1x1_L2 16,16,2,0
+    KERNEL1x1_L2 16,16,3,0  
+    KERNEL1x1_L2 16,16,4,0
+    KERNEL1x1_L2 16,16,5,0 
+    KERNEL1x1_L2 16,16,6,0
+    KERNEL1x1_L2 16,16,7,0
+    KERNEL1x1_L2 16,16,8,0
+    KERNEL1x1_L2 16,16,9,0  
+    KERNEL1x1_L2 16,16,10,0
+    KERNEL1x1_L2 16,16,11,0  
+    KERNEL1x1_L2 16,16,12,0
+    KERNEL1x1_L2 16,16,13,0 
+    KERNEL1x1_L2 16,16,14,0
+    KERNEL1x1_L2 16,16,15,1   
+    bdnz    CGEMM_L1x1_LOOP
+    MY_ALIGN  
+CGEMM_L1x1_LOOP_END:
+/*----------------------------------------*/   
+    END1x1_2 
+    blr
+
+    MY_ALIGN
+CGEMM_1x1_L16_SUB:
+/*----------------------------------------*/   
+    LOAD1x1_2
+    KERNEL1x1_L2 16,16,0,0
+    KERNEL1x1_L2 16,16,1,0  
+    KERNEL1x1_L2 16,16,2,0
+    KERNEL1x1_L2 16,16,3,0  
+    KERNEL1x1_L2 16,16,4,0
+    KERNEL1x1_L2 16,16,5,0 
+    KERNEL1x1_L2 16,16,6,0
+    KERNEL1x1_E2 16,16,7,1
+    blr
+    MY_ALIGN
+
+
+CGEMM_1x1_L8_SUB:
+/*----------------------------------------*/   
+    LOAD1x1_2
+    KERNEL1x1_L2 16,16,0,0
+    KERNEL1x1_L2 16,16,1,0  
+    KERNEL1x1_L2 16,16,2,0
+    KERNEL1x1_E2 16,16,3,1  
+    blr
+
+
+
+/*             MAIN LOOP BEGINS               */   
+    MY_ALIGN
+
+
+CGEMM_L1:
+/*----------------------------------------*/   
+
+    andi.    J,  N,  1
+    ble   CGEMM_L1_END
+
+CGEMM_L1_BEGIN:
+/*----------------------------------------*/   
+    mr    CO, C  
+    add     T2,C,LDC    
+    mr    AO, A  
+    add   C,  C,  T1
+#if defined(TRMMKERNEL) && defined(LEFT)   
+    mr TEMP_REG, OFFSET  /*off = offset;*/
+#endif     
+    srawi.    I,  M,  3
+    ble   CGEMM_L1x8_END
+    dcbt    CO,r0  /*just prefetch*/
+    dcbt    T2,r0    
+
+
+CGEMM_L1x8_BEGIN:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,8,1
+#else    
+    mr    BO, B  
+    dcbt    B,  r0  
+#endif     
+    dcbt    AO, r0
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,8,1
+    mr T1, T6
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512   
+    srawi.   T8, T1, 7 /**(T1-2) % 128x */
+#else   
+    mr T1, K
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512 
+    srawi.   T8, T1, 7 /**(K-2) % 128x */
+#endif   
+    ZERO1x8  
+    ble   CGEMM_L1x8_SUB0
+    bl CGEMM_L1x8_LMAIN_SUB
+    andi.   L,  T1, 127
+    ble   CGEMM_L1x8_SAVE
+    b   CGEMM_L1x8_SUB2
+
+
+CGEMM_L1x8_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 255
+    cmpwi   T6,129
+#else   
+    andi.   L,  K,  255
+    cmpwi   K,129
+#endif       
+    li T8,1
+    bne CMP1x8_128K
+    addi BO,BO,-8
+    addi AO,AO,-64 
+    LOAD1x8O 64,8 
+    END1x8_WITHOUT_ADD   
+    LOAD1x8_2O  128, 16 
+    mtctr   T8    
+    bl CGEMM_L1x8_K128   
+    b CGEMM_L1x8_SAVE  
+    CMP1x8_128K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,128
+#else    
+    cmpwi   K,128
+#endif        
+    bne CGEMM_L1x8_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-16
+    addi AO,AO,-128   
+    LOAD1x8_2O 128,16
+    bl CGEMM_L1x8_K128   
+    b CGEMM_L1x8_SAVE 
+    MY_ALIGN
+
+
+CGEMM_L1x8_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 64
+    ble CGEMM_L1x8_SUB2_32
+    bl  CGEMM_1x8_L64_SUB
+    MY_ALIGN
+
+
+CGEMM_L1x8_SUB2_32:
+/*----------------------------------------*/   
+    andi.      T1,L, 32
+    ble CGEMM_L1x8_SUB2_16    
+    bl  CGEMM_1x8_L32_SUB
+    MY_ALIGN 
+
+
+CGEMM_L1x8_SUB2_16:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L1x8_SUB2_8
+    bl  CGEMM_1x8_L16_SUB  
+    MY_ALIGN    
+
+
+CGEMM_L1x8_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L1x8_SUB2_4
+    LOAD1x8_2
+    KERNEL1x8_L2  128,16, 0,0
+    KERNEL1x8_L2  128,16, 1,0
+    KERNEL1x8_L2  128,16, 2,0
+    KERNEL1x8_E2  128,16, 3,1
+    MY_ALIGN   
+
+
+CGEMM_L1x8_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L1x8_SUB2_2
+    LOAD1x8_2
+    KERNEL1x8_L2  128,16, 0,0
+    KERNEL1x8_E2  128,16, 1,1
+    MY_ALIGN
+
+
+CGEMM_L1x8_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L1x8_SUB2_1
+    LOAD1x8_2 
+    KERNEL1x8_E2  128,16, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L1x8_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L1x8_SAVE 
+    KERNEL1x8
+
+    MY_ALIGN
+CGEMM_L1x8_SAVE:
+/*----------------------------------------*/   
+    addic.    I,  I,  -1
+    MY_ALIGN
+    SAVE1x8
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,8,1
+#endif     
+    bgt   CGEMM_L1x8_BEGIN
+    andi.   T2, M,  7
+    ble   CGEMM_L1x1_END
+    andi.   T1, M,  4
+    ble   CGEMM_L1x4_END
+    b   CGEMM_L1x4_BEGIN
+    MY_ALIGN 
+
+
+CGEMM_L1x8_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L1x4_BEGIN:
+/*----------------------------------------*/   
+    andi.   T2, M,  7
+    ble   CGEMM_L1x1_END
+    andi.   T1, M,  4
+    ble   CGEMM_L1x4_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,4,1
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,4,1
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 31x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 31x */
+#endif     
+    ZERO1x4
+    ble   CGEMM_L1x4_SUB0 
+    bl CGEMM_1x4_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L1x4_SAVE
+    b    CGEMM_L1x4_SUB2
+
+
+CGEMM_L1x4_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP1x4_32K
+    addi BO,BO,-8
+    addi AO,AO,-32  
+    LOAD1x4O 32,8 
+    END1x4_WITHOUT_ADD   
+    LOAD1x4_2O  64, 16 
+    mtctr   T8    
+    bl CGEMM_L1x4_K32   
+    b CGEMM_L1x4_SAVE  
+    CMP1x4_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L1x4_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-16
+    addi AO,AO,-64   
+    LOAD1x4_2O 64,16
+    bl CGEMM_L1x4_K32   
+    b CGEMM_L1x4_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L1x4_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L1x4_SUB2_8
+    bl  CGEMM_1x4_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L1x4_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L1x4_SUB2_4
+    bl CGEMM_1x4_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L1x4_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L1x4_SUB2_2
+    LOAD1x4_2
+    KERNEL1x4_L2  64,16, 0,0
+    KERNEL1x4_E2  64,16, 1,1
+    MY_ALIGN
+
+
+CGEMM_L1x4_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L1x4_SUB2_1
+    LOAD1x4_2
+    KERNEL1x4_E2  64,16, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L1x4_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L1x4_SAVE 
+    KERNEL1x4
+
+
+CGEMM_L1x4_SAVE:
+/*----------------------------------------*/   
+    SAVE1x4
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,4,1
+#endif     
+
+
+CGEMM_L1x4_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L1x2_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M,  2
+    ble   CGEMM_L1x2_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,2,1
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,2,1
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 31x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 31x */
+#endif     
+    ZERO1x2
+    ble   CGEMM_L1x2_SUB0 
+    bl CGEMM_1x2_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L1x2_SAVE
+    b   CGEMM_L1x2_SUB2
+
+
+CGEMM_L1x2_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP1x2_32K
+    addi BO,BO,-8
+    addi AO,AO,-16  
+    LOAD1x2O 16,8 
+    END1x2_WITHOUT_ADD   
+    LOAD1x2_2O  32, 16  
+    mtctr   T8    
+    bl CGEMM_L1x2_K32   
+    b CGEMM_L1x2_SAVE  
+    CMP1x2_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L1x2_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-16
+    addi AO,AO,-32   
+    LOAD1x2_2O 32,16
+    bl CGEMM_L1x2_K32   
+    b CGEMM_L1x2_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L1x2_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L1x2_SUB2_8
+    bl CGEMM_1x2_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L1x2_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L1x2_SUB2_4
+    bl CGEMM_1x2_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L1x2_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L1x2_SUB2_2
+    LOAD1x2_2
+    KERNEL1x2_L2  32,16, 0,0
+    KERNEL1x2_E2  32,16, 1,1
+    MY_ALIGN
+
+
+CGEMM_L1x2_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L1x2_SUB2_1
+    LOAD1x2_2
+    KERNEL1x2_E2  32,16, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L1x2_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L1x2_SAVE 
+    KERNEL1x2
+
+    MY_ALIGN
+CGEMM_L1x2_SAVE:
+/*----------------------------------------*/   
+    SAVE1x2
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,2,1
+#endif     
+
+
+CGEMM_L1x2_END:
+/*----------------------------------------*/   
+
+
+CGEMM_L1x1_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M,  1
+    ble   CGEMM_L1x1_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO,BO,TEMP_REG,B,1,1
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG,1,1
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T1-2) % 31x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 31x */
+#endif     
+    ZERO1x1
+    ble   CGEMM_L1x1_SUB0 
+    bl CGEMM_1x1_LMAIN_SUB
+    andi.   L,  T1, 31
+    ble   CGEMM_L1x1_SAVE
+    b   CGEMM_L1x1_SUB2
+
+
+CGEMM_L1x1_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L,  T6, 63
+    cmpwi   T6,33
+#else   
+    andi.   L,  K,  63
+    cmpwi   K,33
+#endif       
+    li T8,1
+    bne CMP1x1_32K
+    addi BO,BO,-8
+    addi AO,AO,-8  
+    LOAD1x1O 8,8 
+    END1x1_WITHOUT_ADD   
+    LOAD1x1_2O  16, 16  
+    mtctr   T8    
+    bl CGEMM_L1x1_K32   
+    b CGEMM_L1x1_SAVE  
+    CMP1x1_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6,32
+#else    
+    cmpwi   K,32
+#endif        
+    bne CGEMM_L1x1_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO,BO,-16
+    addi AO,AO,-16   
+    LOAD1x1_2O 16,16
+    bl CGEMM_L1x1_K32   
+    b CGEMM_L1x1_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+CGEMM_L1x1_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble CGEMM_L1x1_SUB2_8
+    bl CGEMM_1x1_L16_SUB  
+    MY_ALIGN
+
+
+CGEMM_L1x1_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble CGEMM_L1x1_SUB2_4
+    bl CGEMM_1x1_L8_SUB
+    MY_ALIGN  
+
+
+CGEMM_L1x1_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble CGEMM_L1x1_SUB2_2
+    LOAD1x1_2
+    KERNEL1x1_L2  16,16, 0,0
+    KERNEL1x1_E2  16,16, 1,1
+    MY_ALIGN
+
+
+CGEMM_L1x1_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble CGEMM_L1x1_SUB2_1
+    LOAD1x1_2
+    KERNEL1x1_E2  16,16, 0,1
+    MY_ALIGN    
+
+
+CGEMM_L1x1_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble CGEMM_L1x1_SAVE 
+    KERNEL1x1
+
+    MY_ALIGN
+CGEMM_L1x1_SAVE:
+/*----------------------------------------*/  
+     
+    SAVE1x1
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG,BO,AO,1,1
+#endif   
+
+
+CGEMM_L1x1_END:
+/*----------------------------------------*/   
+    slwi    T1, K,  3
+
+    add   B,  B,  T1
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    addi TEMP_REG, TEMP_REG, 1
+#endif   
+
+CGEMM_L1_END:
+
+
+
+
diff --git a/kernel/power/cgemm_macros_power10.S b/kernel/power/cgemm_macros_power10.S
new file mode 100644
index 000000000..b66e93405
--- /dev/null
+++ b/kernel/power/cgemm_macros_power10.S
@@ -0,0 +1,2131 @@
+/***************************************************************************
+Copyright (c) 2013-2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define unit_size 8
+#define DISP32(ind, disp) (ind*unit_size*32+disp)
+#define DISP16(ind, disp) (ind*unit_size*16+disp)
+#define DISP8(ind, disp) (ind*unit_size*8+disp)
+#define DISP4(ind, disp) (ind*unit_size*4+disp)
+#define DISP2(ind, disp) (ind*unit_size*2+disp)
+#define DISP1(ind, disp) (ind*unit_size+disp)
+#define DISPX(disp)  (disp)
+
+.macro	AGGREGATE_REALS_IMAGES  VSINR_OUT1, VSINR, VSINI_OUT2, VSINI
+#if	defined(NN) || defined(NT) || defined(TN) || defined(TT)
+	xvsubsp  \VSINR_OUT1, \VSINR_OUT1, \VSINR
+	xvaddsp  \VSINI_OUT2, \VSINI_OUT2, \VSINI
+#elif  defined(CN) || defined(CT) || defined(RN) || defined(RT)
+	xvaddsp  \VSINR_OUT1, \VSINR_OUT1, \VSINR
+	xvsubsp  \VSINI_OUT2, \VSINI_OUT2, \VSINI
+#elif  defined(NC) || defined(TC) || defined(NR) || defined(TR)
+	xvaddsp  \VSINR_OUT1, \VSINR_OUT1, \VSINR
+	xvsubsp  \VSINI_OUT2, \VSINI, \VSINI_OUT2
+#else	// CC || CR || RC || RR
+    /*we will assume {-alpha_r,-alpha_i} for this case */
+    /*i1i2-r1r2 so we will negate alpha real instead to fix sign*/
+	xvsubsp  \VSINR_OUT1, \VSINR, \VSINR_OUT1
+    /*we will negate alpha image   instead to fix sign*/
+	xvaddsp  \VSINI_OUT2, \VSINI_OUT2, \VSINI
+#endif
+.endm
+
+.macro	AGGREGATE_REALS_IMAGES_A_PERMUTE  VSINR_OUT1, VSINR, VSINI_OUT2, VSINI
+#if	defined(NN) || defined(NT) || defined(TN) || defined(TT)
+	xvsubsp  \VSINR_OUT1, \VSINR_OUT1, \VSINR
+	xvaddsp  \VSINI_OUT2, \VSINI_OUT2, \VSINI
+#elif  defined(CN) || defined(CT) || defined(RN) || defined(RT)
+	xvaddsp  \VSINR_OUT1, \VSINR_OUT1, \VSINR
+	xvsubsp  \VSINI_OUT2, \VSINI, \VSINI_OUT2
+#elif  defined(NC) || defined(TC) || defined(NR) || defined(TR)
+	xvaddsp  \VSINR_OUT1, \VSINR_OUT1, \VSINR
+	xvsubsp  \VSINI_OUT2, \VSINI_OUT2, \VSINI
+#else	// CC || CR || RC || RR
+    /*we will assume {-alpha_r,-alpha_i} for this case */
+    /*i1i2-r1r2 so we will negate alpha real instead to fix sign*/
+	xvsubsp  \VSINR_OUT1, \VSINR, \VSINR_OUT1
+    /*we will negate alpha image   instead to fix sign*/
+	xvaddsp  \VSINI_OUT2, \VSINI_OUT2, \VSINI
+#endif
+.endm
+
+/* {i0,i1} * {alpha_i,alpha_i} [- VSOUT1] ;[VSOUT2 +] {r0,r1}*{alpha_i,alpha_i} */
+
+.macro MULT_APLHA_PART1  VSINRR, VSINII, VSOUT1, VSOUT2
+	xvmulsp \VSOUT1, \VSINII, alpha_i
+	xvmulsp  \VSOUT2, \VSINRR, alpha_i
+.endm
+
+/*   {r0,r1} * {alpha_r,alpha_r} -  VSOUT1 ;VSOUT2 + {i0,i1} * {alpha_r,alpha_r} */
+
+.macro MULT_APLHA_PART2  VSINRR, VSINII, VSOUT1, VSOUT2
+	xvmsubasp  \VSOUT1, \VSINRR, alpha_r
+	xvmaddasp \VSOUT2, \VSINII, alpha_r
+.endm
+
+.macro	PERMUTE1	OUT, R1, R2, R3, R4
+	xxsel	vs62, \R1, \R2, vs57
+	xxsel	\OUT, \R3, \R4, vs57
+	xxpermdi	\OUT, \OUT, vs62, 1
+.endm
+.macro	PERMUTE2	OUT, R1, R2, R3, R4
+	xxsel	vs62, \R2, \R1, vs57
+	xxsel	\OUT, \R4, \R3, vs57
+	xxpermdi	\OUT, vs62, \OUT, 1
+	xxperm	\OUT, \OUT, permute_mask
+.endm
+.macro PERMUTE3	OUT, R1, R2, R3, R4
+	xxsel	vs62, \R1, \R2, vs57
+	xxsel	\OUT, \R3, \R4, vs57
+	xxpermdi \OUT, vs62, \OUT, 2
+.endm
+.macro PERMUTE4	OUT, R1, R2, R3, R4
+	xxsel	vs62, \R2, \R1, vs57
+	xxsel	\OUT, \R4, \R3, vs57
+	xxpermdi	\OUT, \OUT, vs62, 2
+	xxperm	\OUT, \OUT, permute_mask
+.endm
+.macro	GROUP1
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	xxperm	vs1, vs33, permute_mask
+	xxperm	vs5, vs41, permute_mask
+	xxperm	vs8, vs36, permute_mask
+	xxperm	vs12, vs44, permute_mask
+	xxperm	vs9, vs37, permute_mask
+	xxperm	vs13, vs45, permute_mask
+.endm
+.macro	AGG_GROUP1
+	AGGREGATE_REALS_IMAGES	vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES	vs33, vs1, vs41, vs5
+	AGGREGATE_REALS_IMAGES	vs36, vs8, vs44, vs12
+	AGGREGATE_REALS_IMAGES	vs37, vs9, vs45, vs13
+.endm
+.macro	GROUP2
+	xxperm	vs0, vs34, permute_mask
+	xxperm	vs4, vs42, permute_mask
+	xxperm	vs1, vs35, permute_mask
+	xxperm	vs5, vs43, permute_mask
+	xxperm	vs8, vs38, permute_mask
+	xxperm	vs12, vs46, permute_mask
+	xxperm	vs9, vs39, permute_mask
+	xxperm	vs13, vs47, permute_mask
+.endm
+.macro	AGG_GROUP2
+	AGGREGATE_REALS_IMAGES	vs34, vs0, vs42, vs4
+	AGGREGATE_REALS_IMAGES	vs35, vs1, vs43, vs5
+	AGGREGATE_REALS_IMAGES	vs38, vs8, vs46, vs12
+	AGGREGATE_REALS_IMAGES	vs39, vs9, vs47, vs13
+.endm
+.macro	MULTIPLY_GROUP1
+	MULT_APLHA_PART1	vs32, vs40, vs0, vs1
+	MULT_APLHA_PART1	vs33, vs41, vs2, vs3
+	MULT_APLHA_PART1	vs36, vs44, vs8, vs9
+	MULT_APLHA_PART1	vs37, vs45, vs10, vs11
+	MULT_APLHA_PART2	vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2	vs33, vs41, vs2, vs3
+	MULT_APLHA_PART2	vs36, vs44, vs8, vs9
+	MULT_APLHA_PART2	vs37, vs45, vs10, vs11
+.endm
+.macro	MULTIPLY_GROUP2
+	MULT_APLHA_PART1	vs34, vs42, vs4, vs5
+	MULT_APLHA_PART1	vs35, vs43, vs6, vs7
+	MULT_APLHA_PART1	vs38, vs46, vs12, vs13
+	MULT_APLHA_PART1	vs39, vs47, vs14, vs15
+	MULT_APLHA_PART2	vs34, vs42, vs4, vs5
+	MULT_APLHA_PART2	vs35, vs43, vs6, vs7
+	MULT_APLHA_PART2	vs38, vs46, vs12, vs13
+	MULT_APLHA_PART2	vs39, vs47, vs14, vs15
+.endm
+/* reconstruct r, i pairs*/
+.macro	RECONSTRUCT_PAIR1
+	xxperm	vs0, vs1, save_permute_1
+	xxperm	vs2, vs3, save_permute_1
+	xxperm	vs8, vs9, save_permute_1
+	xxperm	vs10, vs11, save_permute_1
+.endm
+.macro	RECONSTRUCT_PAIR2
+	xxperm	vs4, vs5, save_permute_1
+	xxperm	vs6, vs7, save_permute_1
+	xxperm	vs12, vs13, save_permute_1
+	xxperm	vs14, vs15, save_permute_1
+.endm
+.macro	SHUFFLE_ACC	ACC, R0, R1, R2, R3, O1, O2, O3, O4
+	xxmfacc	\ACC
+	PERMUTE1	\O1, \R3, \R2, \R1, \R0
+	PERMUTE2	\O2, \R1, \R0, \R3, \R2
+	PERMUTE3	\O3, \R1, \R0, \R3, \R2
+	PERMUTE4	\O4, \R3, \R2, \R1, \R0
+.endm
+/*                                             macros for N=4 and M=8
+**********************************************************************************************/
+.macro	ZERO4x8
+	xxsetaccz	0
+	xxsetaccz	1
+	xxsetaccz	2
+	xxsetaccz	3
+	xxsetaccz	4
+	xxsetaccz	5
+	xxsetaccz	6
+	xxsetaccz	7
+.endm
+
+.macro	LOAD4x8
+	LOAD4x8O	0, 0
+.endm
+
+.macro	LOAD4x8O  OffsetA, OffsetB
+	lxvp	vs34, (\OffsetB+0)(BO)
+	lxvp	vs32, (\OffsetA+0)(AO)
+	lxvp	vs36, (\OffsetA+32)(AO)
+.endm
+
+.macro	END4x8_NORMAL
+	END4x8	AO, BO, 64, 32
+.endm
+
+.macro	END4x8_WITHOUT_ADD
+	END4x8	AO, BO, 0, 0
+.endm
+
+.macro	END4x8	AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi	\BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi	\AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	3, 36, 35
+	xvf32gerpp	2, 37, 35
+	xvf32gerpp	1, 32, 35
+	xvf32gerpp	0, 33, 35
+	xvf32gerpp	7, 36, 34
+	xvf32gerpp	6, 37, 34
+	xvf32gerpp	5, 32, 34
+	xvf32gerpp	4, 33, 34
+.endm
+
+.macro	LOAD4x8_2
+	LOAD4x8_2O	0, 0
+.endm
+
+.macro	LOAD4x8_2O  OffsetA, OffsetB
+	lxvp	vs34, (\OffsetB)(BO)
+	lxvp	vs38, (32+\OffsetB)(BO)
+	lxvp	vs32, (0+\OffsetA)(AO)
+	lxvp	vs36, (32+\OffsetA)(AO)
+	lxvp	vs40, (64+\OffsetA)(AO)
+	lxvp	vs42, (64+32+\OffsetA)(AO)
+.endm
+
+.macro	END4x8_2
+	/*for load2 offset will be 128 and 64*/
+	KERNEL4x8_2	AO, BO, 128, 64, 0, 1, 1
+.endm
+
+.macro	KERNEL4x8_E2	OffsetA, OffsetB, Index, IsLast
+	KERNEL4x8_2	AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL4x8_L2	OffsetA, OffsetB, Index, IsLast
+	KERNEL4x8_2	AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL4x8_2	AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	3, 36, 35
+	xvf32gerpp	2, 37, 35
+	xvf32gerpp	1, 32, 35
+	xvf32gerpp	0, 33, 35
+	xvf32gerpp	7, 36, 34
+	xvf32gerpp	6, 37, 34
+	xvf32gerpp	5, 32, 34
+	xvf32gerpp	4, 33, 34
+.if \Complete==0
+	lxvp	vs34, DISP8(\Index, \OffsetB)(\BREG)
+	lxvp	vs32, DISP16(\Index, 0+\OffsetA)(\AREG)
+	lxvp	vs36, DISP16(\Index, 32+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	3, 42, 39
+	xvf32gerpp	2, 43, 39
+	xvf32gerpp	1, 40, 39
+	xvf32gerpp	0, 41, 39
+	xvf32gerpp	7, 42, 38
+	xvf32gerpp	6, 43, 38
+	xvf32gerpp	5, 40, 38
+	xvf32gerpp	4, 41, 38
+.if \Complete==0
+	lxvp	vs40, DISP16(\Index, 64+\OffsetA)(\AREG)
+	lxvp	vs38, DISP8(\Index, 32+\OffsetB)(\BREG)
+	lxvp	vs42, DISP16(\Index, 64+32+\OffsetA)(\AREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi	\BREG, \BREG, DISP8(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP16(\Index, \OffsetA)
+.else
+	addi	\BREG, \BREG, DISP8(\Index, 64)
+	addi    \AREG, \AREG, DISP16(\Index, 128)
+.endif
+.endif
+.endm
+
+.macro	KERNEL4x8
+	LOAD4x8
+	END4x8	AO, BO, 64, 32
+.endm
+
+.macro SAVE4x8
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	SHUFFLE_ACC	2, vs8, vs9, vs10, vs11, vs34, vs42, vs38, vs46
+	SHUFFLE_ACC	3, vs12, vs13, vs14, vs15, vs35, vs43, vs39, vs47
+	SHUFFLE_ACC	4, vs16, vs17, vs18, vs19, vs48, vs56, vs52, vs60
+	SHUFFLE_ACC	5, vs20, vs21, vs22, vs23, vs49, vs16, vs53, vs61
+	SHUFFLE_ACC	7, vs28, vs29, vs30, vs31, vs17, vs19, vs18, vs20
+	SHUFFLE_ACC	6, vs24, vs25, vs26, vs27, vs50, vs58, vs54, vs21
+	add	T4, LDC, LDC
+	add	T1, CO, LDC
+#ifndef TRMMKERNEL
+	lxvp	vs24, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs26, 32(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs28, 0(T1)
+#endif
+	xxperm	vs2, vs34, permute_mask
+	xxperm	vs6, vs42, permute_mask
+#ifndef TRMMKERNEL
+	lxvp	vs30, 32(T1)
+#endif
+	xxperm	vs3, vs35, permute_mask
+	xxperm	vs7, vs43, permute_mask
+	add	T2, CO, T4
+	add	T3, T1, T4
+	GROUP1
+	AGG_GROUP1
+	AGGREGATE_REALS_IMAGES	vs34, vs2, vs42, vs6
+	xxperm	vs10, vs38, permute_mask
+	xxperm	vs14, vs46, permute_mask
+	AGGREGATE_REALS_IMAGES	vs35, vs3, vs43, vs7
+	xxperm	vs11, vs39, permute_mask
+	xxperm	vs15, vs47, permute_mask
+	xxperm	vs0, vs48, permute_mask
+	xxperm	vs4, vs56, permute_mask
+	xxperm	vs1, vs49, permute_mask
+	xxperm	vs5, vs16, permute_mask
+	AGGREGATE_REALS_IMAGES	vs38, vs10, vs46, vs14
+	xxperm	vs2, vs50, permute_mask
+	xxperm	vs6, vs58, permute_mask
+	AGGREGATE_REALS_IMAGES	vs39, vs11, vs47, vs15
+	xxperm	vs3, vs17, permute_mask
+	xxperm	vs7, vs19, permute_mask
+	AGGREGATE_REALS_IMAGES	vs48, vs0, vs56, vs4
+	xxperm	vs8, vs52, permute_mask
+	xxperm	vs12, vs60, permute_mask
+	AGGREGATE_REALS_IMAGES	vs49, vs1, vs16, vs5
+	xxperm	vs9, vs53, permute_mask
+	xxperm	vs13, vs61, permute_mask
+	AGGREGATE_REALS_IMAGES	vs50, vs2, vs58, vs6
+	xxperm	vs10, vs54, permute_mask
+	xxperm	vs14, vs21, permute_mask
+	AGGREGATE_REALS_IMAGES	vs17, vs3, vs19, vs7
+	xxperm	vs11, vs18, permute_mask
+	xxperm	vs15, vs20, permute_mask
+	AGGREGATE_REALS_IMAGES	vs52, vs8, vs60, vs12
+	AGGREGATE_REALS_IMAGES	vs53, vs9, vs61, vs13
+/*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	AGGREGATE_REALS_IMAGES	vs54, vs10, vs21, vs14
+	MULT_APLHA_PART1    vs33, vs41, vs2, vs3
+	AGGREGATE_REALS_IMAGES	vs18, vs11, vs20, vs15
+	MULT_APLHA_PART1    vs34, vs42, vs4, vs5
+	MULT_APLHA_PART1    vs35, vs43, vs6, vs7
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs33, vs41, vs2, vs3
+	MULT_APLHA_PART2    vs34, vs42, vs4, vs5
+	MULT_APLHA_PART2    vs35, vs43, vs6, vs7
+#ifndef TRMMKERNEL
+	lxvp	vs32, 0(T2)
+#endif
+	MULT_APLHA_PART1    vs36, vs44, vs8, vs9
+	MULT_APLHA_PART1    vs37, vs45, vs10, vs11
+#ifndef TRMMKERNEL
+	lxvp	vs40, 32(T2)
+#endif
+	MULT_APLHA_PART1    vs38, vs46, vs12, vs13
+	MULT_APLHA_PART1    vs39, vs47, vs14, vs15
+#ifndef TRMMKERNEL
+	lxvp	vs34, 0(T3)
+#endif
+	MULT_APLHA_PART2    vs36, vs44, vs8, vs9
+	MULT_APLHA_PART2    vs37, vs45, vs10, vs11
+#ifndef TRMMKERNEL
+	lxvp	vs42, 32(T3)
+#endif
+	MULT_APLHA_PART2    vs38, vs46, vs12, vs13
+	MULT_APLHA_PART2    vs39, vs47, vs14, vs15
+	RECONSTRUCT_PAIR1
+	RECONSTRUCT_PAIR2
+#ifndef TRMMKERNEL
+	/* add */
+	xxpermdi	vs1, vs8, vs0, 2
+	xxpermdi	vs3, vs10, vs2, 2
+	xxpermdi	vs5, vs12, vs4, 2
+	xxpermdi	vs7, vs14, vs6, 2
+	xxpermdi	vs9, vs0, vs8, 2
+	xxpermdi	vs11, vs2, vs10, 2
+	xvaddsp	vs24, vs24, vs3
+	xvaddsp	vs25, vs25, vs1
+	xxpermdi	vs13, vs4, vs12, 2
+	xxpermdi	vs15, vs6, vs14, 2
+	xvaddsp	vs26, vs26, vs7
+	xvaddsp	vs27, vs27, vs5
+	xvaddsp	vs28, vs28, vs11
+	xvaddsp	vs29, vs29, vs9
+	xvaddsp	vs30, vs30, vs15
+	xvaddsp	vs31, vs31, vs13
+#else
+	xxpermdi	vs25, vs8, vs0, 2
+	xxpermdi	vs24, vs10, vs2, 2
+	xxpermdi	vs27, vs12, vs4, 2
+	xxpermdi	vs26, vs14, vs6, 2
+	xxpermdi	vs29, vs0, vs8, 2
+	xxpermdi	vs28, vs2, vs10, 2
+	xxpermdi	vs31, vs4, vs12, 2
+	xxpermdi	vs30, vs6, vs14, 2
+#endif
+	stxvp	vs24, 0(CO)
+	MULT_APLHA_PART1    vs48, vs56, vs0, vs1
+	MULT_APLHA_PART1    vs49, vs16, vs2, vs3
+	stxvp	vs26, 32(CO)
+	MULT_APLHA_PART1    vs50, vs58, vs4, vs5
+	MULT_APLHA_PART1    vs17, vs19, vs6, vs7
+	stxvp	vs28, 0(T1)
+	MULT_APLHA_PART2    vs48, vs56, vs0, vs1
+	MULT_APLHA_PART2    vs49, vs16, vs2, vs3
+	stxvp	vs30, 32(T1)
+	MULT_APLHA_PART2    vs50, vs58, vs4, vs5
+	MULT_APLHA_PART2    vs17, vs19, vs6, vs7
+	MULT_APLHA_PART1    vs52, vs60, vs8, vs9
+	MULT_APLHA_PART1    vs53, vs61, vs10, vs11
+	MULT_APLHA_PART1    vs54, vs21, vs12, vs13
+	MULT_APLHA_PART1    vs18, vs20, vs14, vs15
+	MULT_APLHA_PART2    vs52, vs60, vs8, vs9
+	MULT_APLHA_PART2    vs53, vs61, vs10, vs11
+	MULT_APLHA_PART2    vs54, vs21, vs12, vs13
+	MULT_APLHA_PART2    vs18, vs20, vs14, vs15
+	RECONSTRUCT_PAIR1
+	RECONSTRUCT_PAIR2
+#ifndef TRMMKERNEL
+  /* add */
+	xxpermdi	vs1, vs8, vs0, 2
+	xxpermdi	vs3, vs10, vs2, 2
+	xxpermdi	vs5, vs12, vs4, 2
+	xxpermdi	vs7, vs14, vs6, 2
+	xxpermdi	vs9, vs0, vs8, 2
+	xxpermdi	vs11, vs2, vs10, 2
+	xvaddsp	vs32, vs32, vs3
+	xvaddsp	vs33, vs33, vs1
+	xxpermdi	vs13, vs4, vs12, 2
+	xxpermdi	vs15, vs6, vs14, 2
+	xvaddsp	vs40, vs40, vs7
+	xvaddsp vs41, vs41, vs5
+	xvaddsp	vs34, vs34, vs11
+	xvaddsp	vs35, vs35, vs9
+	xvaddsp	vs42, vs42, vs15
+	xvaddsp	vs43, vs43, vs13
+#else
+	xxpermdi	vs33, vs8, vs0, 2
+	xxpermdi	vs32, vs10, vs2, 2
+	xxpermdi	vs41, vs12, vs4, 2
+	xxpermdi	vs40, vs14, vs6, 2
+	xxpermdi	vs35, vs0, vs8, 2
+	xxpermdi	vs34, vs2, vs10, 2
+	xxpermdi	vs43, vs4, vs12, 2
+	xxpermdi	vs42, vs6, vs14, 2
+#endif
+	stxvp	vs32, 0(T2)
+	stxvp	vs40, 32(T2)
+	stxvp	vs34, 0(T3)
+	stxvp	vs42, 32(T3)
+	addi	CO, CO, 64
+.endm
+
+/*                                             macros for N=4 and M=4
+**********************************************************************************************/
+
+.macro	ZERO4x4
+	xxsetaccz	0
+	xxsetaccz	1
+	xxsetaccz	2
+	xxsetaccz	3
+.endm
+
+.macro	LOAD4x4
+	LOAD4x4O 0, 0
+.endm
+
+.macro	LOAD4x4O  OffsetA, OffsetB
+	lxvp	vs34, (\OffsetB+0)(BO)
+	lxvp	vs32, (\OffsetA+0)(AO)
+.endm
+
+.macro	END4x4_NORMAL
+	END4x4 AO, BO, 32, 32
+.endm
+
+.macro	END4x4_WITHOUT_ADD
+	END4x4 AO, BO, 0, 0
+.endm
+
+.macro	END4x4	AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi	\BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi	\AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	3, 32, 34
+	xvf32gerpp	2, 33, 34
+	xvf32gerpp	1, 32, 35
+	xvf32gerpp	0, 33, 35
+.endm
+
+.macro	LOAD4x4_2
+	LOAD4x4_2O 0, 0
+.endm
+
+.macro	LOAD4x4_2O  OffsetA, OffsetB
+	lxvp	vs34, (\OffsetB)(BO)
+	lxvp	vs38, (32+\OffsetB)(BO)
+	lxvp	vs32, (0+\OffsetA)(AO)
+	lxvp	vs36, (32+\OffsetA)(AO)
+.endm
+
+.macro	END4x4_2
+  /*for load2 offset will be 64 and 64*/
+	KERNEL4x4_2	AO, BO, 64, 64, 0, 1, 1
+.endm
+
+.macro	KERNEL4x4_E2	OffsetA, OffsetB, Index, IsLast
+	KERNEL4x4_2	AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL4x4_L2	OffsetA, OffsetB, Index, IsLast
+	KERNEL4x4_2	AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL4x4_2	AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	3, 32, 34
+	xvf32gerpp	2, 33, 34
+	xvf32gerpp	1, 32, 35
+	xvf32gerpp	0, 33, 35
+.if \Complete==0
+	lxvp	vs34, DISP8(\Index, \OffsetB)(\BREG)
+	lxvp	vs32, DISP8(\Index, 0+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	3, 36, 38
+	xvf32gerpp	2, 37, 38
+	xvf32gerpp	1, 36, 39
+	xvf32gerpp	0, 37, 39
+.if \Complete==0
+	lxvp	vs38, DISP8(\Index, 32+\OffsetB)(\BREG)
+	lxvp	vs36, DISP8(\Index, 32+\OffsetA)(\AREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi		\BREG, \BREG, DISP8(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP8(\Index, \OffsetA)
+.else
+	addi		\BREG, \BREG, DISP8(\Index, 64)
+	addi    \AREG, \AREG, DISP8(\Index, 64)
+.endif
+.endif
+.endm
+
+.macro	KERNEL4x4
+	LOAD4x4
+	END4x4  AO, BO, 32, 32
+.endm
+
+.macro SAVE4x4
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	SHUFFLE_ACC	2, vs8, vs9, vs10, vs11, vs34, vs42, vs38, vs46
+	SHUFFLE_ACC	3, vs12, vs13, vs14, vs15, vs35, vs43, vs39, vs47
+	add	T4, LDC, LDC
+	add	T1, CO, LDC
+#ifndef TRMMKERNEL
+	lxvp	vs24, 0(CO)
+#endif
+	add	T2, CO, T4
+	add	T3, T1, T4
+#ifndef TRMMKERNEL
+	lxvp	vs26, 0(T1)
+#endif
+ #ifndef TRMMKERNEL
+	lxvp	vs28, 0(T2)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs30, 0(T3)
+#endif
+	GROUP1
+	AGG_GROUP1
+	GROUP2
+	AGG_GROUP2
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULTIPLY_GROUP1
+	MULTIPLY_GROUP2
+/* reconstruct r, i pairs*/
+	RECONSTRUCT_PAIR1
+	RECONSTRUCT_PAIR2
+#ifndef TRMMKERNEL
+  /* add */
+	xxpermdi	vs1, vs8, vs0, 2
+	xxpermdi	vs3, vs10, vs2, 2
+	xxpermdi	vs9, vs0, vs8, 2
+	xxpermdi	vs11, vs2, vs10, 2
+	xxpermdi	vs5, vs12, vs4, 2
+	xxpermdi	vs7, vs14, vs6, 2
+	xxpermdi	vs13, vs4, vs12, 2
+	xxpermdi	vs15, vs6, vs14, 2
+	xvaddsp	vs24, vs24, vs3
+	xvaddsp	vs25, vs25, vs1
+	xvaddsp	vs26, vs26, vs11
+	xvaddsp	vs27, vs27, vs9
+	xvaddsp	vs28, vs28, vs7
+	xvaddsp	vs29, vs29, vs5
+	xvaddsp	vs30, vs30, vs15
+	xvaddsp	vs31, vs31, vs13
+#else
+	xxpermdi	vs25, vs8, vs0, 2
+	xxpermdi	vs24, vs10, vs2, 2
+	xxpermdi	vs27, vs0, vs8, 2
+	xxpermdi	vs26, vs2, vs10, 2
+	xxpermdi	vs29, vs12, vs4, 2
+	xxpermdi	vs28, vs14, vs6, 2
+	xxpermdi	vs31, vs4, vs12, 2
+	xxpermdi	vs30, vs6, vs14, 2
+#endif
+	stxvp	vs24, 0(CO)
+	stxvp	vs26, 0(T1)
+	stxvp	vs28, 0(T2)
+	stxvp	vs30, 0(T3)
+	addi  CO, CO, 32
+.endm
+
+/*                                             macros for N=4 and M=2
+**********************************************************************************************/
+
+.macro	ZERO4x2
+	xxsetaccz	0
+	xxsetaccz	1
+.endm
+
+.macro	LOAD4x2
+	LOAD4x2O 0, 0
+.endm
+
+.macro	LOAD4x2O  OffsetA, OffsetB
+	lxv	vs32, (\OffsetA+0)(AO)
+	lxvp	vs34, (\OffsetB+0)(BO)
+.endm
+
+.macro	END4x2_NORMAL
+	END4x2 AO, BO, 16, 32
+.endm
+
+.macro	END4x2_WITHOUT_ADD
+	END4x2 AO, BO, 0, 0
+.endm
+
+.macro	END4x2	AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi	\BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi	\AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	1, 34, 32
+	xvf32gerpp	0, 35, 32
+.endm
+
+.macro	LOAD4x2_2
+	LOAD4x2_2O 0, 0
+.endm
+
+.macro	LOAD4x2_2O  OffsetA, OffsetB
+	lxvp	vs32, (\OffsetA)(AO)
+	lxvp	vs34, (0+\OffsetB)(BO)
+	lxvp	vs36, (32+\OffsetB)(BO)
+.endm
+
+.macro	END4x2_2
+  /*for load2 offset will be 32 and 64*/
+	KERNEL4x2_2	AO, BO, 32, 64, 0, 1, 1
+.endm
+
+.macro	KERNEL4x2_E2	OffsetA, OffsetB, Index, IsLast
+	KERNEL4x2_2	AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL4x2_L2	OffsetA, OffsetB, Index, IsLast
+	KERNEL4x2_2	AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL4x2_2	AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	1, 34, 33
+	xvf32gerpp	0, 35, 33
+.if \Complete==0
+	lxvp	vs34, DISP8(\Index, 0+\OffsetB)(\BREG)
+.endif
+	xvf32gerpp	1, 36, 32
+	xvf32gerpp	0, 37, 32
+.if \Complete==0
+	lxvp	vs32, DISP4(\Index, \OffsetA)(\AREG)
+	lxvp	vs36, DISP8(\Index, 32+\OffsetB)(\BREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \AREG, \AREG, DISP4(\Index, \OffsetA)
+	addi		\BREG, \BREG, DISP8(\Index, \OffsetB)
+.else
+	addi    \AREG, \AREG, DISP4(\Index, 32)
+	addi		\BREG, \BREG, DISP8(\Index, 64)
+.endif
+.endif
+.endm
+
+.macro	KERNEL4x2
+	LOAD4x2
+	END4x2  AO, BO, 16, 32
+.endm
+
+.macro SAVE4x2
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	add	T4, LDC, LDC
+	add	T1, CO, LDC
+	add	T2, CO, T4
+	add	T3, T1, T4
+#ifndef TRMMKERNEL
+	lxv	vs24, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxv	vs25, 0(T1)
+#endif
+#ifndef TRMMKERNEL
+	lxv	vs26, 0(T2)
+#endif
+#ifndef TRMMKERNEL
+	lxv	vs27, 0(T3)
+#endif
+	GROUP1
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs33, vs1, vs41, vs5
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs36, vs8, vs44, vs12
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs37, vs9, vs45, vs13
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULTIPLY_GROUP1
+/* reconstruct r, i pairs*/
+	RECONSTRUCT_PAIR1
+#ifndef TRMMKERNEL
+  /* add */
+	xxpermdi	vs1, vs8, vs0, 0
+	xxpermdi	vs9, vs10, vs2, 0
+	xxpermdi	vs3, vs0, vs8, 3
+	xxpermdi	vs11, vs2, vs10, 3
+	xvaddsp	vs24, vs24, vs1
+	xvaddsp	vs26, vs26, vs9
+	xvaddsp	vs25, vs25, vs3
+	xvaddsp	vs27, vs27, vs11
+#else
+	xxpermdi	vs24, vs8, vs0, 0
+	xxpermdi	vs26, vs10, vs2, 0
+	xxpermdi	vs25, vs0, vs8, 3
+	xxpermdi	vs27, vs2, vs10, 3
+#endif
+	stxv	vs24, 0(CO)
+	stxv	vs25, 0(T1)
+	stxv	vs26, 0(T2)
+	stxv	vs27, 0(T3)
+	addi  CO, CO, 16
+.endm
+
+/*                                             macros for N=4 and M=2
+**********************************************************************************************/
+
+.macro	ZERO4x1
+	xxsetaccz	0
+	xxsetaccz	1
+.endm
+
+.macro	LOAD4x1
+	LOAD4x1O 0, 0
+.endm
+
+.macro	LOAD4x1O  OffsetA, OffsetB
+	lxsd	v0, (\OffsetA+0)(AO)
+	lxvp	vs34, (\OffsetB+0)(BO)
+.endm
+
+.macro	END4x1_NORMAL
+	END4x1 AO, BO,8, 32
+.endm
+
+.macro	END4x1_WITHOUT_ADD
+	END4x1 AO, BO, 0, 0
+.endm
+
+.macro	END4x1 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	    0, 35, 32
+	xvf32gerpp	    1, 34, 32
+.endm
+
+.macro	LOAD4x1_2
+	LOAD4x1_2O 0, 0
+.endm
+
+.macro	LOAD4x1_2O  OffsetA, OffsetB
+	lxv	vs32, (\OffsetA)(AO)
+	vspltisb        v6, 0
+	xxpermdi        vs33, vs32, vs38, 0
+	xxpermdi        vs32, vs32, vs38, 2
+	lxvp	vs34, (0+\OffsetB)(BO)
+	lxvp	vs36, (32+\OffsetB)(BO)
+.endm
+
+.macro	END4x1_2
+  /*for load2 offset will be 16 and 64*/
+	KERNEL4x1_2  AO, BO, 16, 64, 0, 1, 1
+.endm
+
+.macro	KERNEL4x1_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL4x1_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL4x1_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL4x1_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL4x1_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	    0, 35, 32
+	xvf32gerpp	    1, 34, 32
+.if \Complete==0
+	lxvp	vs34, DISP8(\Index, 0+\OffsetB)(\BREG)
+.endif
+	xvf32gerpp	    0, 37, 33
+	xvf32gerpp	    1, 36, 33
+.if \Complete==0
+	lxv	vs32, DISP2(\Index, \OffsetA)(\AREG)
+	lxvp	vs36, DISP8(\Index, 32+\OffsetB)(\BREG)
+	xxpermdi        vs33, vs32, vs38, 0
+	xxpermdi        vs32, vs32, vs38, 2
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \AREG, \AREG, DISP2(\Index, \OffsetA)
+	addi    \BREG, \BREG, DISP8(\Index, \OffsetB)
+.else
+	addi    \AREG, \AREG, DISP2(\Index, 16)
+	addi    \BREG, \BREG, DISP8(\Index, 64)
+.endif
+.endif
+.endm
+
+.macro	KERNEL4x1
+	LOAD4x1
+	END4x1  AO, BO, 8, 32
+.endm
+
+.macro SAVE4x1
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	xxpermdi	vs32, vs32, vs36, 1
+	xxpermdi	vs40, vs40, vs44, 1
+	xxpermdi	vs33, vs33, vs37, 1
+	xxpermdi	vs41, vs41, vs45, 1
+	add	T4, LDC, LDC
+	add	T1, CO, LDC
+	add	T2, CO, T4
+	add	T3, T1, T4
+#ifndef TRMMKERNEL
+	lxsd	v4, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxsd	v5, 0(T1)
+#endif
+#ifndef TRMMKERNEL
+	lxsd	v6, 0(T2)
+#endif
+#ifndef TRMMKERNEL
+	lxsd	v7, 0(T3)
+#endif
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	xxperm	vs1, vs33, permute_mask
+	xxperm	vs5, vs41, permute_mask
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs33, vs1, vs41, vs5
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART1    vs33, vs41, vs2, vs3
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs33, vs41, vs2, vs3
+/* reconstruct r, i pairs*/
+	xxperm	vs0, vs1, save_permute_1
+	xxperm	vs2, vs3, save_permute_1
+#ifndef TRMMKERNEL
+  /* add */
+	xxspltd vs1, vs0, 0
+	xxspltd vs3, vs0, 1
+	xxspltd vs9, vs2, 0
+	xxspltd vs11, vs2, 1
+ /*--v4==vs36 v5==vs37 v6==vs38 v7==vs39---*/
+	xvaddsp	vs36, vs36, vs1
+	xvaddsp	vs37, vs37, vs3
+	xvaddsp	vs38, vs38, vs9
+	xvaddsp	vs39, vs39, vs11
+#else
+ /*--v4==vs36 v5==vs37 v6==vs38 v7==vs39---*/
+	xxspltd vs36, vs0, 0
+	xxspltd vs37, vs0, 1
+	xxspltd vs38, vs2, 0
+	xxspltd vs39, vs2, 1
+#endif
+	stxsd	v4, 0(CO)
+	stxsd	v5, 0(T1)
+	stxsd	v6, 0(T2)
+	stxsd	v7, 0(T3)
+	addi  CO, CO, 8
+.endm
+
+/*                                             macros for N=2 and M=8
+**********************************************************************************************/
+
+.macro	ZERO2x8
+	xxsetaccz	0
+	xxsetaccz	1
+	xxsetaccz	2
+	xxsetaccz	3
+.endm
+
+.macro	LOAD2x8
+	LOAD2x8O 0, 0
+.endm
+
+.macro	LOAD2x8O  OffsetA, OffsetB
+	lxv	vs34, (\OffsetB+0)(BO)
+	lxvp	vs32, (\OffsetA+0)(AO)
+	lxvp	vs36, (\OffsetA+32)(AO)
+.endm
+
+.macro	END2x8_NORMAL
+	END2x8 AO, BO, 64, 16
+.endm
+
+.macro	END2x8_WITHOUT_ADD
+	END2x8 AO, BO, 0, 0
+.endm
+
+.macro	END2x8 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	2, 37, 34
+	xvf32gerpp	3, 36, 34
+	xvf32gerpp	0, 33, 34
+	xvf32gerpp	1, 32, 34
+.endm
+
+.macro	LOAD2x8_2
+	LOAD2x8_2O 0, 0
+.endm
+
+.macro	LOAD2x8_2O  OffsetA, OffsetB
+	lxvp	vs34, (\OffsetB)(BO)
+	lxvp	vs32, (0+\OffsetA)(AO)
+	lxvp	vs36, (32+\OffsetA)(AO)
+	lxvp	vs38, (64+\OffsetA)(AO)
+	lxvp	vs40, (64+32+\OffsetA)(AO)
+.endm
+
+.macro	END2x8_2
+  /*for load2 offset will be 128 and 32*/
+	KERNEL2x8_2  AO, BO, 128, 32, 0, 1, 1
+.endm
+
+.macro	KERNEL2x8_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x8_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL2x8_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x8_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL2x8_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	2, 37, 35
+	xvf32gerpp	3, 36, 35
+	xvf32gerpp	0, 33, 35
+	xvf32gerpp	1, 32, 35
+
+.if \Complete==0
+	lxvp	vs32, DISP16(\Index, 0+\OffsetA)(\AREG)
+	lxvp	vs36, DISP16(\Index, 32+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	2, 41, 34
+	xvf32gerpp	3, 40, 34
+	xvf32gerpp	0, 39, 34
+	xvf32gerpp	1, 38, 34
+
+.if \Complete==0
+	lxvp	vs34, DISP4(\Index, \OffsetB)(\BREG)
+	lxvp	vs38, DISP16(\Index, 64+\OffsetA)(\AREG)
+	lxvp	vs40, DISP16(\Index, 64+32+\OffsetA)(\AREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \BREG, \BREG, DISP4(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP16(\Index, \OffsetA)
+.else
+	addi    \BREG, \BREG, DISP4(\Index, 32)
+	addi    \AREG, \AREG, DISP16(\Index, 128)
+.endif
+.endif
+.endm
+
+.macro	KERNEL2x8
+	LOAD2x8
+	END2x8  AO, BO, 64, 16
+.endm
+
+.macro SAVE2x8
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	SHUFFLE_ACC	2, vs8, vs9, vs10, vs11, vs34, vs42, vs38, vs46
+	SHUFFLE_ACC	3, vs12, vs13, vs14, vs15, vs35, vs43, vs39, vs47
+	add	T1, CO, LDC
+#ifndef TRMMKERNEL
+	lxvp	vs24, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs26, 32(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs28, 0(T1)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs30, 32(T1)
+#endif
+	add	T2, CO, T4
+	add	T3, T1, T4
+	GROUP1
+	AGG_GROUP1
+	GROUP2
+	AGG_GROUP2
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULTIPLY_GROUP1
+	MULTIPLY_GROUP2
+/* reconstruct r, i pairs*/
+	RECONSTRUCT_PAIR1
+	RECONSTRUCT_PAIR2
+#ifndef TRMMKERNEL
+  /* add */
+	xxpermdi	vs1, vs8, vs0, 2
+	xxpermdi	vs3, vs10, vs2, 2
+	xxpermdi	vs5, vs12, vs4, 2
+	xxpermdi	vs7, vs14, vs6, 2
+	xxpermdi	vs9, vs0, vs8, 2
+	xxpermdi	vs11, vs2, vs10, 2
+	xvaddsp	vs24, vs24, vs3
+	xvaddsp	vs25, vs25, vs1
+	xxpermdi	vs13, vs4, vs12, 2
+	xxpermdi	vs15, vs6, vs14, 2
+	xvaddsp	vs26, vs26, vs7
+	xvaddsp	vs27, vs27, vs5
+	xvaddsp	vs28, vs28, vs11
+	xvaddsp	vs29, vs29, vs9
+	xvaddsp	vs30, vs30, vs15
+	xvaddsp	vs31, vs31, vs13
+#else
+	xxpermdi	vs25, vs8, vs0, 2
+	xxpermdi	vs24, vs10, vs2, 2
+	xxpermdi	vs27, vs12, vs4, 2
+	xxpermdi	vs26, vs14, vs6, 2
+	xxpermdi	vs29, vs0, vs8, 2
+	xxpermdi	vs28, vs2, vs10, 2
+	xxpermdi	vs31, vs4, vs12, 2
+	xxpermdi	vs30, vs6, vs14, 2
+#endif
+	stxvp	vs24, 0(CO)
+	stxvp	vs26, 32(CO)
+	stxvp	vs28, 0(T1)
+	stxvp	vs30, 32(T1)
+	addi  CO, CO, 64
+.endm
+
+/*                                             macros for N=2 and M=4
+**********************************************************************************************/
+
+.macro	ZERO2x4
+	xxsetaccz	0
+	xxsetaccz	1
+.endm
+
+.macro	LOAD2x4
+	LOAD2x4O 0, 0
+.endm
+
+.macro	LOAD2x4O  OffsetA, OffsetB
+	lxv	vs34, (\OffsetB+0)(BO)
+	lxvp	vs32, (\OffsetA+0)(AO)
+.endm
+
+.macro	END2x4_NORMAL
+	END2x4 AO, BO, 32, 16
+.endm
+
+.macro	END2x4_WITHOUT_ADD
+	END2x4 AO, BO, 0, 0
+.endm
+
+.macro	END2x4 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	0, 33, 34
+	xvf32gerpp	1, 32, 34
+.endm
+
+.macro	LOAD2x4_2
+	LOAD2x4_2O 0, 0
+.endm
+
+.macro	LOAD2x4_2O  OffsetA, OffsetB
+	lxvp	vs34, (\OffsetB)(BO)
+	lxvp	vs32, (0+\OffsetA)(AO)
+	lxvp	vs36, (32+\OffsetA)(AO)
+.endm
+
+.macro	END2x4_2
+  /*for load2 offset will be 64 and 32*/
+	KERNEL2x4_2  AO, BO, 64, 32, 0, 1, 1
+.endm
+
+.macro	KERNEL2x4_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x4_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL2x4_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x4_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL2x4_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	0, 33, 35
+	xvf32gerpp	1, 32, 35
+.if \Complete==0
+	lxvp	vs32, DISP8(\Index, 0+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	0, 37, 34
+	xvf32gerpp	1, 36, 34
+.if \Complete==0
+	lxvp	vs34, DISP4(\Index, \OffsetB)(\BREG)
+	lxvp	vs36, DISP8(\Index, 32+\OffsetA)(\AREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \BREG, \BREG, DISP4(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP8(\Index, \OffsetA)
+.else
+	addi    \BREG, \BREG, DISP4(\Index, 32)
+	addi    \AREG, \AREG, DISP8(\Index, 64)
+.endif
+.endif
+.endm
+
+.macro	KERNEL2x4
+	LOAD2x4
+	END2x4  AO, BO, 32, 16
+.endm
+
+.macro SAVE2x4
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	add	T1, CO, LDC
+#ifndef TRMMKERNEL
+	lxvp	vs24, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxvp	vs26, 0(T1)
+#endif
+	GROUP1
+	AGG_GROUP1
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULTIPLY_GROUP1
+/* reconstruct r, i pairs*/
+	RECONSTRUCT_PAIR1
+#ifndef TRMMKERNEL
+  /* add */
+	xxpermdi	vs1, vs8, vs0, 2
+	xxpermdi	vs3, vs10, vs2, 2
+	xxpermdi	vs9, vs0, vs8, 2
+	xxpermdi	vs11, vs2, vs10, 2
+	xvaddsp	vs24, vs24, vs3
+	xvaddsp	vs25, vs25, vs1
+	xvaddsp	vs26, vs26, vs11
+	xvaddsp	vs27, vs27, vs9
+#else
+	xxpermdi	vs25, vs8, vs0, 2
+	xxpermdi	vs24, vs10, vs2, 2
+	xxpermdi	vs27, vs0, vs8, 2
+	xxpermdi	vs26, vs2, vs10, 2
+#endif
+	stxvp	vs24, 0(CO)
+	stxvp	vs26, 0(T1)
+	addi  CO, CO, 32
+.endm
+
+/*                                             macros for N=2 and M=2
+**********************************************************************************************/
+
+.macro	ZERO2x2
+	xxsetaccz	0
+.endm
+
+.macro	LOAD2x2
+	LOAD2x2O 0, 0
+.endm
+
+.macro	LOAD2x2O  OffsetA, OffsetB
+	lxv	vs32, (\OffsetA+0)(AO)
+	lxv	vs34, (\OffsetB+0)(BO)
+.endm
+
+.macro	END2x2_NORMAL
+	END2x2 AO, BO, 16, 16
+.endm
+
+.macro	END2x2_WITHOUT_ADD
+	END2x2 AO, BO, 0, 0
+.endm
+
+.macro	END2x2 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	0, 34, 32
+.endm
+
+.macro	LOAD2x2_2
+	LOAD2x2_2O 0, 0
+.endm
+
+.macro	LOAD2x2_2O  OffsetA, OffsetB
+	lxvp	vs32, (\OffsetA)(AO)
+	lxvp	vs34, (0+\OffsetB)(BO)
+.endm
+
+.macro	END2x2_2
+  /*for load2 offset will be 32 and 32*/
+	KERNEL2x2_2  AO, BO, 32, 32, 0, 1, 1
+.endm
+
+.macro	KERNEL2x2_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x2_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL2x2_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x2_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL2x2_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	0, 34, 32
+	xvf32gerpp	0, 35, 33
+.if \Complete==0
+	lxvp	vs32, DISP4(\Index, \OffsetA)(\AREG)
+	lxvp	vs34, DISP4(\Index, \OffsetA)(\BREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \AREG, \AREG, DISP4(\Index, \OffsetA)
+	addi    \BREG, \BREG, DISP4(\Index, \OffsetB)
+.else
+	addi    \AREG, \AREG, DISP4(\Index, 32)
+	addi    \BREG, \BREG, DISP4(\Index, 32)
+.endif
+.endif
+.endm
+
+.macro	KERNEL2x2
+	LOAD2x2
+	END2x2  AO, BO, 16, 16
+.endm
+
+.macro SAVE2x2
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	add	T1, CO, LDC
+#ifndef TRMMKERNEL
+	lxv	vs24, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxv	vs26, 0(T1)
+#endif
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	xxperm	vs8, vs36, permute_mask
+	xxperm	vs12, vs44, permute_mask
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs36, vs8, vs44, vs12
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART1    vs36, vs44, vs8, vs9
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs36, vs44, vs8, vs9
+/* reconstruct r, i pairs*/
+	xxperm	vs0, vs1, save_permute_1
+	xxperm	vs8, vs9, save_permute_1
+#ifndef TRMMKERNEL
+  /* add */
+	xxpermdi	vs1, vs8, vs0, 0
+	xxpermdi	vs9, vs0, vs8, 3
+	xvaddsp	vs24, vs24, vs1
+	xvaddsp	vs26, vs26, vs9
+#else
+	xxpermdi	vs24, vs8, vs0, 0
+	xxpermdi	vs26, vs0, vs8, 3
+#endif
+	stxv	vs24, 0(CO)
+	stxv	vs26, 0(T1)
+	addi  CO, CO, 16
+.endm
+
+/*                                             macros for N=2 and M=1
+**********************************************************************************************/
+
+.macro	ZERO2x1
+	xxlxor  vs32, vs32, vs32
+	xxlxor  vs40, vs40, vs40
+.endm
+
+.macro	LOAD2x1
+	LOAD2x1O 0, 0
+.endm
+
+.macro	LOAD2x1O  OffsetA, OffsetB
+	lxsd	v4, (\OffsetA+0)(AO)
+	lxv	vs0, (\OffsetB+0)(BO)
+	xxspltd  vs24, vs36, 0
+	xxperm    vs26, vs24, permute_mask
+.endm
+
+.macro	END2x1_NORMAL
+	END2x1 AO, BO,8, 16
+.endm
+
+.macro	END2x1_WITHOUT_ADD
+	END2x1 AO, BO, 0, 0
+.endm
+
+.macro	END2x1 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvmaddasp	vs32, vs0, vs24
+	xvmaddasp	vs40, vs0, vs26
+.endm
+
+.macro	LOAD2x1_2
+	LOAD2x1_2O 0, 0
+.endm
+
+.macro	LOAD2x1_2O  OffsetA, OffsetB
+	lxv	vs27, (\OffsetA)(AO)
+	lxvp	vs4, (0+\OffsetB)(BO)
+	xxspltd  vs8, vs27, 1
+	xxspltd  vs24, vs27, 0
+	xxperm    vs10, vs8, permute_mask
+	xxperm    vs26, vs24, permute_mask
+.endm
+
+.macro	END2x1_2
+  /*for load2 offset will be 16 and 32*/
+	KERNEL2x1_2  AO, BO, 16, 32, 0, 1, 1
+.endm
+
+.macro	KERNEL2x1_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x1_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL2x1_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL2x1_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL2x1_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvmaddasp	vs32, vs5, vs8
+	xvmaddasp	vs40, vs5, vs10
+.if \Complete==0
+	lxv	vs27, DISP2(\Index, \OffsetA)(\AREG)
+	xxspltd  vs8, vs27, 1
+.endif
+.if \Complete==0
+	xxperm    vs10, vs8, permute_mask
+.endif
+	xvmaddasp	vs32, vs4, vs24
+	xvmaddasp	vs40, vs4, vs26
+.if \Complete==0
+	xxspltd  vs24, vs27, 0
+	xxperm   vs26, vs24, permute_mask
+.endif
+.if \Complete==0
+	lxvp	vs4, DISP4(\Index, 0+\OffsetB)(\BREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \AREG, \AREG, DISP2(\Index, \OffsetA)
+	addi    \BREG, \BREG, DISP4(\Index, \OffsetB)
+.else
+	addi    \AREG, \AREG, DISP2(\Index, 16)
+	addi    \BREG, \BREG, DISP4(\Index, 32)
+.endif
+.endif
+.endm
+
+.macro	KERNEL2x1
+	LOAD2x1
+	END2x1  AO, BO, 8, 16
+.endm
+
+.macro SAVE2x1
+	add	T1, CO, LDC
+#ifndef TRMMKERNEL
+	lxsd	v4, 0(CO)
+#endif
+#ifndef TRMMKERNEL
+	lxsd	v5, 0(T1)
+#endif
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES_A_PERMUTE vs33, vs1, vs41, vs5
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+/* reconstruct r, i pairs*/
+	xxperm	vs0, vs1, save_permute_1
+#ifndef TRMMKERNEL
+  /* add */
+	xxspltd vs1, vs0, 0
+	xxspltd vs3, vs0, 1
+ /*--v4==vs36 v5==vs37---*/
+	xvaddsp	vs36, vs36, vs1
+	xvaddsp	vs37, vs37, vs3
+#else
+ /*--v4==vs36 v5==vs37---*/
+	xxspltd vs36, vs0, 0
+	xxspltd vs37, vs0, 1
+#endif
+	stxsd	v4, 0(CO)
+	stxsd	v5, 0(T1)
+	addi  CO, CO, 8
+.endm
+
+/*                                             macros for N=1 and M=8
+**********************************************************************************************/
+
+.macro	ZERO1x8
+	xxsetaccz	0
+	xxsetaccz	1
+	xxsetaccz	2
+	xxsetaccz	3
+.endm
+
+.macro	LOAD1x8
+	LOAD1x8O 0, 0
+.endm
+
+.macro	LOAD1x8O  OffsetA, OffsetB
+	lxsd	v2, (\OffsetB+0)(BO)
+	lxvp	vs32, (\OffsetA+0)(AO)
+	lxvp	vs36, (\OffsetA+32)(AO)
+.endm
+
+.macro	END1x8_NORMAL
+	END1x8 AO, BO, 64,8
+.endm
+
+.macro	END1x8_WITHOUT_ADD
+	END1x8 AO, BO, 0, 0
+.endm
+
+.macro	END1x8 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	    0, 34, 33
+	xvf32gerpp	    1, 34, 32
+	xvf32gerpp	    2, 34, 37
+	xvf32gerpp	    3, 34, 36
+.endm
+
+.macro	LOAD1x8_2
+	LOAD1x8_2O 0, 0
+.endm
+
+.macro	LOAD1x8_2O  OffsetA, OffsetB
+	lxv	vs34, (\OffsetB)(BO)
+	lxvp	vs32, (0+\OffsetA)(AO)
+	lxvp	vs36, (32+\OffsetA)(AO)
+	vspltisb        v10, 0
+	xxpermdi        vs35, vs34, vs42, 0
+	xxpermdi        vs34, vs34, vs42, 2
+	lxvp	vs38, (64+\OffsetA)(AO)
+	lxvp	vs40, (64+32+\OffsetA)(AO)
+.endm
+
+.macro	END1x8_2
+  /*for load2 offset will be 128 and 16*/
+	KERNEL1x8_2  AO, BO, 128, 16, 0, 1, 1
+.endm
+
+.macro	KERNEL1x8_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x8_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL1x8_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x8_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL1x8_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	    0, 34, 33
+	xvf32gerpp	    1, 34, 32
+.if \Complete==0
+	lxvp	vs32, DISP16(\Index, 0+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	    2, 34, 37
+	xvf32gerpp	    3, 34, 36
+.if \Complete==0
+	lxvp	vs36, DISP16(\Index, 32+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	    0, 35, 39
+	xvf32gerpp	    1, 35, 38
+.if \Complete==0
+	lxvp	vs38, DISP16(\Index, 64+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	    2, 35, 41
+	xvf32gerpp	    3, 35, 40
+.if \Complete==0
+	lxv	vs34, DISP2(\Index, \OffsetB)(\BREG)
+	xxpermdi        vs35, vs34, vs42, 0
+	xxpermdi        vs34, vs34, vs42, 2
+	lxvp	vs40, DISP16(\Index, 64+32+\OffsetA)(\AREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \BREG, \BREG, DISP2(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP16(\Index, \OffsetA)
+.else
+	addi    \BREG, \BREG, DISP2(\Index, 16)
+	addi    \AREG, \AREG, DISP16(\Index, 128)
+.endif
+.endif
+.endm
+
+.macro	KERNEL1x8
+	LOAD1x8
+	END1x8  AO, BO, 64,8
+.endm
+
+.macro SAVE1x8
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	SHUFFLE_ACC	2, vs8, vs9, vs10, vs11, vs34, vs42, vs38, vs46
+	SHUFFLE_ACC	3, vs12, vs13, vs14, vs15, vs35, vs43, vs39, vs47
+	xxpermdi	vs32, vs32, vs36, 0
+	xxpermdi	vs33, vs33, vs37, 0
+	xxpermdi	vs34, vs34, vs38, 0
+	xxpermdi	vs35, vs35, vs39, 0
+	xxpermdi	vs40, vs40, vs44, 0
+	xxperm vs40, vs40, permute_mask
+	xxpermdi	vs41, vs41, vs45, 0
+	xxperm vs41, vs41, permute_mask
+	xxpermdi	vs42, vs42, vs46, 0
+	xxperm vs42, vs42, permute_mask
+	xxpermdi	vs43, vs43, vs47, 0
+	xxperm vs43, vs43, permute_mask
+#ifndef TRMMKERNEL
+	lxvp	vs24, 0(CO)
+#endif
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+#ifndef TRMMKERNEL
+	lxvp	vs26, 32(CO)
+#endif
+	xxperm	vs1, vs33, permute_mask
+	xxperm	vs5, vs41, permute_mask
+	xxperm	vs2, vs34, permute_mask
+	xxperm	vs6, vs42, permute_mask
+	xxperm	vs3, vs35, permute_mask
+	xxperm	vs7, vs43, permute_mask
+	AGGREGATE_REALS_IMAGES	vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES	vs33, vs1, vs41, vs5
+	AGGREGATE_REALS_IMAGES	vs34, vs2, vs42, vs6
+	AGGREGATE_REALS_IMAGES	vs35, vs3, vs43, vs7
+  /*inner reverse save_permute and store vs28 */
+	xxpermdi	vs28,save_permute_1,save_permute_1, 2
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART1    vs33, vs41, vs2, vs3
+	MULT_APLHA_PART1    vs34, vs42, vs4, vs5
+	MULT_APLHA_PART1    vs35, vs43, vs6, vs7
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs33, vs41, vs2, vs3
+	MULT_APLHA_PART2    vs34, vs42, vs4, vs5
+	MULT_APLHA_PART2    vs35, vs43, vs6, vs7
+/* reconstruct r, i pairs*/
+	xxperm	vs0, vs1, vs28
+	xxperm	vs2, vs3, vs28
+	xxperm	vs4, vs5, vs28
+	xxperm	vs6, vs7, vs28
+#ifndef TRMMKERNEL
+  /* add */
+	xvaddsp	vs24, vs24, vs2
+	xvaddsp	vs25, vs25, vs0
+	xvaddsp	vs26, vs26, vs6
+	xvaddsp	vs27, vs27, vs4
+	stxvp	vs24, 0(CO)
+	stxvp	vs26, 32(CO)
+#else
+/* reconstruct r, i pairs*/
+	stxv	vs0, 0(CO)
+	stxv	vs2, 16(CO)
+	stxv	vs4, 32(CO)
+	stxv	vs6, 48(CO)
+#endif
+	addi  CO, CO, 64
+.endm
+
+/*                                             macros for N=1 and M=4
+**********************************************************************************************/
+
+.macro	ZERO1x4
+	xxsetaccz	0
+	xxsetaccz	1
+.endm
+
+.macro	LOAD1x4
+	LOAD1x4O 0, 0
+.endm
+
+.macro	LOAD1x4O  OffsetA, OffsetB
+	lxsd	v2, (\OffsetB+0)(BO)
+	lxvp	vs32, (\OffsetA+0)(AO)
+.endm
+
+.macro	END1x4_NORMAL
+	END1x4 AO, BO, 32,8
+.endm
+
+.macro	END1x4_WITHOUT_ADD
+	END1x4 AO, BO, 0, 0
+.endm
+
+.macro	END1x4 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvf32gerpp	    0, 34, 33
+	xvf32gerpp	    1, 34, 32
+.endm
+
+.macro	LOAD1x4_2
+	LOAD1x4_2O 0, 0
+.endm
+
+.macro	LOAD1x4_2O  OffsetA, OffsetB
+	lxv	vs34, (\OffsetB)(BO)
+	lxvp	vs32, (0+\OffsetA)(AO)
+	vspltisb        v6, 0
+	xxpermdi        vs35, vs34, vs38, 0
+	xxpermdi        vs34, vs34, vs38, 2
+	lxvp	vs36, (32+\OffsetA)(AO)
+.endm
+
+.macro	END1x4_2
+  /*for load2 offset will be 64 and 16*/
+	KERNEL1x4_2  AO, BO, 64, 16, 0, 1, 1
+.endm
+
+.macro	KERNEL1x4_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x4_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL1x4_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x4_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL1x4_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvf32gerpp	    0, 34, 33
+	xvf32gerpp	    1, 34, 32
+.if \Complete==0
+	lxvp	vs32, DISP8(\Index, 0+\OffsetA)(\AREG)
+.endif
+	xvf32gerpp	    0, 35, 37
+	xvf32gerpp	    1, 35, 36
+.if \Complete==0
+	lxv	vs34, DISP2(\Index, \OffsetB)(\BREG)
+	xxpermdi        vs35, vs34, vs38, 0
+	xxpermdi        vs34, vs34, vs38, 2
+	lxvp	vs36, DISP8(\Index, 32+\OffsetA)(\AREG)
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \BREG, \BREG, DISP2(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP8(\Index, \OffsetA)
+.else
+	addi    \BREG, \BREG, DISP2(\Index, 16)
+	addi    \AREG, \AREG, DISP8(\Index, 64)
+.endif
+.endif
+.endm
+
+.macro	KERNEL1x4
+	LOAD1x4
+	END1x4	AO, BO, 32,8
+.endm
+
+.macro SAVE1x4
+	SHUFFLE_ACC	0, vs0, vs1, vs2, vs3, vs32, vs40, vs36, vs44
+	SHUFFLE_ACC	1, vs4, vs5, vs6, vs7, vs33, vs41, vs37, vs45
+	xxpermdi	vs32, vs32, vs36, 0
+	xxpermdi	vs40, vs40, vs44, 0
+	xxpermdi	vs33, vs33, vs37, 0
+	xxpermdi	vs41, vs41, vs45, 0
+	xxperm vs40, vs40, permute_mask
+	xxperm vs41, vs41, permute_mask
+#ifndef TRMMKERNEL
+	lxvp	vs24, 0(CO)
+#endif
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	xxperm	vs1, vs33, permute_mask
+	xxperm	vs5, vs41, permute_mask
+	AGGREGATE_REALS_IMAGES	vs32, vs0, vs40, vs4
+	AGGREGATE_REALS_IMAGES	vs33, vs1, vs41, vs5
+  /*inner reverse save_permute and store vs28 */
+	xxpermdi	vs28,save_permute_1,save_permute_1, 2
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART1    vs33, vs41, vs2, vs3
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs33, vs41, vs2, vs3
+/* reconstruct r, i pairs*/
+	xxperm	vs0, vs1, vs28
+	xxperm	vs2, vs3, vs28
+#ifndef TRMMKERNEL
+  /* add */
+	xvaddsp	vs24, vs24, vs2
+	xvaddsp	vs25, vs25, vs0
+	stxvp	vs24, 0(CO)
+#else
+/* reconstruct r, i pairs*/
+	stxv	vs0, 0(CO)
+	stxv	vs2, 16(CO)
+#endif
+	addi  CO, CO, 32
+.endm
+
+/*                                             macros for N=1 and M=2
+**********************************************************************************************/
+
+.macro	ZERO1x2
+	xxlxor  vs32, vs32, vs32
+	xxlxor  vs40, vs40, vs40
+.endm
+
+.macro	LOAD1x2
+	LOAD1x2O 0, 0
+.endm
+
+.macro	LOAD1x2O  OffsetA, OffsetB
+	lxsd	vs4, (\OffsetB+0)(BO)
+	lxv	vs0, (\OffsetA+0)(AO)
+	xxspltd   vs24, vs36, 0
+	xxperm    vs26, vs24, permute_mask
+.endm
+
+.macro	END1x2_NORMAL
+	END1x2 AO, BO, 16,8
+.endm
+
+.macro	END1x2_WITHOUT_ADD
+	END1x2 AO, BO, 0, 0
+.endm
+
+.macro	END1x2 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvmaddasp	vs32, vs0, vs24
+	xvmaddasp	vs40, vs0, vs26
+.endm
+
+.macro	LOAD1x2_2
+	LOAD1x2_2O 0, 0
+.endm
+
+.macro	LOAD1x2_2O  OffsetA, OffsetB
+	lxv	vs27, (\OffsetB)(BO)
+	lxvp	vs4, (0+\OffsetA)(AO)
+	xxspltd  vs8, vs27, 1
+	xxspltd  vs24, vs27, 0
+	xxperm    vs10, vs8, permute_mask
+	xxperm    vs26, vs24, permute_mask
+.endm
+
+.macro	END1x2_2
+  /*for load2 offset will be 32 and 16*/
+	KERNEL1x2_2  AO, BO, 32, 16, 0, 1, 1
+.endm
+
+.macro	KERNEL1x2_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x2_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL1x2_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x2_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL1x2_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+.if \Complete==0
+	lxv	vs27, DISP2(\Index, \OffsetB)(\BREG)
+.endif
+	xvmaddasp	vs32, vs5, vs8
+	xvmaddasp	vs40, vs5, vs10
+
+.if \Complete==0
+	xxspltd  vs8, vs27, 1
+	xxperm    vs10, vs8, permute_mask
+.endif
+	xvmaddasp	vs32, vs4, vs24
+	xvmaddasp	vs40, vs4, vs26
+.if \Complete==0
+	lxvp	vs4, DISP4(\Index, 0+\OffsetA)(\AREG)
+.endif
+
+.if \Complete==0
+	xxspltd  vs24, vs27, 0
+	xxperm    vs26, vs24, permute_mask
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \BREG, \BREG, DISP2(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP4(\Index, \OffsetA)
+.else
+	addi    \BREG, \BREG, DISP2(\Index, 16)
+	addi    \AREG, \AREG, DISP4(\Index, 32)
+.endif
+.endif
+.endm
+
+.macro	KERNEL1x2
+	LOAD1x2
+	END1x2  AO, BO, 16,8
+.endm
+
+.macro SAVE1x2
+#ifndef TRMMKERNEL
+	lxv	vs24, 0(CO)
+#endif
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	AGGREGATE_REALS_IMAGES	vs32, vs0, vs40, vs4
+  /*inner reverse save_permute and store vs28 */
+	xxpermdi	vs28,save_permute_1,save_permute_1, 2
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs0, vs1
+	MULT_APLHA_PART2    vs32, vs40, vs0, vs1
+/* reconstruct r, i pairs*/
+	xxperm	vs0, vs1, vs28
+#ifndef TRMMKERNEL
+  /* add */
+	xvaddsp	vs24, vs24, vs0
+	stxv	vs24, 0(CO)
+#else
+/* reconstruct r, i pairs*/
+	stxv	vs0, 0(CO)
+#endif
+	addi  CO, CO, 16
+.endm
+
+/*                                             macros for N=1 and M=1
+**********************************************************************************************/
+.macro	ZERO1x1
+	xxlxor  vs32, vs32, vs32
+	xxlxor  vs40, vs40, vs40
+.endm
+
+.macro	LOAD1x1
+	LOAD1x1O 0, 0
+.endm
+
+.macro	LOAD1x1O  OffsetA, OffsetB
+	lxsd	v4, (\OffsetB+0)(BO)
+	lxsd	v5, (\OffsetA+0)(AO)
+	xxperm    vs38, vs36, permute_mask
+.endm
+
+.macro	END1x1_NORMAL
+	END1x1 AO, BO,8,8
+.endm
+
+.macro	END1x1_WITHOUT_ADD
+	END1x1 AO, BO, 0, 0
+.endm
+
+.macro	END1x1 AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi  \BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi  \AREG, \AREG, \OffsetA
+.endif
+	xvmaddasp	vs32, vs37, vs36
+	xvmaddasp	vs40, vs37, vs38
+.endm
+
+.macro	LOAD1x1_2
+	LOAD1x1_2O 0, 0
+.endm
+
+.macro	LOAD1x1_2O  OffsetA, OffsetB
+	lxv	vs8, (\OffsetB)(BO)
+	lxv	vs4, (0+\OffsetA)(AO)
+	xxperm    vs10, vs8, permute_mask
+.endm
+
+.macro	END1x1_2
+  /*for load2 offset will be 16 and 16*/
+	KERNEL1x1_2  AO, BO, 16, 16, 0, 1, 1
+.endm
+
+.macro	KERNEL1x1_E2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x1_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 1
+.endm
+
+.macro	KERNEL1x1_L2 OffsetA, OffsetB, Index, IsLast
+	KERNEL1x1_2 AO, BO, \OffsetA, \OffsetB, \Index, \IsLast, 0
+.endm
+
+.macro	KERNEL1x1_2  AREG, BREG, OffsetA, OffsetB, Index, IsLast, Complete
+	xvmaddasp	vs32, vs4, vs8
+	xvmaddasp	vs40, vs4, vs10
+.if \Complete==0
+	lxv	vs8, DISP2(\Index, \OffsetB)(\BREG)
+	lxv	vs4, DISP2(\Index, \OffsetB)(\AREG)
+	xxperm    vs10, vs8, permute_mask
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi    \BREG, \BREG, DISP2(\Index, \OffsetB)
+	addi    \AREG, \AREG, DISP2(\Index, \OffsetA)
+.else
+	addi    \BREG, \BREG, DISP2(\Index, 16)
+	addi    \AREG, \AREG, DISP2(\Index, 16)
+.endif
+.endif
+.endm
+
+.macro	KERNEL1x1
+	LOAD1x1
+	END1x1  AO, BO, 8,8
+.endm
+
+.macro SAVE1x1
+#ifndef TRMMKERNEL
+	lxsd	v4, 0(CO)
+#endif
+  /*aggregate x2*/
+	xxpermdi	vs33, vs32, vs32, 2
+	xxpermdi	vs41, vs40, vs40, 2
+	xvaddsp	vs32, vs32, vs33
+	xvaddsp	vs40, vs40, vs41
+
+	xxperm	vs0, vs32, permute_mask
+	xxperm	vs4, vs40, permute_mask
+	AGGREGATE_REALS_IMAGES	vs32, vs0, vs40, vs4
+  /*inner reverse save_permute and store vs28 */
+	xxpermdi	vs28,save_permute_1,save_permute_1, 2
+  /*VSINRR, VSINII, VSOUT1, VSOUT2*/
+	MULT_APLHA_PART1    vs32, vs40, vs37, vs1
+	MULT_APLHA_PART2    vs32, vs40, vs37, vs1
+/* reconstruct r, i pairs*/
+	xxperm	vs37, vs1, vs28
+#ifndef TRMMKERNEL
+  /* add */
+	xvaddsp	vs36, vs36, vs37
+	stxsd	v4, 0(CO)
+#else
+/* vs37 is v5 */
+	stxsd	v5, 0(CO)
+#endif
+	addi  CO, CO, 8
+.endm
+
+/****************************TRMM POINTER REFRESH MACROSES*************************/
+.macro SHIFT_REG	REG1,REG2,SHIFT_VAL
+.if \SHIFT_VAL==16
+	slwi		\REG1, \REG2, 7
+.elseif \SHIFT_VAL==8
+	slwi		\REG1, \REG2, 6
+.elseif \SHIFT_VAL==4
+	slwi		\REG1, \REG2, 5
+.elseif \SHIFT_VAL==2
+	slwi		\REG1, \REG2, 4
+.elseif \SHIFT_VAL==1
+	slwi		\REG1, \REG2, 3
+.endif
+.endm
+
+/*
+//#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+// 		ptrbb = bb;
+// #else
+// 		ptrba += off*8;
+// 		ptrbb = bb + off*4;
+// #endif
+*/
+.macro REFRESH_POINTERS  PTR_A,PTR_B, OFF_VAL, B_VAL, C_A, C_B
+#if (defined(LEFT) &&  defined(TRANSA)) ||  (!defined(LEFT) && !defined(TRANSA))
+/* ptrbb = bb;*/
+	mr \PTR_B, \B_VAL     /* refresh BPOINT */
+#else
+/*
+// ptrba  =ptrba+ off*C_A;
+// ptrbb = bb + off*C_B;
+*/
+	SHIFT_REG T4, \OFF_VAL, \C_B	/* Number of values in B shifted  */
+	SHIFT_REG T2, \OFF_VAL, \C_A	/* Number of values in A shifted  */
+	add	\PTR_B, \B_VAL, T4	/* Add values to BO */
+	add	\PTR_A, \PTR_A, T2	/* Add values to AO  */
+#endif
+.endm
+
+/*
+// #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+// 		temp = bk-off;
+// #elif defined(LEFT)
+// 		temp = off+8;	// number of values in A
+// #else
+// 		temp = off+4;	// number of values in B
+// #endif
+*/
+.macro REFRESH_TEMP_BK TEMP_BK, BK_VAL, OFF_VAL, INCR_A, INCR_B
+    #if (defined(LEFT) && !defined(TRANSA)) ||  (!defined(LEFT) && defined(TRANSA))
+	/* temp = bk-off;*/
+	sub \TEMP_BK, \BK_VAL, \OFF_VAL
+    #elif defined(LEFT)
+	/* temp = off+INCR_A;	// number of values in A */
+	addi \TEMP_BK, \OFF_VAL, \INCR_A
+    #else
+	/* temp = off+INCR_B	// number of values in B*/
+	addi \TEMP_BK, \OFF_VAL, \INCR_B
+    #endif
+.endm
+/*
+// #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+// 		temp = bk - off;
+// #ifdef LEFT
+// 		temp -= 8; // number of values in A
+// #else
+// 		temp -= 4; // number of values in B
+// #endif
+// 		ptrba += temp*8;
+// 		ptrbb += temp*4;
+// #endif
+
+// #ifdef LEFT
+// 		off += 8; // number of values in A
+// #endif
+*/
+.macro REFRESH_AFTER_SAVE TEMP_BK, BK_VAL, OFF_VAL,PTR_B,PTR_A, C_A, C_B
+    #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	/*temp = bk - off;*/
+	sub \TEMP_BK, \BK_VAL, \OFF_VAL
+    #ifdef LEFT
+	/*temp -= 8; // number of values in A*/
+	addi \TEMP_BK, \TEMP_BK,-\C_A
+    #else
+	/*temp -= 4; // number of values in B*/
+	addi \TEMP_BK, \TEMP_BK,-\C_B
+    #endif
+	/*ptrba += temp*C_A;
+	ptrbb += temp*C_B;*/
+	SHIFT_REG T4, \TEMP_BK, \C_A
+	SHIFT_REG T2, \TEMP_BK, \C_B
+	add \PTR_A, \PTR_A, T4/*ptrba+temp*C_A*/
+	add \PTR_B, \PTR_B, T2
+    #endif
+    #ifdef LEFT
+	/*off += 8; // number of values in A*/
+	addi \OFF_VAL, \OFF_VAL, \C_A
+    #endif
+.endm
diff --git a/kernel/power/crot.c b/kernel/power/crot.c
index 2a5835546..5c1d44620 100644
--- a/kernel/power/crot.c
+++ b/kernel/power/crot.c
@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  
 #include "common.h"
  
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 
 static void crot_kernel_8 (long n, float *x, float *y, float c, float s)
 {
diff --git a/kernel/power/cswap.c b/kernel/power/cswap.c
index 31e02fe5a..88cb1d638 100644
--- a/kernel/power/cswap.c
+++ b/kernel/power/cswap.c
@@ -36,7 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8)  || defined(POWER9)
+#if defined(POWER8)  || defined(POWER9) || defined(POWER10)
 #include "cswap_microk_power8.c"
 #endif
 
diff --git a/kernel/power/dasum.c b/kernel/power/dasum.c
index d0e060977..09e06d909 100644
--- a/kernel/power/dasum.c
+++ b/kernel/power/dasum.c
@@ -46,7 +46,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "dasum_microk_power8.c"
 #endif
 
diff --git a/kernel/power/daxpy.c b/kernel/power/daxpy.c
index f09611ff0..018beafd1 100644
--- a/kernel/power/daxpy.c
+++ b/kernel/power/daxpy.c
@@ -36,7 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "daxpy_microk_power8.c"
 #endif
 
diff --git a/kernel/power/dcopy.c b/kernel/power/dcopy.c
index 27b39144b..cf203e71e 100644
--- a/kernel/power/dcopy.c
+++ b/kernel/power/dcopy.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "dcopy_microk_power8.c"
 #endif
 
diff --git a/kernel/power/ddot.c b/kernel/power/ddot.c
index f985df1c5..bd9e1fb97 100644
--- a/kernel/power/ddot.c
+++ b/kernel/power/ddot.c
@@ -36,7 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8)  || defined(POWER9)
+#if defined(POWER8)  || defined(POWER9) || defined(POWER10)
 #include "ddot_microk_power8.c"
 #endif
 
diff --git a/kernel/power/dgemm_kernel_power10.c b/kernel/power/dgemm_kernel_power10.c
new file mode 100644
index 000000000..b3ee301be
--- /dev/null
+++ b/kernel/power/dgemm_kernel_power10.c
@@ -0,0 +1,864 @@
+/*********************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+#include "common.h"
+#include <altivec.h>
+
+typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
+typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
+
+#ifdef TRMMKERNEL
+#define SAVE_ACC(ACC, J)  \
+          __builtin_mma_disassemble_acc (result, ACC); \
+          rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] = result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[1*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[3*ldc+J]; \
+          rowC[0] = result[0] * alpha;
+#define SAVE_ACC1(ACC, J)  \
+          __builtin_mma_disassemble_acc (result, ACC); \
+          rowC = (v4sf_t *) &CO[4* ldc+J]; \
+          rowC[0] = result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[5*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[7*ldc+J]; \
+          rowC[0] = result[0] * alpha;
+#define  SAVE2x4_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] = result[3] * alpha; \
+	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
+          rowC[0] = result[2] * alpha;
+#else
+#define SAVE_ACC(ACC, J)  \
+          __builtin_mma_disassemble_acc (result, ACC); \
+          rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[1*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[3*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define SAVE_ACC1(ACC, J)  \
+          __builtin_mma_disassemble_acc (result, ACC); \
+          rowC = (v4sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[5*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[7*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE2x4_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[2] * alpha;
+#endif
+
+#define SET_ACC_ZERO4() \
+          __builtin_mma_xxsetaccz (&acc0); \
+          __builtin_mma_xxsetaccz (&acc1); \
+          __builtin_mma_xxsetaccz (&acc2); \
+          __builtin_mma_xxsetaccz (&acc3);
+
+#define SET_ACC_ZERO8() \
+          __builtin_mma_xxsetaccz (&acc0); \
+          __builtin_mma_xxsetaccz (&acc1); \
+          __builtin_mma_xxsetaccz (&acc2); \
+          __builtin_mma_xxsetaccz (&acc3); \
+          __builtin_mma_xxsetaccz (&acc4); \
+          __builtin_mma_xxsetaccz (&acc5); \
+          __builtin_mma_xxsetaccz (&acc6); \
+          __builtin_mma_xxsetaccz (&acc7);
+
+#define PREFETCH1(x, y) asm volatile ("dcbt %0, %1" : : "r" (x), "b" (y) : "memory");
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+#define REFRESH_TEMP_BK(x, y) \
+            temp = k - off;
+#elif defined(LEFT)
+#define REFRESH_TEMP_BK(x, y) \
+            temp = off + x;
+#else
+#define REFRESH_TEMP_BK(x, y) \
+            temp = off + y;
+#endif
+#if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+#define REFRESH_POINTERS(x, y) \
+          BO = B; \
+          REFRESH_TEMP_BK(x, y)
+#else
+#define REFRESH_POINTERS(x, y) \
+          AO += off * x; \
+          BO = B + off * y; \
+          REFRESH_TEMP_BK(x, y)
+#endif
+
+#ifdef LEFT
+#define REFRESH_OFF(x) \
+            off += x;
+#else
+#define REFRESH_OFF(x)
+#endif
+
+#ifdef LEFT
+#define UPDATE_TEMP(x, y) \
+            temp -= x;
+#else
+#define UPDATE_TEMP(x, y) \
+            temp -= y;
+#endif
+
+#if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+#define REFRESH_TMP_AFTER_SAVE(x, y) \
+            temp = k - off; \
+            UPDATE_TEMP(x, y) \
+            AO += temp * x; \
+            BO += temp * y;
+#else
+#define REFRESH_TMP_AFTER_SAVE(x, y)
+#endif
+
+#define REFRESH_AFTER_SAVE(x,y) \
+        REFRESH_TMP_AFTER_SAVE(x, y) \
+        REFRESH_OFF(x)
+/*************************************************************************************
+* GEMM Kernel
+*************************************************************************************/
+int
+CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
+       FLOAT * C, BLASLONG ldc
+#ifdef TRMMKERNEL
+       , BLASLONG offset
+#endif
+  )
+{
+  BLASLONG N = n;
+  BLASLONG i1;
+#if defined(TRMMKERNEL)
+  BLASLONG off;
+#endif
+#if defined(TRMMKERNEL) && !defined(LEFT)
+  off = -offset;
+#endif
+  v4sf_t valpha = { alpha, alpha };
+  N = n >> 2;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j, temp;
+      FLOAT *CO;
+      FLOAT *AO;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      CO = C;
+      C += ldc << 2;
+      AO = A;
+      PREFETCH1 (A, 128);
+      PREFETCH1 (A, 256);
+      i = m >> 4;
+      for (j = 0; j < i; j++)
+	{
+          FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (16, 4);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  BLASLONG l = 0;
+	  PREFETCH1 (CO, 0);
+	  PREFETCH1 (CO + ldc, 0);
+	  PREFETCH1 (CO + ldc + ldc, 0);
+	  PREFETCH1 (CO + ldc + ldc + ldc, 0);
+	  PREFETCH1 (CO, 128);
+	  PREFETCH1 (CO + ldc, 128);
+	  PREFETCH1 (CO + ldc + ldc, 128);
+	  PREFETCH1 (CO + ldc + ldc + ldc, 128);
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & BO[l << 2];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
+	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[2]);
+	      __builtin_mma_xvf64gerpp (&acc3, rowB, rowA[3]);
+	      __builtin_mma_xvf64gerpp (&acc4, rowB, rowA[4]);
+	      __builtin_mma_xvf64gerpp (&acc5, rowB, rowA[5]);
+	      __builtin_mma_xvf64gerpp (&acc6, rowB, rowA[6]);
+	      __builtin_mma_xvf64gerpp (&acc7, rowB, rowA[7]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC (&acc1, 2);
+	  SAVE_ACC (&acc3, 6);
+	  SAVE_ACC (&acc4, 8);
+	  SAVE_ACC (&acc6, 12);
+	  SAVE_ACC (&acc5, 10);
+	  SAVE_ACC (&acc7, 14);
+	  AO += temp << 4;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (16, 4)
+#endif
+	  CO += 16;
+	}
+      i = (m & 15) >> 3;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (8, 4);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & BO[l << 2];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
+	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[2]);
+	      __builtin_mma_xvf64gerpp (&acc3, rowB, rowA[3]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC (&acc1, 2);
+	  SAVE_ACC (&acc3, 6);
+	  CO += 8;
+	  AO += temp << 3;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (8, 4)
+#endif
+	}
+      i = (m & 7) >> 2;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (4, 4);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & BO[l << 2];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 2);
+	  CO += 4;
+	  AO += temp << 2;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (4, 4)
+#endif
+	}
+      i = (m & 3) >> 1;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (2, 4);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 1];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & BO[l << 2];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  CO += 2;
+	  AO += temp << 1;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (2, 4)
+#endif
+	}
+      i = (m & 1) >> 0;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (1, 4);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0 };
+	  v4sf_t t1 = { 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowA = { AO[l], AO[l] };
+	      v4sf_t rowB = { BO[l << 2], BO[(l << 2) + 1] };
+	      v4sf_t rowB1 = { BO[(l << 2) + 2], BO[(l << 2) + 3] };
+	      t += rowA * rowB;
+	      t1 += rowA * rowB1;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0 * ldc] = t[0];
+	  CO[1 * ldc] = t[1];
+	  CO[2 * ldc] = t1[0];
+	  CO[3 * ldc] = t1[1];
+#else
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t1[0];
+	  CO[3 * ldc] += t1[1];
+#endif
+	  CO += 1;
+	  AO += temp;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (1, 4)
+#endif
+	}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 4;                 // number of values in A
+#endif
+      B += k << 2;
+    }
+  N = (n & 3) >> 1;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j, temp;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      FLOAT *CO;
+      FLOAT *AO;
+      CO = C;
+      C += ldc << 1;
+      AO = A;
+      i = m >> 4;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (16, 2);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0, 0, 0, 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & t[0];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
+	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[2]);
+	      __builtin_mma_xvf64gerpp (&acc3, rowB, rowA[3]);
+	      __builtin_mma_xvf64gerpp (&acc4, rowB, rowA[4]);
+	      __builtin_mma_xvf64gerpp (&acc5, rowB, rowA[5]);
+	      __builtin_mma_xvf64gerpp (&acc6, rowB, rowA[6]);
+	      __builtin_mma_xvf64gerpp (&acc7, rowB, rowA[7]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 2);
+	  SAVE2x4_ACC (&acc2, 4);
+	  SAVE2x4_ACC (&acc3, 6);
+	  SAVE2x4_ACC (&acc4, 8);
+	  SAVE2x4_ACC (&acc5, 10);
+	  SAVE2x4_ACC (&acc6, 12);
+	  SAVE2x4_ACC (&acc7, 14);
+	  CO += 16;
+	  AO += temp << 4;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (16, 2)
+#endif
+	}
+      i = (m & 15) >> 3;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (8, 2);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0, 0, 0, 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & t[0];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
+	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[2]);
+	      __builtin_mma_xvf64gerpp (&acc3, rowB, rowA[3]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 2);
+	  SAVE2x4_ACC (&acc2, 4);
+	  SAVE2x4_ACC (&acc3, 6);
+	  CO += 8;
+	  AO += temp << 3;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (8, 2)
+#endif
+	}
+      i = (m & 7) >> 2;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (4, 2);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0, 0, 0, 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & t[0];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 2);
+	  CO += 4;
+	  AO += temp << 2;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (4, 2)
+#endif
+	}
+      i = (m & 3) >> 1;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (2, 2);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0, 0, 0, 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      __vector_pair rowB;
+	      vec_t *rb = (vec_t *) & t[0];
+	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      vec_t *rowA = (vec_t *) & AO[l << 1];
+	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  CO += 2;
+	  AO += temp << 1;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (2, 2)
+#endif
+	}
+      i = (m & 1) >> 0;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (1, 2);
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowA = { AO[l], AO[l] };
+	      v4sf_t rowB = { BO[l << 1], BO[(l << 1) + 1] };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0 * ldc] = t[0];
+	  CO[1 * ldc] = t[1];
+#else
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+#endif
+	  CO += 1;
+	  AO += temp;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (1, 2)
+#endif
+	}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 2;                 // number of values in A
+#endif
+      B += k << 1;
+    }
+  N = (n & 1) >> 0;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, temp;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      FLOAT *CO;
+      FLOAT *AO;
+      CO = C;
+      C += ldc;
+      AO = A;
+      i = m;
+      while (i >= 16)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (16, 1)
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0 };
+	  v4sf_t t1 = { 0, 0 };
+	  v4sf_t t2 = { 0, 0 };
+	  v4sf_t t3 = { 0, 0 };
+	  v4sf_t t4 = { 0, 0 };
+	  v4sf_t t5 = { 0, 0 };
+	  v4sf_t t6 = { 0, 0 };
+	  v4sf_t t7 = { 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 4], AO[(l << 4) + 1] };
+	      v4sf_t rowA1 = { AO[(l << 4) + 2], AO[(l << 4) + 3] };
+	      v4sf_t rowA2 = { AO[(l << 4) + 4], AO[(l << 4) + 5] };
+	      v4sf_t rowA3 = { AO[(l << 4) + 6], AO[(l << 4) + 7] };
+	      v4sf_t rowA4 = { AO[(l << 4) + 8], AO[(l << 4) + 9] };
+	      v4sf_t rowA5 = { AO[(l << 4) + 10], AO[(l << 4) + 11] };
+	      v4sf_t rowA6 = { AO[(l << 4) + 12], AO[(l << 4) + 13] };
+	      v4sf_t rowA7 = { AO[(l << 4) + 14], AO[(l << 4) + 15] };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	      t2 += rowA2 * rowB;
+	      t3 += rowA3 * rowB;
+	      t4 += rowA4 * rowB;
+	      t5 += rowA5 * rowB;
+	      t6 += rowA6 * rowB;
+	      t7 += rowA7 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  t2 = t2 * valpha;
+	  t3 = t3 * valpha;
+	  t4 = t4 * valpha;
+	  t5 = t5 * valpha;
+	  t6 = t6 * valpha;
+	  t7 = t7 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+	  CO[2] = t1[0];
+	  CO[3] = t1[1];
+	  CO[4] = t2[0];
+	  CO[5] = t2[1];
+	  CO[6] = t3[0];
+	  CO[7] = t3[1];
+	  CO[8] = t4[0];
+	  CO[9] = t4[1];
+	  CO[10] = t5[0];
+	  CO[11] = t5[1];
+	  CO[12] = t6[0];
+	  CO[13] = t6[1];
+	  CO[14] = t7[0];
+	  CO[15] = t7[1];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t1[0];
+	  CO[3] += t1[1];
+	  CO[4] += t2[0];
+	  CO[5] += t2[1];
+	  CO[6] += t3[0];
+	  CO[7] += t3[1];
+	  CO[8] += t4[0];
+	  CO[9] += t4[1];
+	  CO[10] += t5[0];
+	  CO[11] += t5[1];
+	  CO[12] += t6[0];
+	  CO[13] += t6[1];
+	  CO[14] += t7[0];
+	  CO[15] += t7[1];
+#endif
+	  AO += temp << 4;
+	  BO += temp;
+	  CO += 16;
+	  i -= 16;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (16, 1)
+#endif
+	}
+      while (i >= 8)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (8, 1)
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0 };
+	  v4sf_t t1 = { 0, 0 };
+	  v4sf_t t2 = { 0, 0 };
+	  v4sf_t t3 = { 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 3], AO[(l << 3) + 1] };
+	      v4sf_t rowA1 = { AO[(l << 3) + 2], AO[(l << 3) + 3] };
+	      v4sf_t rowA2 = { AO[(l << 3) + 4], AO[(l << 3) + 5] };
+	      v4sf_t rowA3 = { AO[(l << 3) + 6], AO[(l << 3) + 7] };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	      t2 += rowA2 * rowB;
+	      t3 += rowA3 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  t2 = t2 * valpha;
+	  t3 = t3 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+	  CO[2] = t1[0];
+	  CO[3] = t1[1];
+	  CO[4] = t2[0];
+	  CO[5] = t2[1];
+	  CO[6] = t3[0];
+	  CO[7] = t3[1];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t1[0];
+	  CO[3] += t1[1];
+	  CO[4] += t2[0];
+	  CO[5] += t2[1];
+	  CO[6] += t3[0];
+	  CO[7] += t3[1];
+#endif
+	  AO += temp << 3;
+	  BO += temp;
+	  CO += 8;
+	  i -= 8;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (8, 1)
+#endif
+	}
+      while (i >= 4)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (4, 1)
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0 };
+	  v4sf_t t1 = { 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 2], AO[(l << 2) + 1] };
+	      v4sf_t rowA1 = { AO[(l << 2) + 2], AO[(l << 2) + 3] };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+	  CO[2] = t1[0];
+	  CO[3] = t1[1];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t1[0];
+	  CO[3] += t1[1];
+#endif
+	  AO += temp << 2;
+	  BO += temp;
+	  CO += 4;
+	  i -= 4;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (4, 1)
+#endif
+	}
+      while (i >= 2)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (2, 1)
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 1], AO[(l << 1) + 1] };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+#endif
+	  AO += temp << 1;
+	  BO += temp;
+	  CO += 2;
+	  i -= 2;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (2, 1)
+#endif
+	}
+      while (i >= 1)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+          REFRESH_POINTERS (1, 1)
+#else
+          BO = B;
+          temp = k;
+#endif
+	  BLASLONG l = 0;
+	  FLOAT t = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      t += AO[l] * BO[l];
+	    }
+	  AO += temp;
+	  BO += temp;
+#if defined(TRMMKERNEL)
+	  CO[0] = t * alpha;
+#else
+	  CO[0] += t * alpha;
+#endif
+	  CO += 1;
+	  i -= 1;
+#if defined(TRMMKERNEL)
+          REFRESH_AFTER_SAVE (1, 1)
+#endif
+	}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 1;                 // number of values in A
+#endif
+      B += k;
+    }
+  return 0;
+}
diff --git a/kernel/power/dgemv_n.c b/kernel/power/dgemv_n.c
index 1a3d7669c..b4dfda550 100644
--- a/kernel/power/dgemv_n.c
+++ b/kernel/power/dgemv_n.c
@@ -38,7 +38,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "dgemv_n_microk_power8.c"
 #endif
 
diff --git a/kernel/power/drot.c b/kernel/power/drot.c
index baeb54205..b808ab566 100644
--- a/kernel/power/drot.c
+++ b/kernel/power/drot.c
@@ -39,7 +39,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #pragma GCC optimize "O1"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "drot_microk_power8.c"
 #endif
 
diff --git a/kernel/power/dscal.c b/kernel/power/dscal.c
index 779a08e9c..7e0fe48c0 100644
--- a/kernel/power/dscal.c
+++ b/kernel/power/dscal.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "dscal_microk_power8.c"
 #endif
 
diff --git a/kernel/power/dswap.c b/kernel/power/dswap.c
index 52b7f50da..795bb10b4 100644
--- a/kernel/power/dswap.c
+++ b/kernel/power/dswap.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "dswap_microk_power8.c"
 #endif
 
diff --git a/kernel/power/sasum.c b/kernel/power/sasum.c
index 5908347d3..b259d7d76 100644
--- a/kernel/power/sasum.c
+++ b/kernel/power/sasum.c
@@ -46,7 +46,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "sasum_microk_power8.c"
 #endif
 
diff --git a/kernel/power/scopy.c b/kernel/power/scopy.c
index 5e3fe45a5..5207d386e 100644
--- a/kernel/power/scopy.c
+++ b/kernel/power/scopy.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "scopy_microk_power8.c"
 #endif
 
diff --git a/kernel/power/sdot.c b/kernel/power/sdot.c
index ae527dde9..8de434e41 100644
--- a/kernel/power/sdot.c
+++ b/kernel/power/sdot.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8)  || defined(POWER9)
+#if defined(POWER8)  || defined(POWER9) || defined(POWER10)
 #include "sdot_microk_power8.c"
 #endif
 
diff --git a/kernel/power/sgemm_kernel_power10.c b/kernel/power/sgemm_kernel_power10.c
new file mode 100644
index 000000000..01c122c6d
--- /dev/null
+++ b/kernel/power/sgemm_kernel_power10.c
@@ -0,0 +1,1334 @@
+/*********************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+#include "common.h"
+#include <altivec.h>
+
+typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
+typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
+#if defined(TRMMKERNEL)
+#define SAVE_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] = result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[1*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[3*ldc+J]; \
+          rowC[0] = result[0] * alpha;
+#define SAVE_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
+          rowC[0] = result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[5*ldc+J]; \
+          rowC[0] = result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] = result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[7*ldc+J]; \
+          rowC[0] = result[0] * alpha;
+#define  SAVE4x2_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
+          rowC[0] = result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
+          rowC[0] = result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] = result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
+          rowC[0] = result[0] * alpha;
+#define  SAVE4x2_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
+          rowC[0] = result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
+          rowC[0] = result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] = result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
+          rowC[0] = result[0] * alpha;
+#define  SAVE2x4_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] = result[3] * alpha; \
+	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
+          rowC[0] = result[2] * alpha;
+#else
+#define SAVE_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[1*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[3*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define SAVE_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[5*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[7*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE4x2_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE4x2_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE2x4_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[2] * alpha;
+#endif
+#define KERNEL(i, j) \
+          __builtin_mma_xvf32gerpp (&acc0, rowB[i], rowA[j]); \
+          __builtin_mma_xvf32gerpp (&acc1, rowB[i+1], rowA[j]); \
+          __builtin_mma_xvf32gerpp (&acc2, rowB[i], rowA[j+1]); \
+          __builtin_mma_xvf32gerpp (&acc3, rowB[i+1], rowA[j+1]); \
+          __builtin_mma_xvf32gerpp (&acc4, rowB[i], rowA[j+2]); \
+          __builtin_mma_xvf32gerpp (&acc5, rowB[i+1], rowA[j+2]); \
+          __builtin_mma_xvf32gerpp (&acc6, rowB[i], rowA[j+3]); \
+          __builtin_mma_xvf32gerpp (&acc7, rowB[i+1], rowA[j+3]);
+#define SET_ACC_ZERO4() \
+          __builtin_mma_xxsetaccz (&acc0); \
+          __builtin_mma_xxsetaccz (&acc1); \
+          __builtin_mma_xxsetaccz (&acc2); \
+          __builtin_mma_xxsetaccz (&acc3);
+
+#define SET_ACC_ZERO8() \
+          __builtin_mma_xxsetaccz (&acc0); \
+          __builtin_mma_xxsetaccz (&acc1); \
+          __builtin_mma_xxsetaccz (&acc2); \
+          __builtin_mma_xxsetaccz (&acc3); \
+          __builtin_mma_xxsetaccz (&acc4); \
+          __builtin_mma_xxsetaccz (&acc5); \
+          __builtin_mma_xxsetaccz (&acc6); \
+          __builtin_mma_xxsetaccz (&acc7);
+
+#define PREFETCH1(x, y) asm volatile ("dcbt %0, %1" : : "r" (x), "b" (y) : "memory");
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+#define REFRESH_TEMP_BK(x, y) \
+            temp = k - off;
+#elif defined(LEFT)
+#define REFRESH_TEMP_BK(x, y) \
+            temp = off + x;
+#else
+#define REFRESH_TEMP_BK(x, y) \
+            temp = off + y;
+#endif
+#if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+#define REFRESH_POINTERS(x, y) \
+	  BO = B; \
+          REFRESH_TEMP_BK(x, y)
+#else
+#define REFRESH_POINTERS(x, y) \
+          AO += off * x; \
+          BO = B + off * y; \
+          REFRESH_TEMP_BK(x, y)
+#endif
+
+#ifdef LEFT
+#define REFRESH_OFF(x) \
+            off += x;
+#else
+#define REFRESH_OFF(x)
+#endif
+
+#ifdef LEFT
+#define UPDATE_TEMP(x, y) \
+            temp -= x;
+#else
+#define UPDATE_TEMP(x, y) \
+            temp -= y;
+#endif
+
+#if (defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+#define REFRESH_TMP_AFTER_SAVE(x, y) \
+            temp = k - off; \
+            UPDATE_TEMP(x, y) \
+            AO += temp * x; \
+            BO += temp * y;
+#else
+#define REFRESH_TMP_AFTER_SAVE(x, y)
+#endif
+
+#define REFRESH_AFTER_SAVE(x,y) \
+        REFRESH_TMP_AFTER_SAVE(x, y) \
+	REFRESH_OFF(x)
+/*************************************************************************************
+* GEMM Kernel
+*************************************************************************************/
+int
+CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
+       FLOAT * C, BLASLONG ldc
+#ifdef TRMMKERNEL
+       , BLASLONG offset
+#endif
+  )
+{
+  BLASLONG N = n;
+  BLASLONG i1;
+#if defined(TRMMKERNEL)
+  BLASLONG off;
+#endif
+#if defined(TRMMKERNEL) && !defined(LEFT)
+  off = -offset;
+#endif
+
+  v4sf_t valpha = { alpha, alpha, alpha, alpha };
+  N = n >> 3;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j, temp;
+      FLOAT *CO;
+      FLOAT *AO;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      CO = C;
+      C += ldc << 3;
+      AO = A;
+      PREFETCH1 (A, 128);
+      PREFETCH1 (A, 256);
+      i = m >> 4;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (16, 8);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  BLASLONG K = temp / 64;
+	  for (l = 0; l < K; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      KERNEL (2, 4);
+	      KERNEL (4, 8);
+	      KERNEL (6, 12);
+	      KERNEL (8, 16);
+	      KERNEL (10, 20);
+	      KERNEL (12, 24);
+	      KERNEL (14, 28);
+	      KERNEL (16, 32);
+	      KERNEL (18, 36);
+	      KERNEL (20, 40);
+	      KERNEL (22, 44);
+	      KERNEL (24, 48);
+	      KERNEL (26, 52);
+	      KERNEL (28, 56);
+	      KERNEL (30, 60);
+	      KERNEL (32, 64);
+	      KERNEL (34, 68);
+	      KERNEL (36, 72);
+	      KERNEL (38, 76);
+	      KERNEL (40, 80);
+	      KERNEL (42, 84);
+	      KERNEL (44, 88);
+	      KERNEL (46, 92);
+	      KERNEL (48, 96);
+	      KERNEL (50, 100);
+	      KERNEL (52, 104);
+	      KERNEL (54, 108);
+	      KERNEL (56, 112);
+	      KERNEL (58, 116);
+	      KERNEL (60, 120);
+	      KERNEL (62, 124);
+	      KERNEL (64, 128);
+	      KERNEL (66, 132);
+	      KERNEL (68, 136);
+	      KERNEL (70, 140);
+	      KERNEL (72, 144);
+	      KERNEL (74, 148);
+	      KERNEL (76, 152);
+	      KERNEL (78, 156);
+	      KERNEL (80, 160);
+	      KERNEL (82, 164);
+	      KERNEL (84, 168);
+	      KERNEL (86, 172);
+	      KERNEL (88, 176);
+	      KERNEL (90, 180);
+	      KERNEL (92, 184);
+	      KERNEL (94, 188);
+	      KERNEL (96, 192);
+	      KERNEL (98, 196);
+	      KERNEL (100, 200);
+	      KERNEL (102, 204);
+	      KERNEL (104, 208);
+	      KERNEL (106, 212);
+	      KERNEL (108, 216);
+	      KERNEL (110, 220);
+	      KERNEL (112, 224);
+	      KERNEL (114, 228);
+	      KERNEL (116, 232);
+	      KERNEL (118, 236);
+	      KERNEL (120, 240);
+	      KERNEL (122, 244);
+	      KERNEL (124, 248);
+	      KERNEL (126, 252);
+	      AO += 1024;
+	      BO += 512;
+	    }
+	  if ((temp & 63) >> 5)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      KERNEL (2, 4);
+	      KERNEL (4, 8);
+	      KERNEL (6, 12);
+	      KERNEL (8, 16);
+	      KERNEL (10, 20);
+	      KERNEL (12, 24);
+	      KERNEL (14, 28);
+	      KERNEL (16, 32);
+	      KERNEL (18, 36);
+	      KERNEL (20, 40);
+	      KERNEL (22, 44);
+	      KERNEL (24, 48);
+	      KERNEL (26, 52);
+	      KERNEL (28, 56);
+	      KERNEL (30, 60);
+	      KERNEL (32, 64);
+	      KERNEL (34, 68);
+	      KERNEL (36, 72);
+	      KERNEL (38, 76);
+	      KERNEL (40, 80);
+	      KERNEL (42, 84);
+	      KERNEL (44, 88);
+	      KERNEL (46, 92);
+	      KERNEL (48, 96);
+	      KERNEL (50, 100);
+	      KERNEL (52, 104);
+	      KERNEL (54, 108);
+	      KERNEL (56, 112);
+	      KERNEL (58, 116);
+	      KERNEL (60, 120);
+	      KERNEL (62, 124);
+	      AO += 512;
+	      BO += 256;
+	    }
+	  if ((temp & 31) >> 4)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      KERNEL (2, 4);
+	      KERNEL (4, 8);
+	      KERNEL (6, 12);
+	      KERNEL (8, 16);
+	      KERNEL (10, 20);
+	      KERNEL (12, 24);
+	      KERNEL (14, 28);
+	      KERNEL (16, 32);
+	      KERNEL (18, 36);
+	      KERNEL (20, 40);
+	      KERNEL (22, 44);
+	      KERNEL (24, 48);
+	      KERNEL (26, 52);
+	      KERNEL (28, 56);
+	      KERNEL (30, 60);
+	      AO += 256;
+	      BO += 128;
+	    }
+	  if ((temp & 15) >> 3)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      KERNEL (2, 4);
+	      KERNEL (4, 8);
+	      KERNEL (6, 12);
+	      KERNEL (8, 16);
+	      KERNEL (10, 20);
+	      KERNEL (12, 24);
+	      KERNEL (14, 28);
+	      AO += 128;
+	      BO += 64;
+	    }
+	  if ((temp & 7) >> 2)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      KERNEL (2, 4);
+	      KERNEL (4, 8);
+	      KERNEL (6, 12);
+	      AO += 64;
+	      BO += 32;
+	    }
+	  if ((temp & 3) >> 1)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      KERNEL (2, 4);
+	      AO += 32;
+	      BO += 16;
+	    }
+	  if ((temp & 1) >> 0)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[0];
+	      vec_t *rowB = (vec_t *) & BO[0];
+	      KERNEL (0, 0);
+	      AO += 16;
+	      BO += 8;
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC1 (&acc1, 0);
+	  SAVE_ACC1 (&acc3, 4);
+	  SAVE_ACC (&acc4, 8);
+	  SAVE_ACC (&acc6, 12);
+	  SAVE_ACC1 (&acc5, 8);
+	  SAVE_ACC1 (&acc7, 12);
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (16, 8)
+#endif
+	    CO += 16;
+	}
+      i = (m & 15) >> 3;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (8, 8);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      vec_t *rowB = (vec_t *) & BO[l << 3];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[1], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[1]);
+	      __builtin_mma_xvf32gerpp (&acc3, rowB[1], rowA[1]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC1 (&acc1, 0);
+	  SAVE_ACC1 (&acc3, 4);
+	  AO += (temp << 3);
+	  BO += (temp << 3);
+	  CO += 8;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (8, 8)
+#endif
+	}
+      i = (m & 7) >> 2;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (4, 8);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      vec_t *rowB = (vec_t *) & BO[l << 3];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[1], rowA[0]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC1 (&acc1, 0);
+	  CO += 4;
+	  AO += (temp << 2);
+	  BO += (temp << 3);
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (4, 8)
+#endif
+	}
+      i = (m & 3) >> 1;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (2, 8);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+
+	  v2sf_t *rowC;
+	  v2sf_t result[8];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0 };
+	      t[0] = AO[l << 1], t[1] = AO[(l << 1) + 1];
+	      vec_t *rowA = (vec_t *) & t[0];
+	      vec_t *rowB = (vec_t *) & BO[l << 3];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[1], rowA[0]);
+	    }
+	  SAVE4x2_ACC (&acc0, 0);
+	  SAVE4x2_ACC1 (&acc1, 0);
+	  CO += 2;
+	  AO += (temp << 1);
+	  BO += (temp << 3);
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (2, 8)
+#endif
+	}
+      i = (m & 1) >> 0;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (1, 8);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowA = { AO[l], AO[l], AO[l], AO[l] };
+	      v4sf_t rowB = { BO[l << 3], BO[(l << 3) + 1], BO[(l << 3) + 2],
+		BO[(l << 3) + 3]
+	      };
+	      v4sf_t rowB1 =
+		{ BO[(l << 3) + 4], BO[(l << 3) + 5], BO[(l << 3) + 6],
+		BO[(l << 3) + 7]
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA * rowB1;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0 * ldc] = t[0];
+	  CO[1 * ldc] = t[1];
+	  CO[2 * ldc] = t[2];
+	  CO[3 * ldc] = t[3];
+	  CO[4 * ldc] = t1[0];
+	  CO[5 * ldc] = t1[1];
+	  CO[6 * ldc] = t1[2];
+	  CO[7 * ldc] = t1[3];
+#else
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t[2];
+	  CO[3 * ldc] += t[3];
+	  CO[4 * ldc] += t1[0];
+	  CO[5 * ldc] += t1[1];
+	  CO[6 * ldc] += t1[2];
+	  CO[7 * ldc] += t1[3];
+#endif
+	  CO += 1;
+	  AO += temp;
+	  BO += (temp << 3);
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (1, 8)
+#endif
+	}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 8;			// number of values in A
+#endif
+
+      B += k << 3;
+    }
+  N = (n & 7) >> 2;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j, temp;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      FLOAT *CO;
+      FLOAT *AO;
+      CO = C;
+      C += ldc << 2;
+      AO = A;
+#if !defined(TRMMKERNEL)
+      i = m >> 5;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  FLOAT *A1;
+	  A1 = AO + (16 * k);
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      vec_t *rowA1 = (vec_t *) & A1[l << 4];
+	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
+	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
+	      __builtin_mma_xvf32gerpp (&acc3, rowB[0], rowA[3]);
+	      __builtin_mma_xvf32gerpp (&acc4, rowB[0], rowA1[0]);
+	      __builtin_mma_xvf32gerpp (&acc5, rowB[0], rowA1[1]);
+	      __builtin_mma_xvf32gerpp (&acc6, rowB[0], rowA1[2]);
+	      __builtin_mma_xvf32gerpp (&acc7, rowB[0], rowA1[3]);
+	    }
+
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc2, 0);
+	  SAVE_ACC (&acc3, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc4, 0);
+	  SAVE_ACC (&acc5, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc6, 0);
+	  SAVE_ACC (&acc7, 4);
+	  CO += 8;
+	  AO += k << 5;
+	  BO += k << 2;
+	}
+      i = (m & 31) >> 4;
+#else
+      i = m >> 4;
+#endif
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (16, 4);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
+	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
+	      __builtin_mma_xvf32gerpp (&acc3, rowB[0], rowA[3]);
+	    }
+
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc2, 0);
+	  SAVE_ACC (&acc3, 4);
+	  CO += 8;
+	  AO += temp << 4;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (16, 4)
+#endif
+	}
+      i = (m & 15) >> 3;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (8, 4);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  AO += temp << 3;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (8, 4)
+#endif
+	}
+      i = (m & 7) >> 2;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (4, 4);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t *rowC;
+	  __vector_quad acc0;
+	  v4sf_t result[4];
+	  __builtin_mma_xxsetaccz (&acc0);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  CO += 4;
+	  AO += temp << 2;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (4, 4)
+#endif
+	}
+      i = (m & 3) >> 1;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (2, 4);
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v2sf_t *rowC;
+	  v2sf_t result[8];
+	  __vector_quad acc0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0 };
+	      t[0] = AO[l << 1], t[1] = AO[(l << 1) + 1];
+	      vec_t *rowA = (vec_t *) & t[0];
+	      vec_t *rowB = (vec_t *) & BO[l << 2];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	    }
+	  SAVE4x2_ACC (&acc0, 0);
+	  CO += 2;
+	  AO += temp << 1;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (2, 4)
+#endif
+	}
+      i = (m & 1) >> 0;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (1, 4)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowA = { AO[l], AO[l], AO[l], AO[l] };
+	      v4sf_t rowB = { BO[l << 2], BO[(l << 2) + 1], BO[(l << 2) + 2],
+		BO[(l << 2) + 3]
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0 * ldc] = t[0];
+	  CO[1 * ldc] = t[1];
+	  CO[2 * ldc] = t[2];
+	  CO[3 * ldc] = t[3];
+#else
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t[2];
+	  CO[3 * ldc] += t[3];
+#endif
+	  CO += 1;
+	  AO += temp;
+	  BO += temp << 2;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (1, 4)
+#endif
+	}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 4;			// number of values in A
+#endif
+
+      B += k << 2;
+    }
+  N = (n & 3) >> 1;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j, temp;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      FLOAT *CO;
+      FLOAT *AO;
+      CO = C;
+      C += ldc << 1;
+      AO = A;
+#if !defined(TRMMKERNEL)
+      i = m >> 5;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  FLOAT *A1;
+	  A1 = AO + (16 * k);
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k; l++)
+	    {
+	      FLOAT t[4] = { 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      vec_t *rowB = (vec_t *) & t[0];
+	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      vec_t *rowA1 = (vec_t *) & A1[l << 4];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
+	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
+	      __builtin_mma_xvf32gerpp (&acc3, rowB[0], rowA[3]);
+	      __builtin_mma_xvf32gerpp (&acc4, rowB[0], rowA1[0]);
+	      __builtin_mma_xvf32gerpp (&acc5, rowB[0], rowA1[1]);
+	      __builtin_mma_xvf32gerpp (&acc6, rowB[0], rowA1[2]);
+	      __builtin_mma_xvf32gerpp (&acc7, rowB[0], rowA1[3]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  SAVE2x4_ACC (&acc2, 8);
+	  SAVE2x4_ACC (&acc3, 12);
+	  CO += 16;
+	  SAVE2x4_ACC (&acc4, 0);
+	  SAVE2x4_ACC (&acc5, 4);
+	  SAVE2x4_ACC (&acc6, 8);
+	  SAVE2x4_ACC (&acc7, 12);
+	  CO += 16;
+	  AO += k << 5;
+	  BO += k << 1;
+	}
+      i = (m & 31) >> 4;
+#else
+      i = m >> 4;
+#endif
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (16, 2)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      vec_t *rowB = (vec_t *) & t[0];
+	      vec_t *rowA = (vec_t *) & AO[l << 4];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
+	      __builtin_mma_xvf32gerpp (&acc2, rowB[0], rowA[2]);
+	      __builtin_mma_xvf32gerpp (&acc3, rowB[0], rowA[3]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  SAVE2x4_ACC (&acc2, 8);
+	  SAVE2x4_ACC (&acc3, 12);
+	  CO += 16;
+	  AO += temp << 4;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (16, 2)
+#endif
+	}
+      i = (m & 15) >> 3;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (8, 2)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      vec_t *rowB = (vec_t *) & t[0];
+	      vec_t *rowA = (vec_t *) & AO[l << 3];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	      __builtin_mma_xvf32gerpp (&acc1, rowB[0], rowA[1]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  CO += 8;
+	  AO += temp << 3;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (8, 2)
+#endif
+	}
+      i = (m & 7) >> 2;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0;
+	  __builtin_mma_xxsetaccz (&acc0);
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (4, 2)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  BLASLONG l = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      FLOAT t[4] = { 0 };
+	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
+	      vec_t *rowB = (vec_t *) & t[0];
+	      vec_t *rowA = (vec_t *) & AO[l << 2];
+	      __builtin_mma_xvf32gerpp (&acc0, rowB[0], rowA[0]);
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  CO += 4;
+	  AO += temp << 2;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (4, 2)
+#endif
+	}
+      i = (m & 3) >> 1;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+	  BLASLONG l = 0;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (2, 2)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < (temp << 1); l += 2)
+	    {
+	      v4sf_t rowA = { AO[l], AO[l], AO[l + 1], AO[l + 1] };
+	      v4sf_t rowB = { BO[l], BO[l + 1], BO[l], BO[l + 1] };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0 * ldc] = t[0];
+	  CO[1 * ldc] = t[1];
+	  CO[0 * ldc + 1] = t[2];
+	  CO[1 * ldc + 1] = t[3];
+#else
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[0 * ldc + 1] += t[2];
+	  CO[1 * ldc + 1] += t[3];
+#endif
+	  CO += 2;
+	  AO += temp << 1;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (2, 2)
+#endif
+	}
+      i = (m & 1) >> 0;
+      for (j = 0; j < i; j++)
+	{
+	  FLOAT *BO;
+	  BLASLONG l = 0;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (1, 2)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowA = { AO[l], AO[l], 0, 0 };
+	      v4sf_t rowB = { BO[l << 1], BO[(l << 1) + 1], 0, 0 };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0 * ldc] = t[0];
+	  CO[1 * ldc] = t[1];
+#else
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+#endif
+	  CO += 1;
+	  AO += temp;
+	  BO += temp << 1;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (1, 2)
+#endif
+	}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 2;			// number of values in A
+#endif
+
+      B += k << 1;
+    }
+  N = (n & 1) >> 0;
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, temp;
+#if defined(TRMMKERNEL) && defined(LEFT)
+      off = offset;
+#endif
+      FLOAT *CO;
+      FLOAT *AO;
+      CO = C;
+      C += ldc;
+      AO = A;
+      i = m;
+      while (i >= 16)
+	{
+	  FLOAT *BO;
+	  BLASLONG l = 0;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (16, 1)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+	  v4sf_t t2 = { 0, 0, 0, 0 };
+	  v4sf_t t3 = { 0, 0, 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l], BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 4], AO[(l << 4) + 1], AO[(l << 4) + 2],
+		AO[(l << 4) + 3]
+	      };
+	      v4sf_t rowA1 =
+		{ AO[(l << 4) + 4], AO[(l << 4) + 5], AO[(l << 4) + 6],
+		AO[(l << 4) + 7]
+	      };
+	      v4sf_t rowA2 =
+		{ AO[(l << 4) + 8], AO[(l << 4) + 9], AO[(l << 4) + 10],
+		AO[(l << 4) + 11]
+	      };
+	      v4sf_t rowA3 =
+		{ AO[(l << 4) + 12], AO[(l << 4) + 13], AO[(l << 4) + 14],
+		AO[(l << 4) + 15]
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	      t2 += rowA2 * rowB;
+	      t3 += rowA3 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  t2 = t2 * valpha;
+	  t3 = t3 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+	  CO[2] = t[2];
+	  CO[3] = t[3];
+	  CO[4] = t1[0];
+	  CO[5] = t1[1];
+	  CO[6] = t1[2];
+	  CO[7] = t1[3];
+	  CO[8] = t2[0];
+	  CO[9] = t2[1];
+	  CO[10] = t2[2];
+	  CO[11] = t2[3];
+	  CO[12] = t3[0];
+	  CO[13] = t3[1];
+	  CO[14] = t3[2];
+	  CO[15] = t3[3];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  CO[4] += t1[0];
+	  CO[5] += t1[1];
+	  CO[6] += t1[2];
+	  CO[7] += t1[3];
+	  CO[8] += t2[0];
+	  CO[9] += t2[1];
+	  CO[10] += t2[2];
+	  CO[11] += t2[3];
+	  CO[12] += t3[0];
+	  CO[13] += t3[1];
+	  CO[14] += t3[2];
+	  CO[15] += t3[3];
+#endif
+	  AO += temp << 4;
+	  BO += temp;
+	  CO += 16;
+	  i -= 16;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (16, 1)
+#endif
+	}
+      while (i >= 8)
+	{
+	  FLOAT *BO;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (8, 1)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l], BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 3], AO[(l << 3) + 1], AO[(l << 3) + 2],
+		AO[(l << 3) + 3]
+	      };
+	      v4sf_t rowA1 =
+		{ AO[(l << 3) + 4], AO[(l << 3) + 5], AO[(l << 3) + 6],
+		AO[(l << 3) + 7]
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+	  CO[2] = t[2];
+	  CO[3] = t[3];
+	  CO[4] = t1[0];
+	  CO[5] = t1[1];
+	  CO[6] = t1[2];
+	  CO[7] = t1[3];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  CO[4] += t1[0];
+	  CO[5] += t1[1];
+	  CO[6] += t1[2];
+	  CO[7] += t1[3];
+#endif
+	  AO += temp << 3;
+	  BO += temp;
+	  CO += 8;
+	  i -= 8;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (8, 1)
+#endif
+	}
+      while (i >= 4)
+	{
+	  FLOAT *BO;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (4, 1)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l], BO[l], BO[l] };
+	      v4sf_t rowA = { AO[l << 2], AO[(l << 2) + 1], AO[(l << 2) + 2],
+		AO[(l << 2) + 3]
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+	  CO[2] = t[2];
+	  CO[3] = t[3];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+#endif
+	  AO += temp << 2;
+	  BO += temp;
+	  CO += 4;
+	  i -= 4;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (4, 1)
+#endif
+	}
+      while (i >= 2)
+	{
+	  FLOAT *BO;
+	  BLASLONG l = 0;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (2, 1)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < temp; l++)
+	    {
+	      v4sf_t rowB = { BO[l], BO[l], 0, 0 };
+	      v4sf_t rowA = { AO[l << 1], AO[(l << 1) + 1], 0, 0 };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+#if defined(TRMMKERNEL)
+	  CO[0] = t[0];
+	  CO[1] = t[1];
+#else
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+#endif
+	  AO += temp << 1;
+	  BO += temp;
+	  CO += 2;
+	  i -= 2;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (2, 1)
+#endif
+	}
+      while (i >= 1)
+	{
+	  FLOAT *BO;
+#if defined(TRMMKERNEL)
+	  REFRESH_POINTERS (1, 1)
+#else
+	  BO = B;
+	  temp = k;
+#endif
+
+	  BLASLONG l = 0;
+	  FLOAT t = 0;
+	  for (l = 0; l < temp; l++)
+	    {
+	      t += AO[l] * BO[l];
+	    }
+	  AO += temp;
+	  BO += temp;
+#if defined(TRMMKERNEL)
+	  CO[0] = t * alpha;
+#else
+	  CO[0] += t * alpha;
+#endif
+	  CO += 1;
+	  i -= 1;
+#if defined(TRMMKERNEL)
+	  REFRESH_AFTER_SAVE (1, 1)
+#endif
+	}
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+      off += 1;			// number of values in A
+#endif
+      B += k;
+    }
+  return 0;
+}
diff --git a/kernel/power/shgemm_kernel_power10.c b/kernel/power/shgemm_kernel_power10.c
new file mode 100644
index 000000000..7455f925c
--- /dev/null
+++ b/kernel/power/shgemm_kernel_power10.c
@@ -0,0 +1,1044 @@
+/*********************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+#include "common.h"
+#include <altivec.h>
+#if defined(HALF) && defined(HALFCONVERSION)
+static float
+bfloat16tof32 (bfloat16 f16)
+{
+  float result = 0;
+  unsigned short *q = (unsigned short *) (&result);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  q[0] = f16;
+#else
+  q[1] = f16;
+#endif
+  return result;
+}
+
+#define BF16TOF32(x) (bfloat16tof32(x))
+#else
+#define BF16TOF32(x) x
+#endif
+
+typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
+typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
+
+vector char mask =
+  { 0x0, 0x1, 0x8, 0x9, 0x2, 0x3, 0xa, 0xb, 0x4, 0x5, 0xc, 0xd, 0x6, 0x7, 0xe,
+  0xf
+};
+
+/* 
+ * BFLOAT16 xvbf16ger2pp instruction needs 4×2 matrix of
+ * bfloat16 floating-point values as input. Hence this
+ * merging is needed on A and B matrices. 
+ */
+#define MERGE_ROW(x) vec_perm(x, x, mask)
+#define MERGE_HIGH(x, y) (vec_t) vec_mergeh ((vector short)x, (vector short)y)
+#define MERGE_LOW(x, y) (vec_t) vec_mergel ((vector short)x, (vector short)y)
+
+#define SAVE_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[1*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[3*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define SAVE_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[5*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[7*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE4x2_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE4x2_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
+          rowC[0] += result[0] * alpha;
+
+#define MMA __builtin_mma_xvbf16ger2pp
+
+#define  SAVE2x4_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[2] * alpha;
+
+#define SET_ACC_ZERO4() \
+	  __builtin_mma_xxsetaccz (&acc0); \
+	  __builtin_mma_xxsetaccz (&acc1); \
+	  __builtin_mma_xxsetaccz (&acc2); \
+	  __builtin_mma_xxsetaccz (&acc3);
+
+#define SET_ACC_ZERO8() \
+	  __builtin_mma_xxsetaccz (&acc0); \
+	  __builtin_mma_xxsetaccz (&acc1); \
+	  __builtin_mma_xxsetaccz (&acc2); \
+	  __builtin_mma_xxsetaccz (&acc3); \
+	  __builtin_mma_xxsetaccz (&acc4); \
+	  __builtin_mma_xxsetaccz (&acc5); \
+	  __builtin_mma_xxsetaccz (&acc6); \
+	  __builtin_mma_xxsetaccz (&acc7);
+
+#define PREFETCH1(x, y) asm volatile ("dcbt %0, %1" : : "r" (x), "b" (y) : "memory");
+/*************************************************************************************
+* SHGEMM Kernel
+*************************************************************************************/
+int
+CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
+       IFLOAT * B, FLOAT * C, BLASLONG ldc)
+{
+  BLASLONG N = n;
+  BLASLONG i1;
+  v4sf_t valpha = { alpha, alpha, alpha, alpha };
+  vector short vzero = { 0, 0, 0, 0, 0, 0, 0, 0 };
+  N = n >> 3;
+  /* Loop for n >= 8. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc << 3;
+      AO = A;
+      PREFETCH1 (A, 128);
+      PREFETCH1 (A, 256);
+      i = m >> 4;
+      /* Loop for m >= 16. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], rowA[2]);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], rowA[2]);
+	      vec_t rowA2_h = MERGE_HIGH (rowA[1], rowA[3]);
+	      vec_t rowA2_l = MERGE_LOW (rowA[1], rowA[3]);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	      MMA (&acc4, rowB_h, rowA2_h);
+	      MMA (&acc5, rowB_l, rowA2_h);
+	      MMA (&acc6, rowB_h, rowA2_l);
+	      MMA (&acc7, rowB_l, rowA2_l);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 4;
+	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], vzero);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], vzero);
+	      vec_t rowA2_h = MERGE_HIGH (rowA[1], vzero);
+	      vec_t rowA2_l = MERGE_LOW (rowA[1], vzero);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	      MMA (&acc4, rowB_h, rowA2_h);
+	      MMA (&acc5, rowB_l, rowA2_h);
+	      MMA (&acc6, rowB_h, rowA2_l);
+	      MMA (&acc7, rowB_l, rowA2_l);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC1 (&acc1, 0);
+	  SAVE_ACC1 (&acc3, 4);
+	  SAVE_ACC (&acc4, 8);
+	  SAVE_ACC (&acc6, 12);
+	  SAVE_ACC1 (&acc5, 8);
+	  SAVE_ACC1 (&acc7, 12);
+	  CO += 16;
+
+	  AO += (k << 4);
+	  BO += (k << 3);
+	}
+      i = (m & 15) >> 3;
+      /* Loop for m >= 8. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 4]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], rowA[1]);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], rowA[1]);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 4;
+	      vec_t *rowA = (vec_t *) & (AO[l]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], vzero);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], vzero);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC1 (&acc1, 0);
+	  SAVE_ACC1 (&acc3, 4);
+	  CO += 8;
+	  AO += (k << 3);
+	  BO += (k << 3);
+	}
+      i = (m & 7) >> 2;
+      /* Loop for m >= 4. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      vec_t rowA_mrg = MERGE_ROW (rowA[0]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), rowA_mrg);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), rowA_mrg);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vector short rowA =
+		{ AO[l + 0], 0, AO[l + 1], 0, AO[l + 2], 0, AO[l + 3], 0 };
+	      vec_t *rowB = (vec_t *) & (BO[l << 1]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC1 (&acc1, 0);
+	  CO += 4;
+	  AO += (k << 2);
+	  BO += (k << 3);
+	}
+      i = (m & 3) >> 1;
+      /* Loop for m >= 2. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v2sf_t *rowC;
+	  v2sf_t result[8];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowA =
+		{ AO[(l << 2) + 0], AO[(l << 2) + 2], AO[(l << 2) + 1],
+		AO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowA = { AO[l + 0], 0, AO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowB = (vec_t *) & (BO[(l << 2)]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	    }
+	  SAVE4x2_ACC (&acc0, 0);
+	  SAVE4x2_ACC1 (&acc1, 0);
+	  CO += 2;
+	  AO += (k << 1);
+	  BO += (k << 3);
+	}
+      i = (m & 1) >> 0;
+      /* Loop for m = 1. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 }
+	  , t1 =
+	  {
+	  0, 0, 0, 0};
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), BF16TOF32 (AO[l]),
+		BF16TOF32 (AO[l])
+	      };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l << 3]), BF16TOF32 (BO[(l << 3) + 1]),
+		BF16TOF32 (BO[(l << 3) + 2]),
+		BF16TOF32 (BO[(l << 3) + 3])
+	      };
+	      v4sf_t rowB1 =
+		{ BF16TOF32 (BO[(l << 3) + 4]), BF16TOF32 (BO[(l << 3) + 5]),
+		BF16TOF32 (BO[(l << 3) + 6]),
+		BF16TOF32 (BO[(l << 3) + 7])
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA * rowB1;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t[2];
+	  CO[3 * ldc] += t[3];
+	  CO[4 * ldc] += t1[0];
+	  CO[5 * ldc] += t1[1];
+	  CO[6 * ldc] += t1[2];
+	  CO[7 * ldc] += t1[3];
+	  CO += 1;
+	  AO += k;
+	  BO += (k << 3);
+	}
+      B += k << 3;
+    }
+  N = (n & 7) >> 2;
+  /* Loop for n >= 4. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc << 2;
+      AO = A;
+      i = m >> 5;
+      /* Loop for m >= 32. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  IFLOAT *A1 = AO + (16 * k);
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowA1 = (vec_t *) & (A1[l << 5]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc4, rowB_mrg, MERGE_HIGH (rowA1[0], rowA1[2]));
+	      MMA (&acc5, rowB_mrg, MERGE_LOW (rowA1[0], rowA1[2]));
+	      MMA (&acc6, rowB_mrg, MERGE_HIGH (rowA1[1], rowA1[3]));
+	      MMA (&acc7, rowB_mrg, MERGE_LOW (rowA1[1], rowA1[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
+	      vec_t *rowA1 = (vec_t *) & (A1[(l << 2)]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], vzero));
+	      MMA (&acc4, rowB_mrg, MERGE_HIGH (rowA1[0], vzero));
+	      MMA (&acc5, rowB_mrg, MERGE_LOW (rowA1[0], vzero));
+	      MMA (&acc6, rowB_mrg, MERGE_HIGH (rowA1[1], vzero));
+	      MMA (&acc7, rowB_mrg, MERGE_LOW (rowA1[1], vzero));
+	    }
+
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc2, 0);
+	  SAVE_ACC (&acc3, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc4, 0);
+	  SAVE_ACC (&acc5, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc6, 0);
+	  SAVE_ACC (&acc7, 4);
+	  CO += 8;
+	  AO += k << 5;
+	  BO += k << 2;
+	}
+      i = (m & 31) >> 4;
+      /* Loop for m >= 16. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], rowA[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], vzero));
+	    }
+
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc2, 0);
+	  SAVE_ACC (&acc3, 4);
+	  CO += 8;
+	  AO += k << 4;
+	  BO += k << 2;
+	}
+      i = (m & 15) >> 3;
+      /* Loop for m >= 8. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 4]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], rowA[1]));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], rowA[1]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  AO += k << 3;
+	  BO += k << 2;
+	}
+      i = (m & 7) >> 2;
+      /* Loop for m >= 4. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  __vector_quad acc0;
+	  v4sf_t result[4];
+	  BLASLONG l = 0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), MERGE_ROW (rowA[0]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vector short rowA =
+		{ AO[l], 0, AO[l + 1], 0, AO[l + 2], 0, AO[l + 3], 0 };
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  CO += 4;
+	  AO += k << 2;
+	  BO += k << 2;
+	}
+      i = (m & 3) >> 1;
+      /* Loop for m >= 2. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v2sf_t *rowC;
+	  v2sf_t result[8];
+	  __vector_quad acc0;
+	  BLASLONG l = 0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowA =
+		{ AO[(l << 2) + 0], AO[(l << 2) + 2], AO[(l << 2) + 1],
+		AO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowA = { AO[l], 0, AO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowB = (vec_t *) & (BO[l << 1]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	    }
+	  SAVE4x2_ACC (&acc0, 0);
+	  CO += 2;
+	  AO += k << 1;
+	  BO += k << 2;
+	}
+      i = (m & 1) >> 0;
+      /* Loop for m = 1. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), BF16TOF32 (AO[l]),
+		BF16TOF32 (AO[l])
+	      };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l << 2]), BF16TOF32 (BO[(l << 2) + 1]),
+		BF16TOF32 (BO[(l << 2) + 2]),
+		BF16TOF32 (BO[(l << 2) + 3])
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t[2];
+	  CO[3 * ldc] += t[3];
+	  AO += k;
+	  BO += (k << 2);
+	  CO += 1;
+	}
+
+      B += k << 2;
+    }
+  N = (n & 3) >> 1;
+  /* Loop for n >= 2. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc << 1;
+      AO = A;
+      i = m >> 5;
+      /* Loop for m >= 32. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  IFLOAT *A1 = AO + (16 * k);
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowA1 = (vec_t *) & (A1[l << 5]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc4, (vec_t) rowB, MERGE_HIGH (rowA1[0], rowA1[2]));
+	      MMA (&acc5, (vec_t) rowB, MERGE_LOW (rowA1[0], rowA1[2]));
+	      MMA (&acc6, (vec_t) rowB, MERGE_HIGH (rowA1[1], rowA1[3]));
+	      MMA (&acc7, (vec_t) rowB, MERGE_LOW (rowA1[1], rowA1[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      vec_t *rowA1 = (vec_t *) & (A1[l << 3]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc4, (vec_t) rowB, MERGE_HIGH (rowA1[0], rowA1[2]));
+	      MMA (&acc5, (vec_t) rowB, MERGE_LOW (rowA1[0], rowA1[2]));
+	      MMA (&acc6, (vec_t) rowB, MERGE_HIGH (rowA1[1], rowA1[3]));
+	      MMA (&acc7, (vec_t) rowB, MERGE_LOW (rowA1[1], rowA1[3]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  SAVE2x4_ACC (&acc2, 8);
+	  SAVE2x4_ACC (&acc3, 12);
+	  CO += 16;
+	  SAVE2x4_ACC (&acc4, 0);
+	  SAVE2x4_ACC (&acc5, 4);
+	  SAVE2x4_ACC (&acc6, 8);
+	  SAVE2x4_ACC (&acc7, 12);
+	  CO += 16;
+	  AO += k << 5;
+	  BO += k << 1;
+	}
+      i = (m & 31) >> 4;
+      /* Loop for m >= 16. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  SAVE2x4_ACC (&acc2, 8);
+	  SAVE2x4_ACC (&acc3, 12);
+	  CO += 16;
+	  AO += k << 4;
+	  BO += k << 1;
+	}
+      i = (m & 15) >> 3;
+      /* Loop for m >= 8. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 4]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[1]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[1]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[1]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[1]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  CO += 8;
+	  AO += k << 3;
+	  BO += k << 1;
+	}
+      i = (m & 7) >> 2;
+      /* Loop for m >= 4. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_ROW (rowA[0]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_ROW (rowA[0]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  CO += 4;
+	  AO += k << 2;
+	  BO += k << 1;
+	}
+      i = (m & 3) >> 1;
+      /* Loop for m >= 2. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < (k << 1); l += 2)
+	    {
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), BF16TOF32 (AO[l + 1]),
+		BF16TOF32 (AO[l + 1])
+	      };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l + 1]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l + 1])
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[0 * ldc + 1] += t[2];
+	  CO[1 * ldc + 1] += t[3];
+	  CO += 2;
+	  AO += k << 1;
+	  BO += k << 1;
+	}
+      i = (m & 1) >> 0;
+      /* Loop for m = 1. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowA = { BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), 0, 0 };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l << 1]), BF16TOF32 (BO[(l << 1) + 1]), 0,
+		0
+	      };
+	      t += rowA * rowB;
+	    }
+	  CO[0 * ldc] += t[0] * alpha;
+	  CO[1 * ldc] += t[1] * alpha;
+	  CO += 1;
+	  AO += k;
+	  BO += k << 1;
+	}
+      B += k << 1;
+    }
+  N = (n & 1) >> 0;
+  /* Loop for n = 1. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc;
+      AO = A;
+      i = m;
+      /* Loop for m >= 16. */
+      while (i >= 16)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+	  v4sf_t t2 = { 0, 0, 0, 0 };
+	  v4sf_t t3 = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l])
+	      };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 4]), BF16TOF32 (AO[(l << 4) + 1]),
+		BF16TOF32 (AO[(l << 4) + 2]),
+		BF16TOF32 (AO[(l << 4) + 3])
+	      };
+	      v4sf_t rowA1 =
+		{ BF16TOF32 (AO[(l << 4) + 4]), BF16TOF32 (AO[(l << 4) + 5]),
+		BF16TOF32 (AO[(l << 4) + 6]),
+		BF16TOF32 (AO[(l << 4) + 7])
+	      };
+	      v4sf_t rowA2 =
+		{ BF16TOF32 (AO[(l << 4) + 8]), BF16TOF32 (AO[(l << 4) + 9]),
+		BF16TOF32 (AO[(l << 4) + 10]),
+		BF16TOF32 (AO[(l << 4) + 11])
+	      };
+	      v4sf_t rowA3 = { BF16TOF32 (AO[(l << 4) + 12]),
+		BF16TOF32 (AO[(l << 4) + 13]), BF16TOF32 (AO[(l << 4) + 14]),
+		BF16TOF32 (AO[(l << 4) + 15])
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	      t2 += rowA2 * rowB;
+	      t3 += rowA3 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  t2 = t2 * valpha;
+	  t3 = t3 * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  CO[4] += t1[0];
+	  CO[5] += t1[1];
+	  CO[6] += t1[2];
+	  CO[7] += t1[3];
+	  CO[8] += t2[0];
+	  CO[9] += t2[1];
+	  CO[10] += t2[2];
+	  CO[11] += t2[3];
+	  CO[12] += t3[0];
+	  CO[13] += t3[1];
+	  CO[14] += t3[2];
+	  CO[15] += t3[3];
+	  AO += k << 4;
+	  BO += k;
+	  CO += 16;
+	  i -= 16;
+	}
+      /* Loop for m >= 8. */
+      while (i >= 8)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l])
+	      };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 3]), BF16TOF32 (AO[(l << 3) + 1]),
+		BF16TOF32 (AO[(l << 3) + 2]),
+		BF16TOF32 (AO[(l << 3) + 3])
+	      };
+	      v4sf_t rowA1 =
+		{ BF16TOF32 (AO[(l << 3) + 4]), BF16TOF32 (AO[(l << 3) + 5]),
+		BF16TOF32 (AO[(l << 3) + 6]),
+		BF16TOF32 (AO[(l << 3) + 7])
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  CO[4] += t1[0];
+	  CO[5] += t1[1];
+	  CO[6] += t1[2];
+	  CO[7] += t1[3];
+	  AO += k << 3;
+	  BO += k;
+	  CO += 8;
+	  i -= 8;
+	}
+      /* Loop for m >= 4. */
+      while (i >= 4)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l])
+	      };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 2]), BF16TOF32 (AO[(l << 2) + 1]),
+		BF16TOF32 (AO[(l << 2) + 2]),
+		BF16TOF32 (AO[(l << 2) + 3])
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  AO += k << 2;
+	  BO += k;
+	  CO += 4;
+	  i -= 4;
+	}
+      /* Loop for m >= 2. */
+      while (i >= 2)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB = { BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), 0, 0 };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 1]), BF16TOF32 (AO[(l << 1) + 1]), 0,
+		0
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  AO += k << 1;
+	  BO += k;
+	  CO += 2;
+	  i -= 2;
+	}
+      /* Loop for m = 1. */
+      while (i >= 1)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  FLOAT t = 0;
+	  for (l = 0; l < k; l++)
+	    {
+	      t += BF16TOF32 (AO[l]) * BF16TOF32 (BO[l]);
+	    }
+	  AO += k;
+	  BO += k;
+	  CO[0] += t * alpha;
+	  CO += 1;
+	  i -= 1;
+	}
+
+      B += k;
+    }
+
+  return 0;
+}
diff --git a/kernel/power/srot.c b/kernel/power/srot.c
index 6af813c16..9638a59eb 100644
--- a/kernel/power/srot.c
+++ b/kernel/power/srot.c
@@ -39,7 +39,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #pragma GCC optimize "O1"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "srot_microk_power8.c"
 #endif
 
diff --git a/kernel/power/sscal.c b/kernel/power/sscal.c
index 4f3ba5698..ddd5b2c5b 100644
--- a/kernel/power/sscal.c
+++ b/kernel/power/sscal.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "sscal_microk_power8.c"
 #endif
 
diff --git a/kernel/power/sswap.c b/kernel/power/sswap.c
index 23d13280f..a56434444 100644
--- a/kernel/power/sswap.c
+++ b/kernel/power/sswap.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "sswap_microk_power8.c"
 #endif
 
diff --git a/kernel/power/zasum.c b/kernel/power/zasum.c
index f61c62e75..8383e39ab 100644
--- a/kernel/power/zasum.c
+++ b/kernel/power/zasum.c
@@ -46,7 +46,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "zasum_microk_power8.c"
 #endif
 
diff --git a/kernel/power/zaxpy.c b/kernel/power/zaxpy.c
index f0f8c6910..4a7c26c69 100644
--- a/kernel/power/zaxpy.c
+++ b/kernel/power/zaxpy.c
@@ -36,7 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "zaxpy_microk_power8.c"
 #endif
 
diff --git a/kernel/power/zcopy.c b/kernel/power/zcopy.c
index b21d6ef15..bb80decd2 100644
--- a/kernel/power/zcopy.c
+++ b/kernel/power/zcopy.c
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "zcopy_microk_power8.c"
 #endif
 
diff --git a/kernel/power/zdot.c b/kernel/power/zdot.c
index fd36c7f44..9086ef35b 100644
--- a/kernel/power/zdot.c
+++ b/kernel/power/zdot.c
@@ -36,7 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "zdot_microk_power8.c"
 #endif
 
diff --git a/kernel/power/zgemm_kernel_power10.S b/kernel/power/zgemm_kernel_power10.S
new file mode 100644
index 000000000..fca389e69
--- /dev/null
+++ b/kernel/power/zgemm_kernel_power10.S
@@ -0,0 +1,245 @@
+/***************************************************************************
+Copyright (c) 2013-2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+#include "def_vsx.h"
+
+#define LOAD	ld
+ 
+#define STACKSIZE 512
+
+#define FZERO	312+192(SP)
+
+#define FLINK_SAVE (STACKSIZE+16) /* 16($r12) */
+
+#define	M	r3
+#define	N	r4
+#define	K	r5
+
+ 
+#define A	r8
+#define	B	r9
+#define	C	r10
+#define	LDC	r6
+#define OFFSET	r7
+ 
+ 
+
+#define o0	0
+#define alpha_r vs62
+#define alpha_i vs63
+
+#define VECSAVE r11
+
+#define FRAMEPOINTER r12
+
+#define T10 r14
+
+#define L	r15
+#define T8	r16
+#define T5	r17
+#define T2	r19
+#define TEMP_REG	r20
+#define	T6	r21
+#define	I	r22
+#define J	r23
+#define AO	r24
+#define	BO	r25
+#define	CO	r26
+#define T7	r27
+#define	T3	r28
+#define T4	r29
+
+#define PRE	r30
+#define T1  	r31
+
+#ifndef NEEDPARAM
+
+	PROLOGUE
+	PROFCODE
+
+	mr      FRAMEPOINTER, SP
+    addi    SP, SP, -STACKSIZE 
+    mflr    r0
+	stfd	f14,    0(SP)
+	stfd	f15,    8(SP)
+	stfd	f16,   16(SP)
+	stfd	f17,   24(SP)
+
+	stfd	f18,   32(SP)
+	stfd	f19,   40(SP)
+	stfd	f20,   48(SP)
+	stfd	f21,   56(SP)
+
+	stfd	f22,   64(SP)
+	stfd	f23,   72(SP)
+	stfd	f24,   80(SP)
+	stfd	f25,   88(SP)
+
+	stfd	f26,   96(SP)
+	stfd	f27,  104(SP)
+	stfd	f28,  112(SP)
+	stfd	f29,  120(SP)
+
+	stfd	f30,  128(SP)
+	stfd	f31,  136(SP)
+
+    xxspltd  alpha_r,vs1,0  /*copy from register f1 */
+    xxspltd  alpha_i,vs2,0  /*copy from register f2 */
+ 
+	std	r31,  144(SP)
+	std	r30,  152(SP)
+	std	r29,  160(SP)
+	std	r28,  168(SP)
+	std	r27,  176(SP)
+	std	r26,  184(SP)
+	std	r25,  192(SP)
+	std	r24,  200(SP)
+	std	r23,  208(SP)
+	std	r22,  216(SP)
+	std	r21,  224(SP)
+	std	r20,  232(SP)
+	std	r19,  240(SP)
+	std	r18,  248(SP)
+	std	r17,  256(SP)
+	std	r16,  264(SP)
+	std	r15,  272(SP)
+	std	r14,  280(SP)
+ 
+ 
+    stxv    vs20,  288(SP)
+    stxv    vs21,  304(SP)
+    stxv    vs22,  320(SP)
+    stxv    vs23,  336(SP)
+    stxv    vs24,  352(SP)
+    stxv    vs25,  368(SP)
+    stxv    vs26,  384(SP)
+    stxv    vs27,  400(SP)
+    stxv    vs28,  416(SP)
+    stxv    vs29,  432(SP)
+    stxv    vs30,  448(SP)
+    stxv    vs31,  464(SP)
+
+    std    r0, FLINK_SAVE(SP)
+ 
+
+#if defined(linux) || defined(__FreeBSD__)
+	ld	LDC, FRAMESLOT(0) + 0(FRAMEPOINTER)
+#endif
+
+
+#ifdef TRMMKERNEL
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
+	ld	OFFSET,  FRAMESLOT(1) + 0(FRAMEPOINTER)
+#endif 
+#endif
+
+
+#include "zgemm_macros_power10.S"
+
+ 
+
+	slwi	LDC, LDC, ZBASE_SHIFT
+	li	PRE,  512 
+    li  r0,   0
+ 
+
+#if defined(CC) || defined(CR) || defined(RC) || defined(RR) 
+/*negate for this case as we will use addition -1*(a+b) */
+  xvnegdp alpha_r,alpha_r
+  xvnegdp alpha_i,alpha_i
+#endif
+	.align 4
+
+#include "zgemm_logic_power10.S"
+
+L999:
+ 
+	lfd	f14,    0(SP)
+	lfd	f15,    8(SP)
+	lfd	f16,   16(SP)
+	lfd	f17,   24(SP)
+
+	lfd	f18,   32(SP)
+	lfd	f19,   40(SP)
+	lfd	f20,   48(SP)
+	lfd	f21,   56(SP)
+
+	lfd	f22,   64(SP)
+	lfd	f23,   72(SP)
+	lfd	f24,   80(SP)
+	lfd	f25,   88(SP)
+
+	lfd	f26,   96(SP)
+	lfd	f27,  104(SP)
+	lfd	f28,  112(SP)
+	lfd	f29,  120(SP)
+
+	lfd	f30,  128(SP)
+	lfd	f31,  136(SP)
+
+ 
+	ld	r31,  144(SP)
+	ld	r30,  152(SP)
+	ld	r29,  160(SP)
+	ld	r28,  168(SP)
+	ld	r27,  176(SP)
+	ld	r26,  184(SP)
+	ld	r25,  192(SP)
+	ld	r24,  200(SP)
+	ld	r23,  208(SP)
+	ld	r22,  216(SP)
+	ld	r21,  224(SP)
+	ld	r20,  232(SP)
+	ld	r19,  240(SP)
+	ld	r18,  248(SP)
+	ld	r17,  256(SP)
+	ld	r16,  264(SP)
+	ld	r15,  272(SP)
+	ld	r14,  280(SP)
+
+	ld    r0, 	 FLINK_SAVE(SP)	
+ 
+    lxv    vs20,  288(SP)
+    lxv    vs21,  304(SP)
+    lxv    vs22,  320(SP)
+    lxv    vs23,  336(SP)
+    lxv    vs24,  352(SP)
+    lxv    vs25,  368(SP)
+    lxv    vs26,  384(SP) 
+    lxv    vs27,  400(SP)
+	mtlr r0
+    lxv    vs28,  416(SP)
+    lxv    vs29,  432(SP) 
+    lxv    vs30,  448(SP)
+    lxv    vs31,  464(SP)
+
+	addi	SP, SP, STACKSIZE 
+	blr
+
+	EPILOGUE
+#endif
diff --git a/kernel/power/zgemm_logic_power10.S b/kernel/power/zgemm_logic_power10.S
new file mode 100644
index 000000000..1143733e0
--- /dev/null
+++ b/kernel/power/zgemm_logic_power10.S
@@ -0,0 +1,1735 @@
+/***************************************************************************
+Copyright (c) 2013-2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#define MY_ALIGN .align 3
+b ZGEMM_L2
+/*                MINI SUBROUTINES                            */      
+/*                2x8 MAIN 128x+2 LOOP                     */      
+
+
+ZGEMM_L2x8_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    MY_ALIGN
+ZGEMM_L2x8_LOOP:
+/*----------------------------------------*/   
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_2 0, 0
+ZGEMM_L2x8_K128:
+/*----------------------------------------*/   
+    KERNEL2x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL2x8_2 2, 0
+    KERNEL2x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_2 4, 0
+    KERNEL2x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL2x8_2 6, 0
+    KERNEL2x8_2 7, 0
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL2x8_2 8, 0
+    KERNEL2x8_2 9, 0
+    KERNEL2x8_2 10, 0
+    KERNEL2x8_2 11, 0
+    dcbt    BO, T4
+    KERNEL2x8_2 12, 0
+    KERNEL2x8_2 13, 0
+    KERNEL2x8_2 14, 0
+    KERNEL2x8_2 15, 0
+    KERNEL2x8_2 16, 0
+    KERNEL2x8_2 17, 0
+    KERNEL2x8_2 18, 0
+    KERNEL2x8_2 19, 0
+    KERNEL2x8_2 20, 0
+    KERNEL2x8_2 21, 0
+    KERNEL2x8_2 22, 0
+    KERNEL2x8_2 23, 0
+    KERNEL2x8_2 24, 0
+    KERNEL2x8_2 25, 0
+    KERNEL2x8_2 26, 0
+    KERNEL2x8_2 27, 0
+    KERNEL2x8_2 28, 0
+    KERNEL2x8_2 29, 0
+    KERNEL2x8_2 30, 0
+    KERNEL2x8_2 31, 0
+    KERNEL2x8_2 32, 0
+    KERNEL2x8_2 33, 0
+    KERNEL2x8_2 34, 0
+    KERNEL2x8_2 35, 0
+    KERNEL2x8_2 36, 0
+    KERNEL2x8_2 37, 0
+    KERNEL2x8_2 38, 0
+    KERNEL2x8_2 39, 0
+    KERNEL2x8_2 40, 0
+    KERNEL2x8_2 41, 0
+    KERNEL2x8_2 42, 0
+    KERNEL2x8_2 43, 0
+    KERNEL2x8_2 44, 0
+    KERNEL2x8_2 45, 0
+    KERNEL2x8_2 46, 0
+    KERNEL2x8_2 47, 0
+    KERNEL2x8_2 48, 0
+    KERNEL2x8_2 49, 0
+    KERNEL2x8_2 50, 0
+    KERNEL2x8_2 51, 0
+    KERNEL2x8_2 52, 0
+    KERNEL2x8_2 53, 0
+    KERNEL2x8_2 54, 0
+    KERNEL2x8_2 55, 0
+    KERNEL2x8_2 56, 0
+    KERNEL2x8_2 57, 0
+    KERNEL2x8_2 58, 0
+    KERNEL2x8_2 59, 0
+    KERNEL2x8_2 60, 0
+    KERNEL2x8_2 61, 0
+    KERNEL2x8_2 62, 0
+    KERNEL2x8_2 63, 1
+    bdz     ZGEMM_L2x8_LOOP_END
+    b       ZGEMM_L2x8_LOOP
+    MY_ALIGN  
+
+ZGEMM_L2x8_LOOP_END:
+/*----------------------------------------*/   
+    KERNEL2x8_2 0, 1
+    blr
+    MY_ALIGN
+
+
+ZGEMM_2x4_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    MY_ALIGN
+ZGEMM_L2x4_LOOP:
+/*----------------------------------------*/   
+    KERNEL2x4_2 0, 0
+ZGEMM_L2x4_K32:
+/*----------------------------------------*/   
+    KERNEL2x4_2 1, 0
+    KERNEL2x4_2 2, 0
+    KERNEL2x4_2 3, 0
+    KERNEL2x4_2 4, 0
+    KERNEL2x4_2 5, 0
+    KERNEL2x4_2 6, 0
+    KERNEL2x4_2 7, 0
+    KERNEL2x4_2 8, 0
+    KERNEL2x4_2 9, 0
+    KERNEL2x4_2 10, 0
+    KERNEL2x4_2 11, 0
+    KERNEL2x4_2 12, 0
+    KERNEL2x4_2 13, 0
+    KERNEL2x4_2 14, 0
+    KERNEL2x4_2 15, 1
+    bdnz    ZGEMM_L2x4_LOOP
+    MY_ALIGN  
+ZGEMM_L2x4_LOOP_END:
+/*----------------------------------------*/   
+    KERNEL2x4_2 0, 1
+    blr
+    MY_ALIGN
+
+
+ZGEMM_2x2_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    MY_ALIGN 
+ZGEMM_L2x2_LOOP:
+/*----------------------------------------*/   
+    KERNEL2x2_2 0, 0 
+ZGEMM_L2x2_K32:
+/*----------------------------------------*/   
+    KERNEL2x2_2 1, 0  
+    KERNEL2x2_2 2, 0
+    KERNEL2x2_2 3, 0  
+    KERNEL2x2_2 4, 0
+    KERNEL2x2_2 5, 0 
+    KERNEL2x2_2 6, 0
+    KERNEL2x2_2 7, 0
+    KERNEL2x2_2 8, 0
+    KERNEL2x2_2 9, 0  
+    KERNEL2x2_2 10, 0
+    KERNEL2x2_2 11, 0  
+    KERNEL2x2_2 12, 0
+    KERNEL2x2_2 13, 0 
+    KERNEL2x2_2 14, 0
+    KERNEL2x2_2 15, 1   
+    bdnz    ZGEMM_L2x2_LOOP
+    MY_ALIGN  
+
+
+ZGEMM_L2x2_LOOP_END:
+/*----------------------------------------*/   
+    KERNEL2x2_2 0, 1
+    blr
+    MY_ALIGN
+
+ZGEMM_2x1_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD2x1_2  
+    MY_ALIGN
+ZGEMM_L2x1_LOOP:
+/*----------------------------------------*/   
+    KERNEL2x1_L2 32, 64, 0, 0 
+ZGEMM_L2x1_K32:
+/*----------------------------------------*/   
+    KERNEL2x1_L2 32, 64, 1, 0  
+    KERNEL2x1_L2 32, 64, 2, 0
+    KERNEL2x1_L2 32, 64, 3, 0  
+    KERNEL2x1_L2 32, 64, 4, 0
+    KERNEL2x1_L2 32, 64, 5, 0 
+    KERNEL2x1_L2 32, 64, 6, 0
+    KERNEL2x1_L2 32, 64, 7, 0
+    KERNEL2x1_L2 32, 64, 8, 0
+    KERNEL2x1_L2 32, 64, 9, 0  
+    KERNEL2x1_L2 32, 64, 10, 0
+    KERNEL2x1_L2 32, 64, 11, 0  
+    KERNEL2x1_L2 32, 64, 12, 0
+    KERNEL2x1_L2 32, 64, 13, 0 
+    KERNEL2x1_L2 32, 64, 14, 0
+    KERNEL2x1_L2 32, 64, 15, 1   
+    bdnz    ZGEMM_L2x1_LOOP
+    MY_ALIGN  
+ZGEMM_L2x1_LOOP_END:
+/*----------------------------------------*/   
+    END2x1_2 
+    blr
+
+    MY_ALIGN
+
+
+/*             MAIN LOOP BEGINS               */   
+    MY_ALIGN
+
+
+ZGEMM_L2:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    neg TEMP_REG, OFFSET 
+#endif   
+    srawi.    J, N, 1
+    bgt   ZGEMM_L2_BEGIN
+    b     ZGEMM_L2_END
+
+ZGEMM_L2_BEGIN:
+/*----------------------------------------*/   
+    mr    CO, C
+    slwi    T1, LDC, 1     
+    add     T2,C,LDC    
+    mr    AO, A  
+    add   C, C, T1
+#if defined(TRMMKERNEL) && defined(LEFT)   
+    mr TEMP_REG, OFFSET  /*off = offset;*/
+#endif     
+    srawi.    I, M, 3
+    bgt   ZGEMM_L2_BEGIN_CONTINUE
+    b     ZGEMM_L2x8_END
+
+ZGEMM_L2_BEGIN_CONTINUE:
+    dcbt    CO,r0  /*just prefetch*/
+    dcbt    T2,r0
+
+
+ZGEMM_L2x8_BEGIN:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 8, 2
+#else    
+    mr    BO, B  
+    dcbt    B, r0  
+#endif     
+    dcbt    AO, r0
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 8, 2
+    mr T1, T6
+#else   
+    mr T1, K
+#endif   
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512 
+    srawi.   T8, T1, 7 /* T8 <- T1 % 128 */
+
+    KERNEL2x8_PRELOAD
+    KERNEL2x8_ZERO_AND_PRIME_MMA
+    ble   ZGEMM_L2x8_SUB0
+    bl ZGEMM_L2x8_LMAIN_SUB
+    andi.   L, T1, 127
+
+    bgt   ZGEMM_L2x8_BEGIN_CONTINUE
+    b     ZGEMM_L2x8_SAVE
+
+ZGEMM_L2x8_BEGIN_CONTINUE:
+    b   ZGEMM_L2x8_SUB2
+
+
+ZGEMM_L2x8_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 255
+    cmpwi   T6, 129
+#else   
+    andi.   L, K, 255
+    cmpwi   K, 129
+#endif       
+    li T8, 1
+    bne CMP2x8_128K
+    LOAD_END_2x8 128, 32
+    KERNEL2x8_PRELOAD
+    addi BO, BO, -64
+    addi AO,AO, -256   
+    mtctr   T8    
+    bl ZGEMM_L2x8_K128
+    b ZGEMM_L2x8_SAVE  
+
+CMP2x8_128K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 128
+#else    
+    cmpwi   K, 128
+#endif        
+    bne ZGEMM_L2x8_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -64
+    addi AO,AO, -256   
+    bl ZGEMM_L2x8_K128
+    b ZGEMM_L2x8_SAVE 
+    MY_ALIGN
+
+
+ZGEMM_L2x8_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 64
+    ble ZGEMM_L2x8_SUB2_32
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_2 0, 0
+    KERNEL2x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL2x8_2 2, 0
+    KERNEL2x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_2 4, 0
+    KERNEL2x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL2x8_2 6, 0
+    KERNEL2x8_2 7, 0
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL2x8_2 8, 0
+    KERNEL2x8_2 9, 0
+    KERNEL2x8_2 10, 0
+    KERNEL2x8_2 11, 0
+    dcbt    BO, T4
+    KERNEL2x8_2 12, 0
+    KERNEL2x8_2 13, 0
+    KERNEL2x8_2 14, 0
+    KERNEL2x8_2 15, 0
+    KERNEL2x8_2 16, 0
+    KERNEL2x8_2 17, 0
+    KERNEL2x8_2 18, 0
+    KERNEL2x8_2 19, 0
+    KERNEL2x8_2 20, 0
+    KERNEL2x8_2 21, 0
+    KERNEL2x8_2 22, 0
+    KERNEL2x8_2 23, 0
+    KERNEL2x8_2 24, 0
+    KERNEL2x8_2 25, 0
+    KERNEL2x8_2 26, 0
+    KERNEL2x8_2 27, 0
+    KERNEL2x8_2 28, 0
+    KERNEL2x8_2 29, 0
+    KERNEL2x8_2 30, 0
+    KERNEL2x8_2 31, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x8_SUB2_32:
+/*----------------------------------------*/   
+    andi.      T1,L, 32
+    ble ZGEMM_L2x8_SUB2_16    
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_2 0, 0
+    KERNEL2x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL2x8_2 2, 0
+    KERNEL2x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_2 4, 0
+    KERNEL2x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL2x8_2 6, 0
+    KERNEL2x8_2 7, 0
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL2x8_2 8, 0
+    KERNEL2x8_2 9, 0
+    KERNEL2x8_2 10, 0
+    KERNEL2x8_2 11, 0
+    dcbt    BO, T4
+    KERNEL2x8_2 12, 0
+    KERNEL2x8_2 13, 0
+    KERNEL2x8_2 14, 0
+    KERNEL2x8_2 15, 1
+    MY_ALIGN 
+
+
+ZGEMM_L2x8_SUB2_16:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L2x8_SUB2_8
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL2x8_2 0, 0
+    KERNEL2x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL2x8_2 2, 0
+    KERNEL2x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL2x8_2 4, 0
+    KERNEL2x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL2x8_2 6, 0
+    KERNEL2x8_2 7, 1
+    MY_ALIGN    
+
+
+ZGEMM_L2x8_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L2x8_SUB2_4
+    KERNEL2x8_2 0, 0
+    KERNEL2x8_2 1, 0
+    KERNEL2x8_2 2, 0
+    KERNEL2x8_2 3, 1
+    MY_ALIGN   
+
+
+ZGEMM_L2x8_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L2x8_SUB2_2
+    KERNEL2x8_2 0, 0
+    KERNEL2x8_2 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x8_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L2x8_SUB2_1
+    KERNEL2x8_2 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L2x8_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L2x8_SAVE 
+    LOAD_END_2x8 128, 32
+
+
+ZGEMM_L2x8_SAVE:
+/*----------------------------------------*/   
+    addic.    I, I, -1
+    KERNEL2x8_UNPRIME_MMA
+    SAVE2x8
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 8, 2
+#endif     
+
+    ble   ZGEMM_L2x8_SAVE_CONTINUE
+    b     ZGEMM_L2x8_BEGIN
+
+ZGEMM_L2x8_SAVE_CONTINUE:
+    andi.   T2, M, 7
+    ble   ZGEMM_L2x1_END
+    andi.   T1, M, 4
+    ble   ZGEMM_L2x4_END
+    b   ZGEMM_L2x4_BEGIN
+    MY_ALIGN 
+
+
+ZGEMM_L2x8_END:
+/*----------------------------------------*/   
+
+
+ZGEMM_L2x4_BEGIN:
+/*----------------------------------------*/   
+    andi.   T2, M, 7
+    ble   ZGEMM_L2x1_END
+    andi.   T1, M, 4
+    ble   ZGEMM_L2x4_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 4, 2
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 4, 2
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T11-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    KERNEL2x4_PRELOAD
+    KERNEL2x4_ZERO_AND_PRIME_MMA
+    ble   ZGEMM_L2x4_SUB0 
+    bl ZGEMM_2x4_LMAIN_SUB
+    andi.   L, T1, 31
+    ble   ZGEMM_L2x4_SAVE
+    b    ZGEMM_L2x4_SUB2
+
+
+ZGEMM_L2x4_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 63
+    cmpwi   T6, 33
+#else   
+    andi.   L, K, 63
+    cmpwi   K, 33
+#endif       
+    li T8, 1
+    bne CMP2x4_32K
+    LOAD_END_2x4 64, 32
+    KERNEL2x4_PRELOAD
+    addi BO, BO, -64
+    addi AO,AO, -128
+    mtctr   T8    
+    bl ZGEMM_L2x4_K32   
+    b ZGEMM_L2x4_SAVE  
+    CMP2x4_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 32
+#else    
+    cmpwi   K, 32
+#endif        
+    bne ZGEMM_L2x4_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -64
+    addi AO,AO, -128
+    bl ZGEMM_L2x4_K32   
+    b ZGEMM_L2x4_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+ZGEMM_L2x4_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L2x4_SUB2_8
+    KERNEL2x4_2 0, 0
+    KERNEL2x4_2 1, 0
+    KERNEL2x4_2 2, 0
+    KERNEL2x4_2 3, 0
+    KERNEL2x4_2 4, 0
+    KERNEL2x4_2 5, 0
+    KERNEL2x4_2 6, 0
+    KERNEL2x4_2 7, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x4_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L2x4_SUB2_4
+    KERNEL2x4_2 0, 0
+    KERNEL2x4_2 1, 0
+    KERNEL2x4_2 2, 0
+    KERNEL2x4_2 3, 1
+    MY_ALIGN  
+
+
+ZGEMM_L2x4_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L2x4_SUB2_2
+    KERNEL2x4_2 0, 0
+    KERNEL2x4_2 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x4_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L2x4_SUB2_1
+    KERNEL2x4_2 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L2x4_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L2x4_SAVE 
+    LOAD_END_2x4 64, 32
+
+
+ZGEMM_L2x4_SAVE:
+/*----------------------------------------*/   
+    KERNEL2x4_UNPRIME_MMA
+    SAVE2x4
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 4, 2
+#endif     
+
+
+ZGEMM_L2x4_END:
+/*----------------------------------------*/   
+
+
+ZGEMM_L2x2_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M, 2
+    ble   ZGEMM_L2x2_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 2, 2
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 2, 2
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T11-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    KERNEL2x2_PRELOAD
+    KERNEL2x2_ZERO_AND_PRIME_MMA
+    ble   ZGEMM_L2x2_SUB0 
+    bl ZGEMM_2x2_LMAIN_SUB
+    andi.   L, T1, 31
+    ble   ZGEMM_L2x2_SAVE
+    b   ZGEMM_L2x2_SUB2
+
+
+ZGEMM_L2x2_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 63
+    cmpwi   T6, 33
+#else   
+    andi.   L, K, 63
+    cmpwi   K, 33
+#endif       
+    li T8, 1
+    bne CMP2x2_32K
+    LOAD_END_2x2 32, 32
+    KERNEL2x2_PRELOAD
+    addi BO, BO, -64
+    addi AO,AO, -64
+    mtctr   T8    
+    bl ZGEMM_L2x2_K32   
+    b ZGEMM_L2x2_SAVE  
+    CMP2x2_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 32
+#else    
+    cmpwi   K, 32
+#endif        
+    bne ZGEMM_L2x2_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -64
+    addi AO,AO, -64
+    bl ZGEMM_L2x2_K32   
+    b ZGEMM_L2x2_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+ZGEMM_L2x2_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L2x2_SUB2_8
+    KERNEL2x2_2 0, 0
+    KERNEL2x2_2 1, 0  
+    KERNEL2x2_2 2, 0
+    KERNEL2x2_2 3, 0  
+    KERNEL2x2_2 4, 0
+    KERNEL2x2_2 5, 0 
+    KERNEL2x2_2 6, 0
+    KERNEL2x2_2 7, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x2_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L2x2_SUB2_4
+    KERNEL2x2_2 0, 0
+    KERNEL2x2_2 1, 0  
+    KERNEL2x2_2 2, 0
+    KERNEL2x2_2 3, 1  
+    MY_ALIGN  
+
+
+ZGEMM_L2x2_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L2x2_SUB2_2
+    KERNEL2x2_2 0, 0
+    KERNEL2x2_2 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x2_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L2x2_SUB2_1
+    KERNEL2x2_2 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L2x2_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L2x2_SAVE 
+    LOAD_END_2x2 32, 32
+
+
+ZGEMM_L2x2_SAVE:
+/*----------------------------------------*/   
+    KERNEL2x2_UNPRIME_MMA
+    SAVE2x2
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 2, 2
+#endif     
+
+
+ZGEMM_L2x2_END:
+/*----------------------------------------*/   
+
+
+ZGEMM_L2x1_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M, 1
+    ble   ZGEMM_L2x1_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 1, 2
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 1, 2
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T11-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO2x1
+    ble   ZGEMM_L2x1_SUB0 
+    bl ZGEMM_2x1_LMAIN_SUB
+    andi.   L, T1, 31
+    ble   ZGEMM_L2x1_SAVE
+    b   ZGEMM_L2x1_SUB2
+
+
+ZGEMM_L2x1_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 63
+    cmpwi   T6, 33
+#else   
+    andi.   L, K, 63
+    cmpwi   K, 33
+#endif       
+    li T8, 1
+    bne CMP2x1_32K
+    addi BO, BO, -32
+    addi AO,AO, -16  
+    LOAD2x1O 16, 32 
+    END2x1_WITHOUT_ADD   
+    LOAD2x1_2O  32, 64  
+    mtctr   T8    
+    bl ZGEMM_L2x1_K32   
+    b ZGEMM_L2x1_SAVE  
+    CMP2x1_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 32
+#else    
+    cmpwi   K, 32
+#endif        
+    bne ZGEMM_L2x1_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -64
+    addi AO,AO, -32   
+    LOAD2x1_2O 32, 64
+    bl ZGEMM_L2x1_K32   
+    b ZGEMM_L2x1_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+ZGEMM_L2x1_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L2x1_SUB2_8
+    LOAD2x1_2
+    KERNEL2x1_L2 32, 64, 0, 0
+    KERNEL2x1_L2 32, 64, 1, 0  
+    KERNEL2x1_L2 32, 64, 2, 0
+    KERNEL2x1_L2 32, 64, 3, 0  
+    KERNEL2x1_L2 32, 64, 4, 0
+    KERNEL2x1_L2 32, 64, 5, 0 
+    KERNEL2x1_L2 32, 64, 6, 0
+    KERNEL2x1_E2 32, 64, 7, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x1_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L2x1_SUB2_4
+    LOAD2x1_2
+    KERNEL2x1_L2 32, 64, 0, 0
+    KERNEL2x1_L2 32, 64, 1, 0  
+    KERNEL2x1_L2 32, 64, 2, 0
+    KERNEL2x1_E2 32, 64, 3, 1  
+    MY_ALIGN  
+
+
+ZGEMM_L2x1_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L2x1_SUB2_2
+    LOAD2x1_2
+    KERNEL2x1_L2  32, 64, 0, 0
+    KERNEL2x1_E2  32, 64, 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L2x1_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L2x1_SUB2_1
+    LOAD2x1_2
+    KERNEL2x1_E2  32, 64, 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L2x1_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L2x1_SAVE 
+    KERNEL2x1
+
+
+ZGEMM_L2x1_SAVE:
+/*----------------------------------------*/   
+    SAVE2x1
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 1, 2
+#endif   
+
+
+ZGEMM_L2x1_END:
+/*----------------------------------------*/   
+    slwi    T1, K, 5
+    addic.    J, J, -1
+    add   B, B, T1
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    addi TEMP_REG, TEMP_REG, 2 
+#endif   
+    ble   ZGEMM_L2_END 
+    b     ZGEMM_L2_BEGIN
+
+ZGEMM_L2_END:
+
+b ZGEMM_L1
+/*                MINI SUBROUTINES                            */      
+/*                1x8 MAIN 128x+2 LOOP                     */      
+
+
+ZGEMM_L1x8_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    MY_ALIGN
+ZGEMM_L1x8_LOOP:
+/*----------------------------------------*/   
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_2 0, 0
+ZGEMM_L1x8_K128:
+/*----------------------------------------*/   
+    KERNEL1x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL1x8_2 2, 0
+    KERNEL1x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_2 4, 0
+    KERNEL1x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL1x8_2 6, 0
+    KERNEL1x8_2 7, 0
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL1x8_2 8, 0
+    KERNEL1x8_2 9, 0
+    KERNEL1x8_2 10, 0
+    KERNEL1x8_2 11, 0
+    dcbt    BO, T4
+    KERNEL1x8_2 12, 0
+    KERNEL1x8_2 13, 0
+    KERNEL1x8_2 14, 0
+    KERNEL1x8_2 15, 0
+    KERNEL1x8_2 16, 0
+    KERNEL1x8_2 17, 0
+    KERNEL1x8_2 18, 0
+    KERNEL1x8_2 19, 0
+    KERNEL1x8_2 20, 0
+    KERNEL1x8_2 21, 0
+    KERNEL1x8_2 22, 0
+    KERNEL1x8_2 23, 0
+    KERNEL1x8_2 24, 0
+    KERNEL1x8_2 25, 0
+    KERNEL1x8_2 26, 0
+    KERNEL1x8_2 27, 0
+    KERNEL1x8_2 28, 0
+    KERNEL1x8_2 29, 0
+    KERNEL1x8_2 30, 0
+    KERNEL1x8_2 31, 0
+    KERNEL1x8_2 32, 0
+    KERNEL1x8_2 33, 0
+    KERNEL1x8_2 34, 0
+    KERNEL1x8_2 35, 0
+    KERNEL1x8_2 36, 0
+    KERNEL1x8_2 37, 0
+    KERNEL1x8_2 38, 0
+    KERNEL1x8_2 39, 0
+    KERNEL1x8_2 40, 0
+    KERNEL1x8_2 41, 0
+    KERNEL1x8_2 42, 0
+    KERNEL1x8_2 43, 0
+    KERNEL1x8_2 44, 0
+    KERNEL1x8_2 45, 0
+    KERNEL1x8_2 46, 0
+    KERNEL1x8_2 47, 0
+    KERNEL1x8_2 48, 0
+    KERNEL1x8_2 49, 0
+    KERNEL1x8_2 50, 0
+    KERNEL1x8_2 51, 0
+    KERNEL1x8_2 52, 0
+    KERNEL1x8_2 53, 0
+    KERNEL1x8_2 54, 0
+    KERNEL1x8_2 55, 0
+    KERNEL1x8_2 56, 0
+    KERNEL1x8_2 57, 0
+    KERNEL1x8_2 58, 0
+    KERNEL1x8_2 59, 0
+    KERNEL1x8_2 60, 0
+    KERNEL1x8_2 61, 0
+    KERNEL1x8_2 62, 0
+    KERNEL1x8_2 63, 1
+    bdnz    ZGEMM_L1x8_LOOP
+    MY_ALIGN  
+ZGEMM_L1x8_LOOP_END:
+/*----------------------------------------*/   
+    KERNEL1x8_2 0, 1
+    blr
+    MY_ALIGN
+
+
+ZGEMM_1x4_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    MY_ALIGN
+
+
+ZGEMM_L1x4_LOOP:
+/*----------------------------------------*/   
+    KERNEL1x4_2 0, 0
+
+
+ZGEMM_L1x4_K32:
+/*----------------------------------------*/   
+    KERNEL1x4_2 1, 0
+    KERNEL1x4_2 2, 0
+    KERNEL1x4_2 3, 0
+    KERNEL1x4_2 4, 0
+    KERNEL1x4_2 5, 0
+    KERNEL1x4_2 6, 0
+    KERNEL1x4_2 7, 0
+    KERNEL1x4_2 8, 0
+    KERNEL1x4_2 9, 0
+    KERNEL1x4_2 10, 0
+    KERNEL1x4_2 11, 0
+    KERNEL1x4_2 12, 0
+    KERNEL1x4_2 13, 0
+    KERNEL1x4_2 14, 0
+    KERNEL1x4_2 15, 1
+    bdnz    ZGEMM_L1x4_LOOP
+    MY_ALIGN  
+
+
+ZGEMM_L1x4_LOOP_END:
+/*----------------------------------------*/   
+    KERNEL1x4_2 0, 1
+    blr
+    MY_ALIGN
+
+
+ZGEMM_1x2_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    MY_ALIGN
+
+
+ZGEMM_L1x2_LOOP:
+/*----------------------------------------*/   
+    KERNEL1x2_2 0, 0
+
+
+ZGEMM_L1x2_K32:
+/*----------------------------------------*/   
+    KERNEL1x2_2 1, 0
+    KERNEL1x2_2 2, 0
+    KERNEL1x2_2 3, 0
+    KERNEL1x2_2 4, 0
+    KERNEL1x2_2 5, 0
+    KERNEL1x2_2 6, 0
+    KERNEL1x2_2 7, 0
+    KERNEL1x2_2 8, 0
+    KERNEL1x2_2 9, 0
+    KERNEL1x2_2 10, 0
+    KERNEL1x2_2 11, 0
+    KERNEL1x2_2 12, 0
+    KERNEL1x2_2 13, 0
+    KERNEL1x2_2 14, 0
+    KERNEL1x2_2 15, 1
+    bdnz    ZGEMM_L1x2_LOOP
+    MY_ALIGN  
+
+
+ZGEMM_L1x2_LOOP_END:
+/*----------------------------------------*/   
+    KERNEL1x2_2 0, 1
+    blr
+    MY_ALIGN
+
+
+ZGEMM_1x1_LMAIN_SUB:
+/*----------------------------------------*/   
+    mtctr   T8
+    LOAD1x1_2  
+    MY_ALIGN
+
+
+ZGEMM_L1x1_LOOP:
+/*----------------------------------------*/   
+    KERNEL1x1_L2 32, 32, 0, 0
+
+
+ZGEMM_L1x1_K32:
+/*----------------------------------------*/   
+    KERNEL1x1_L2 32, 32, 1, 0  
+    KERNEL1x1_L2 32, 32, 2, 0
+    KERNEL1x1_L2 32, 32, 3, 0  
+    KERNEL1x1_L2 32, 32, 4, 0
+    KERNEL1x1_L2 32, 32, 5, 0 
+    KERNEL1x1_L2 32, 32, 6, 0
+    KERNEL1x1_L2 32, 32, 7, 0
+    KERNEL1x1_L2 32, 32, 8, 0
+    KERNEL1x1_L2 32, 32, 9, 0  
+    KERNEL1x1_L2 32, 32, 10, 0
+    KERNEL1x1_L2 32, 32, 11, 0  
+    KERNEL1x1_L2 32, 32, 12, 0
+    KERNEL1x1_L2 32, 32, 13, 0 
+    KERNEL1x1_L2 32, 32, 14, 0
+    KERNEL1x1_L2 32, 32, 15, 1   
+    bdnz    ZGEMM_L1x1_LOOP
+    MY_ALIGN  
+
+
+ZGEMM_L1x1_LOOP_END:
+/*----------------------------------------*/   
+    END1x1_2 
+    blr
+    MY_ALIGN
+
+
+/*----------------------N1 BEGINS---------*/
+ZGEMM_L1:
+/*----------------------------------------*/   
+    andi.   T1, N, 1
+    ble   ZGEMM_L1_END
+		
+ZGEMM_L1_BEGIN:
+/*----------------------------------------*/   
+    mr    CO, C
+   
+    add     T2,C,LDC    
+    mr    AO, A  
+    add   C, C, T1
+#if defined(TRMMKERNEL) && defined(LEFT)   
+    mr TEMP_REG, OFFSET  /*off = offset;*/
+#endif     
+    srawi.    I, M, 3
+    ble   ZGEMM_L1x8_END
+    dcbt    CO,r0  /*just prefetch*/
+    dcbt    T2,r0    
+
+
+ZGEMM_L1x8_BEGIN:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 8, 1
+#else    
+    mr    BO, B  
+    dcbt    B, r0  
+#endif     
+    dcbt    AO, r0
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 8, 1
+    mr T1, T6
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512   
+    srawi.   T8, T1, 7 /**(T11-2) % 128x */
+#else   
+    mr T1, K
+/* TEMPS FOR PREFETCH */   
+    li T2, 1024
+    li T3, 1024+512
+    addi T1,T1, -2
+/* TEMPS FOR PREFETCH */     
+    li T4, 2048
+    li T5, 2048+512 
+    srawi.   T8, T1, 7 /**(K-2) % 128x */
+#endif   
+    KERNEL1x8_ZERO_AND_PRIME_MMA
+    ble   ZGEMM_L1x8_SUB0
+    bl ZGEMM_L1x8_LMAIN_SUB
+    andi.   L, T1, 127
+    ble   ZGEMM_L1x8_SAVE
+    b   ZGEMM_L1x8_SUB2
+
+
+ZGEMM_L1x8_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 255
+    cmpwi   T6, 129
+#else   
+    andi.   L, K, 255
+    cmpwi   K, 129
+#endif       
+    li T8, 1
+    bne CMP1x8_128K
+    LOAD_END_1x8 -128, -16
+    mtctr   T8    
+    bl ZGEMM_L1x8_K128   
+    b ZGEMM_L1x8_SAVE  
+    CMP1x8_128K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 128
+#else    
+    cmpwi   K, 128
+#endif        
+    bne ZGEMM_L1x8_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -32
+    addi AO,AO, -256   
+    bl ZGEMM_L1x8_K128   
+    b ZGEMM_L1x8_SAVE 
+    MY_ALIGN
+
+
+ZGEMM_L1x8_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 64
+    ble ZGEMM_L1x8_SUB2_32
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_2 0, 0
+    KERNEL1x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL1x8_2 2, 0
+    KERNEL1x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_2 4, 0
+    KERNEL1x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL1x8_2 6, 0
+    KERNEL1x8_2 7, 0
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL1x8_2 8, 0
+    KERNEL1x8_2 9, 0
+    KERNEL1x8_2 10, 0
+    KERNEL1x8_2 11, 0
+    dcbt    BO, T4
+    KERNEL1x8_2 12, 0
+    KERNEL1x8_2 13, 0
+    KERNEL1x8_2 14, 0
+    KERNEL1x8_2 15, 0
+    KERNEL1x8_2 16, 0
+    KERNEL1x8_2 17, 0
+    KERNEL1x8_2 18, 0
+    KERNEL1x8_2 19, 0
+    KERNEL1x8_2 20, 0
+    KERNEL1x8_2 21, 0
+    KERNEL1x8_2 22, 0
+    KERNEL1x8_2 23, 0
+    KERNEL1x8_2 24, 0
+    KERNEL1x8_2 25, 0
+    KERNEL1x8_2 26, 0
+    KERNEL1x8_2 27, 0
+    KERNEL1x8_2 28, 0
+    KERNEL1x8_2 29, 0
+    KERNEL1x8_2 30, 0
+    KERNEL1x8_2 31, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x8_SUB2_32:
+/*----------------------------------------*/   
+    andi.      T1,L, 32
+    ble ZGEMM_L1x8_SUB2_16    
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_2 0, 0
+    KERNEL1x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL1x8_2 2, 0
+    KERNEL1x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_2 4, 0
+    KERNEL1x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL1x8_2 6, 0
+    KERNEL1x8_2 7, 0
+    dcbt    AO, T5  
+    dcbt    BO, T3
+    KERNEL1x8_2 8, 0
+    KERNEL1x8_2 9, 0
+    KERNEL1x8_2 10, 0
+    KERNEL1x8_2 11, 0
+    dcbt    BO, T4
+    KERNEL1x8_2 12, 0
+    KERNEL1x8_2 13, 0
+    KERNEL1x8_2 14, 0
+    KERNEL1x8_2 15, 1
+    MY_ALIGN 
+
+
+ZGEMM_L1x8_SUB2_16:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L1x8_SUB2_8
+    dcbt    AO, PRE
+    dcbt    BO, PRE
+    KERNEL1x8_2 0, 0
+    KERNEL1x8_2 1, 0
+    dcbt    AO, T2  
+    KERNEL1x8_2 2, 0
+    KERNEL1x8_2 3, 0
+    dcbt    AO, T3
+    dcbt    BO, T2
+    KERNEL1x8_2 4, 0
+    KERNEL1x8_2 5, 0
+    dcbt    AO, T4  
+    KERNEL1x8_2 6, 0
+    KERNEL1x8_2 7, 1
+    MY_ALIGN    
+
+
+ZGEMM_L1x8_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L1x8_SUB2_4
+    KERNEL1x8_2 0, 0
+    KERNEL1x8_2 1, 0
+    KERNEL1x8_2 2, 0
+    KERNEL1x8_2 3, 1
+    MY_ALIGN   
+
+
+ZGEMM_L1x8_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L1x8_SUB2_2
+    KERNEL1x8_2 0, 0
+    KERNEL1x8_2 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x8_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L1x8_SUB2_1
+    KERNEL1x8_2 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L1x8_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L1x8_SAVE 
+    LOAD_END_1x8 128, 16
+
+
+ZGEMM_L1x8_SAVE:
+/*----------------------------------------*/   
+    addic.    I, I, -1
+    KERNEL1x8_UNPRIME_MMA
+    SAVE1x8
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 8, 1
+#endif     
+    bgt   ZGEMM_L1x8_BEGIN
+    andi.   T2, M, 7
+    ble   ZGEMM_L1x1_END
+    andi.   T1, M, 4
+    ble   ZGEMM_L1x4_END
+    b   ZGEMM_L1x4_BEGIN
+    MY_ALIGN 
+
+
+ZGEMM_L1x8_END:
+/*----------------------------------------*/   
+
+
+ZGEMM_L1x4_BEGIN:
+/*----------------------------------------*/   
+    andi.   T2, M, 7
+    ble   ZGEMM_L1x1_END
+    andi.   T1, M, 4
+    ble   ZGEMM_L1x4_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 4, 1
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 4, 1
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T11-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    KERNEL1x4_ZERO_AND_PRIME_MMA
+    ble   ZGEMM_L1x4_SUB0 
+    bl ZGEMM_1x4_LMAIN_SUB
+    andi.   L, T1, 31
+    ble   ZGEMM_L1x4_SAVE
+    b   ZGEMM_L1x4_SUB2
+
+
+ZGEMM_L1x4_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 63
+    cmpwi   T6, 33
+#else   
+    andi.   L, K, 63
+    cmpwi   K, 33
+#endif       
+    li T8, 1
+    bne CMP1x4_32K
+    LOAD_END_1x4 -64, -16 
+    mtctr   T8    
+    bl ZGEMM_L1x4_K32   
+    b ZGEMM_L1x4_SAVE  
+    CMP1x4_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 32
+#else    
+    cmpwi   K, 32
+#endif        
+    bne ZGEMM_L1x4_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -32
+    addi AO,AO, -128   
+    bl ZGEMM_L1x4_K32   
+    b ZGEMM_L1x4_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+ZGEMM_L1x4_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L1x4_SUB2_8
+    KERNEL1x4_2 0, 0
+    KERNEL1x4_2 1, 0
+    KERNEL1x4_2 2, 0
+    KERNEL1x4_2 3, 0
+    KERNEL1x4_2 4, 0
+    KERNEL1x4_2 5, 0
+    KERNEL1x4_2 6, 0
+    KERNEL1x4_2 7, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x4_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L1x4_SUB2_4
+    KERNEL1x4_2 0, 0
+    KERNEL1x4_2 1, 0
+    KERNEL1x4_2 2, 0
+    KERNEL1x4_2 3, 1
+    MY_ALIGN  
+
+
+ZGEMM_L1x4_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L1x4_SUB2_2
+    KERNEL1x4_2 0, 0
+    KERNEL1x4_2 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x4_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L1x4_SUB2_1
+    KERNEL1x4_2 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L1x4_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L1x4_SAVE 
+    LOAD_END_1x4 64,16
+
+
+
+ZGEMM_L1x4_SAVE:
+/*----------------------------------------*/   
+    KERNEL1x4_UNPRIME_MMA
+    SAVE1x4
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 4, 1
+#endif     
+
+
+ZGEMM_L1x4_END:
+/*----------------------------------------*/   
+
+
+ZGEMM_L1x2_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M, 2
+    ble   ZGEMM_L1x2_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 2, 1
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 2, 1
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T11-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    KERNEL1x2_ZERO_AND_PRIME_MMA
+    ble   ZGEMM_L1x2_SUB0 
+    bl ZGEMM_1x2_LMAIN_SUB
+    andi.   L, T1, 31
+    ble   ZGEMM_L1x2_SAVE
+    b   ZGEMM_L1x2_SUB2
+
+
+ZGEMM_L1x2_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 63
+    cmpwi   T6, 33
+#else   
+    andi.   L, K, 63
+    cmpwi   K, 33
+#endif       
+    li T8, 1
+    bne CMP1x2_32K
+    LOAD_END_1x2 -32, -16 
+    mtctr   T8    
+    bl ZGEMM_L1x2_K32   
+    b ZGEMM_L1x2_SAVE  
+    CMP1x2_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 32
+#else    
+    cmpwi   K, 32
+#endif        
+    bne ZGEMM_L1x2_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -32
+    addi AO,AO, -64   
+    bl ZGEMM_L1x2_K32   
+    b ZGEMM_L1x2_SAVE 
+    MY_ALIGN 
+    MY_ALIGN 
+
+
+ZGEMM_L1x2_SUB2:
+/*----------------------------------------*/   
+    andi.      T1,L, 16
+    ble ZGEMM_L1x2_SUB2_8
+    KERNEL1x2_2 0, 0
+    KERNEL1x2_2 1, 0
+    KERNEL1x2_2 2, 0
+    KERNEL1x2_2 3, 0
+    KERNEL1x2_2 4, 0
+    KERNEL1x2_2 5, 0
+    KERNEL1x2_2 6, 0
+    KERNEL1x2_2 7, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x2_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1,L, 8
+    ble ZGEMM_L1x2_SUB2_4
+    KERNEL1x2_2 0, 0
+    KERNEL1x2_2 1, 0
+    KERNEL1x2_2 2, 0
+    KERNEL1x2_2 3, 1
+    MY_ALIGN  
+
+
+ZGEMM_L1x2_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L1x2_SUB2_2
+    KERNEL1x2_2 0, 0
+    KERNEL1x2_2 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x2_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L1x2_SUB2_1
+    KERNEL1x2_2 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L1x2_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L1x2_SAVE 
+    LOAD_END_1x2 32,16
+
+
+ZGEMM_L1x2_SAVE:
+/*----------------------------------------*/   
+    KERNEL1x2_UNPRIME_MMA
+    SAVE1x2
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 2, 1
+#endif     
+
+
+ZGEMM_L1x2_END:
+/*----------------------------------------*/   
+
+
+ZGEMM_L1x1_BEGIN:
+/*----------------------------------------*/   
+    andi.   T1, M, 1
+    ble   ZGEMM_L1x1_END
+#if defined(TRMMKERNEL)   
+    REFRESH_POINTERS  AO, BO,TEMP_REG, B, 1, 1
+#else    
+    mr    BO, B   
+#endif        
+#if defined(TRMMKERNEL)   
+    REFRESH_TEMP_BK T6,K,TEMP_REG, 1, 1
+    mr T1, T6 
+    addi T1,T1, -2 
+    srawi.   T8, T1, 5 /**(T11-2) % 32x */
+#else   
+    mr T1, K 
+    addi T1,T1, -2
+    srawi.   T8, T1, 5 /**(K-2) % 32x */
+#endif     
+    ZERO1x1
+    ble   ZGEMM_L1x1_SUB0 
+    bl ZGEMM_1x1_LMAIN_SUB
+    andi.   L, T1, 31
+    ble   ZGEMM_L1x1_SAVE
+    b   ZGEMM_L1x1_SUB2
+
+
+ZGEMM_L1x1_SUB0:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)   
+    andi.   L, T6, 63
+    cmpwi   T6, 33
+#else   
+    andi.   L, K, 63
+    cmpwi   K, 33
+#endif       
+    li T8, 1
+    bne CMP1x1_32K
+    addi BO, BO, -16
+    addi AO,AO, -16  
+    LOAD1x1O 16, 16 
+    END1x1_WITHOUT_ADD   
+    LOAD1x1_2O  32, 32  
+    mtctr   T8    
+    bl ZGEMM_L1x1_K32   
+    b ZGEMM_L1x1_SAVE  
+    CMP1x1_32K:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL)    
+    cmpwi   T6, 32
+#else    
+    cmpwi   K, 32
+#endif        
+    bne ZGEMM_L1x1_SUB2 
+    MY_ALIGN   
+    mtctr   T8
+    addi BO, BO, -32
+    addi AO,AO, -32   
+    LOAD1x1_2O 32, 32
+    bl ZGEMM_L1x1_K32   
+    b ZGEMM_L1x1_SAVE 
+    MY_ALIGN 
+
+
+ZGEMM_L1x1_SUB2:
+/*----------------------------------------*/   
+    andi.      T1, L, 16
+    ble ZGEMM_L1x1_SUB2_8
+    LOAD1x1_2
+    KERNEL1x1_L2 32, 32, 0, 0
+    KERNEL1x1_L2 32, 32, 1, 0  
+    KERNEL1x1_L2 32, 32, 2, 0
+    KERNEL1x1_L2 32, 32, 3, 0  
+    KERNEL1x1_L2 32, 32, 4, 0
+    KERNEL1x1_L2 32, 32, 5, 0 
+    KERNEL1x1_L2 32, 32, 6, 0
+    KERNEL1x1_E2 32, 32, 7, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x1_SUB2_8:
+/*----------------------------------------*/   
+    andi.      T1, L, 8
+    ble ZGEMM_L1x1_SUB2_4
+    LOAD1x1_2
+    KERNEL1x1_L2 32, 32, 0, 0
+    KERNEL1x1_L2 32, 32, 1, 0  
+    KERNEL1x1_L2 32, 32, 2, 0
+    KERNEL1x1_E2 32, 32, 3, 1  
+    MY_ALIGN  
+
+
+ZGEMM_L1x1_SUB2_4:
+/*----------------------------------------*/   
+    andi.      T1,L, 4
+    ble ZGEMM_L1x1_SUB2_2
+    LOAD1x1_2
+    KERNEL1x1_L2  32, 32, 0, 0
+    KERNEL1x1_E2  32, 32, 1, 1
+    MY_ALIGN
+
+
+ZGEMM_L1x1_SUB2_2:
+/*----------------------------------------*/   
+    andi.      T1,L, 2
+    ble ZGEMM_L1x1_SUB2_1
+    LOAD1x1_2
+    KERNEL1x1_E2  32, 32, 0, 1
+    MY_ALIGN    
+
+
+ZGEMM_L1x1_SUB2_1:
+/*----------------------------------------*/   
+    andi.      T1,L, 1
+    ble ZGEMM_L1x1_SAVE 
+    KERNEL1x1
+
+
+ZGEMM_L1x1_SAVE:
+/*----------------------------------------*/   
+    SAVE1x1
+#if defined(TRMMKERNEL)    
+    REFRESH_AFTER_SAVE T6,K,TEMP_REG, BO,AO, 1, 1
+#endif   
+
+
+ZGEMM_L1x1_END:
+/*----------------------------------------*/   
+#if defined(TRMMKERNEL) && !defined(LEFT)   
+    addi TEMP_REG, TEMP_REG, 1
+#endif   
+
+
+ZGEMM_L1_END:
+/*----------------------------------------*/   
diff --git a/kernel/power/zgemm_macros_power10.S b/kernel/power/zgemm_macros_power10.S
new file mode 100644
index 000000000..42f9c5ad4
--- /dev/null
+++ b/kernel/power/zgemm_macros_power10.S
@@ -0,0 +1,1138 @@
+/***************************************************************************
+Copyright (c) 2013-2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define unit_size 16
+#define DISP32(ind,disp) (ind*unit_size*32+disp)
+#define DISP16(ind,disp) (ind*unit_size*16+disp)
+#define DISP8(ind,disp) (ind*unit_size*8+disp)
+#define DISP4(ind,disp) (ind*unit_size*4+disp)
+#define DISP2(ind,disp) (ind*unit_size*2+disp)
+#define DISP1(ind,disp) (ind*unit_size+disp)
+#define DISPX(disp)  (disp)
+/*	HELPERS FOR SAVE	*/
+/* {r0,i0} and {r1,i1} into  {r0,r1} {i0,i1} */
+
+
+.macro LOAD_COUPLE_AS_RR_II  VS_OUT1,VS_OUT2,VS_TEMP1,VS_TEMP2,REG,LOFFSET 
+#ifndef TRMMKERNEL 
+  lxv	\VS_TEMP1,	DISPX(\LOFFSET)(\REG)
+  lxv	\VS_TEMP2,	DISPX(\LOFFSET+16)(\REG)
+  xxmrgld  \VS_OUT1,\VS_TEMP1,\VS_TEMP2
+  xxmrghd  \VS_OUT2,\VS_TEMP1,\VS_TEMP2	
+#endif	
+.endm
+/*from 2 result {a0r*br,a0i*bi} and {a1r*br,a1i*bi} pack into {a0r*br,a1r*br} and {a0i*bi,a1i*bi}*/
+
+
+.macro RESULT_INTO_REALREAL_IMAGEIMAGE VSIN1,VSIN2,VSOUT1,VSOUT2
+	xxmrgld	\VSOUT1, \VSIN1,\VSIN2 /*  real*real from 2 results*/
+	xxmrghd	\VSOUT2, \VSIN1,\VSIN2 /*  imag*imag from 2 results*/
+.endm 
+/*from 2 result {a0r*bi,a0i*br} and {a1r*bi,a1i*br} pack into {a0r*bi,a1r*bi} and {a0i*br,a1i*br}*/
+
+
+.macro RESULT_INTO_REALIMAG_IMAGREAL VSIN1,VSIN2,VSOUT1,VSOUT2 
+	xxmrgld	\VSOUT1, \VSIN1,\VSIN2 /*  real*imag */
+	xxmrghd	\VSOUT2, \VSIN1,\VSIN2 /*  imag*real*/
+.endm
+/* {a0r*br op a0i*bi ,a1r*br op a1i*bi} ~ {r0,r1}; {a0r*bi op a0i*br ,a1r*bi op a1i*br} ~ {i0,i1}*/
+
+
+.macro  AGGREGATE_REALS_IMAGES  VSINR_OUT1,VSINR,VSINI_OUT2,VSINI
+#if	defined(NN) || defined(NT) || defined(TN) || defined(TT) 
+	xvsubdp  \VSINR_OUT1,\VSINR_OUT1,\VSINR
+	xvadddp  \VSINI_OUT2,\VSINI_OUT2,\VSINI  
+#elif  defined(CN) || defined(CT) || defined(RN) || defined(RT) 
+	xvadddp  \VSINR_OUT1,\VSINR_OUT1,\VSINR
+	xvsubdp  \VSINI_OUT2,\VSINI_OUT2,\VSINI 
+#elif  defined(NC) || defined(TC) || defined(NR) || defined(TR) 
+	xvadddp  \VSINR_OUT1,\VSINR_OUT1,\VSINR
+	xvsubdp  \VSINI_OUT2,\VSINI,\VSINI_OUT2  
+#else	// CC || CR || RC || RR 
+    /*we will assume {-alpha_r,-alpha_i} for this case */
+    /*i1i2-r1r2 so we will negate alpha real instead to fix sign*/
+	xvsubdp  \VSINR_OUT1,\VSINR,\VSINR_OUT1
+    /*we will negate alpha image instead  instead to fix sign*/
+	xvadddp  \VSINI_OUT2,\VSINI_OUT2,\VSINI 
+#endif
+.endm 
+/* {i0,i1} * {alpha_i,alpha_i} - VSOUT1 ;VSOUT2 + {r0,r1}*{alpha_i,alpha_i} */
+
+
+.macro MULT_APLHA_PART1  VSINRR,VSINII,VSOUT1,VSOUT2
+#ifndef TRMMKERNEL  
+	xvmsubadp \VSOUT1,\VSINII, alpha_i
+	xvmaddadp  \VSOUT2,\VSINRR, alpha_i
+#else 
+	xvmuldp \VSOUT1,\VSINII, alpha_i 
+	xvmuldp  \VSOUT2,\VSINRR, alpha_i
+#endif 
+.endm
+/*   {r0,r1} * {alpha_r,alpha_r} -  VSOUT1 ;VSOUT2 + {i0,i1} * {alpha_r,alpha_r} */
+
+
+.macro MULT_APLHA_PART2  VSINRR,VSINII,VSOUT1,VSOUT2 
+	xvmsubadp  \VSOUT1,\VSINRR, alpha_r
+	xvmaddadp \VSOUT2,\VSINII, alpha_r
+.endm
+/* unpack to store 2{r,r} {i,i} into  {r,i} {r,i} (big endian because of stxv) */
+
+
+.macro UNPACK_FOR_STORE VSIN1,VSIN2,VSOUT1,VSOUT2 
+	xxmrghd  \VSOUT1,\VSIN2,\VSIN1
+	xxmrgld  \VSOUT2,\VSIN2,\VSIN1
+.endm
+
+
+.macro STORE_COUPLE REG,LOFFSET,VSIN1,VSIN2
+	stxv	\VSIN1,	DISPX(\LOFFSET)(\REG)
+	stxv	\VSIN2,	DISPX(\LOFFSET+16)(\REG)
+.endm
+
+
+.macro SAVE8 VSRes1,VSRes2,VSRes3,VSRes4,VSRes5,VSRes6,VSRes7,VSRes8,VSRes9,VSRes10,VSRes11,VSRes12,VSRes13,VSRes14,VSRes15,VSRes16,BASE_REG,LOFFSET
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes1,\VSRes3,vs34,vs35
+  LOAD_COUPLE_AS_RR_II	vs46,vs47,vs50,vs51,\BASE_REG,\LOFFSET
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes2,\VSRes4,vs36,vs37
+  LOAD_COUPLE_AS_RR_II	vs48,vs49,vs52,vs53,\BASE_REG,(\LOFFSET+32)
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes5,\VSRes7,vs38,vs39
+  LOAD_COUPLE_AS_RR_II	vs56,vs57,vs50,vs51,\BASE_REG,(\LOFFSET +64)
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes6,\VSRes8,vs40,vs41 
+  LOAD_COUPLE_AS_RR_II	vs58,vs59,vs52,vs53,\BASE_REG,(\LOFFSET+96)
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes9,\VSRes11,vs42,vs43
+  AGGREGATE_REALS_IMAGES	vs34,vs35,vs36,vs37
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes10,\VSRes12,vs44,vs45 
+  AGGREGATE_REALS_IMAGES	vs38,vs39,vs40,vs41  
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes13,\VSRes15,\VSRes1,\VSRes2
+  MULT_APLHA_PART1	vs34,vs36, vs46,vs47
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes14,\VSRes16,\VSRes3,\VSRes4
+  MULT_APLHA_PART1	vs38,vs40,vs48,vs49
+  MULT_APLHA_PART2  vs34,vs36,vs46,vs47 
+  AGGREGATE_REALS_IMAGES	vs42,vs43,vs44,vs45
+  MULT_APLHA_PART2	vs38,vs40,vs48,vs49
+  AGGREGATE_REALS_IMAGES	\VSRes1,\VSRes2,\VSRes3,\VSRes4	
+  UNPACK_FOR_STORE	vs46,vs47,vs39,vs41
+  MULT_APLHA_PART1	vs42,vs44, vs56,vs57
+  UNPACK_FOR_STORE	vs48,vs49,vs35,vs37 
+  MULT_APLHA_PART1	\VSRes1,\VSRes3, vs58,vs59
+  STORE_COUPLE	\BASE_REG,\LOFFSET,vs39,vs41
+  MULT_APLHA_PART2	vs42,vs44,vs56,vs57
+  STORE_COUPLE	\BASE_REG,(\LOFFSET+32),vs35,vs37 
+  MULT_APLHA_PART2	\VSRes1,\VSRes3, vs58,vs59
+  UNPACK_FOR_STORE	vs56,vs57,vs42,vs44
+  UNPACK_FOR_STORE	vs58,vs59,\VSRes1,\VSRes3
+  STORE_COUPLE	\BASE_REG,(\LOFFSET +64),vs42,vs44
+  STORE_COUPLE	\BASE_REG,(\LOFFSET+96),\VSRes1,\VSRes3
+.endm
+
+
+.macro SAVE4  VSRes1,VSRes2,VSRes3,VSRes4,VSRes5,VSRes6,VSRes7,VSRes8,BASE_REG,LOFFSET
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes1,\VSRes3,vs34,vs35
+  LOAD_COUPLE_AS_RR_II	vs46,vs47,vs50,vs51,\BASE_REG,\LOFFSET
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes2,\VSRes4,vs36,vs37
+  LOAD_COUPLE_AS_RR_II	vs48,vs49,vs52,vs53,\BASE_REG,(\LOFFSET+32)
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes5,\VSRes7,vs38,vs39
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes6,\VSRes8,vs40,vs41 
+  AGGREGATE_REALS_IMAGES	vs34,vs35,vs36,vs37	
+  AGGREGATE_REALS_IMAGES	vs38,vs39,vs40,vs41  
+  MULT_APLHA_PART1	vs34,vs36, vs46,vs47
+  MULT_APLHA_PART1	vs38,vs40, vs48,vs49
+  MULT_APLHA_PART2	vs34,vs36, vs46,vs47 
+  MULT_APLHA_PART2	vs38,vs40,vs48,vs49
+  UNPACK_FOR_STORE	vs46,vs47,vs39,vs41
+  UNPACK_FOR_STORE	vs48,vs49,vs35,vs37
+  STORE_COUPLE	\BASE_REG,\LOFFSET,vs39,vs41
+  STORE_COUPLE	\BASE_REG,(\LOFFSET+32),vs35,vs37
+.endm
+
+
+.macro SAVE2  VSRes1,VSRes2,VSRes3,VSRes4,BASE_REG,LOFFSET
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes1,\VSRes3,vs34,vs35
+  LOAD_COUPLE_AS_RR_II	vs46,vs47,vs50,vs51,\BASE_REG,\LOFFSET
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes2,\VSRes4,vs36,vs37	
+  AGGREGATE_REALS_IMAGES	vs34,vs35,vs36,vs37	
+  MULT_APLHA_PART1	vs34,vs36, vs46,vs47	
+  MULT_APLHA_PART2	vs34,vs36, vs46,vs47  
+  UNPACK_FOR_STORE	vs46,vs47,vs39,vs41	
+  STORE_COUPLE	\BASE_REG,\LOFFSET,vs39,vs41  
+.endm
+
+
+.macro SAVE1  VSRes1,VSRes2,BASE_REG,LOFFSET
+  RESULT_INTO_REALREAL_IMAGEIMAGE \VSRes1,\VSRes1,vs34,vs35
+#ifndef TRMMKERNEL 
+  lxv	vs50,	(\LOFFSET)(\BASE_REG) 
+  xxmrgld  vs46,vs50,vs50
+  xxmrghd  vs47,vs50,vs50	
+#endif	
+  RESULT_INTO_REALIMAG_IMAGREAL	\VSRes2,\VSRes2,vs36,vs37	
+  AGGREGATE_REALS_IMAGES	vs34,vs35,vs36,vs37	
+  MULT_APLHA_PART1	vs34,vs36, vs46,vs47	
+  MULT_APLHA_PART2	vs34,vs36, vs46,vs47  
+  UNPACK_FOR_STORE	vs46,vs47,vs39,vs41 
+  xxmrghd  vs39,vs47,vs46	
+  stxv	vs39,	(\LOFFSET)(\BASE_REG) 
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=2 and M=8
+**********************************************************************************************/
+
+.macro  KERNEL2x8_ZERO_AND_PRIME_MMA
+        /* zero out and prime the MMA accumulators */
+        xxsetaccz 0
+        xxsetaccz 1
+        xxsetaccz 2
+        xxsetaccz 3
+        xxsetaccz 4
+        xxsetaccz 5
+        xxsetaccz 6
+        xxsetaccz 7
+.endm
+
+
+.macro KERNEL2x8_PRELOAD
+	lxvp	vs32,	 0(AO)	// load real,imag from A
+	lxvp	vs34,	32(AO)	// load real,imag from A
+	lxvp	vs36,	64(AO)	// load real,imag from A
+	lxvp	vs38,	96(AO)	// load real,imag from A
+	lxvp	vs48,	 0(BO)	// load real imag from B
+.endm
+
+
+.macro KERNEL2x8_2	Index, IsLast
+	lxvp	vs40,	DISP16(\Index,128)(AO)	// load real,imag from A
+	lxvp	vs42,	DISP16(\Index,160)(AO)	// load real,imag from A
+	lxvp	vs44,	DISP16(\Index,192)(AO)	// load real,imag from A
+	lxvp	vs46,	DISP16(\Index,224)(AO)	// load real,imag from A
+ 	lxvp	vs50,	DISP4(\Index,  32)(BO)	// load real,imag from B
+	xvf64gerpp	0,	vs32,	vs49
+	xvf64gerpp	1,	vs34,	vs49
+	xvf64gerpp	2,	vs36,	vs49
+	xvf64gerpp	3,	vs38,	vs49
+	xvf64gerpp	4,	vs32,	vs48
+	xvf64gerpp	5,	vs34,	vs48
+	xvf64gerpp	6,	vs36,	vs48
+	xvf64gerpp	7,	vs38,	vs48
+	lxvp	vs32,	DISP16(\Index, 256)(AO)	// load real,imag from A
+	lxvp	vs34,	DISP16(\Index, 288)(AO)	// load real,imag from A
+	lxvp	vs36,	DISP16(\Index, 320)(AO)	// load real,imag from A
+	lxvp	vs38,	DISP16(\Index, 352)(AO)	// load real,imag from A
+	lxvp	vs48,	DISP4(\Index,  64)(BO)	// load real imag from B
+	xvf64gerpp	0,	vs40,	vs51
+	xvf64gerpp	1,	vs42,	vs51
+	xvf64gerpp	2,	vs44,	vs51
+	xvf64gerpp	3,	vs46,	vs51
+	xvf64gerpp	4,	vs40,	vs50
+	xvf64gerpp	5,	vs42,	vs50
+	xvf64gerpp	6,	vs44,	vs50
+	xvf64gerpp	7,	vs46,	vs50
+.if \IsLast==1
+	addi	AO, AO,  DISP16(\Index,256)
+	addi	BO, BO,  DISP4(\Index,64)
+.endif 
+.endm
+
+
+.macro LOAD_END_2x8  OffsetA,OffsetB
+	xvf64gerpp	0,	vs32,	vs49
+	xvf64gerpp	1,	vs34,	vs49
+	xvf64gerpp	2,	vs36,	vs49
+	xvf64gerpp	3,	vs38,	vs49
+	xvf64gerpp	4,	vs32,	vs48
+	xvf64gerpp	5,	vs34,	vs48
+	xvf64gerpp	6,	vs36,	vs48
+	xvf64gerpp	7,	vs38,	vs48
+	addi	BO, BO, \OffsetB
+	addi	AO, AO, \OffsetA
+.endm
+
+
+.macro  KERNEL2x8_UNPRIME_MMA
+        /* "unprime" MMA accumulators */
+        xxmfacc 0
+        xxmfacc 1
+        xxmfacc 2
+        xxmfacc 3
+        xxmfacc 4
+        xxmfacc 5
+        xxmfacc 6
+        xxmfacc 7
+.endm
+
+
+.macro SAVE2x8
+	add	T1, CO ,LDC 
+        xxpermdi vs32, vs0, vs1, 0b01
+        xxpermdi vs33, vs0, vs1, 0b10
+        xxpermdi vs34, vs2, vs3, 0b01
+        xxpermdi vs35, vs2, vs3, 0b10
+        xxpermdi vs36, vs4, vs5, 0b01
+        xxpermdi vs37, vs4, vs5, 0b10
+        xxpermdi vs38, vs6, vs7, 0b01
+        xxpermdi vs39, vs6, vs7, 0b10
+        xxpermdi vs40, vs8, vs9, 0b01
+        xxpermdi vs41, vs8, vs9, 0b10
+        xxpermdi vs42, vs10, vs11, 0b01
+        xxpermdi vs43, vs10, vs11, 0b10
+        xxpermdi vs44, vs12, vs13, 0b01
+        xxpermdi vs45, vs12, vs13, 0b10
+        xxpermdi vs46, vs14, vs15, 0b01
+        xxpermdi vs47, vs14, vs15, 0b10
+
+	xxlor vs2, vs32, vs32
+	xxlor vs3, vs33, vs33
+	xxlor vs0, vs34, vs34
+	xxlor vs1, vs35, vs35
+	xxlor vs6, vs36, vs36
+	xxlor vs7, vs37, vs37
+	xxlor vs4, vs38, vs38
+	xxlor vs5, vs39, vs39
+	xxlor vs10, vs40, vs40
+	xxlor vs11, vs41, vs41
+	xxlor vs8, vs42, vs42
+	xxlor vs9, vs43, vs43
+	xxlor vs14, vs44, vs44
+	xxlor vs15, vs45, vs45
+	xxlor vs12, vs46, vs46
+	xxlor vs13, vs47, vs47
+
+        xxpermdi vs32, vs16, vs17, 0b01
+        xxpermdi vs33, vs16, vs17, 0b10
+        xxpermdi vs34, vs18, vs19, 0b01
+        xxpermdi vs35, vs18, vs19, 0b10
+        xxpermdi vs36, vs20, vs21, 0b01
+        xxpermdi vs37, vs20, vs21, 0b10
+        xxpermdi vs38, vs22, vs23, 0b01
+        xxpermdi vs39, vs22, vs23, 0b10
+        xxpermdi vs40, vs24, vs25, 0b01
+        xxpermdi vs41, vs24, vs25, 0b10
+        xxpermdi vs42, vs26, vs27, 0b01
+        xxpermdi vs43, vs26, vs27, 0b10
+        xxpermdi vs44, vs28, vs29, 0b01
+        xxpermdi vs45, vs28, vs29, 0b10
+        xxpermdi vs46, vs30, vs31, 0b01
+        xxpermdi vs47, vs30, vs31, 0b10
+       
+	xxlor vs18, vs32, vs32
+	xxlor vs19, vs33, vs33
+	xxlor vs16, vs34, vs34
+	xxlor vs17, vs35, vs35
+	xxlor vs22, vs36, vs36
+	xxlor vs23, vs37, vs37
+	xxlor vs20, vs38, vs38
+	xxlor vs21, vs39, vs39
+	xxlor vs26, vs40, vs40
+	xxlor vs27, vs41, vs41
+	xxlor vs24, vs42, vs42
+	xxlor vs25, vs43, vs43
+	xxlor vs30, vs44, vs44
+	xxlor vs31, vs45, vs45
+	xxlor vs28, vs46, vs46
+	xxlor vs29, vs47, vs47
+
+	SAVE8  vs0,vs1,vs2,vs3,vs4,vs5,vs6,vs7,vs8,vs9,vs10,vs11,vs12,vs13,vs14,vs15,CO,0
+	SAVE8  vs16,vs17,vs18,vs19,vs20,vs21,vs22,vs23,vs24,vs25,vs26,vs27,vs28,vs29,vs30,vs31,T1,0  
+	addi	CO, CO, 128
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=2 and M=4
+**********************************************************************************************/
+
+.macro  KERNEL2x4_ZERO_AND_PRIME_MMA
+        /* zero out and prime the MMA accumulators */
+        xxsetaccz 0
+        xxsetaccz 1
+        xxsetaccz 2
+        xxsetaccz 3
+.endm
+
+
+.macro KERNEL2x4_PRELOAD
+	lxvp	vs32,	 0(AO)	// load real,imag from A
+	lxvp	vs34,	32(AO)	// load real,imag from A
+	lxvp	vs48,	 0(BO)	// load real imag from B
+.endm
+
+
+.macro KERNEL2x4_2 Index, IsLast
+	lxvp	vs40,	DISP8(\Index,  64)(AO)	// load real,imag from A
+	lxvp	vs42,	DISP8(\Index,  96)(AO)	// load real,imag from A
+ 	lxvp	vs50,	DISP4(\Index,  32)(BO)  // load real,imag from B
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      1,      vs34,   vs49
+        xvf64gerpp      2,      vs32,   vs48
+        xvf64gerpp      3,      vs34,   vs48
+	lxvp	vs32,	DISP8(\Index, 128)(AO)	// load real,imag from A
+	lxvp	vs34,	DISP8(\Index, 160)(AO)	// load real,imag from A
+ 	lxvp	vs48,	DISP4(\Index,  64)(BO)  // load real,imag from B
+        xvf64gerpp      0,      vs40,   vs51 
+        xvf64gerpp      1,      vs42,   vs51
+        xvf64gerpp      2,      vs40,   vs50
+        xvf64gerpp      3,      vs42,   vs50
+.if \IsLast==1
+	addi	AO, AO, DISP8(\Index,128)
+	addi	BO, BO, DISP4(\Index,64)
+.endif 
+.endm
+ 
+
+.macro LOAD_END_2x4	OffsetA, OffsetB
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      1,      vs34,   vs49
+        xvf64gerpp      2,      vs32,   vs48
+        xvf64gerpp      3,      vs34,   vs48
+	addi	BO, BO, \OffsetB
+	addi	AO, AO, \OffsetA
+.endm
+
+
+.macro  KERNEL2x4_UNPRIME_MMA
+        /* "unprime" MMA accumulators */
+        xxmfacc 0
+        xxmfacc 1
+        xxmfacc 2
+        xxmfacc 3
+.endm
+
+
+.macro SAVE2x4 
+	add	T1, CO ,LDC 
+        xxpermdi vs32, vs0, vs1, 0b01
+        xxpermdi vs33, vs0, vs1, 0b10
+        xxpermdi vs34, vs2, vs3, 0b01
+        xxpermdi vs35, vs2, vs3, 0b10
+        xxpermdi vs36, vs4, vs5, 0b01
+        xxpermdi vs37, vs4, vs5, 0b10
+        xxpermdi vs38, vs6, vs7, 0b01
+        xxpermdi vs39, vs6, vs7, 0b10
+        xxpermdi vs40, vs8, vs9, 0b01
+        xxpermdi vs41, vs8, vs9, 0b10
+        xxpermdi vs42, vs10, vs11, 0b01
+        xxpermdi vs43, vs10, vs11, 0b10
+        xxpermdi vs44, vs12, vs13, 0b01
+        xxpermdi vs45, vs12, vs13, 0b10
+        xxpermdi vs46, vs14, vs15, 0b01
+        xxpermdi vs47, vs14, vs15, 0b10
+
+	xxlor vs2, vs32, vs32
+	xxlor vs3, vs33, vs33
+	xxlor vs0, vs34, vs34
+	xxlor vs1, vs35, vs35
+	xxlor vs6, vs36, vs36
+	xxlor vs7, vs37, vs37
+	xxlor vs4, vs38, vs38
+	xxlor vs5, vs39, vs39
+	xxlor vs10, vs40, vs40
+	xxlor vs11, vs41, vs41
+	xxlor vs8, vs42, vs42
+	xxlor vs9, vs43, vs43
+	xxlor vs14, vs44, vs44
+	xxlor vs15, vs45, vs45
+	xxlor vs12, vs46, vs46
+	xxlor vs13, vs47, vs47
+
+	SAVE4  vs0,vs1,vs2,vs3,vs4,vs5,vs6,vs7,CO,0
+	SAVE4  vs8,vs9,vs10,vs11,vs12,vs13,vs14,vs15,T1,0  
+	addi	CO, CO, 64
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=2 and M=2
+**********************************************************************************************/
+
+.macro  KERNEL2x2_ZERO_AND_PRIME_MMA
+        /* zero out and prime the MMA accumulators */
+        xxsetaccz 0
+        xxsetaccz 1
+.endm
+
+
+.macro KERNEL2x2_PRELOAD
+	lxvp	vs32,	 0(AO)	// load real,imag from A
+	lxvp	vs48,	 0(BO)	// load real imag from B
+.endm
+
+
+.macro KERNEL2x2_2 Index, IsLast
+	lxvp	vs40,	DISP4(\Index, 32)(AO)	// load real,imag from A
+ 	lxvp	vs50,	DISP4(\Index, 32)(BO)	// load real,imag from B
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      1,      vs32,   vs48
+	lxvp	vs32,	DISP4(\Index, 64)(AO)	// load real,imag from A
+	lxvp	vs48,	DISP4(\Index, 64)(BO)	// load real imag from B
+        xvf64gerpp      0,      vs40,   vs51
+        xvf64gerpp      1,      vs40,   vs50
+.if \IsLast==1
+	addi	AO, AO, DISP4(\Index,64)
+	addi	BO, BO, DISP4(\Index,64)
+.endif 
+.endm
+
+ 
+.macro LOAD_END_2x2  OffsetA,OffsetB
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      1,      vs32,   vs48
+	addi	BO, BO, \OffsetB
+	addi	AO, AO, \OffsetA
+.endm
+
+
+.macro  KERNEL2x2_UNPRIME_MMA
+        /* "unprime" MMA accumulators */
+        xxmfacc 0
+        xxmfacc 1
+.endm
+
+
+.macro SAVE2x2 
+	add	T1, CO ,LDC 
+        xxpermdi vs32, vs0, vs1, 0b01
+        xxpermdi vs33, vs0, vs1, 0b10
+        xxpermdi vs34, vs2, vs3, 0b01
+        xxpermdi vs35, vs2, vs3, 0b10
+        xxpermdi vs36, vs4, vs5, 0b01
+        xxpermdi vs37, vs4, vs5, 0b10
+        xxpermdi vs38, vs6, vs7, 0b01
+        xxpermdi vs39, vs6, vs7, 0b10
+
+	xxlor vs2, vs32, vs32
+	xxlor vs3, vs33, vs33
+	xxlor vs0, vs34, vs34
+	xxlor vs1, vs35, vs35
+	xxlor vs6, vs36, vs36
+	xxlor vs7, vs37, vs37
+	xxlor vs4, vs38, vs38
+	xxlor vs5, vs39, vs39
+
+	SAVE2  vs0,vs1,vs2,vs3,CO,0
+	SAVE2  vs4,vs5,vs6,vs7,T1,0 
+	addi	CO, CO, 32 
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=2 and M=1
+**********************************************************************************************/
+
+.macro ZERO2x1
+	xxlxor	vs0,	vs0,	vs0
+	xxlxor	vs1,	vs1,	vs1
+	xxlxor	vs2,	vs2,	vs2
+	xxlxor	vs3,	vs3,	vs3
+ 
+.endm
+
+
+.macro LOAD2x1   
+	LOAD2x1O 0,0 
+.endm
+
+
+.macro LOAD2x1O  OffsetA,OffsetB
+	lxv	vs48,(\OffsetB+	0)(BO)	// load real imag from B
+	lxv	vs50,	(\OffsetB+16)(BO)	// load real,imag from B 
+	xxswapd	vs49, vs48
+	xxswapd	vs51, vs50
+	lxv	vs32,	(0+\OffsetA)(AO)	// load real,imag from A 
+.endm
+
+
+.macro END2x1_WITHOUT_ADD
+	END2x1 AO,BO,0,0
+.endm
+
+
+.macro END2x1	AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi	\BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi	\AREG, \AREG, \OffsetA
+.endif
+	xvmaddadp	vs0,	vs32,	vs48
+	xvmaddadp	vs2,	vs32,	vs50
+	xvmaddadp	vs1,	vs32,	vs49
+	xvmaddadp	vs3,	vs32,	vs51 
+.endm
+
+
+.macro LOAD2x1_2
+    LOAD2x1_2O 0,0
+.endm	
+
+
+.macro LOAD2x1_2O  OffsetA,OffsetB
+	lxv	vs48,(\OffsetB+	0)(BO)	// load real imag from B
+	lxv	vs50,	(\OffsetB+16)(BO)	// load real,imag from B
+	lxv	vs52,	(\OffsetB+32)(BO)	// load real,imag	from B
+	lxv	vs54,	(\OffsetB+48)(BO)	// load real,imag  from B	
+	xxswapd	vs49, vs48
+	xxswapd	vs51, vs50
+	lxv	vs32,	(0+\OffsetA)(AO)	// load real,imag from A
+	lxv	vs40,	(16+\OffsetA)(AO)	// load real,imag from A 
+.endm	
+
+
+.macro END2x1_2	  
+  /*for load2 offset will be 32 and 64*/
+   KERNEL2x1_2	AO,BO,	32,64,0 ,1,1 
+.endm
+
+
+.macro KERNEL2x1_E2	OffsetA,OffsetB, Index,IsLast 
+  KERNEL2x1_2	AO,BO,	\OffsetA,\OffsetB, \Index,\IsLast ,1 
+.endm
+
+
+.macro KERNEL2x1_L2	OffsetA,OffsetB, Index,IsLast
+  KERNEL2x1_2	AO,BO,	\OffsetA,\OffsetB, \Index,\IsLast ,0 
+.endm
+
+
+.macro KERNEL2x1_2	AREG,BREG,	OffsetA,OffsetB, Index,IsLast ,Complete
+  xxswapd	vs53, vs52
+  xxswapd	vs55, vs54 
+	xvmaddadp	vs0,	vs32,	vs48
+	xvmaddadp	vs2,	vs32,	vs50
+	xvmaddadp	vs1,	vs32,	vs49
+	xvmaddadp	vs3,	vs32,	vs51
+.if \Complete==0	
+	lxv	vs32,	DISP2(\Index, 0 + \OffsetA)(\AREG)	// load real,imag from A 
+.endif	 
+.if \Complete==0		
+	lxv	vs48,	DISP4(\Index, 0+\OffsetB)(\BREG)	// load real imag from B
+	lxv	vs50,	DISP4(\Index, 16+\OffsetB)(\BREG)	// load real,imag from B
+.endif
+.if \Complete==0		
+  xxswapd	vs49, vs48
+  xxswapd	vs51, vs50
+.endif 
+	xvmaddadp	vs0,	vs40,	vs52
+	xvmaddadp	vs2,	vs40,	vs54 
+	xvmaddadp	vs1,	vs40,	vs53
+	xvmaddadp	vs3,	vs40,	vs55
+.if \Complete==0		
+	lxv	vs40,	DISP2(\Index,16+0+ \OffsetA)(\AREG)	// load real,imag from A 
+.endif
+ 
+.if \Complete==0	 
+ 	lxv	vs52,	DISP4(\Index, 32+\OffsetB)(\BREG)	// load real,imag	from B
+	lxv	vs54,	DISP4(\Index, 48+\OffsetB)(\BREG)	// load real,imag  from B
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi	\AREG, \AREG,  DISP2(\Index,\OffsetA)
+	addi	\BREG, \BREG,  DISP4(\Index,\OffsetB)
+.else
+	addi	\AREG, \AREG, DISP2(\Index,32)
+	addi	\BREG, \BREG,  DISP4(\Index,64)
+.endif
+.endif 
+.endm
+ 
+
+.macro KERNEL2x1
+  LOAD2x1
+  END2x1  AO, BO, 16,32
+.endm
+
+
+.macro SAVE2x1
+	add	T1, CO ,LDC 
+	SAVE1  vs0,vs1,CO,0
+	SAVE1  vs2,vs3,T1,0  
+	addi	CO, CO, 16 
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=1 and M=8
+**********************************************************************************************/
+
+.macro  KERNEL1x8_ZERO_AND_PRIME_MMA
+        /* zero out and prime the MMA accumulators */
+        xxsetaccz 0
+        xxsetaccz 1
+        xxsetaccz 2
+        xxsetaccz 3
+.endm
+
+
+.macro KERNEL1x8_2	Index,IsLast
+	lxvp	vs32,	DISP16(\Index,   0)(AO)	// load real,imag from A
+	lxvp	vs34,	DISP16(\Index,  32)(AO)	// load real,imag from A
+	lxvp	vs36,	DISP16(\Index,  64)(AO)	// load real,imag from A
+	lxvp	vs38,	DISP16(\Index,  96)(AO)	// load real,imag from A
+	lxvp	vs40,	DISP16(\Index, 128)(AO)	// load real,imag from A
+	lxvp	vs42,	DISP16(\Index, 160)(AO)	// load real,imag from A
+	lxvp	vs44,	DISP16(\Index, 192)(AO)	// load real,imag from A
+	lxvp	vs46,	DISP16(\Index, 224)(AO)	// load real,imag from A
+	lxvp	vs48,	DISP2(\Index,    0)(BO)	// load real imag from B
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      1,      vs34,   vs49
+        xvf64gerpp      2,      vs36,   vs49
+        xvf64gerpp      3,      vs38,   vs49
+        xvf64gerpp      0,      vs40,   vs48
+        xvf64gerpp      1,      vs42,   vs48
+        xvf64gerpp      2,      vs44,   vs48
+        xvf64gerpp      3,      vs46,   vs48
+.if \IsLast==1
+	addi	AO, AO, DISP16(\Index,256)
+	addi	BO, BO,  DISP2(\Index,32)
+.endif 
+.endm
+
+
+.macro LOAD_END_1x8  OffsetA,OffsetB
+	lxvp	vs32,	0(AO)	// load real,imag from A
+	lxvp	vs34,	32(AO)	// load real,imag from A
+	lxvp	vs36,	64(AO)	// load real,imag from A
+	lxvp	vs38,	96(AO)	// load real,imag from A
+	lxv	vs48,	0(BO)	// load real imag from B 
+        xvf64gerpp      0,      vs32,   vs48
+        xvf64gerpp      1,      vs34,   vs48
+        xvf64gerpp      2,      vs36,   vs48
+        xvf64gerpp      3,      vs38,   vs48
+	addi	BO, BO, \OffsetB
+	addi	AO, AO, \OffsetA
+.endm
+
+
+.macro  KERNEL1x8_UNPRIME_MMA
+        /* "unprime" MMA accumulators */
+        xxmfacc 0
+        xxmfacc 1
+        xxmfacc 2
+        xxmfacc 3
+.endm
+
+
+.macro SAVE1x8
+        xxpermdi vs32, vs0, vs1, 0b01
+        xxpermdi vs33, vs0, vs1, 0b10
+        xxpermdi vs34, vs2, vs3, 0b01
+        xxpermdi vs35, vs2, vs3, 0b10
+        xxpermdi vs36, vs4, vs5, 0b01
+        xxpermdi vs37, vs4, vs5, 0b10
+        xxpermdi vs38, vs6, vs7, 0b01
+        xxpermdi vs39, vs6, vs7, 0b10
+        xxpermdi vs40, vs8, vs9, 0b01
+        xxpermdi vs41, vs8, vs9, 0b10
+        xxpermdi vs42, vs10, vs11, 0b01
+        xxpermdi vs43, vs10, vs11, 0b10
+        xxpermdi vs44, vs12, vs13, 0b01
+        xxpermdi vs45, vs12, vs13, 0b10
+        xxpermdi vs46, vs14, vs15, 0b01
+        xxpermdi vs47, vs14, vs15, 0b10
+
+	xxlor vs2, vs32, vs32
+	xxlor vs3, vs33, vs33
+	xxlor vs0, vs34, vs34
+	xxlor vs1, vs35, vs35
+	xxlor vs6, vs36, vs36
+	xxlor vs7, vs37, vs37
+	xxlor vs4, vs38, vs38
+	xxlor vs5, vs39, vs39
+	xxlor vs10, vs40, vs40
+	xxlor vs11, vs41, vs41
+	xxlor vs8, vs42, vs42
+	xxlor vs9, vs43, vs43
+	xxlor vs14, vs44, vs44
+	xxlor vs15, vs45, vs45
+	xxlor vs12, vs46, vs46
+	xxlor vs13, vs47, vs47
+
+	SAVE8  vs0,vs1,vs2,vs3,vs4,vs5,vs6,vs7,vs8,vs9,vs10,vs11,vs12,vs13,vs14,vs15,CO,0
+	addi	CO, CO, 128
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=1 and M=4
+**********************************************************************************************/
+
+.macro  KERNEL1x4_ZERO_AND_PRIME_MMA
+        /* zero out and prime the MMA accumulators */
+        xxsetaccz 0
+        xxsetaccz 1
+.endm
+
+
+.macro KERNEL1x4_2	Index,IsLast
+	lxvp	vs32,	DISP8(\Index,  0)(AO)	// load real,imag from A
+	lxvp	vs34,	DISP8(\Index, 32)(AO)	// load real,imag from A
+	lxvp	vs40,	DISP8(\Index, 64)(AO)	// load real,imag from A
+	lxvp	vs42,	DISP8(\Index, 96)(AO)	// load real,imag from A
+	lxvp	vs48,	DISP2(\Index,  0)(BO)	// load real imag from B
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      1,      vs34,   vs49
+        xvf64gerpp      0,      vs40,   vs48
+        xvf64gerpp      1,      vs42,   vs48
+.if \IsLast==1
+	addi	AO, AO, DISP8(\Index,128)
+	addi	BO, BO,  DISP2(\Index,32)
+.endif 
+.endm
+ 
+
+.macro LOAD_END_1x4  OffsetA,OffsetB
+	lxvp	vs32,	0(AO)	// load real,imag from A
+	lxvp	vs34,	32(AO)	// load real,imag from A
+	lxv	vs48,	0(BO)	// load real imag from B
+        xvf64gerpp      0,      vs32,   vs48
+        xvf64gerpp      1,      vs34,   vs48
+	addi	BO, BO, \OffsetB
+	addi	AO, AO, \OffsetA
+.endm
+
+
+.macro  KERNEL1x4_UNPRIME_MMA
+        /* "unprime" MMA accumulators */
+        xxmfacc 0
+        xxmfacc 1
+.endm
+
+
+.macro SAVE1x4 
+        xxpermdi vs32, vs0, vs1, 0b01
+        xxpermdi vs33, vs0, vs1, 0b10
+        xxpermdi vs34, vs2, vs3, 0b01
+        xxpermdi vs35, vs2, vs3, 0b10
+        xxpermdi vs36, vs4, vs5, 0b01
+        xxpermdi vs37, vs4, vs5, 0b10
+        xxpermdi vs38, vs6, vs7, 0b01
+        xxpermdi vs39, vs6, vs7, 0b10
+
+	xxlor vs2, vs32, vs32
+	xxlor vs3, vs33, vs33
+	xxlor vs0, vs34, vs34
+	xxlor vs1, vs35, vs35
+	xxlor vs6, vs36, vs36
+	xxlor vs7, vs37, vs37
+	xxlor vs4, vs38, vs38
+	xxlor vs5, vs39, vs39
+
+	SAVE4  vs0,vs1,vs2,vs3,vs4,vs5,vs6,vs7,CO,0
+	addi	CO, CO, 64
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=1 and M=2
+**********************************************************************************************/
+
+.macro  KERNEL1x2_ZERO_AND_PRIME_MMA
+        /* zero out and prime the MMA accumulators */
+        xxsetaccz 0
+.endm
+
+
+.macro KERNEL1x2_2	Index,IsLast
+	lxvp	vs32,	DISP4(\Index,  0)(AO)	// load real,imag from A
+	lxvp	vs40,	DISP4(\Index, 32)(AO)	// load real,imag from A
+	lxvp	vs48,	DISP2(\Index,  0)(BO)	// load real imag from B
+        xvf64gerpp      0,      vs32,   vs49
+        xvf64gerpp      0,      vs40,   vs48
+.if \IsLast==1
+	addi	AO, AO, DISP4(\Index,64)
+	addi	BO, BO, DISP2(\Index,32)
+.endif 
+.endm
+ 
+
+.macro LOAD_END_1x2  OffsetA,OffsetB
+	lxvp	vs32,	0(AO)	// load real,imag from A
+	lxv	vs48,	0(BO)	// load real imag from B
+        xvf64gerpp      0,      vs32,   vs48
+	addi	BO, BO, \OffsetB
+	addi	AO, AO, \OffsetA
+.endm
+
+
+.macro  KERNEL1x2_UNPRIME_MMA
+        /* "unprime" MMA accumulators */
+        xxmfacc 0
+.endm
+
+
+.macro SAVE1x2 
+        xxpermdi vs32, vs0, vs1, 0b01
+        xxpermdi vs33, vs0, vs1, 0b10
+        xxpermdi vs34, vs2, vs3, 0b01
+        xxpermdi vs35, vs2, vs3, 0b10
+
+	xxlor vs2, vs32, vs32
+	xxlor vs3, vs33, vs33
+	xxlor vs0, vs34, vs34
+	xxlor vs1, vs35, vs35
+
+	SAVE2  vs0,vs1,vs2,vs3,CO,0
+	addi	CO, CO, 32 
+.endm
+
+/**********************************************************************************************
+*
+
+.macros for N=1 and M=1
+**********************************************************************************************/
+
+.macro ZERO1x1
+	xxlxor	vs0,	vs0,	vs0
+	xxlxor	vs1,	vs1,	vs1 
+.endm
+
+
+.macro LOAD1x1   
+	LOAD1x1O 0,0 
+.endm
+
+
+.macro LOAD1x1O  OffsetA,OffsetB
+	lxv	vs48,(\OffsetB+	0)(BO)	// load real imag from B
+	lxv	vs32,	(0+\OffsetA)(AO)	// load real,imag from A 
+	xxswapd	vs49, vs48
+
+.endm
+
+
+.macro END1x1_WITHOUT_ADD
+	END1x1 AO,BO,0,0
+.endm
+
+
+.macro END1x1	AREG, BREG, OffsetA, OffsetB
+.if \OffsetB != 0
+	addi	\BREG, \BREG, \OffsetB
+.endif
+.if \OffsetA != 0
+	addi	\AREG, \AREG, \OffsetA
+.endif
+	xvmaddadp	vs0,	vs32,	vs48 
+	xvmaddadp	vs1,	vs32,	vs49 
+.endm
+
+
+.macro LOAD1x1_2
+    LOAD1x1_2O 0,0
+.endm	
+
+
+.macro LOAD1x1_2O  OffsetA,OffsetB
+	lxv	vs48,(\OffsetB+	0)(BO)	// load real imag from B
+	lxv	vs52,	(\OffsetB+16)(BO)	// load real,imag	from B
+	xxswapd	vs49, vs48
+
+	lxv	vs32,	(0+\OffsetA)(AO)	// load real,imag from A
+	lxv	vs40,	(16+\OffsetA)(AO)	// load real,imag from A 
+.endm	
+
+
+.macro END1x1_2	  
+  /*for load2 offset will be 32 and 32*/
+   KERNEL1x1_2	AO,BO,	32,32,0 ,1,1 
+.endm
+ 
+
+
+.macro KERNEL1x1_E2	OffsetA,OffsetB, Index,IsLast 
+  KERNEL1x1_2	AO,BO,	\OffsetA,\OffsetB, \Index,\IsLast ,1 
+.endm
+
+
+.macro KERNEL1x1_L2	OffsetA,OffsetB, Index,IsLast
+  KERNEL1x1_2	AO,BO,	\OffsetA,\OffsetB, \Index,\IsLast ,0 
+.endm
+
+
+.macro KERNEL1x1_2	AREG,BREG,	OffsetA,OffsetB, Index,IsLast ,Complete
+  xxswapd	vs53, vs52
+	xvmaddadp	vs0,	vs32,	vs48 
+	xvmaddadp	vs1,	vs32,	vs49 
+.if \Complete==0	
+	lxv	vs32,	DISP2(\Index, 0 + \OffsetA)(\AREG)	// load real,imag from A 
+.endif	 
+.if \Complete==0		
+	lxv	vs48,	DISP2(\Index, 0+\OffsetB)(\BREG)	// load real imag from B
+.endif
+.if \Complete==0		
+  xxswapd	vs49, vs48
+.endif 
+	xvmaddadp	vs0,	vs40,	vs52
+	xvmaddadp	vs1,	vs40,	vs53 
+.if \Complete==0		
+	lxv	vs40,	DISP2(\Index,16+0+ \OffsetA)(\AREG)	// load real,imag from A 
+.endif
+ 
+.if \Complete==0	 
+ 	lxv	vs52,	DISP2(\Index, 16+\OffsetB)(\BREG)	// load real,imag	from B
+.endif
+.if \IsLast==1
+.if \Complete==1
+	addi	\AREG, \AREG,  DISP2(\Index,\OffsetA)
+	addi	\BREG, \BREG,  DISP2(\Index,\OffsetB)
+.else
+	addi	\AREG, \AREG, DISP2(\Index,32)
+	addi	\BREG, \BREG,  DISP2(\Index,32)
+.endif
+.endif 
+.endm
+ 
+
+
+.macro KERNEL1x1
+  LOAD1x1
+  END1x1  AO, BO, 16,16
+.endm
+
+
+
+.macro SAVE1x1
+	SAVE1  vs0,vs1,CO,0
+	addi	CO, CO, 16 
+.endm
+
+/****************************TRMM POINTER REFRESH
+
+.macroSES*************************/
+
+
+.macro SHIFT_REG  REG1,REG2,SHIFT_VAL
+		.if \SHIFT_VAL==16 
+			slwi		\REG1,	\REG2,	8			
+		.elseif \SHIFT_VAL==8  
+			slwi		\REG1,	\REG2,	7			 
+		.elseif \SHIFT_VAL==4
+			slwi		\REG1,	\REG2,	6			  
+		.elseif \SHIFT_VAL==2
+			slwi		\REG1,	\REG2,	5			 
+		.elseif \SHIFT_VAL==1
+			slwi		\REG1,	\REG2,	4			 
+		.endif
+.endm
+/*
+//#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+// 		ptrbb = bb;
+// #else
+// 		ptrba += off*16;
+// 		ptrbb = bb + off*2;
+// #endif
+*/
+
+
+.macro REFRESH_POINTERS  PTR_A,PTR_B,OFF_VAL,B_VAL,C_A,C_B
+    #if (defined(LEFT) &&  defined(TRANSA)) ||  (!defined(LEFT) && !defined(TRANSA))
+        /* ptrbb = bb;*/
+        mr \PTR_B,\B_VAL     /* refresh BPOINT */
+    #else
+		    /*
+        // ptrba  =ptrba+ off*C_A;
+        // ptrbb = bb + off*C_B; 
+				*/
+		SHIFT_REG T4,\OFF_VAL,\C_B		/* Number of values in B shifted  */
+		SHIFT_REG T2,\OFF_VAL,\C_A		/* Number of values in A shifted  */
+		add		\PTR_B,	\B_VAL ,	T4				/* Add values to BO */
+		add		\PTR_A,	\PTR_A,	T2				/* Add values to AO  */
+    #endif 
+.endm
+
+/*
+// #if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+// 		temp = bk-off;
+// #elif defined(LEFT)
+// 		temp = off+16;	// number of values in A
+// #else
+// 		temp = off+2;	// number of values in B
+// #endif
+*/
+
+
+.macro REFRESH_TEMP_BK TEMP_BK,BK_VAL,OFF_VAL,INCR_A,INCR_B
+    #if (defined(LEFT) && !defined(TRANSA)) ||  (!defined(LEFT) && defined(TRANSA))
+                            /* temp = bk-off;*/
+           sub \TEMP_BK,\BK_VAL,\OFF_VAL
+    #elif defined(LEFT)
+                            /* temp = off+INCR_A;	// number of values in A */
+           addi \TEMP_BK, \OFF_VAL, \INCR_A
+    #else
+                            /* temp = off+INCR_B	// number of values in B*/
+           addi \TEMP_BK,\OFF_VAL, \INCR_B
+    #endif
+.endm
+/*
+// #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+// 		temp = bk - off;
+// #ifdef LEFT
+// 		temp -= 16; // number of values in A
+// #else
+// 		temp -= 2; // number of values in B
+// #endif
+// 		ptrba += temp*16;
+// 		ptrbb += temp*2;
+// #endif
+// #ifdef LEFT
+// 		off += 16; // number of values in A
+// #endif
+*/
+ 
+
+
+.macro REFRESH_AFTER_SAVE TEMP_BK,BK_VAL,OFF_VAL,PTR_B,PTR_A,C_A,C_B
+    #if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+                    /*temp = bk - off;*/
+                sub \TEMP_BK,\BK_VAL,\OFF_VAL
+    #ifdef LEFT
+                    /*temp -= 8; // number of values in A*/
+                addi \TEMP_BK,\TEMP_BK,-\C_A
+    #else
+                    /*temp -= 4; // number of values in B*/
+                addi \TEMP_BK,\TEMP_BK,-\C_B 
+    #endif
+                    /*ptrba += temp*C_A;
+                    ptrbb += temp*C_B;*/ 
+                SHIFT_REG T4,\TEMP_BK,\C_A
+								SHIFT_REG T2,\TEMP_BK,\C_B
+                add \PTR_A, \PTR_A,T4/*ptrba+temp*C_A*/ 
+								add \PTR_B, \PTR_B,T2 
+    #endif
+    #ifdef LEFT
+                    /*off += 8; // number of values in A*/
+                 addi \OFF_VAL,\OFF_VAL,\C_A
+    #endif
+.endm
+
diff --git a/kernel/power/zscal.c b/kernel/power/zscal.c
index a1b441d2c..16b584bca 100644
--- a/kernel/power/zscal.c
+++ b/kernel/power/zscal.c
@@ -38,7 +38,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #pragma GCC optimize "O1"
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #if defined(DOUBLE)
 #include "zscal_microk_power8.c"
 #endif
diff --git a/kernel/power/zswap.c b/kernel/power/zswap.c
index 1d8826f41..c6508f032 100644
--- a/kernel/power/zswap.c
+++ b/kernel/power/zswap.c
@@ -36,7 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8) || defined(POWER9)
+#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #include "zswap_microk_power8.c"
 #endif
 
diff --git a/kernel/setparam-ref.c b/kernel/setparam-ref.c
index 3c71c778e..d3aa030c1 100644
--- a/kernel/setparam-ref.c
+++ b/kernel/setparam-ref.c
@@ -53,6 +53,65 @@ gotoblas_t TABLE_NAME = {
 
   GEMM_DEFAULT_OFFSET_A, GEMM_DEFAULT_OFFSET_B, GEMM_DEFAULT_ALIGN,
 
+#ifdef BUILD_HALF
+  0, 0, 0,
+  SHGEMM_DEFAULT_UNROLL_M, SHGEMM_DEFAULT_UNROLL_N,
+#ifdef SHGEMM_DEFAULT_UNROLL_MN
+ SHGEMM_DEFAULT_UNROLL_MN,
+#else
+ MAX(SHGEMM_DEFAULT_UNROLL_M, SHGEMM_DEFAULT_UNROLL_N),
+#endif
+
+  samax_kTS,  samin_kTS,  smax_kTS,  smin_kTS,
+  isamax_kTS, isamin_kTS, ismax_kTS, ismin_kTS,
+  snrm2_kTS,  sasum_kTS, ssum_kTS, scopy_kTS, sdot_kTS,
+  dsdot_kTS,
+  srot_kTS,   saxpy_kTS,  sscal_kTS, sswap_kTS,
+  sgemv_nTS,  sgemv_tTS, sger_kTS,
+  ssymv_LTS, ssymv_UTS,
+
+  shgemm_kernelTS, shgemm_betaTS,
+#if SHGEMM_DEFAULT_UNROLL_M != SHGEMM_DEFAULT_UNROLL_N
+  shgemm_incopyTS, shgemm_itcopyTS,
+#else
+  shgemm_oncopyTS, shgemm_otcopyTS,
+#endif
+  shgemm_oncopyTS, shgemm_otcopyTS,
+
+  strsm_kernel_LNTS, strsm_kernel_LTTS, strsm_kernel_RNTS, strsm_kernel_RTTS,
+#if SGEMM_DEFAULT_UNROLL_M != SGEMM_DEFAULT_UNROLL_N
+  strsm_iunucopyTS, strsm_iunncopyTS, strsm_iutucopyTS, strsm_iutncopyTS,
+  strsm_ilnucopyTS, strsm_ilnncopyTS, strsm_iltucopyTS, strsm_iltncopyTS,
+#else
+  strsm_ounucopyTS, strsm_ounncopyTS, strsm_outucopyTS, strsm_outncopyTS,
+  strsm_olnucopyTS, strsm_olnncopyTS, strsm_oltucopyTS, strsm_oltncopyTS,
+#endif
+  strsm_ounucopyTS, strsm_ounncopyTS, strsm_outucopyTS, strsm_outncopyTS,
+  strsm_olnucopyTS, strsm_olnncopyTS, strsm_oltucopyTS, strsm_oltncopyTS,
+  strmm_kernel_RNTS, strmm_kernel_RTTS, strmm_kernel_LNTS, strmm_kernel_LTTS,
+#if SGEMM_DEFAULT_UNROLL_M != SGEMM_DEFAULT_UNROLL_N
+  strmm_iunucopyTS, strmm_iunncopyTS, strmm_iutucopyTS, strmm_iutncopyTS,
+  strmm_ilnucopyTS, strmm_ilnncopyTS, strmm_iltucopyTS, strmm_iltncopyTS,
+#else
+  strmm_ounucopyTS, strmm_ounncopyTS, strmm_outucopyTS, strmm_outncopyTS,
+  strmm_olnucopyTS, strmm_olnncopyTS, strmm_oltucopyTS, strmm_oltncopyTS,
+#endif
+  strmm_ounucopyTS, strmm_ounncopyTS, strmm_outucopyTS, strmm_outncopyTS,
+  strmm_olnucopyTS, strmm_olnncopyTS, strmm_oltucopyTS, strmm_oltncopyTS,
+#if SGEMM_DEFAULT_UNROLL_M != SGEMM_DEFAULT_UNROLL_N
+  ssymm_iutcopyTS, ssymm_iltcopyTS,
+#else
+  ssymm_outcopyTS, ssymm_oltcopyTS,
+#endif
+  ssymm_outcopyTS, ssymm_oltcopyTS,
+
+#ifndef NO_LAPACK
+  sneg_tcopyTS, slaswp_ncopyTS,
+#else
+  NULL,NULL,
+#endif
+#endif
+
   0, 0, 0,
   SGEMM_DEFAULT_UNROLL_M, SGEMM_DEFAULT_UNROLL_N,
 #ifdef SGEMM_DEFAULT_UNROLL_MN
@@ -648,16 +707,25 @@ gotoblas_t TABLE_NAME = {
 
 #if defined(ARCH_ARM64)
 static void init_parameter(void) {
+#if defined(BUILD_HALF)
+  TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+#endif
   TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
   TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
   TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
   TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
 
+#if defined(BUILD_HALF)
+  TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
   TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
   TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
   TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
   TABLE_NAME.zgemm_q = ZGEMM_DEFAULT_Q;
 
+#if defined(BUILD_HALF)
+  TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+#endif
   TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
   TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
   TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
@@ -721,17 +789,26 @@ static void init_parameter(void) {
 #if defined(ARCH_POWER)
 static void init_parameter(void) {
 
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+#endif
   TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
   TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
   TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
   TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
 
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+#endif
   TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
   TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
   TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
   TABLE_NAME.zgemm_r = ZGEMM_DEFAULT_R;
 
 
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
   TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
   TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
   TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
@@ -741,17 +818,26 @@ static void init_parameter(void) {
 
 #if defined(ARCH_ZARCH)
 static void init_parameter(void) {
+#ifdef BUILD_HALF
+	TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+#endif
 	TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
 	TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
 	TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
 	TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
 
+#ifdef BUILD_HALF
+	TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+#endif
 	TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
 	TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
 	TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
 	TABLE_NAME.zgemm_r = ZGEMM_DEFAULT_R;
 
 
+#ifdef BUILD_HALF
+	TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
 	TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
 	TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
 	TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
@@ -891,6 +977,11 @@ static void init_parameter(void) {
   (void) l2; /* dirty trick to suppress unused variable warning for targets */
              /* where the GEMM unrolling parameters do not depend on l2 */
   
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+  TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+  TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
   TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
   TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
   TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
diff --git a/kernel/x86_64/KERNEL.SKYLAKEX b/kernel/x86_64/KERNEL.SKYLAKEX
index 65f031d03..9b8b84c30 100644
--- a/kernel/x86_64/KERNEL.SKYLAKEX
+++ b/kernel/x86_64/KERNEL.SKYLAKEX
@@ -14,7 +14,7 @@ STRSMKERNEL_RT =  ../generic/trsm_kernel_RT.c
 DGEMMKERNEL    =  dgemm_kernel_16x2_skylakex.c
 DTRMMKERNEL    =  dgemm_kernel_16x2_skylakex.c
 DGEMMINCOPY    =  ../generic/gemm_ncopy_16.c
-DGEMMITCOPY    =  ../generic/gemm_tcopy_16.c
+DGEMMITCOPY    =  dgemm_tcopy_16_skylakex.c
 DGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
 DGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
 DTRSMKERNEL_RN =  ../generic/trsm_kernel_RN.c
@@ -24,3 +24,6 @@ DGEMM_BETA = dgemm_beta_skylakex.c
 
 CGEMMKERNEL    =  cgemm_kernel_8x2_skylakex.c
 ZGEMMKERNEL    =  zgemm_kernel_4x2_skylakex.c
+
+CSCALKERNEL    = ../arm/zscal.c
+ZSCALKERNEL    = ../arm/zscal.c
diff --git a/kernel/x86_64/dgemm_kernel_16x2_skylakex.c b/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
index 416ace59b..9f2bf24e2 100644
--- a/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
+++ b/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
@@ -54,40 +54,40 @@
     #define kernel_kstart_n10(mdim,updk) ""
     #define kernel_kstart_n12(mdim,updk) ""
     #define kernel_kend_n4(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16)
     #define kernel_kend_n6(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48)
     #define kernel_kend_n8(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8) acc_kend_nc4_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24) acc_kend_nc4_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40) acc_kend_nc4_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56) acc_kend_nc4_k1m##mdim(48,56)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64,72)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80,88)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0) acc_kend_nc4_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16) acc_kend_nc4_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32) acc_kend_nc4_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48) acc_kend_nc4_k1m##mdim(48)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80)
     #define kernel_kend_n10(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8) acc_kend_nc4_k1m##mdim(0,8) acc_kend_nc5_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24) acc_kend_nc4_k1m##mdim(16,24) acc_kend_nc5_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40) acc_kend_nc4_k1m##mdim(32,40) acc_kend_nc5_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56) acc_kend_nc4_k1m##mdim(48,56) acc_kend_nc5_k1m##mdim(48,56)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64,72) acc_kend_nc5_k1m##mdim(64,72)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80,88) acc_kend_nc5_k1m##mdim(80,88)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96,104)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112,120)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0) acc_kend_nc4_k1m##mdim(0) acc_kend_nc5_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16) acc_kend_nc4_k1m##mdim(16) acc_kend_nc5_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32) acc_kend_nc4_k1m##mdim(32) acc_kend_nc5_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48) acc_kend_nc4_k1m##mdim(48) acc_kend_nc5_k1m##mdim(48)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64) acc_kend_nc5_k1m##mdim(64)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80) acc_kend_nc5_k1m##mdim(80)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112)
     #define kernel_kend_n12(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8) acc_kend_nc4_k1m##mdim(0,8) acc_kend_nc5_k1m##mdim(0,8) acc_kend_nc6_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24) acc_kend_nc4_k1m##mdim(16,24) acc_kend_nc5_k1m##mdim(16,24) acc_kend_nc6_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40) acc_kend_nc4_k1m##mdim(32,40) acc_kend_nc5_k1m##mdim(32,40) acc_kend_nc6_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56) acc_kend_nc4_k1m##mdim(48,56) acc_kend_nc5_k1m##mdim(48,56) acc_kend_nc6_k1m##mdim(48,56)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64,72) acc_kend_nc5_k1m##mdim(64,72) acc_kend_nc6_k1m##mdim(64,72)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80,88) acc_kend_nc5_k1m##mdim(80,88) acc_kend_nc6_k1m##mdim(80,88)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96,104) acc_kend_nc6_k1m##mdim(96,104)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112,120) acc_kend_nc6_k1m##mdim(112,120)\
-      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(128,136)\
-      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(144,152)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0) acc_kend_nc4_k1m##mdim(0) acc_kend_nc5_k1m##mdim(0) acc_kend_nc6_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16) acc_kend_nc4_k1m##mdim(16) acc_kend_nc5_k1m##mdim(16) acc_kend_nc6_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32) acc_kend_nc4_k1m##mdim(32) acc_kend_nc5_k1m##mdim(32) acc_kend_nc6_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48) acc_kend_nc4_k1m##mdim(48) acc_kend_nc5_k1m##mdim(48) acc_kend_nc6_k1m##mdim(48)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64) acc_kend_nc5_k1m##mdim(64) acc_kend_nc6_k1m##mdim(64)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80) acc_kend_nc5_k1m##mdim(80) acc_kend_nc6_k1m##mdim(80)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96) acc_kend_nc6_k1m##mdim(96)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112) acc_kend_nc6_k1m##mdim(112)\
+      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(128)\
+      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(144)
   #endif
 #else
   #define HEAD_SET_OFF(ndim) {}
@@ -129,18 +129,28 @@
   #define init_update_k(mdim) ""
   #define save_update_k(mdim) ""
 #endif
-    
+
 #define KERNEL_h_k1m16n1 \
   "vmovupd (%0),%%zmm1; vmovupd 64(%0),%%zmm2; addq $128,%0;"\
   "vbroadcastsd (%1),%%zmm3; vfmadd231pd %%zmm1,%%zmm3,%%zmm8; vfmadd231pd %%zmm2,%%zmm3,%%zmm9;"
 #define KERNEL_k1m16n1 KERNEL_h_k1m16n1 "addq $8,%1;"
-#define KERNEL_h_k1m16n2 KERNEL_h_k1m16n1\
+#ifdef BROADCAST_KERNEL
+ #define KERNEL_h_k1m16n2 KERNEL_h_k1m16n1\
   "vbroadcastsd 8(%1),%%zmm4; vfmadd231pd %%zmm1,%%zmm4,%%zmm10; vfmadd231pd %%zmm2,%%zmm4,%%zmm11;"
-#define KERNEL_k1m16n2 KERNEL_h_k1m16n2 "addq $16,%1;"
-#define unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,boff1,boff2,...)\
+ #define unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,boff1,...)\
   "vbroadcastsd "#boff1"("#__VA_ARGS__"),%%zmm3; vfmadd231pd %%zmm1,%%zmm3,%%zmm"#c1_no"; vfmadd231pd %%zmm2,%%zmm3,%%zmm"#c2_no";"\
-  "vbroadcastsd "#boff2"("#__VA_ARGS__"),%%zmm4; vfmadd231pd %%zmm1,%%zmm4,%%zmm"#c3_no"; vfmadd231pd %%zmm2,%%zmm4,%%zmm"#c4_no";"
-#define unit_acc_m16n2(c1_no,c2_no,c3_no,c4_no,...) unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,0,8,__VA_ARGS__)
+  "vbroadcastsd "#boff1"+8("#__VA_ARGS__"),%%zmm4; vfmadd231pd %%zmm1,%%zmm4,%%zmm"#c3_no"; vfmadd231pd %%zmm2,%%zmm4,%%zmm"#c4_no";"
+ #define unit_acc_m16n2(c1_no,c2_no,c3_no,c4_no,...) unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,0,__VA_ARGS__)
+#else
+ #define unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,boff1,...)\
+  "vbroadcastf32x4 "#boff1"("#__VA_ARGS__"),%%zmm5; vfmadd231pd %%zmm1,%%zmm5,%%zmm"#c1_no"; vfmadd231pd %%zmm2,%%zmm5,%%zmm"#c2_no";"\
+  "vfmadd231pd %%zmm3,%%zmm5,%%zmm"#c3_no"; vfmadd231pd %%zmm4,%%zmm5,%%zmm"#c4_no";"
+ #define unit_acc_m16n2(c1_no,c2_no,c3_no,c4_no,...) unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,0,__VA_ARGS__)
+ #define KERNEL_h_k1m16n2 \
+  "vmovddup (%0),%%zmm1; vmovddup 8(%0),%%zmm2; vmovddup 64(%0),%%zmm3; vmovddup 72(%0),%%zmm4; addq $128,%0;"\
+  unit_acc_m16n2(8,9,10,11,%1)
+#endif
+#define KERNEL_k1m16n2 KERNEL_h_k1m16n2 "addq $16,%1;"
 #define KERNEL_h_k1m16n4 KERNEL_h_k1m16n2 "prefetcht0 384(%0);" unit_acc_m16n2(12,13,14,15,%1,%%r12,1)
 #define KERNEL_k1m16n4 KERNEL_h_k1m16n4 "addq $16,%1;"
 #define KERNEL_k1m16n6 KERNEL_h_k1m16n4 unit_acc_m16n2(16,17,18,19,%1,%%r12,2) "addq $16,%1;"
@@ -151,24 +161,42 @@
 #define KERNEL_h_k1m16n12 KERNEL_h_k1m16n10 unit_acc_m16n2(28,29,30,31,%%r15,%%r12,2)
 #define KERNEL_k1m16n12 KERNEL_h_k1m16n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
+ #ifdef BROADCAST_KERNEL
   #define loada_kend_k1m16 "vmovupd (%0,%3,1),%%zmm1; vmovupd 64(%0,%3,1),%%zmm2; addq $128,%3;"
-  #define acc_kend_nc2_k1m16(boff1,boff2) unit_acc_gen_m16n2(12,13,14,15,boff1,boff2,%1,%%r12,1)
-  #define acc_kend_nc3_k1m16(boff1,boff2) unit_acc_gen_m16n2(16,17,18,19,boff1,boff2,%1,%%r12,2)
-  #define acc_kend_nc4_k1m16(boff1,boff2) unit_acc_gen_m16n2(20,21,22,23,boff1,boff2,%%r15)
-  #define acc_kend_nc5_k1m16(boff1,boff2) unit_acc_gen_m16n2(24,25,26,27,boff1,boff2,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m16(boff1,boff2) unit_acc_gen_m16n2(28,29,30,31,boff1,boff2,%%r15,%%r12,2)
+ #else
+  #define loada_kend_k1m16 "vmovddup (%0,%3,1),%%zmm1; vmovddup 8(%0,%3,1),%%zmm2; vmovddup 64(%0,%3,1),%%zmm3; vmovddup 72(%0,%3,1),%%zmm4; addq $128,%3;"
+ #endif
+ #define acc_kend_nc2_k1m16(boff1) unit_acc_gen_m16n2(12,13,14,15,boff1,%1,%%r12,1)
+ #define acc_kend_nc3_k1m16(boff1) unit_acc_gen_m16n2(16,17,18,19,boff1,%1,%%r12,2)
+ #define acc_kend_nc4_k1m16(boff1) unit_acc_gen_m16n2(20,21,22,23,boff1,%%r15)
+ #define acc_kend_nc5_k1m16(boff1) unit_acc_gen_m16n2(24,25,26,27,boff1,%%r15,%%r12,1)
+ #define acc_kend_nc6_k1m16(boff1) unit_acc_gen_m16n2(28,29,30,31,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m16 "movq %2,%3; addq $128,%2;"
 #ifdef TRMMKERNEL
   #define SAVE_m16n1 "vmulpd %%zmm8,%%zmm0,%%zmm8; vmovupd %%zmm8,(%2); vmulpd %%zmm9,%%zmm0,%%zmm9; vmovupd %%zmm9,64(%2); addq $128,%2;"
+ #ifdef BROADCAST_KERNEL
   #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
     "vmulpd %%zmm"#c1_no",%%zmm0,%%zmm"#c1_no"; vmovupd %%zmm"#c1_no",(%3); vmulpd %%zmm"#c2_no",%%zmm0,%%zmm"#c2_no"; vmovupd %%zmm"#c2_no",64(%3);"\
     "vmulpd %%zmm"#c3_no",%%zmm0,%%zmm"#c3_no"; vmovupd %%zmm"#c3_no",(%3,%4,1); vmulpd %%zmm"#c4_no",%%zmm0,%%zmm"#c4_no"; vmovupd %%zmm"#c4_no",64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #else
+  #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
+    "vunpcklpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm1; vunpcklpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm2; vunpckhpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm3; vunpckhpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm4;"\
+    "vmulpd %%zmm1,%%zmm0,%%zmm1; vmovupd %%zmm1,(%3); vmulpd %%zmm2,%%zmm0,%%zmm2; vmovupd %%zmm2,64(%3);"\
+    "vmulpd %%zmm3,%%zmm0,%%zmm3; vmovupd %%zmm3,(%3,%4,1); vmulpd %%zmm4,%%zmm0,%%zmm4; vmovupd %%zmm4,64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #endif
 #else
   #define SAVE_m16n1 "vfmadd213pd (%2),%%zmm0,%%zmm8; vmovupd %%zmm8,(%2); vfmadd213pd 64(%2),%%zmm0,%%zmm9; vmovupd %%zmm9,64(%2); addq $128,%2;"
+ #ifdef BROADCAST_KERNEL
   #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
     "vfmadd213pd (%3),%%zmm0,%%zmm"#c1_no"; vmovupd %%zmm"#c1_no",(%3); vfmadd213pd 64(%3),%%zmm0,%%zmm"#c2_no"; vmovupd %%zmm"#c2_no",64(%3);"\
     "vfmadd213pd (%3,%4,1),%%zmm0,%%zmm"#c3_no"; vmovupd %%zmm"#c3_no",(%3,%4,1); vfmadd213pd 64(%3,%4,1),%%zmm0,%%zmm"#c4_no"; vmovupd %%zmm"#c4_no",64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #else
+  #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
+    "vunpcklpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm1; vunpcklpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm2; vunpckhpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm3; vunpckhpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm4;"\
+    "vfmadd213pd (%3),%%zmm0,%%zmm1; vmovupd %%zmm1,(%3); vfmadd213pd 64(%3),%%zmm0,%%zmm2; vmovupd %%zmm2,64(%3);"\
+    "vfmadd213pd (%3,%4,1),%%zmm0,%%zmm3; vmovupd %%zmm3,(%3,%4,1); vfmadd213pd 64(%3,%4,1),%%zmm0,%%zmm4; vmovupd %%zmm4,64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #endif
 #endif
 #define SAVE_m16n2 save_init_m16 unit_save_m16n2(8,9,10,11)
 #define SAVE_m16n4 SAVE_m16n2 unit_save_m16n2(12,13,14,15)
@@ -206,11 +234,11 @@
 #define KERNEL_k1m8n12 KERNEL_h_k1m8n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m8 "vmovddup (%0,%3,1),%%zmm1; vmovddup 8(%0,%3,1),%%zmm2; addq $64,%3;"
-  #define acc_kend_nc2_k1m8(boff1,boff2) unit_acc_gen_m8n2(10,11,boff1,%1,%%r12,1)
-  #define acc_kend_nc3_k1m8(boff1,boff2) unit_acc_gen_m8n2(12,13,boff1,%1,%%r12,2)
-  #define acc_kend_nc4_k1m8(boff1,boff2) unit_acc_gen_m8n2(14,15,boff1,%%r15)
-  #define acc_kend_nc5_k1m8(boff1,boff2) unit_acc_gen_m8n2(16,17,boff1,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m8(boff1,boff2) unit_acc_gen_m8n2(18,19,boff1,%%r15,%%r12,2)
+  #define acc_kend_nc2_k1m8(boff1) unit_acc_gen_m8n2(10,11,boff1,%1,%%r12,1)
+  #define acc_kend_nc3_k1m8(boff1) unit_acc_gen_m8n2(12,13,boff1,%1,%%r12,2)
+  #define acc_kend_nc4_k1m8(boff1) unit_acc_gen_m8n2(14,15,boff1,%%r15)
+  #define acc_kend_nc5_k1m8(boff1) unit_acc_gen_m8n2(16,17,boff1,%%r15,%%r12,1)
+  #define acc_kend_nc6_k1m8(boff1) unit_acc_gen_m8n2(18,19,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m8 "movq %2,%3; addq $64,%2;"
 #ifdef TRMMKERNEL
@@ -258,11 +286,11 @@
 #define KERNEL_k1m4n12 KERNEL_h_k1m4n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m4 "vmovddup (%0,%3,1),%%ymm1; vmovddup 8(%0,%3,1),%%ymm2; addq $32,%3;"
-  #define acc_kend_nc2_k1m4(boff1,boff2) unit_acc_gen_m4n2(6,7,boff1,%1,%%r12,1)
-  #define acc_kend_nc3_k1m4(boff1,boff2) unit_acc_gen_m4n2(8,9,boff1,%1,%%r12,2)
-  #define acc_kend_nc4_k1m4(boff1,boff2) unit_acc_gen_m4n2(10,11,boff1,%%r15)
-  #define acc_kend_nc5_k1m4(boff1,boff2) unit_acc_gen_m4n2(12,13,boff1,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m4(boff1,boff2) unit_acc_gen_m4n2(14,15,boff1,%%r15,%%r12,2)
+  #define acc_kend_nc2_k1m4(boff1) unit_acc_gen_m4n2(6,7,boff1,%1,%%r12,1)
+  #define acc_kend_nc3_k1m4(boff1) unit_acc_gen_m4n2(8,9,boff1,%1,%%r12,2)
+  #define acc_kend_nc4_k1m4(boff1) unit_acc_gen_m4n2(10,11,boff1,%%r15)
+  #define acc_kend_nc5_k1m4(boff1) unit_acc_gen_m4n2(12,13,boff1,%%r15,%%r12,1)
+  #define acc_kend_nc6_k1m4(boff1) unit_acc_gen_m4n2(14,15,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m4 "movq %2,%3; addq $32,%2;"
 #ifdef TRMMKERNEL
@@ -311,11 +339,11 @@
 #define KERNEL_k1m2n12 KERNEL_h_k1m2n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m2 "vmovddup (%0,%3,1),%%xmm1; vmovddup 8(%0,%3,1),%%xmm2; addq $16,%3;"
-  #define acc_kend_nc2_k1m2(boff1,boff2) unit_acc_gen_m2n2(6,7,boff1,%1,%%r12,1)
-  #define acc_kend_nc3_k1m2(boff1,boff2) unit_acc_gen_m2n2(8,9,boff1,%1,%%r12,2)
-  #define acc_kend_nc4_k1m2(boff1,boff2) unit_acc_gen_m2n2(10,11,boff1,%%r15)
-  #define acc_kend_nc5_k1m2(boff1,boff2) unit_acc_gen_m2n2(12,13,boff1,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m2(boff1,boff2) unit_acc_gen_m2n2(14,15,boff1,%%r15,%%r12,2)
+  #define acc_kend_nc2_k1m2(boff1) unit_acc_gen_m2n2(6,7,boff1,%1,%%r12,1)
+  #define acc_kend_nc3_k1m2(boff1) unit_acc_gen_m2n2(8,9,boff1,%1,%%r12,2)
+  #define acc_kend_nc4_k1m2(boff1) unit_acc_gen_m2n2(10,11,boff1,%%r15)
+  #define acc_kend_nc5_k1m2(boff1) unit_acc_gen_m2n2(12,13,boff1,%%r15,%%r12,1)
+  #define acc_kend_nc6_k1m2(boff1) unit_acc_gen_m2n2(14,15,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m2 "movq %2,%3; addq $16,%2;"
 #ifdef TRMMKERNEL
@@ -362,11 +390,11 @@
 #define KERNEL_k1m1n12 KERNEL_h_k1m1n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m1 "vmovddup (%0,%3,1),%%xmm1; addq $8,%3;"
-  #define acc_kend_nc2_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%1,%%r12,1),%%xmm1,%%xmm5;"
-  #define acc_kend_nc3_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%1,%%r12,2),%%xmm1,%%xmm6;"
-  #define acc_kend_nc4_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%%r15),%%xmm1,%%xmm7;"
-  #define acc_kend_nc5_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%%r15,%%r12,1),%%xmm1,%%xmm8;"
-  #define acc_kend_nc6_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%%r15,%%r12,2),%%xmm1,%%xmm9;"
+  #define acc_kend_nc2_k1m1(boff1) "vfmadd231pd "#boff1"(%1,%%r12,1),%%xmm1,%%xmm5;"
+  #define acc_kend_nc3_k1m1(boff1) "vfmadd231pd "#boff1"(%1,%%r12,2),%%xmm1,%%xmm6;"
+  #define acc_kend_nc4_k1m1(boff1) "vfmadd231pd "#boff1"(%%r15),%%xmm1,%%xmm7;"
+  #define acc_kend_nc5_k1m1(boff1) "vfmadd231pd "#boff1"(%%r15,%%r12,1),%%xmm1,%%xmm8;"
+  #define acc_kend_nc6_k1m1(boff1) "vfmadd231pd "#boff1"(%%r15,%%r12,2),%%xmm1,%%xmm9;"
 #endif
 #define save_init_m1 "movq %2,%3; addq $8,%2;"
 #ifdef TRMMKERNEL
diff --git a/kernel/x86_64/dgemm_tcopy_16_skylakex.c b/kernel/x86_64/dgemm_tcopy_16_skylakex.c
new file mode 100644
index 000000000..a1da60f8f
--- /dev/null
+++ b/kernel/x86_64/dgemm_tcopy_16_skylakex.c
@@ -0,0 +1,129 @@
+#include <stdio.h>
+#include "common.h"
+#include <immintrin.h>
+
+int CNAME(BLASLONG dim_second, BLASLONG dim_first, double *src, BLASLONG lead_dim, double *dst){
+  double *src1, *src2, *src3, *src4, *dst1;
+  __m512d z1,z2,z3,z4,z5,z6,z7,z8; __m256d y1,y2,y3,y4; __m128d x1,x2,x3,x4; double s1,s2,s3,s4;
+  BLASLONG dim1_count, dim2_count, src_inc;
+  src_inc = 4 * lead_dim - dim_first;
+  src1 = src; src2 = src + lead_dim; src3 = src2 + lead_dim; src4 = src3 + lead_dim;
+  for(dim2_count=dim_second; dim2_count>3; dim2_count-=4){
+    dst1 = dst + 16 * (dim_second - dim2_count);
+    for(dim1_count=dim_first; dim1_count>15; dim1_count-=16){
+      z1 = _mm512_loadu_pd(src1); z2 = _mm512_loadu_pd(src1+8); src1 += 16;
+      z3 = _mm512_loadu_pd(src2); z4 = _mm512_loadu_pd(src2+8); src2 += 16;
+      z5 = _mm512_loadu_pd(src3); z6 = _mm512_loadu_pd(src3+8); src3 += 16;
+      z7 = _mm512_loadu_pd(src4); z8 = _mm512_loadu_pd(src4+8); src4 += 16;
+      _mm512_storeu_pd(dst1+ 0,z1); _mm512_storeu_pd(dst1+ 8,z2);
+      _mm512_storeu_pd(dst1+16,z3); _mm512_storeu_pd(dst1+24,z4);
+      _mm512_storeu_pd(dst1+32,z5); _mm512_storeu_pd(dst1+40,z6);
+      _mm512_storeu_pd(dst1+48,z7); _mm512_storeu_pd(dst1+56,z8); dst1 += 16 * dim_second;
+    }
+    dst1 -= 8 * (dim_second - dim2_count);
+    if(dim1_count>7){
+      z1 = _mm512_loadu_pd(src1); src1 += 8;
+      z2 = _mm512_loadu_pd(src2); src2 += 8;
+      z3 = _mm512_loadu_pd(src3); src3 += 8;
+      z4 = _mm512_loadu_pd(src4); src4 += 8;
+      _mm512_storeu_pd(dst1+ 0,z1); _mm512_storeu_pd(dst1+ 8,z2);
+      _mm512_storeu_pd(dst1+16,z3); _mm512_storeu_pd(dst1+24,z4); dst1 += 8 * dim_second;
+      dim1_count -= 8;
+    }
+    dst1 -= 4 * (dim_second - dim2_count);
+    if(dim1_count>3){
+      y1 = _mm256_loadu_pd(src1); src1 += 4;
+      y2 = _mm256_loadu_pd(src2); src2 += 4;
+      y3 = _mm256_loadu_pd(src3); src3 += 4;
+      y4 = _mm256_loadu_pd(src4); src4 += 4;
+      _mm256_storeu_pd(dst1+ 0,y1); _mm256_storeu_pd(dst1+ 4,y2);
+      _mm256_storeu_pd(dst1+ 8,y3); _mm256_storeu_pd(dst1+12,y4); dst1 += 4 * dim_second;
+      dim1_count -= 4;
+    }
+    dst1 -= 2 * (dim_second - dim2_count);
+    if(dim1_count>1){
+      x1 = _mm_loadu_pd(src1); src1 += 2;
+      x2 = _mm_loadu_pd(src2); src2 += 2;
+      x3 = _mm_loadu_pd(src3); src3 += 2;
+      x4 = _mm_loadu_pd(src4); src4 += 2;
+      _mm_storeu_pd(dst1+0,x1); _mm_storeu_pd(dst1+2,x2);
+      _mm_storeu_pd(dst1+4,x3); _mm_storeu_pd(dst1+6,x4); dst1 += 2 * dim_second;
+      dim1_count -= 2;
+    }
+    dst1 -= dim_second - dim2_count;
+    if(dim1_count>0){
+      s1 = *src1; src1++; s2 = *src2; src2++; s3 = *src3; src3++; s4 = *src4; src4++;
+      dst1[0] = s1; dst1[1] = s2; dst1[2] = s3; dst1[3] = s4;
+    }
+    src1 += src_inc; src2 += src_inc; src3 += src_inc; src4 += src_inc;
+  }
+  src_inc -= 2 * lead_dim;
+  for(; dim2_count>1; dim2_count-=2){
+    dst1 = dst + 16 * (dim_second - dim2_count);
+    for(dim1_count=dim_first; dim1_count>15; dim1_count-=16){
+      z1 = _mm512_loadu_pd(src1); z2 = _mm512_loadu_pd(src1+8); src1 += 16;
+      z3 = _mm512_loadu_pd(src2); z4 = _mm512_loadu_pd(src2+8); src2 += 16;
+      _mm512_storeu_pd(dst1+ 0,z1); _mm512_storeu_pd(dst1+ 8,z2);
+      _mm512_storeu_pd(dst1+16,z3); _mm512_storeu_pd(dst1+24,z4); dst1 += 16 * dim_second;
+    }
+    dst1 -= 8 * (dim_second - dim2_count);
+    if(dim1_count>7){
+      z1 = _mm512_loadu_pd(src1); src1 += 8;
+      z2 = _mm512_loadu_pd(src2); src2 += 8;
+      _mm512_storeu_pd(dst1+ 0,z1); _mm512_storeu_pd(dst1+ 8,z2); dst1 += 8 * dim_second;
+      dim1_count -= 8;
+    }
+    dst1 -= 4 * (dim_second - dim2_count);
+    if(dim1_count>3){
+      y1 = _mm256_loadu_pd(src1); src1 += 4;
+      y2 = _mm256_loadu_pd(src2); src2 += 4;
+      _mm256_storeu_pd(dst1+ 0,y1); _mm256_storeu_pd(dst1+ 4,y2); dst1 += 4 * dim_second;
+      dim1_count -= 4;
+    }
+    dst1 -= 2 * (dim_second - dim2_count);
+    if(dim1_count>1){
+      x1 = _mm_loadu_pd(src1); src1 += 2;
+      x2 = _mm_loadu_pd(src2); src2 += 2;
+      _mm_storeu_pd(dst1+0,x1); _mm_storeu_pd(dst1+2,x2); dst1 += 2 * dim_second;
+      dim1_count -= 2;
+    }
+    dst1 -= dim_second - dim2_count;
+    if(dim1_count>0){
+      s1 = *src1; src1++; s2 = *src2; src2++;
+      dst1[0] = s1; dst1[1] = s2;
+    }
+    src1 += src_inc; src2 += src_inc;
+  }
+  src_inc -= lead_dim;
+  for(; dim2_count>0; dim2_count--){
+    dst1 = dst + 16 * (dim_second - dim2_count);
+    for(dim1_count=dim_first; dim1_count>15; dim1_count-=16){
+      z1 = _mm512_loadu_pd(src1); z2 = _mm512_loadu_pd(src1+8); src1 += 16;
+      _mm512_storeu_pd(dst1+ 0,z1); _mm512_storeu_pd(dst1+ 8,z2); dst1 += 16 * dim_second;
+    }
+    dst1 -= 8 * (dim_second - dim2_count);
+    if(dim1_count>7){
+      z1 = _mm512_loadu_pd(src1); src1 += 8;
+      _mm512_storeu_pd(dst1+ 0,z1); dst1 += 8 * dim_second;
+      dim1_count -= 8;
+    }
+    dst1 -= 4 * (dim_second - dim2_count);
+    if(dim1_count>3){
+      y1 = _mm256_loadu_pd(src1); src1 += 4;
+      _mm256_storeu_pd(dst1+ 0,y1); dst1 += 4 * dim_second;
+      dim1_count -= 4;
+    }
+    dst1 -= 2 * (dim_second - dim2_count);
+    if(dim1_count>1){
+      x1 = _mm_loadu_pd(src1); src1 += 2;
+      _mm_storeu_pd(dst1+0,x1); dst1 += 2 * dim_second;
+      dim1_count -= 2;
+    }
+    dst1 -= dim_second - dim2_count;
+    if(dim1_count>0){
+      s1 = *src1; src1++;
+      dst1[0] = s1;
+    }
+    src1 += src_inc;
+  }
+}
diff --git a/kernel/x86_64/dsymv_L_microk_skylakex-2.c b/kernel/x86_64/dsymv_L_microk_skylakex-2.c
index bdcd914fb..f0df5aaa8 100644
--- a/kernel/x86_64/dsymv_L_microk_skylakex-2.c
+++ b/kernel/x86_64/dsymv_L_microk_skylakex-2.c
@@ -36,7 +36,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(__clang_patchlevel__) && __clang_major__ == 9 && __clang_minor__ == 0 && __clang_patchlevel__ == 0
 #pragma clang optimize off
 #endif
-
+#if defined(__apple_build_version__) && __clang_major__ == 11 && __clang_minor__ == 0 && __clang_patchlevel__ == 3
+#pragma clang optimize off
+#endif
 static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
 {
 
@@ -164,6 +166,9 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 #if defined(__clang_patchlevel__) && __clang_major__ == 9 && __clang_minor__ == 0 && __clang_patchlevel__ == 0
 #pragma clang optimize on
 #endif
+#if defined(__apple_build_version__) && __clang_major__ == 11 && __clang_minor__ == 0 && __clang_patchlevel__ == 3
+#pragma clang optimize on
+#endif
 
 #else
 #include "dsymv_L_microk_haswell-2.c"
diff --git a/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c b/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c
index 5ab3e6d1f..a2e78c58d 100644
--- a/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c
+++ b/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c
@@ -1,4 +1,4 @@
-/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store, %6 = b_pref */
+/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store, %6 = b_pref */
 /* r10 = tmp, r11 = m_counter, r12 = k << 2(const), r13 = tmp, r14 = b_head_pos(const), r15 = tmp */
 
 /* m = 8 *//* ymm0 for alpha, ymm1-ymm3 for temporary use, ymm4-ymm15 for accumulators */
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c b/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
index 4131debb1..5410bd4ae 100644
--- a/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
@@ -1,4 +1,4 @@
-#include "common.h"
+#include "common.h"
 #include <stdint.h>
 #include "strsm_kernel_8x4_haswell_L_common.h"
 
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h b/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
index cfa56da97..2862a5b8d 100644
--- a/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
@@ -1,4 +1,4 @@
-/* r11 = m_counter, r12 = size_of_k_elements, r13 = kk, r14 = b_head, r15 = a_head */
+/* r11 = m_counter, r12 = size_of_k_elements, r13 = kk, r14 = b_head, r15 = a_head */
 /* register i/o: %0 = a_ptr, %1 = b_ptr, %2 = c_ptr, %3 = c_tmp, %4 = ldc, %5 = k_counter */
 /* memory input: %6 = K, %7 = offset, %8 = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}, %9 = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}, %10 = M */
 
diff --git a/kernel/zarch/KERNEL.Z14 b/kernel/zarch/KERNEL.Z14
index f6e3bec23..3510938a7 100644
--- a/kernel/zarch/KERNEL.Z14
+++ b/kernel/zarch/KERNEL.Z14
@@ -86,24 +86,24 @@ DGEMVTKERNEL = dgemv_t_4.c
 CGEMVTKERNEL = cgemv_t_4.c
 ZGEMVTKERNEL = zgemv_t_4.c
 
-STRMMKERNEL	= strmm8x4V.S
-DTRMMKERNEL	= trmm8x4V.S
+STRMMKERNEL	= gemm_vec.c
+DTRMMKERNEL	= gemm_vec.c
 CTRMMKERNEL	= ctrmm4x4V.S
 ZTRMMKERNEL	= ztrmm4x4V.S
 
-SGEMMKERNEL    =  strmm8x4V.S
-SGEMMINCOPY    = ../generic/gemm_ncopy_8.c
-SGEMMITCOPY    = ../generic/gemm_tcopy_8.c
-SGEMMONCOPY    = ../generic/gemm_ncopy_4.c
-SGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
+SGEMMKERNEL    = gemm_vec.c
+ifneq ($(SGEMM_UNROLL_M),$(SGEMM_UNROLL_N))
+SGEMMINCOPY    = ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+SGEMMITCOPY    = ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
 SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+SGEMMONCOPY    = ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
+SGEMMOTCOPY    = ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
 SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-
- 
-DGEMMKERNEL    =  gemm8x4V.S
+DGEMMKERNEL    = gemm_vec.c
 DGEMMINCOPY    = ../generic/gemm_ncopy_8.c
 DGEMMITCOPY    = ../generic/gemm_tcopy_8.c
 DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
@@ -145,7 +145,3 @@ ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
 ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
 ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
 
-
-
-
-
diff --git a/kernel/zarch/gemm_vec.c b/kernel/zarch/gemm_vec.c
new file mode 100644
index 000000000..eb6d7700b
--- /dev/null
+++ b/kernel/zarch/gemm_vec.c
@@ -0,0 +1,475 @@
+/*
+ * Copyright (c) IBM Corporation 2020.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    1. Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *    3. Neither the name of the OpenBLAS project nor the names of
+ *       its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written
+ *       permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include "common.h"
+#include <vecintrin.h>
+
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#ifdef COMPLEX
+#error "Handling for complex numbers is not supported in this kernel"
+#endif
+
+#ifdef DOUBLE
+#define UNROLL_M DGEMM_DEFAULT_UNROLL_M
+#define UNROLL_N DGEMM_DEFAULT_UNROLL_N
+#else
+#define UNROLL_M SGEMM_DEFAULT_UNROLL_M
+#define UNROLL_N SGEMM_DEFAULT_UNROLL_N
+#endif
+
+static const size_t unroll_m = UNROLL_M;
+static const size_t unroll_n = UNROLL_N;
+
+/* Handling of triangular matrices */
+#ifdef TRMMKERNEL
+static const bool trmm = true;
+static const bool left =
+#ifdef LEFT
+	true;
+#else
+	false;
+#endif
+
+static const bool backwards =
+#if defined(LEFT) != defined(TRANSA)
+	true;
+#else
+	false;
+#endif
+
+#else
+static const bool trmm = false;
+static const bool left = false;
+static const bool backwards = false;
+#endif /* TRMMKERNEL */
+
+/*
+ * Background:
+ *
+ * The algorithm of GotoBLAS / OpenBLAS breaks down the matrix multiplication
+ * problem by splitting all matrices into partitions multiple times, so that the
+ * submatrices fit into the L1 or L2 caches. As a result, each multiplication of
+ * submatrices can stream data fast from L1 and L2 caches. Inbetween, it copies
+ * and rearranges the submatrices to enable contiguous memory accesses to
+ * improve locality in both caches and TLBs.
+ *
+ * At the heart of the algorithm is this kernel, which multiplies, a "Block
+ * matrix" A (small dimensions) with a "Panel matrix" B (number of rows is
+ * small) and adds the result into a "Panel matrix" C; GotoBLAS calls this
+ * operation GEBP. This kernel further partitions GEBP twice, such that (1)
+ * submatrices of C and B fit into the L1 caches (GEBP_column_block) and (2) a
+ * block of C fits into the registers, while multiplying panels from A and B
+ * streamed from the L2 and L1 cache, respectively (GEBP_block).
+ *
+ *
+ * Algorithm GEBP(A, B, C, m, n, k, alpha):
+ *
+ * The problem is calculating C += alpha * (A * B)
+ * C is an m x n matrix, A is an m x k matrix, B is an k x n matrix.
+ *
+ * - C is in column-major-order, with an offset of ldc to the element in the
+ *   next column (same row).
+ * - A is in row-major-order yet stores SGEMM_UNROLL_M elements of each column
+ *   contiguously while walking along rows.
+ * - B is in column-major-order but packs SGEMM_UNROLL_N elements of a row
+ *   contiguously.
+ * If the numbers of rows and columns are not multiples of SGEMM_UNROLL_M or
+ * SGEMM_UNROLL_N, the remaining elements are arranged in blocks with power-of-2
+ * dimensions (e.g., 5 remaining columns would be in a block-of-4 and a
+ * block-of-1).
+ *
+ * Note that packing A and B into that form is taken care of by the caller in
+ * driver/level3/level3.c (actually done by "copy kernels").
+ *
+ * Steps:
+ * - Partition C and B into blocks of n_r (SGEMM_UNROLL_N) columns, C_j and B_j.
+ *   Now, B_j should fit into the L1 cache.
+ * - For each partition, calculate C_j += alpha * (A * B_j) by
+ *     (1) Calculate C_aux := A * B_j (see below)
+ *     (2) unpack C_j = C_j + alpha * C_aux
+ *
+ *
+ * Algorithm for Calculating C_aux:
+ *
+ * - Further partition C_aux and A into groups of m_r (SGEMM_UNROLL_M) rows,
+ *   such that the m_r x n_r-submatrix of C_aux can be held in registers. Each
+ *   submatrix of C_aux can be calculated independently, and the registers are
+ *   added back into C_j.
+ *
+ * - For each row-block of C_aux:
+ *   (uses a row block of A and full B_j)
+ *    - stream over all columns of A, multiply with elements from B and
+ *      accumulate in registers. (use different inner-kernels to exploit
+ *      vectorization for varying block sizes)
+ *    - add alpha * row block of C_aux back into C_j.
+ *
+ * Note that there are additional mechanics for handling triangular matrices,
+ * calculating B := alpha (A * B) where either of the matrices A or B can be
+ * triangular. In case of A, the macro "LEFT" is defined. In addition, A can
+ * optionally be transposed.
+ * The code effectively skips an "offset" number of columns in A and rows of B
+ * in each block, to save unnecessary work by exploiting the triangular nature.
+ * To handle all cases, the code discerns (1) a "left" mode when A is triangular
+ * and (2) "forward" / "backwards" modes where only the first "offset"
+ * columns/rows of A/B are used or where the first "offset" columns/rows are
+ * skipped, respectively.
+ *
+ * Reference:
+ *
+ * The summary above is based on staring at various kernel implementations and:
+ * K. Goto and R. A. Van de Geijn, Anatomy of High-Performance Matrix
+ * Multiplication, in ACM Transactions of Mathematical Software, Vol.  34, No.
+ * 3, May 2008.
+ */
+
+#define VLEN_BYTES 16
+#define VLEN_FLOATS (VLEN_BYTES / sizeof(FLOAT))
+
+typedef FLOAT vector_float __attribute__ ((vector_size (16)));
+
+/**
+ * Load a vector into register, and hint on 8-byte alignment to improve
+ * performance. gcc-9 and newer will create these hints by itself. For older
+ * compiler versions, use inline assembly to explicitly express the hint.
+ * Provide explicit hex encoding to cater for binutils versions that do not know
+ * about vector-load with alignment hints yet.
+ *
+ * Note that, for block sizes where we apply vectorization, vectors in A will
+ * always be 8-byte aligned.
+ */
+static inline vector_float vec_load_hinted(FLOAT const *restrict a) {
+	vector_float const *restrict addr = (vector_float const *restrict)a;
+	vector_float y;
+
+#if __GNUC__ < 9
+	// hex-encode vl %[out],%[addr],3
+	asm(".insn vrx,0xe70000003006,%[out],%[addr],3"
+	    : [ out ] "=v"(y)
+	    : [ addr ] "R"(*addr));
+#else
+	y = *addr;
+#endif
+
+	return y;
+}
+
+/**
+ * Calculate for a row-block in C_i of size ROWSxCOLS using vector intrinsics.
+ *
+ * @param[in] 	A	Pointer current block of input matrix A.
+ * @param[in]	k	Number of columns in A.
+ * @param[in]	B	Pointer current block of input matrix B.
+ * @param[inout] C	Pointer current block of output matrix C.
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	alpha	Scalar factor.
+ */
+#define VECTOR_BLOCK(ROWS, COLS)                                              \
+	static inline void GEBP_block_##ROWS##_##COLS(                        \
+	    FLOAT const *restrict A, BLASLONG bk, FLOAT const *restrict B,    \
+	    FLOAT *restrict C, BLASLONG ldc, FLOAT alpha) {                   \
+		_Static_assert(                                               \
+		    ROWS % VLEN_FLOATS == 0,                                  \
+		    "rows in block must be multiples of vector length");      \
+		vector_float Caux[ROWS / VLEN_FLOATS][COLS];                  \
+                                                                              \
+		for (BLASLONG i = 0; i < ROWS / VLEN_FLOATS; i++) {           \
+			vector_float A0 =                                     \
+			    vec_load_hinted(A + i * VLEN_FLOATS);             \
+			for (BLASLONG j = 0; j < COLS; j++)                   \
+				Caux[i][j] = A0 * B[j];                       \
+		}                                                             \
+                                                                              \
+		/*                                                            \
+		 * Stream over the row-block of A, which is packed            \
+		 * column-by-column, multiply by coefficients in B and add up \
+		 * into temporaries Caux (which the compiler will hold in     \
+		 * registers). Vectorization: Multiply column vectors from A  \
+		 * with scalars from B and add up in column vectors of Caux.  \
+		 * That equates to unrolling the loop over rows (in i) and    \
+		 * executing each unrolled iteration as a vector element.     \
+		 */                                                           \
+		for (BLASLONG k = 1; k < bk; k++) {                           \
+			for (BLASLONG i = 0; i < ROWS / VLEN_FLOATS; i++) {   \
+				vector_float Ak = vec_load_hinted(            \
+				    A + i * VLEN_FLOATS + k * ROWS);          \
+                                                                              \
+				for (BLASLONG j = 0; j < COLS; j++)           \
+					Caux[i][j] += Ak * B[j + k * COLS];   \
+			}                                                     \
+		}                                                             \
+                                                                              \
+		/*                                                            \
+		 * Unpack row-block of C_aux into outer C_i, multiply by      \
+		 * alpha and add up.                                          \
+		 */                                                           \
+		for (BLASLONG j = 0; j < COLS; j++) {                         \
+			for (BLASLONG i = 0; i < ROWS / VLEN_FLOATS; i++) {   \
+				vector_float *C_ij =                          \
+				    (vector_float *)(C + i * VLEN_FLOATS +    \
+						     j * ldc);                \
+				if (trmm) {                                   \
+					*C_ij = alpha * Caux[i][j];           \
+				} else {                                      \
+					*C_ij += alpha * Caux[i][j];          \
+				}                                             \
+			}                                                     \
+		}                                                             \
+	}
+
+
+#if UNROLL_M == 16
+VECTOR_BLOCK(16, 4)
+VECTOR_BLOCK(16, 2)
+VECTOR_BLOCK(16, 1)
+#endif
+#if UNROLL_N == 8
+VECTOR_BLOCK(8, 8)
+VECTOR_BLOCK(4, 8)
+#endif
+VECTOR_BLOCK(8, 4)
+VECTOR_BLOCK(8, 2)
+VECTOR_BLOCK(8, 1)
+VECTOR_BLOCK(4, 4)
+VECTOR_BLOCK(4, 2)
+VECTOR_BLOCK(4, 1)
+
+#ifdef DOUBLE
+VECTOR_BLOCK(2, 4)
+VECTOR_BLOCK(2, 2)
+#endif
+
+/**
+ * Handle calculation for row blocks in C_i of any size by dispatching into
+ * macro-defined (inline) functions or by deferring to a simple generic
+ * implementation. Note that the compiler can remove this awkward-looking
+ * dispatching code while inlineing.
+ *
+ * @param[in]	m	Number of rows in block C_i.
+ * @param[in]	n	Number of columns in block C_i.
+ * @param[in]	first_row Index of first row of the block C_i (relative to C).
+ * @param[in]	A	Pointer to input matrix A (note: all of it).
+ * @param[in]	k	Number of columns in A and rows in B.
+ * @param[in]	B	Pointer to current column block (panel) of input matrix B.
+ * @param[inout] C	Pointer to current column block (panel) of output matrix C.
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	alpha	Scalar factor.
+ * @param[in]	offset  Number of columns of A and rows of B to skip (for triangular matrices).
+ * @param[in]	off	Running offset for handling triangular matrices.
+ */
+static inline void GEBP_block(BLASLONG m, BLASLONG n,
+		       BLASLONG first_row,
+		       const FLOAT * restrict A, BLASLONG k,
+		       const FLOAT * restrict B,
+		       FLOAT *restrict C, BLASLONG ldc,
+		       FLOAT alpha,
+		       BLASLONG offset, BLASLONG off)
+{
+	if (trmm && left)
+		off = offset + first_row;
+
+	A += first_row * k;
+	C += first_row;
+
+	if (trmm) {
+		if (backwards) {
+			A += off * m;
+			B += off * n;
+			k -= off;
+		} else {
+			if (left) {
+				k = off + m;
+			} else {
+				k = off + n;
+			}
+		}
+	}
+
+#define BLOCK(bm, bn)                                           \
+	if (m == bm && n == bn) {                               \
+		GEBP_block_##bm##_##bn(A, k, B, C, ldc, alpha); \
+		return;                                         \
+	}
+
+#if UNROLL_M == 16
+	BLOCK(16, 4); BLOCK(16, 2); BLOCK(16, 1);
+#endif
+#if UNROLL_N == 8
+	BLOCK(8, 8); BLOCK(4, 8);
+#endif
+	BLOCK(8, 4); BLOCK(8, 2); BLOCK(8, 1);
+	BLOCK(4, 4); BLOCK(4, 2); BLOCK(4, 1);
+
+	#ifdef DOUBLE
+	BLOCK(2, 4);
+	BLOCK(2, 2);
+	#endif
+
+#undef BLOCK
+
+	/* simple implementation for smaller block sizes: */
+	FLOAT Caux[m][n] __attribute__ ((aligned (16)));
+
+	/*
+	 * Peel off first iteration (i.e., column of A) for initializing Caux
+	 */
+	for (BLASLONG i = 0; i < m; i++)
+		for (BLASLONG j = 0; j < n; j++)
+			Caux[i][j] = A[i] * B[j];
+
+	for (BLASLONG kk = 1; kk < k; kk++)
+		for (BLASLONG i = 0; i < m; i++)
+			for (BLASLONG j = 0; j < n; j++)
+				Caux[i][j] += A[i + kk * m] * B[j + kk * n];
+
+	for (BLASLONG i = 0; i < m; i++)
+		for (BLASLONG j = 0; j < n; j++)
+			if (trmm) {
+				C[i + j * ldc] = alpha * Caux[i][j];
+			} else {
+				C[i + j * ldc] += alpha * Caux[i][j];
+			}
+}
+
+/**
+ * Handle a column block (panel) of C and B while calculating C += alpha(A * B).
+ *
+ * @param[in]	num_cols	Number of columns in the block (in C and B).
+ * @param[in]	first_col	First column of the current block (in C and B).
+ * @param[in]	A	Pointer to input matrix A.
+ * @param[in]	bk	Number of columns in A and rows in B.
+ * @param[in]	B	Pointer to input matrix B (note: all of it).
+ * @param[in]	bm	Number of rows in C and A.
+ * @param[inout] C	Pointer to output matrix C (note: all of it).
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	alpha	Scalar factor.
+ * @param[in]	offset	Number of columns of A and rows of B to skip (for triangular matrices).
+ */
+static inline void GEBP_column_block(BLASLONG num_cols, BLASLONG first_col,
+			const FLOAT *restrict A, BLASLONG bk,
+			const FLOAT *restrict B, BLASLONG bm,
+			FLOAT *restrict C, BLASLONG ldc,
+			FLOAT alpha,
+			BLASLONG const offset) {
+
+	FLOAT *restrict C_i = C + first_col * ldc;
+	/*
+	 * B is in column-order with n_r packed row elements, which does
+	 * not matter -- we always move in full such blocks of
+	 * column*pack
+	 */
+	const FLOAT *restrict B_i = B + first_col * bk;
+
+	BLASLONG off = 0;
+	if (trmm) {
+		if (left) {
+			off = offset;
+		} else {
+			off = -offset + first_col;
+		}
+	}
+
+	/*
+	 * Calculate C_aux := A * B_j
+	 * then unpack C_i += alpha * C_aux.
+	 *
+	 * For that purpose, further partition C_aux and A into blocks
+	 * of m_r (unroll_m) rows, or powers-of-2 if smaller.
+	 */
+	BLASLONG row = 0;
+	for (BLASLONG block_size = unroll_m; block_size > 0; block_size /= 2)
+		for (; bm - row >= block_size; row += block_size)
+			GEBP_block(block_size, num_cols, row, A, bk, B_i, C_i,
+				   ldc, alpha, offset, off);
+}
+
+/**
+ * Inner kernel for matrix-matrix multiplication. C += alpha (A * B)
+ * where C is an m-by-n matrix, A is m-by-k and B is k-by-n. Note that A, B, and
+ * C are pointers to submatrices of the actual matrices.
+ *
+ * For triangular matrix multiplication, calculate B := alpha (A * B) where A
+ * or B can be triangular (in case of A, the macro LEFT will be defined).
+ *
+ * @param[in]	bm	Number of rows in C and A.
+ * @param[in]	bn	Number of columns in C and B.
+ * @param[in]	bk	Number of columns in A and rows in B.
+ * @param[in]	alpha	Scalar factor.
+ * @param[in]	ba	Pointer to input matrix A.
+ * @param[in]	bb	Pointer to input matrix B.
+ * @param[inout] C	Pointer to output matrix C.
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	offset	Number of columns of A and rows of B to skip (for triangular matrices).
+ * @returns 0 on success.
+ */
+int CNAME(BLASLONG bm, BLASLONG bn, BLASLONG bk, FLOAT alpha,
+	  FLOAT *restrict ba, FLOAT *restrict bb,
+	  FLOAT *restrict C, BLASLONG ldc
+#ifdef TRMMKERNEL
+	  , BLASLONG offset
+#endif
+	  )
+{
+	if ( (bm == 0) || (bn == 0) || (bk == 0) || (alpha == ZERO))
+		return 0;
+
+	/*
+	 * interface code allocates buffers for ba and bb at page
+	 * granularity (i.e., using mmap(MAP_ANONYMOUS), so enable the compiler
+	 * to make use of the fact in vector load operations.
+	 */
+	ba = __builtin_assume_aligned(ba, 16);
+	bb = __builtin_assume_aligned(bb, 16);
+
+	/*
+	 * Use offset and off even when compiled as SGEMMKERNEL to simplify
+	 * function signatures and function calls.
+	 */
+#ifndef TRMMKERNEL
+	BLASLONG const offset = 0;
+#endif
+
+	/*
+	 * Partition B and C into blocks of n_r (unroll_n) columns, called B_i
+	 * and C_i. For each partition, calculate C_i += alpha * (A * B_j).
+	 *
+	 * For remaining columns that do not fill up a block of n_r, iteratively
+	 * use smaller block sizes of powers of 2.
+	 */
+	BLASLONG col = 0;
+	for (BLASLONG block_size = unroll_n; block_size > 0; block_size /= 2)
+		for (; bn - col >= block_size; col += block_size)
+			GEBP_column_block(block_size, col, ba, bk, bb, bm, C, ldc, alpha, offset);
+
+   return 0;
+}
diff --git a/lapack-netlib/LAPACKE/include/lapacke.h b/lapack-netlib/LAPACKE/include/lapacke.h
index 6eb0b696b..012c104bb 100644
--- a/lapack-netlib/LAPACKE/include/lapacke.h
+++ b/lapack-netlib/LAPACKE/include/lapacke.h
@@ -12575,7 +12575,7 @@ lapack_int LAPACKE_zhetrs_aa_2stage_work( int matrix_layout, char uplo, lapack_i
                    
 /* APIs for set/get nancheck flags */
 void LAPACKE_set_nancheck( int flag );
-int LAPACKE_get_nancheck( );
+int LAPACKE_get_nancheck( void );
 
 #ifdef __cplusplus
 }
diff --git a/lapack-netlib/SRC/clargv.f b/lapack-netlib/SRC/clargv.f
index ba53cae6f..36c5108df 100644
--- a/lapack-netlib/SRC/clargv.f
+++ b/lapack-netlib/SRC/clargv.f
@@ -200,7 +200,7 @@
             FS = FS*SAFMN2
             GS = GS*SAFMN2
             SCALE = SCALE*SAFMN2
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $         GO TO 10
          ELSE IF( SCALE.LE.SAFMN2 ) THEN
             IF( G.EQ.CZERO ) THEN
diff --git a/lapack-netlib/SRC/clartg.f b/lapack-netlib/SRC/clartg.f
index da9a1cdef..baa68b657 100644
--- a/lapack-netlib/SRC/clartg.f
+++ b/lapack-netlib/SRC/clartg.f
@@ -161,7 +161,7 @@
          FS = FS*SAFMN2
          GS = GS*SAFMN2
          SCALE = SCALE*SAFMN2
-         IF( SCALE.GE.SAFMX2 )
+         IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $      GO TO 10
       ELSE IF( SCALE.LE.SAFMN2 ) THEN
          IF( G.EQ.CZERO.OR.SISNAN( ABS( G ) ) ) THEN
diff --git a/lapack-netlib/SRC/dlartg.f b/lapack-netlib/SRC/dlartg.f
index 1c7c46f63..dc49986a0 100644
--- a/lapack-netlib/SRC/dlartg.f
+++ b/lapack-netlib/SRC/dlartg.f
@@ -163,7 +163,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/dlartgp.f b/lapack-netlib/SRC/dlartgp.f
index 0cb0d2d13..334e416e8 100644
--- a/lapack-netlib/SRC/dlartgp.f
+++ b/lapack-netlib/SRC/dlartgp.f
@@ -161,7 +161,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/slartg.f b/lapack-netlib/SRC/slartg.f
index 784d4bc36..307c9c83a 100644
--- a/lapack-netlib/SRC/slartg.f
+++ b/lapack-netlib/SRC/slartg.f
@@ -163,7 +163,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/slartgp.f b/lapack-netlib/SRC/slartgp.f
index ad76c94b4..f8be5f52b 100644
--- a/lapack-netlib/SRC/slartgp.f
+++ b/lapack-netlib/SRC/slartgp.f
@@ -161,7 +161,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/zheequb.f b/lapack-netlib/SRC/zheequb.f
index d698232e8..7d719f41e 100644
--- a/lapack-netlib/SRC/zheequb.f
+++ b/lapack-netlib/SRC/zheequb.f
@@ -271,7 +271,7 @@
          AVG = AVG / N
 
          STD = 0.0D0
-         DO I = N+1, N
+         DO I = N+1, 2*N
             WORK( I ) = S( I-N ) * WORK( I-N ) - AVG
          END DO
          CALL ZLASSQ( N, WORK( N+1 ), 1, SCALE, SUMSQ )
diff --git a/lapack-netlib/SRC/zlargv.f b/lapack-netlib/SRC/zlargv.f
index 1e17983d5..f83ca1851 100644
--- a/lapack-netlib/SRC/zlargv.f
+++ b/lapack-netlib/SRC/zlargv.f
@@ -201,7 +201,7 @@
             FS = FS*SAFMN2
             GS = GS*SAFMN2
             SCALE = SCALE*SAFMN2
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $         GO TO 10
          ELSE IF( SCALE.LE.SAFMN2 ) THEN
             IF( G.EQ.CZERO ) THEN
diff --git a/lapack-netlib/SRC/zlartg.f b/lapack-netlib/SRC/zlartg.f
index 8989bb896..894b4ded0 100644
--- a/lapack-netlib/SRC/zlartg.f
+++ b/lapack-netlib/SRC/zlartg.f
@@ -161,7 +161,7 @@
          FS = FS*SAFMN2
          GS = GS*SAFMN2
          SCALE = SCALE*SAFMN2
-         IF( SCALE.GE.SAFMX2 )
+         IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $      GO TO 10
       ELSE IF( SCALE.LE.SAFMN2 ) THEN
          IF( G.EQ.CZERO.OR.DISNAN( ABS( G ) ) ) THEN
diff --git a/lapack/CMakeLists.txt b/lapack/CMakeLists.txt
index e21a9aabb..778e6f8fa 100644
--- a/lapack/CMakeLists.txt
+++ b/lapack/CMakeLists.txt
@@ -2,6 +2,7 @@
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_BINARY_DIR})
 
+list (REMOVE_ITEM FLOAT_TYPES "HALF")
 
 set(LAPACK_SOURCES
   potrf/potrf_U_single.c
@@ -45,6 +46,9 @@ GenerateNamedObjects("laswp/generic/laswp_k_4.c" "" "laswp_plus" false "" ""  fa
 GenerateNamedObjects("laswp/generic/laswp_k_4.c" "MINUS" "laswp_minus" false "" ""  false 3)
 
 foreach (float_type ${FLOAT_TYPES})
+if (${float_type} STREQUAL "HALF")
+ continue()
+endif()
 GenerateNamedObjects("getrf/getrf_single.c" "UNIT" "getrf_single" false "" "" false ${float_type})
 endforeach ()
 
diff --git a/lapack/getrf/potrf_parallel.c b/lapack/getrf/potrf_parallel.c
index c2fee6bd1..312509685 100644
--- a/lapack/getrf/potrf_parallel.c
+++ b/lapack/getrf/potrf_parallel.c
@@ -380,6 +380,9 @@ static int thread_driver(blas_arg_t *args, FLOAT *sa, FLOAT *sb){
 #elif defined(DOUBLE)
   mode  =  BLAS_DOUBLE  | BLAS_REAL;
   mask  = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
+#elif defined(HALF)
+  mode  =  BLAS_HALF  | BLAS_REAL;
+  mask  = MAX(SHGEMM_UNROLL_M, SHGEMM_UNROLL_N) - 1;
 #else
   mode  =  BLAS_SINGLE  | BLAS_REAL;
   mask  = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;
diff --git a/param.h b/param.h
index d6cbe544a..e8cf53f0a 100644
--- a/param.h
+++ b/param.h
@@ -72,6 +72,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef PARAM_H
 #define PARAM_H
 
+#define SHGEMM_DEFAULT_UNROLL_N 4
+#define SHGEMM_DEFAULT_UNROLL_M 8
+#define SHGEMM_DEFAULT_UNROLL_MN 32
+#define SHGEMM_DEFAULT_P 256
+#define SHGEMM_DEFAULT_R 256
+#define SHGEMM_DEFAULT_Q 256
 #ifdef OPTERON
 
 #define SNUMOPT		4
@@ -1968,7 +1974,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_UNROLL_N 4
 #define DGEMM_DEFAULT_UNROLL_M 4
 #define DGEMM_DEFAULT_UNROLL_N 4
-#define CGEMM_DEFAULT_UNROLL_M 8
+#define CGEMM_DEFAULT_UNROLL_M 2
 #define CGEMM_DEFAULT_UNROLL_N 2
 #define ZGEMM_DEFAULT_UNROLL_M 2
 #define ZGEMM_DEFAULT_UNROLL_N 2
@@ -2248,13 +2254,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_R 4096
 #define DGEMM_DEFAULT_R 4096
 #define CGEMM_DEFAULT_R 4096
-#define ZGEMM_DEFAULT_R 512
+#define ZGEMM_DEFAULT_R 4096
 
 #define SYMV_P	 8
 
 #endif
 
-#if defined(POWER9)
+#if defined(POWER9) || defined(POWER10)
 
 #define SNUMOPT		16
 #define DNUMOPT		8
@@ -2282,10 +2288,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_Q  1026
 #define ZGEMM_DEFAULT_Q 1026
 
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 4096
+
 #define SYMV_P	 8
 
 #endif
 
+#if defined(POWER10)
+#undef SHGEMM_DEFAULT_UNROLL_N
+#undef SHGEMM_DEFAULT_UNROLL_M
+#undef SHGEMM_DEFAULT_P
+#undef SHGEMM_DEFAULT_R
+#undef SHGEMM_DEFAULT_Q
+#define SHGEMM_DEFAULT_UNROLL_M 16
+#define SHGEMM_DEFAULT_UNROLL_N 8
+#define SHGEMM_DEFAULT_P 832
+#define SHGEMM_DEFAULT_Q 1026
+#define SHGEMM_DEFAULT_R 4096
+#endif
+
 #if defined(SPARC) && defined(V7)
 
 #define SNUMOPT		4
@@ -2468,7 +2492,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SYMV_P	16
 #endif
 
-#if defined(P5600) || defined(MIPS1004K) || defined(I6400) || defined(P6600) || defined(I6500)
+#if defined(P5600) || defined(MIPS1004K) || defined(MIPS24K) || defined(I6400) || defined(P6600) || defined(I6500)
 #define SNUMOPT  2
 #define DNUMOPT  2
 
@@ -2612,7 +2636,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define SYMV_P	16
 
-#if defined(CORTEXA53) || defined(CORTEXA57) || \
+#if defined(CORTEXA57) || \
     defined(CORTEXA72) || defined(CORTEXA73) || \
     defined(FALKOR)    || defined(TSV110) || defined(EMAG8180)
 
@@ -2658,6 +2682,35 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define CGEMM_DEFAULT_R 4096
 #define ZGEMM_DEFAULT_R 2048
 
+#elif defined(CORTEXA53)
+
+#define SGEMM_DEFAULT_UNROLL_M  8
+#define SGEMM_DEFAULT_UNROLL_N  8
+
+#define DGEMM_DEFAULT_UNROLL_M  8
+#define DGEMM_DEFAULT_UNROLL_N  4
+
+#define CGEMM_DEFAULT_UNROLL_M  8
+#define CGEMM_DEFAULT_UNROLL_N  4
+
+#define ZGEMM_DEFAULT_UNROLL_M  4
+#define ZGEMM_DEFAULT_UNROLL_N  4
+
+#define SGEMM_DEFAULT_P 256
+#define DGEMM_DEFAULT_P 160
+#define CGEMM_DEFAULT_P 128
+#define ZGEMM_DEFAULT_P 128
+
+#define SGEMM_DEFAULT_Q 256
+#define DGEMM_DEFAULT_Q 128
+#define CGEMM_DEFAULT_Q 224
+#define ZGEMM_DEFAULT_Q 112
+
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 2048
+
 #elif defined(THUNDERX)
 
 #define SGEMM_DEFAULT_UNROLL_M  4
@@ -2988,7 +3041,7 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define GEMM_DEFAULT_OFFSET_B 0
 #define GEMM_DEFAULT_ALIGN 0x03fffUL
 
-#define SGEMM_DEFAULT_UNROLL_M  8
+#define SGEMM_DEFAULT_UNROLL_M  16
 #define SGEMM_DEFAULT_UNROLL_N  4
 
 #define DGEMM_DEFAULT_UNROLL_M  8
diff --git a/test/Makefile b/test/Makefile
index 7a873b7e5..45f9821ec 100644
--- a/test/Makefile
+++ b/test/Makefile
@@ -64,9 +64,17 @@ endif
 endif
 endif
 
+ifeq ($(BUILD_HALF),1)
+level3 : test_shgemm sblat3 dblat3 cblat3 zblat3
+else
 level3 : sblat3 dblat3 cblat3 zblat3
+endif
 ifndef CROSS
 	rm -f ?BLAT3.SUMM
+ifeq ($(BUILD_HALF),1)
+	OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./test_shgemm > SHBLAT3.SUMM
+	@$(GREP) -q FATAL SHBLAT3.SUMM && cat SHBLAT3.SUMM || exit 0
+endif
 	OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./sblat3 < ./sblat3.dat
 	@$(GREP) -q FATAL SBLAT3.SUMM && cat SBLAT3.SUMM || exit 0
 	OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./dblat3 < ./dblat3.dat
@@ -78,6 +86,10 @@ ifndef CROSS
 ifdef SMP
 	rm -f ?BLAT3.SUMM
 ifeq ($(USE_OPENMP), 1)
+ifeq ($(BUILD_HALF),1)
+	OMP_NUM_THREADS=2 ./test_shgemm > SHBLAT3.SUMM
+	@$(GREP) -q FATAL SHBLAT3.SUMM && cat SHBLAT3.SUMM || exit 0
+endif
 	OMP_NUM_THREADS=2 ./sblat3 < ./sblat3.dat
 	@$(GREP) -q FATAL SBLAT3.SUMM && cat SBLAT3.SUMM || exit 0
 	OMP_NUM_THREADS=2 ./dblat3 < ./dblat3.dat
@@ -87,6 +99,10 @@ ifeq ($(USE_OPENMP), 1)
 	OMP_NUM_THREADS=2 ./zblat3 < ./zblat3.dat
 	@$(GREP) -q FATAL ZBLAT3.SUMM && cat ZBLAT3.SUMM || exit 0
 else
+ifeq ($(BUILD_HALF),1)
+	OPENBLAS_NUM_THREADS=2 ./test_shgemm > SHBLAT3.SUMM
+	@$(GREP) -q FATAL SHBLAT3.SUMM && cat SHBLAT3.SUMM || exit 0
+endif
 	OPENBLAS_NUM_THREADS=2 ./sblat3 < ./sblat3.dat
 	@$(GREP) -q FATAL SBLAT3.SUMM && cat SBLAT3.SUMM || exit 0
 	OPENBLAS_NUM_THREADS=2 ./dblat3 < ./dblat3.dat
@@ -165,6 +181,11 @@ zblat2 : zblat2.$(SUFFIX) ../$(LIBNAME)
 sblat3 : sblat3.$(SUFFIX) ../$(LIBNAME)
 	$(FC) $(FLDFLAGS) -o sblat3 sblat3.$(SUFFIX) ../$(LIBNAME) $(EXTRALIB) $(CEXTRALIB)
 
+ifeq ($(BUILD_HALF),1)
+test_shgemm : compare_sgemm_shgemm.c ../$(LIBNAME)
+	$(FC) $(FLDFLAGS) -o test_shgemm compare_sgemm_shgemm.c ../$(LIBNAME) $(EXTRALIB) $(CEXTRALIB)
+endif
+
 dblat3 : dblat3.$(SUFFIX) ../$(LIBNAME)
 	$(FC) $(FLDFLAGS) -o dblat3 dblat3.$(SUFFIX) ../$(LIBNAME) $(EXTRALIB) $(CEXTRALIB)
 
@@ -187,7 +208,7 @@ clean:
 	@rm -f *.$(SUFFIX) *.$(PSUFFIX) gmon.$(SUFFIX)ut *.SUMM *.cxml *.exe *.pdb *.dwf \
 	sblat1 dblat1 cblat1 zblat1 \
 	sblat2 dblat2 cblat2 zblat2 \
-	sblat3 dblat3 cblat3 zblat3 \
+	test_shgemm sblat3 dblat3 cblat3 zblat3 \
 	sblat1p dblat1p cblat1p zblat1p \
 	sblat2p dblat2p cblat2p zblat2p \
 	sblat3p dblat3p cblat3p zblat3p \
diff --git a/test/cblat1.f b/test/cblat1.f
index d6b53d105..ecf2a44cb 100644
--- a/test/cblat1.f
+++ b/test/cblat1.f
@@ -1,7 +1,49 @@
+*> \brief \b CBLAT1
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM CBLAT1
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*>    Test program for the COMPLEX Level 1 BLAS.
+*>    Based upon the original BLAS test routine together with:
+*>
+*>    F06GAF Example Program Text
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex_blas_testing
+*
+*  =====================================================================
       PROGRAM CBLAT1
-*     Test program for the COMPLEX    Level 1 BLAS.
-*     Based upon the original BLAS test routine together with:
-*     F06GAF Example Program Text
+*
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
+*
+*  =====================================================================
+*
 *     .. Parameters ..
       INTEGER          NOUT
       PARAMETER        (NOUT=6)
@@ -114,8 +156,8 @@
      +                  (5.0E0,6.0E0), (5.0E0,6.0E0), (0.1E0,0.1E0),
      +                  (-0.6E0,0.1E0), (0.1E0,-0.3E0), (7.0E0,8.0E0),
      +                  (7.0E0,8.0E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
-     +                  (7.0E0,8.0E0), (0.3E0,0.1E0), (0.1E0,0.4E0),
-     +                  (0.4E0,0.1E0), (0.1E0,0.2E0), (2.0E0,3.0E0),
+     +                  (7.0E0,8.0E0), (0.3E0,0.1E0), (0.5E0,0.0E0),
+     +                  (0.0E0,0.5E0), (0.0E0,0.2E0), (2.0E0,3.0E0),
      +                  (2.0E0,3.0E0), (2.0E0,3.0E0), (2.0E0,3.0E0)/
       DATA              ((CV(I,J,2),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (4.0E0,5.0E0), (4.0E0,5.0E0), (4.0E0,5.0E0),
@@ -129,10 +171,10 @@
      +                  (3.0E0,6.0E0), (-0.6E0,0.1E0), (4.0E0,7.0E0),
      +                  (0.1E0,-0.3E0), (7.0E0,2.0E0), (7.0E0,2.0E0),
      +                  (7.0E0,2.0E0), (0.3E0,0.1E0), (5.0E0,8.0E0),
-     +                  (0.1E0,0.4E0), (6.0E0,9.0E0), (0.4E0,0.1E0),
-     +                  (8.0E0,3.0E0), (0.1E0,0.2E0), (9.0E0,4.0E0)/
-      DATA              STRUE2/0.0E0, 0.5E0, 0.6E0, 0.7E0, 0.7E0/
-      DATA              STRUE4/0.0E0, 0.7E0, 1.0E0, 1.3E0, 1.7E0/
+     +                  (0.5E0,0.0E0), (6.0E0,9.0E0), (0.0E0,0.5E0),
+     +                  (8.0E0,3.0E0), (0.0E0,0.2E0), (9.0E0,4.0E0)/
+      DATA              STRUE2/0.0E0, 0.5E0, 0.6E0, 0.7E0, 0.8E0/
+      DATA              STRUE4/0.0E0, 0.7E0, 1.0E0, 1.3E0, 1.6E0/
       DATA              ((CTRUE5(I,J,1),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
@@ -145,8 +187,8 @@
      +                  (0.11E0,-0.03E0), (-0.17E0,0.46E0),
      +                  (-0.17E0,-0.19E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
      +                  (7.0E0,8.0E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
-     +                  (0.19E0,-0.17E0), (0.32E0,0.09E0),
-     +                  (0.23E0,-0.24E0), (0.18E0,0.01E0),
+     +                  (0.19E0,-0.17E0), (0.20E0,-0.35E0),
+     +                  (0.35E0,0.20E0), (0.14E0,0.08E0),
      +                  (2.0E0,3.0E0), (2.0E0,3.0E0), (2.0E0,3.0E0),
      +                  (2.0E0,3.0E0)/
       DATA              ((CTRUE5(I,J,2),I=1,8),J=1,5)/(0.1E0,0.1E0),
@@ -162,9 +204,9 @@
      +                  (-0.17E0,0.46E0), (4.0E0,7.0E0),
      +                  (-0.17E0,-0.19E0), (7.0E0,2.0E0), (7.0E0,2.0E0),
      +                  (7.0E0,2.0E0), (0.19E0,-0.17E0), (5.0E0,8.0E0),
-     +                  (0.32E0,0.09E0), (6.0E0,9.0E0),
-     +                  (0.23E0,-0.24E0), (8.0E0,3.0E0),
-     +                  (0.18E0,0.01E0), (9.0E0,4.0E0)/
+     +                  (0.20E0,-0.35E0), (6.0E0,9.0E0),
+     +                  (0.35E0,0.20E0), (8.0E0,3.0E0),
+     +                  (0.14E0,0.08E0), (9.0E0,4.0E0)/
       DATA              ((CTRUE6(I,J,1),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
@@ -177,8 +219,8 @@
      +                  (0.03E0,0.03E0), (-0.18E0,0.03E0),
      +                  (0.03E0,-0.09E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
      +                  (7.0E0,8.0E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
-     +                  (0.09E0,0.03E0), (0.03E0,0.12E0),
-     +                  (0.12E0,0.03E0), (0.03E0,0.06E0), (2.0E0,3.0E0),
+     +                  (0.09E0,0.03E0), (0.15E0,0.00E0),
+     +                  (0.00E0,0.15E0), (0.00E0,0.06E0), (2.0E0,3.0E0),
      +                  (2.0E0,3.0E0), (2.0E0,3.0E0), (2.0E0,3.0E0)/
       DATA              ((CTRUE6(I,J,2),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (4.0E0,5.0E0), (4.0E0,5.0E0), (4.0E0,5.0E0),
@@ -193,8 +235,8 @@
      +                  (-0.18E0,0.03E0), (4.0E0,7.0E0),
      +                  (0.03E0,-0.09E0), (7.0E0,2.0E0), (7.0E0,2.0E0),
      +                  (7.0E0,2.0E0), (0.09E0,0.03E0), (5.0E0,8.0E0),
-     +                  (0.03E0,0.12E0), (6.0E0,9.0E0), (0.12E0,0.03E0),
-     +                  (8.0E0,3.0E0), (0.03E0,0.06E0), (9.0E0,4.0E0)/
+     +                  (0.15E0,0.00E0), (6.0E0,9.0E0), (0.00E0,0.15E0),
+     +                  (8.0E0,3.0E0), (0.00E0,0.06E0), (9.0E0,4.0E0)/
       DATA              ITRUE3/0, 1, 2, 2, 2/
 *     .. Executable Statements ..
       DO 60 INCX = 1, 2
@@ -529,7 +571,8 @@
 *
 *     .. Parameters ..
       INTEGER          NOUT
-      PARAMETER        (NOUT=6)
+      REAL             ZERO
+      PARAMETER        (NOUT=6, ZERO=0.0E0)
 *     .. Scalar Arguments ..
       REAL             SFAC
       INTEGER          LEN
@@ -552,7 +595,7 @@
 *
       DO 40 I = 1, LEN
          SD = SCOMP(I) - STRUE(I)
-         IF (SDIFF(ABS(SSIZE(I))+ABS(SFAC*SD),ABS(SSIZE(I))).EQ.0.0E0)
+         IF (ABS(SFAC*SD) .LE. ABS(SSIZE(I))*EPSILON(ZERO))
      +       GO TO 40
 *
 *                             HERE    SCOMP(I) IS NOT CLOSE TO STRUE(I).
diff --git a/test/cblat2.f b/test/cblat2.f
index 20f188100..8c7bac48e 100644
--- a/test/cblat2.f
+++ b/test/cblat2.f
@@ -1,68 +1,114 @@
+*> \brief \b CBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM CBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX          Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 17 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 35 lines:
+*> 'cblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> CGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CGERC  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CGERU  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHER2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex_blas_testing
+*
+*  =====================================================================
       PROGRAM CBLAT2
 *
-*  Test program for the COMPLEX          Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 17 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 35 lines:
-*  'CBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  CGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CGERC  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CGERU  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHER2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -71,8 +117,8 @@
       PARAMETER          ( NSUBS = 17 )
       COMPLEX            ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0, 0.0 ), ONE = ( 1.0, 0.0 ) )
-      REAL               RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0, RHALF = 0.5, RONE = 1.0 )
+      REAL               RZERO
+      PARAMETER          ( RZERO = 0.0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -126,7 +172,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -135,7 +181,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -240,14 +286,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   90 CONTINUE
-      IF( SDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 100
-      EPS = RHALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of CMVCH using exact data.
@@ -3079,7 +3118,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LCERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/cblat3.f b/test/cblat3.f
index 5df1ddd64..a65e1364c 100644
--- a/test/cblat3.f
+++ b/test/cblat3.f
@@ -1,50 +1,96 @@
+*> \brief \b CBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM CBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX          Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 9 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 23 lines:
+*> 'cblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'CBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> CGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHERK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHER2K T PUT F FOR NO TEST. SAME COLUMNS.
+*> CSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex_blas_testing
+*
+*  =====================================================================
       PROGRAM CBLAT3
 *
-*  Test program for the COMPLEX          Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 9 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 23 lines:
-*  'CBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'CBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  CGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHERK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHER2K T PUT F FOR NO TEST. SAME COLUMNS.
-*  CSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -53,8 +99,8 @@
       PARAMETER          ( NSUBS = 9 )
       COMPLEX            ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0, 0.0 ), ONE = ( 1.0, 0.0 ) )
-      REAL               RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0, RHALF = 0.5, RONE = 1.0 )
+      REAL               RZERO
+      PARAMETER          ( RZERO = 0.0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -103,7 +149,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -112,7 +158,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -189,14 +235,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   70 CONTINUE
-      IF( SDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 80
-      EPS = RHALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of CMMCH using exact data.
@@ -1301,8 +1340,6 @@
       NC = 0
       RESET = .TRUE.
       ERRMAX = RZERO
-      RALS = RONE
-      RBETS = RONE
 *
       DO 100 IN = 1, NIDIM
          N = IDIM( IN )
@@ -1948,7 +1985,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, RALPHA, BETA, RBETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1958,12 +1995,19 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA, BETA, RALPHA, and RBETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to CSYMM and CHEMM
+*            with INFOT = 9  (eca)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      REAL               ONE, TWO
+      PARAMETER          ( ONE = 1.0E0, TWO = 2.0E0 )
 *     .. Local Scalars ..
       COMPLEX            ALPHA, BETA
       REAL               RALPHA, RBETA
@@ -1981,6 +2025,14 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA, BETA, RALPHA, and RBETA.
+*
+      ALPHA = CMPLX( ONE, -ONE )
+      BETA = CMPLX( TWO, -TWO )
+      RALPHA = ONE
+      RBETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60, 70, 80,
      $        90 )ISNUM
    10 INFOT = 1
@@ -2207,16 +2259,16 @@
       CALL CHEMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL CHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2274,16 +2326,16 @@
       CALL CSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL CSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -3270,7 +3322,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LCERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/compare_sgemm_shgemm.c b/test/compare_sgemm_shgemm.c
new file mode 100644
index 000000000..57aee7b8f
--- /dev/null
+++ b/test/compare_sgemm_shgemm.c
@@ -0,0 +1,131 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#include <stdio.h>
+#include <stdint.h>
+#include "../common.h"
+#define SGEMM   BLASFUNC(sgemm)
+#define SHGEMM   BLASFUNC(shgemm)
+typedef union
+{
+  unsigned short v;
+  struct
+  {
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    unsigned short s:1;
+    unsigned short e:8;
+    unsigned short m:7;
+#else
+    unsigned short m:7;
+    unsigned short e:8;
+    unsigned short s:1;
+#endif
+  } bits;
+} bfloat16_bits;
+
+typedef union
+{
+  float v;
+  struct
+  {
+    uint32_t m:23;
+    uint32_t e:8;
+    uint32_t s:1;
+  } bits;
+} float32_bits;
+
+float
+float16to32 (bfloat16_bits f16)
+{
+  float32_bits f32;
+  f32.bits.s = f16.bits.s;
+  f32.bits.e = f16.bits.e;
+  f32.bits.m = (uint32_t) f16.bits.m << 16;
+  return f32.v;
+}
+
+int
+main (int argc, char *argv[])
+{
+  int m, n, k;
+  int i, j, l;
+  int x;
+  int ret = 0;
+  int loop = 100;
+  char transA = 'N', transB = 'N';
+  float alpha = 1.0, beta = 0.0;
+
+  for (x = 0; x <= loop; x++)
+    {
+      m = k = n = x;
+      float A[m * k];
+      float B[k * n];
+      float C[m * n];
+      bfloat16_bits AA[m * k], BB[k * n];
+      float DD[m * n], CC[m * n];
+
+      for (j = 0; j < m; j++)
+	{
+	  for (i = 0; i < m; i++)
+	    {
+	      A[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
+	      B[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
+	      C[j * k + i] = 0;
+	      AA[j * k + i].v = *(uint32_t *) & A[j * k + i] >> 16;
+	      BB[j * k + i].v = *(uint32_t *) & B[j * k + i] >> 16;
+	      CC[j * k + i] = 0;
+	      DD[j * k + i] = 0;
+	    }
+	}
+      SGEMM (&transA, &transB, &m, &n, &k, &alpha, A,
+	     &m, B, &k, &beta, C, &m);
+      SHGEMM (&transA, &transB, &m, &n, &k, &alpha, AA,
+	      &m, BB, &k, &beta, CC, &m);
+      for (i = 0; i < n; i++)
+	for (j = 0; j < m; j++)
+	  for (l = 0; l < k; l++)
+	    if (fabs (CC[i * m + j] - C[i * m + j]) > 1.0)
+	      ret++;
+      if (transA == 'N' && transB == 'N')
+	{
+	  for (i = 0; i < n; i++)
+	    for (j = 0; j < m; j++)
+	      for (l = 0; l < k; l++)
+		{
+		  DD[i * m + j] +=
+		    float16to32 (AA[l * m + j]) * float16to32 (BB[l + k * i]);
+		}
+	  for (i = 0; i < n; i++)
+	    for (j = 0; j < m; j++)
+	      for (l = 0; l < k; l++)
+		if (CC[i * m + j] != DD[i * m + j])
+		  ret++;
+	}
+    }
+  if (ret != 0)
+    fprintf (stderr, "FATAL ERROR SHGEMM - Return code: %d\n", ret);
+  return ret;
+}
diff --git a/test/dblat2.f b/test/dblat2.f
index 4002d4368..9bbbe9792 100644
--- a/test/dblat2.f
+++ b/test/dblat2.f
@@ -1,75 +1,121 @@
+*> \brief \b DBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM DBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the DOUBLE PRECISION Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 16 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 34 lines:
+*> 'dblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'DBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 0.9       VALUES OF BETAC
+*> DGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DGER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup double_blas_testing
+*
+*  =====================================================================
       PROGRAM DBLAT2
 *
-*  Test program for the DOUBLE PRECISION Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 16 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 34 lines:
-*  'DBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'DBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 0.9       VALUES OF BETA
-*  DGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DGER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 16 )
-      DOUBLE PRECISION   ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0D0, HALF = 0.5D0, ONE = 1.0D0 )
+      DOUBLE PRECISION   ZERO, ONE
+      PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -121,7 +167,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -130,7 +176,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -235,14 +281,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   90 CONTINUE
-      IF( DDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 100
-      EPS = HALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of DMVCH using exact data.
@@ -2982,7 +3021,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LDERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/dblat3.f b/test/dblat3.f
index 082e03e5e..1ebec4ffa 100644
--- a/test/dblat3.f
+++ b/test/dblat3.f
@@ -1,55 +1,101 @@
+*> \brief \b DBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM DBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the DOUBLE PRECISION Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 6 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 20 lines:
+*> 'dblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'DBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 1.3       VALUES OF BETA
+*> DGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup double_blas_testing
+*
+*  =====================================================================
       PROGRAM DBLAT3
 *
-*  Test program for the DOUBLE PRECISION Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 6 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 20 lines:
-*  'DBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'DBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 1.3       VALUES OF BETA
-*  DGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 6 )
-      DOUBLE PRECISION   ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0D0, HALF = 0.5D0, ONE = 1.0D0 )
+      DOUBLE PRECISION   ZERO, ONE
+      PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -96,7 +142,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -105,7 +151,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -182,14 +228,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   70 CONTINUE
-      IF( DDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 80
-      EPS = HALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of DMMCH using exact data.
@@ -1802,7 +1841,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, BETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1812,12 +1851,18 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA and BETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to SSYMM with INFOT = 9  (eca)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      DOUBLE PRECISION   ONE, TWO
+      PARAMETER          ( ONE = 1.0D0, TWO = 2.0D0 )
 *     .. Local Scalars ..
       DOUBLE PRECISION   ALPHA, BETA
 *     .. Local Arrays ..
@@ -1834,6 +1879,12 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA and BETA.
+*
+      ALPHA = ONE
+      BETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60 )ISNUM
    10 INFOT = 1
       CALL DGEMM( '/', 'N', 0, 0, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
@@ -1963,16 +2014,16 @@
       CALL DSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL DSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2660,7 +2711,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LDERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/sblat2.f b/test/sblat2.f
index a1074be52..56ead8640 100644
--- a/test/sblat2.f
+++ b/test/sblat2.f
@@ -1,75 +1,121 @@
+*> \brief \b SBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM SBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the REAL Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 16 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 34 lines:
+*> 'sblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'SBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 0.9       VALUES OF BETA
+*> SGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SGER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup single_blas_testing
+*
+*  =====================================================================
       PROGRAM SBLAT2
 *
-*  Test program for the REAL             Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 16 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 34 lines:
-*  'SBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'SBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 0.9       VALUES OF BETA
-*  SGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SGER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 16 )
-      REAL               ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0, HALF = 0.5, ONE = 1.0 )
+      REAL               ZERO, ONE
+      PARAMETER          ( ZERO = 0.0, ONE = 1.0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -121,7 +167,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -130,7 +176,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -235,14 +281,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   90 CONTINUE
-      IF( SDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 100
-      EPS = HALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of SMVCH using exact data.
@@ -2982,7 +3021,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LSERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/sblat3.f b/test/sblat3.f
index 325a9eb92..66edac14e 100644
--- a/test/sblat3.f
+++ b/test/sblat3.f
@@ -1,55 +1,101 @@
+*> \brief \b SBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM SBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the REAL             Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 6 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 20 lines:
+*> 'sblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'SBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 1.3       VALUES OF BETA
+*> SGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup single_blas_testing
+*
+*  =====================================================================
       PROGRAM SBLAT3
 *
-*  Test program for the REAL             Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 6 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 20 lines:
-*  'SBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'SBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 1.3       VALUES OF BETA
-*  SGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 6 )
-      REAL               ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0, HALF = 0.5, ONE = 1.0 )
+      REAL               ZERO, ONE
+      PARAMETER          ( ZERO = 0.0, ONE = 1.0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -96,7 +142,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -105,7 +151,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -182,14 +228,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   70 CONTINUE
-      IF( SDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 80
-      EPS = HALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of SMMCH using exact data.
@@ -1802,7 +1841,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, BETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1812,12 +1851,18 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA and BETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to SSYMM with INFOT = 9  (eca)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      REAL               ONE, TWO
+      PARAMETER          ( ONE = 1.0E0, TWO = 2.0E0 )
 *     .. Local Scalars ..
       REAL               ALPHA, BETA
 *     .. Local Arrays ..
@@ -1834,6 +1879,12 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA and BETA.
+*
+      ALPHA = ONE
+      BETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60 )ISNUM
    10 INFOT = 1
       CALL SGEMM( '/', 'N', 0, 0, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
@@ -1963,16 +2014,16 @@
       CALL SSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL SSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2660,7 +2711,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LSERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/zblat1.f b/test/zblat1.f
index 8b4b8d21e..2d7b88490 100644
--- a/test/zblat1.f
+++ b/test/zblat1.f
@@ -1,7 +1,49 @@
+*> \brief \b ZBLAT1
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM ZBLAT1
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*>    Test program for the COMPLEX*16 Level 1 BLAS.
+*>
+*>    Based upon the original BLAS test routine together with:
+*>    F06GAF Example Program Text
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex16_blas_testing
+*
+*  =====================================================================
       PROGRAM ZBLAT1
-*     Test program for the COMPLEX*16 Level 1 BLAS.
-*     Based upon the original BLAS test routine together with:
-*     F06GAF Example Program Text
+*
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
+*
+*  =====================================================================
+*
 *     .. Parameters ..
       INTEGER          NOUT
       PARAMETER        (NOUT=6)
@@ -114,8 +156,8 @@
      +                  (5.0D0,6.0D0), (5.0D0,6.0D0), (0.1D0,0.1D0),
      +                  (-0.6D0,0.1D0), (0.1D0,-0.3D0), (7.0D0,8.0D0),
      +                  (7.0D0,8.0D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
-     +                  (7.0D0,8.0D0), (0.3D0,0.1D0), (0.1D0,0.4D0),
-     +                  (0.4D0,0.1D0), (0.1D0,0.2D0), (2.0D0,3.0D0),
+     +                  (7.0D0,8.0D0), (0.3D0,0.1D0), (0.5D0,0.0D0),
+     +                  (0.0D0,0.5D0), (0.0D0,0.2D0), (2.0D0,3.0D0),
      +                  (2.0D0,3.0D0), (2.0D0,3.0D0), (2.0D0,3.0D0)/
       DATA              ((CV(I,J,2),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (4.0D0,5.0D0), (4.0D0,5.0D0), (4.0D0,5.0D0),
@@ -129,10 +171,10 @@
      +                  (3.0D0,6.0D0), (-0.6D0,0.1D0), (4.0D0,7.0D0),
      +                  (0.1D0,-0.3D0), (7.0D0,2.0D0), (7.0D0,2.0D0),
      +                  (7.0D0,2.0D0), (0.3D0,0.1D0), (5.0D0,8.0D0),
-     +                  (0.1D0,0.4D0), (6.0D0,9.0D0), (0.4D0,0.1D0),
-     +                  (8.0D0,3.0D0), (0.1D0,0.2D0), (9.0D0,4.0D0)/
-      DATA              STRUE2/0.0D0, 0.5D0, 0.6D0, 0.7D0, 0.7D0/
-      DATA              STRUE4/0.0D0, 0.7D0, 1.0D0, 1.3D0, 1.7D0/
+     +                  (0.5D0,0.0D0), (6.0D0,9.0D0), (0.0D0,0.5D0),
+     +                  (8.0D0,3.0D0), (0.0D0,0.2D0), (9.0D0,4.0D0)/
+      DATA              STRUE2/0.0D0, 0.5D0, 0.6D0, 0.7D0, 0.8D0/
+      DATA              STRUE4/0.0D0, 0.7D0, 1.0D0, 1.3D0, 1.6D0/
       DATA              ((CTRUE5(I,J,1),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
@@ -145,8 +187,8 @@
      +                  (0.11D0,-0.03D0), (-0.17D0,0.46D0),
      +                  (-0.17D0,-0.19D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
      +                  (7.0D0,8.0D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
-     +                  (0.19D0,-0.17D0), (0.32D0,0.09D0),
-     +                  (0.23D0,-0.24D0), (0.18D0,0.01D0),
+     +                  (0.19D0,-0.17D0), (0.20D0,-0.35D0),
+     +                  (0.35D0,0.20D0), (0.14D0,0.08D0),
      +                  (2.0D0,3.0D0), (2.0D0,3.0D0), (2.0D0,3.0D0),
      +                  (2.0D0,3.0D0)/
       DATA              ((CTRUE5(I,J,2),I=1,8),J=1,5)/(0.1D0,0.1D0),
@@ -162,9 +204,9 @@
      +                  (-0.17D0,0.46D0), (4.0D0,7.0D0),
      +                  (-0.17D0,-0.19D0), (7.0D0,2.0D0), (7.0D0,2.0D0),
      +                  (7.0D0,2.0D0), (0.19D0,-0.17D0), (5.0D0,8.0D0),
-     +                  (0.32D0,0.09D0), (6.0D0,9.0D0),
-     +                  (0.23D0,-0.24D0), (8.0D0,3.0D0),
-     +                  (0.18D0,0.01D0), (9.0D0,4.0D0)/
+     +                  (0.20D0,-0.35D0), (6.0D0,9.0D0),
+     +                  (0.35D0,0.20D0), (8.0D0,3.0D0),
+     +                  (0.14D0,0.08D0), (9.0D0,4.0D0)/
       DATA              ((CTRUE6(I,J,1),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
@@ -177,8 +219,8 @@
      +                  (0.03D0,0.03D0), (-0.18D0,0.03D0),
      +                  (0.03D0,-0.09D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
      +                  (7.0D0,8.0D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
-     +                  (0.09D0,0.03D0), (0.03D0,0.12D0),
-     +                  (0.12D0,0.03D0), (0.03D0,0.06D0), (2.0D0,3.0D0),
+     +                  (0.09D0,0.03D0), (0.15D0,0.00D0),
+     +                  (0.00D0,0.15D0), (0.00D0,0.06D0), (2.0D0,3.0D0),
      +                  (2.0D0,3.0D0), (2.0D0,3.0D0), (2.0D0,3.0D0)/
       DATA              ((CTRUE6(I,J,2),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (4.0D0,5.0D0), (4.0D0,5.0D0), (4.0D0,5.0D0),
@@ -193,8 +235,8 @@
      +                  (-0.18D0,0.03D0), (4.0D0,7.0D0),
      +                  (0.03D0,-0.09D0), (7.0D0,2.0D0), (7.0D0,2.0D0),
      +                  (7.0D0,2.0D0), (0.09D0,0.03D0), (5.0D0,8.0D0),
-     +                  (0.03D0,0.12D0), (6.0D0,9.0D0), (0.12D0,0.03D0),
-     +                  (8.0D0,3.0D0), (0.03D0,0.06D0), (9.0D0,4.0D0)/
+     +                  (0.15D0,0.00D0), (6.0D0,9.0D0), (0.00D0,0.15D0),
+     +                  (8.0D0,3.0D0), (0.00D0,0.06D0), (9.0D0,4.0D0)/
       DATA              ITRUE3/0, 1, 2, 2, 2/
 *     .. Executable Statements ..
       DO 60 INCX = 1, 2
@@ -529,7 +571,8 @@
 *
 *     .. Parameters ..
       INTEGER          NOUT
-      PARAMETER        (NOUT=6)
+      DOUBLE PRECISION ZERO
+      PARAMETER        (NOUT=6, ZERO=0.0D0)
 *     .. Scalar Arguments ..
       DOUBLE PRECISION SFAC
       INTEGER          LEN
@@ -552,7 +595,7 @@
 *
       DO 40 I = 1, LEN
          SD = SCOMP(I) - STRUE(I)
-         IF (SDIFF(ABS(SSIZE(I))+ABS(SFAC*SD),ABS(SSIZE(I))).EQ.0.0D0)
+         IF (ABS(SFAC*SD) .LE. ABS(SSIZE(I))*EPSILON(ZERO))
      +       GO TO 40
 *
 *                             HERE    SCOMP(I) IS NOT CLOSE TO STRUE(I).
diff --git a/test/zblat2.f b/test/zblat2.f
index e65cdcc70..4a20ac567 100644
--- a/test/zblat2.f
+++ b/test/zblat2.f
@@ -1,68 +1,114 @@
+*> \brief \b ZBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM ZBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX*16       Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 17 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 35 lines:
+*> 'zblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> ZGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZGERC  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZGERU  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHER2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex16_blas_testing
+*
+*  =====================================================================
       PROGRAM ZBLAT2
 *
-*  Test program for the COMPLEX*16       Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 17 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 35 lines:
-*  'ZBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  ZGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZGERC  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZGERU  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHER2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -72,8 +118,8 @@
       COMPLEX*16         ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0D0, 0.0D0 ),
      $                   ONE = ( 1.0D0, 0.0D0 ) )
-      DOUBLE PRECISION   RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0D0, RHALF = 0.5D0, RONE = 1.0D0 )
+      DOUBLE PRECISION   RZERO
+      PARAMETER          ( RZERO = 0.0D0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -127,7 +173,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -136,7 +182,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -241,14 +287,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   90 CONTINUE
-      IF( DDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 100
-      EPS = RHALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of ZMVCH using exact data.
@@ -3087,7 +3126,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LZERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/zblat3.f b/test/zblat3.f
index f03b1a617..0e38334e9 100644
--- a/test/zblat3.f
+++ b/test/zblat3.f
@@ -1,50 +1,97 @@
+*> \brief \b ZBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM ZBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX*16       Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 9 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 23 lines:
+*> 'zblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'ZBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> ZGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHERK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHER2K T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex16_blas_testing
+*
+*  =====================================================================
       PROGRAM ZBLAT3
 *
-*  Test program for the COMPLEX*16       Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 9 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 23 lines:
-*  'ZBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'ZBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  ZGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHERK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHER2K T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -54,8 +101,8 @@
       COMPLEX*16         ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0D0, 0.0D0 ),
      $                   ONE = ( 1.0D0, 0.0D0 ) )
-      DOUBLE PRECISION   RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0D0, RHALF = 0.5D0, RONE = 1.0D0 )
+      DOUBLE PRECISION   RZERO
+      PARAMETER          ( RZERO = 0.0D0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -104,7 +151,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -113,7 +160,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -190,14 +237,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   70 CONTINUE
-      IF( DDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 80
-      EPS = RHALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of ZMMCH using exact data.
@@ -1303,8 +1343,6 @@
       NC = 0
       RESET = .TRUE.
       ERRMAX = RZERO
-      RALS = RONE
-      RBETS = RONE
 *
       DO 100 IN = 1, NIDIM
          N = IDIM( IN )
@@ -1951,7 +1989,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, RALPHA, BETA, RBETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1961,12 +1999,20 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA, BETA, RALPHA, and RBETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to ZSYMM and ZHEMM
+*            with INFOT = 9  (eca)
+*  10-9-00:  Declared INTRINSIC DCMPLX (susan)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      REAL               ONE, TWO
+      PARAMETER          ( ONE = 1.0D0, TWO = 2.0D0 )
 *     .. Local Scalars ..
       COMPLEX*16         ALPHA, BETA
       DOUBLE PRECISION   RALPHA, RBETA
@@ -1975,6 +2021,8 @@
 *     .. External Subroutines ..
       EXTERNAL           ZGEMM, ZHEMM, ZHER2K, ZHERK, CHKXER, ZSYMM,
      $                   ZSYR2K, ZSYRK, ZTRMM, ZTRSM
+*     .. Intrinsic Functions ..
+      INTRINSIC          DCMPLX
 *     .. Common blocks ..
       COMMON             /INFOC/INFOT, NOUTC, OK, LERR
 *     .. Executable Statements ..
@@ -1984,6 +2032,14 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA, BETA, RALPHA, and RBETA.
+*
+      ALPHA = DCMPLX( ONE, -ONE )
+      BETA = DCMPLX( TWO, -TWO )
+      RALPHA = ONE
+      RBETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60, 70, 80,
      $        90 )ISNUM
    10 INFOT = 1
@@ -2210,16 +2266,16 @@
       CALL ZHEMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL ZHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2277,16 +2333,16 @@
       CALL ZSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL ZSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -3276,7 +3332,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LZERES = .TRUE.
       GO TO 80
    70 CONTINUE