Fix compilation on older OSX versions

Add OSX build to Azure

.drone.yml

@@ -190,3 +190,27 @@ steps:
     - make -C ctest $COMMON_FLAGS
     - make -C utest $COMMON_FLAGS
     - make -C cpp_thread_test dgemm_tester
+---
+kind: pipeline
+name: arm64_gcc10
+
+platform:
+  os: linux
+  arch: arm64
+
+steps:
+- name: Build and Test
+  image: ubuntu:20.04
+  environment:
+    CC: gcc-10
+    FC: gfortran-10
+    COMMON_FLAGS: 'TARGET=ARMV8 DYNAMIC_ARCH=1'
+  commands:
+    - echo "MAKE_FLAGS:= $COMMON_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC gfortran-10 perl python g++
+    - $CC --version
+    - make QUIET_MAKE=1 $COMMON_FLAGS
+    - make -C utest $COMMON_FLAGS
+    - make -C test $COMMON_FLAGS
+    

.github/workflows/nightly-Homebrew-build.yml

@@ -44,6 +44,11 @@ jobs:
         if: github.event_name != 'pull_request'
         run: brew update || true
 
+      - name: unlink installed gcc to allow updating
+        run: |
+          brew unlink gcc@8
+          brew unlink gcc@9
+          
       - name: Install prerequisites
         run: brew install --fetch-HEAD --HEAD --only-dependencies --keep-tmp openblas
 

.gitignore

@@ -89,5 +89,7 @@ build.*
 *.swp
 benchmark/*.goto
 benchmark/smallscaling
+.vscode
 CMakeCache.txt
 CMakeFiles/*
+.vscode

.travis.yml

@@ -224,12 +224,21 @@ matrix:
       before_script:
         - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
         - brew update
-        - brew install gcc@10
       script:
         - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
       env:
-        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-10"
-        
+        - BTYPE="TARGET=HASWELL USE_OPENMP=1 BINARY=64 INTERFACE64=1 CC=gcc-10 FC=gfortran-10"
+
+    - <<: *test-macos
+      osx_image: xcode12
+      before_script:
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+        - brew update
+      script:
+        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+      env:
+        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-10"      
+
   #  - <<: *test-macos
   #    osx_image: xcode10
   #    env:

CMakeLists.txt

@@ -6,7 +6,7 @@ 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 12.dev)
+set(OpenBLAS_PATCH_VERSION 14.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -14,6 +14,9 @@ include(GNUInstallDirs)
 
 include(CMakePackageConfigHelpers)
 
+if(MSVC AND NOT DEFINED NOFORTRAN)
+  set(NOFORTRAN ON)
+endif()
 
 #######
 if(MSVC)
@@ -229,7 +232,7 @@ if (NOT NO_CBLAS)
   add_subdirectory(utest)
 endif()
 
-if (NOT MSVC AND NOT NOFORTRAN)
+if (NOT NOFORTRAN)
   # Build test and ctest
   add_subdirectory(test)
   if(NOT NO_CBLAS)

Changelog.txt

@@ -1,4 +1,102 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.14
+ 17-Mar-2021
+ 
+ common:
+	* Fixed a race condition on thread shutdown in non-OpenMP builds
+	* Fixed custom BUFFERSIZE option getting ignored in gmake builds
+	* Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
+	* Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
+	* Improved performance of OMATCOPY_RT across all platforms
+	* Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
+	* Fixed potential misreading of the GCC compiler version in the build scripts
+	* Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
+	* Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
+
+ RISCV:
+ 	* Fixed compilation on RISCV (missing entry in getarch)
+	
+ POWER:
+ 	* Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
+	* Added support for compilation on FreeBSD/ppc64le
+ 	* Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
+	* Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
+	* Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
+	* Improved SCOPY and CCOPY performance on POWER10
+	* Improved SGEMM and DGEMM performance on POWER10
+	* Added support for compilation with the NVIDIA HPC compiler
+
+ x86_64:
+	* Added an optimized bfloat16 GEMM kernel for Cooperlake
+	* Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
+	* Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
+	* Added support for compilation with the NAG Fortran compiler
+	* Fixed recognition of the AMD AOCC compiler
+	* Fixed compilation for DYNAMIC_ARCH with clang on Windows
+	* Added support for running the BLAS/CBLAS tests on Windows
+	* Fixed signatures of the tls callback functions for Windows x64
+	* Fixed various issues with fma intrinsics support handling 
+
+ ARM:
+	* Added support for embedded Cortex M targets via a new option EMBEDDED
+
+ ARMV8:
+ 	* Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
+	* Added support for the DYNAMIC_LIST option
+	* Added support for compilation with the NVIDIA HPC compiler
+ 	* Added support for compiling with the NAG Fortran compiler
+
+====================================================================
+Version 0.3.13
+ 12-Dec-2020
+ 
+ common:
+	* Added a generic bfloat16 SBGEMV kernel
+	* Fixed a potentially severe memory leak after fork in OpenMP builds
+	  that was introduced in 0.3.12
+	* Added detection of the Fujitsu Fortran compiler
+	* Added detection of the (e)gfortran compiler on OpenBSD
+	* Added support for overriding the default name of the library independently
+	  from symbol suffixing in the gmake builds (already supported in cmake)
+
+RISCV:
+	* Added a RISC V port optimized for C910V
+
+POWER:
+	* Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
+	* Improved DGEMM performance on POWER10
+	* Improved STRSM and DTRSM performance on POWER9 and POWER10
+	* Fixed segmemtation faults in DYNAMIC_ARCH builds
+ 	* Fixed compilation with the PGI compiler
+
+x86:
+	* Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
+	
+x86_64:
+	* Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
+	* Improved the performance of SASUM and DASUM kernels through parallelization
+	* Improved the performance of SROT and DROT kernels
+	* Improved the performance of multithreaded xSYRK
+	* Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
+	  (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
+	  wrong results)
+	* Fixed miscompilations by old gcc 4.6
+	* Fixed misdetection of AVX2 capability in some Sandybridge cpus
+	* Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
+
+ARM64:
+	* Fixed segmemtation faults in DYNAMIC_ARCH builds
+
+MIPS:
+	* Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
+	* Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
+	* Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
+	* Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
+
+SPARC:
+	* Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
+
 ====================================================================
 Version 0.3.12
  24-Oct-2020

Makefile

@@ -59,6 +59,9 @@ endif
 	@$(CC) --version > /dev/null 2>&1;\
 	if [ $$? -eq 0 ]; then \
 	   cverinfo=`$(CC) --version | sed -n '1p'`; \
+	   if [ -z "$${cverinfo}" ]; then \
+	   cverinfo=`$(CC) --version | sed -n '2p'`; \
+	   fi; \
 	   echo "  C compiler       ... $(C_COMPILER)  (cmd & version : $${cverinfo})";\
 	else  \
 	   echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))";\
@@ -67,6 +70,9 @@ ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	@$(FC) --version > /dev/null 2>&1;\
 	if [ $$? -eq 0 ]; then \
 	   fverinfo=`$(FC) --version | sed -n '1p'`; \
+	   if [ -z "$${fverinfo}" ]; then \
+	   fverinfo=`$(FC) --version | sed -n '2p'`; \
+	   fi; \
 	   echo "  Fortran compiler ... $(F_COMPILER)  (cmd & version : $${fverinfo})";\
 	else \
 	   echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))";\

Makefile.arm64

@@ -1,28 +1,38 @@
-
+ifneq ($(C_COMPILER), PGI)
 ifeq ($(CORE), ARMV8)
 CCOMMON_OPT += -march=armv8-a
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a
 endif
+endif
 
 ifeq ($(CORE), CORTEXA53)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
 endif
+endif
 
 ifeq ($(CORE), CORTEXA57)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a57
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a57
 endif
+endif
 
 ifeq ($(CORE), CORTEXA72)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
 endif
+endif
 
 ifeq ($(CORE), CORTEXA73)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a73
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a73
 endif
+endif
 
 # Use a72 tunings because Neoverse-N1 is only available
 # in GCC>=9
@@ -30,51 +40,71 @@ ifeq ($(CORE), NEOVERSEN1)
 ifeq ($(GCCVERSIONGTEQ7), 1)
 ifeq ($(GCCVERSIONGTEQ9), 1)
 CCOMMON_OPT += -march=armv8.2-a -mtune=neoverse-n1
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=neoverse-n1
+endif
 else
 CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
 endif
+endif
 else
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
 endif
 endif
+endif
 
 ifeq ($(CORE), THUNDERX)
 CCOMMON_OPT += -march=armv8-a -mtune=thunderx
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=thunderx
 endif
+endif
 
 ifeq ($(CORE), FALKOR)
 CCOMMON_OPT += -march=armv8-a -mtune=falkor
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=falkor
 endif
+endif
 
 ifeq ($(CORE), THUNDERX2T99)
 CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
+endif
 
 ifeq ($(CORE), THUNDERX3T110)
 ifeq ($(GCCVERSIONGTEQ10), 1)
 CCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+endif
 else
 CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
 endif
+endif
 
 ifeq ($(CORE), VORTEX)
 CCOMMON_OPT += -march=armv8.3-a
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.3-a
 endif
+endif
 
 ifeq ($(GCCVERSIONGTEQ9), 1)
 ifeq ($(CORE), TSV110)
 CCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
 endif
 endif
-
+endif
+endif

Makefile.power

@@ -10,9 +10,11 @@ USE_OPENMP = 1
 endif
 
 ifeq ($(CORE), POWER10)
+ifneq ($(C_COMPILER), PGI)
 CCOMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
 FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -fno-fast-math
 endif
+endif
 
 ifeq ($(CORE), POWER9)
 ifneq ($(C_COMPILER), PGI)

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.12.dev
+VERSION = 0.3.14.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library

Makefile.system

@@ -21,6 +21,8 @@ ifeq ($(ARCH), amd64)
 override ARCH=x86_64
 else ifeq ($(ARCH), powerpc64)
 override ARCH=power
+else ifeq ($(ARCH), powerpc64le)
+override ARCH=power
 else ifeq ($(ARCH), powerpc)
 override ARCH=power
 else ifeq ($(ARCH), i386)
@@ -181,7 +183,7 @@ endif
 
 # On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
 ifeq ($(HOSTARCH), x86_64)
-ifeq ($(findstring pgcc,$(HOSTCC)),)
+ifeq ($(findstring pgcc,$(HOSTCC))$(findstring nvc,$(HOSTCC)),)
 GETARCH_FLAGS += -march=native
 endif
 endif
@@ -623,6 +625,15 @@ DYNAMIC_CORE += THUNDERX2T99
 DYNAMIC_CORE += TSV110
 DYNAMIC_CORE += EMAG8180
 DYNAMIC_CORE += THUNDERX3T110
+ifdef DYNAMIC_LIST
+override DYNAMIC_CORE = ARMV8 $(DYNAMIC_LIST)
+XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_ARMV8
+XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
+endif
+endif
+
+ifeq ($(ARCH), mips64)
+DYNAMIC_CORE = LOONGSON3R3 LOONGSON3R4
 endif
 
 ifeq ($(ARCH), zarch)
@@ -659,6 +670,7 @@ endif
 endif # ARCH zarch
 
 ifeq ($(ARCH), power)
+ifneq ($(C_COMPILER), PGI)
 DYNAMIC_CORE = POWER6
 DYNAMIC_CORE += POWER8
 ifneq ($(C_COMPILER), GCC)
@@ -672,7 +684,7 @@ DYNAMIC_CORE += POWER9
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
 endif
-LDVERSIONGTEQ35 := $(shell expr `ld --version | head -1 | cut -f2 -d "." | cut -f1 -d "-"` >= 35)
+LDVERSIONGTEQ35 := $(shell expr `$(CC) -Wl,--version 2> /dev/null | head -1 | cut -f2 -d "." | cut -f1 -d "-"` \>= 35)
 ifeq ($(GCCVERSIONGTEQ11)$(LDVERSIONGTEQ35), 11)
 DYNAMIC_CORE += POWER10
 CCOMMON_OPT += -DHAVE_P10_SUPPORT
@@ -685,6 +697,10 @@ else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)
 endif
 endif
+else
+DYNAMIC_CORE = POWER8
+DYNAMIC_CORE += POWER9
+endif
 endif
 
 # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
@@ -787,14 +803,9 @@ CCOMMON_OPT += -mabi=32
 BINARY_DEFINED = 1
 endif
 
-ifeq ($(CORE), LOONGSON3A)
-CCOMMON_OPT += -march=mips64
-FCOMMON_OPT += -march=mips64
-endif
-
-ifeq ($(CORE), LOONGSON3B)
-CCOMMON_OPT += -march=mips64
-FCOMMON_OPT += -march=mips64
+ifeq ($(CORE), $(filter $(CORE),LOONGSON3R3 LOONGSON3R4))
+CCOMMON_OPT += -march=loongson3a
+FCOMMON_OPT += -march=loongson3a
 endif
 
 ifeq ($(CORE), MIPS24K)
@@ -848,9 +859,19 @@ endif
 endif
 
 ifeq ($(C_COMPILER), PGI)
+PGCVERSIONGT20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 20)
+PGCVERSIONGTEQ20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \>= 20)
+PGCMINORVERSIONGE11 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -c 4-5` == 11)
+PGCVERSIONCHECK := $(PGCVERSIONGT20)$(PGCVERSIONEQ20)$(PGCMINORVERSIONGE11)
+ifeq ($(PGCVERSIONCHECK), $(filter $(PGCVERSIONCHECK), 110 111 011))
+NEWPGI := 1
+endif
 ifdef BINARY64
 ifeq ($(ARCH), x86_64)
-CCOMMON_OPT += -tp p7-64 -D__MMX__ -Mnollvm
+CCOMMON_OPT += -tp p7-64
+ifneq ($(NEWPGI),1)
+CCOMMON_OPT +=  -D__MMX__ -Mnollvm
+endif
 else
 ifeq ($(ARCH), power)
 ifeq ($(CORE), POWER8)
@@ -878,13 +899,25 @@ endif
 #  Fortran Compiler dependent settings
 #
 
+ifeq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -dcfuns -recursive -ieee=full -w=obs -thread_safe
+ifdef INTERFACE64
+ifneq ($(INTERFACE64), 0)
+FCOMMON_OPT += -i8
+endif
+endif
+ifeq ($(USE_OPENMP), 1)
+FCOMMON_OPT += -openmp
+endif
+endif
+
 ifeq ($(F_COMPILER), FLANG)
 CCOMMON_OPT += -DF_INTERFACE_FLANG
 FCOMMON_OPT += -Mrecursive -Kieee
 ifeq ($(OSNAME), Linux)
 ifeq ($(ARCH), x86_64)
-FLANG_VENDOR := $(shell `$(FC) --version|cut -f 1 -d "."|head -1`)
-ifeq ($(FLANG_VENDOR),AOCC)
+FLANG_VENDOR := $(shell $(FC) --version|head -1 |cut -f 1 -d " ")
+ifeq ($(FLANG_VENDOR), AMD)
 FCOMMON_OPT += -fno-unroll-loops
 endif
 endif
@@ -1030,18 +1063,24 @@ ifeq ($(ARCH), x86_64)
 FCOMMON_OPT += -tp p7-64
 else
 ifeq ($(ARCH), power)
+ifeq ($(CORE), POWER6)
+$(warning NVIDIA HPC compilers do not support POWER6.)
+endif
 ifeq ($(CORE), POWER8)
 FCOMMON_OPT += -tp pwr8
 endif
 ifeq ($(CORE), POWER9)
 FCOMMON_OPT += -tp pwr9
 endif
+ifeq ($(CORE), POWER10)
+$(warning NVIDIA HPC compilers do not support POWER10.)
+endif
 endif
 endif
 else
 FCOMMON_OPT += -tp p7
 endif
-FCOMMON_OPT += -Mrecursive
+FCOMMON_OPT += -Mrecursive -Kieee
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -mp
 endif
@@ -1078,11 +1117,11 @@ FCOMMON_OPT += -n32
 else
 FCOMMON_OPT += -n64
 endif
-ifeq ($(CORE), LOONGSON3A)
+ifeq ($(CORE), LOONGSON3R3)
 FCOMMON_OPT += -loongson3 -static
 endif
 
-ifeq ($(CORE), LOONGSON3B)
+ifeq ($(CORE), LOONGSON3R4)
 FCOMMON_OPT += -loongson3 -static
 endif
 
@@ -1108,11 +1147,11 @@ CCOMMON_OPT += -n32
 else
 CCOMMON_OPT += -n64
 endif
-ifeq ($(CORE), LOONGSON3A)
+ifeq ($(CORE), LOONGSON3R3)
 CCOMMON_OPT += -loongson3 -static
 endif
 
-ifeq ($(CORE), LOONGSON3B)
+ifeq ($(CORE), LOONGSON3R4)
 CCOMMON_OPT += -loongson3 -static
 endif
 
@@ -1180,6 +1219,8 @@ CCOMMON_OPT += -fPIC
 endif
 ifeq ($(F_COMPILER), SUN)
 FCOMMON_OPT  += -pic
+else ifeq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -PIC
 else
 FCOMMON_OPT += -fPIC
 endif
@@ -1223,10 +1264,8 @@ ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
 ifeq ($(ARCH), mips64)
-ifneq ($(CORE), LOONGSON3B)
 USE_SIMPLE_THREADED_LEVEL3 = 1
 endif
-endif
 
 ifeq ($(USE_OPENMP), 1)
 # USE_SIMPLE_THREADED_LEVEL3 = 1
@@ -1259,6 +1298,10 @@ CCOMMON_OPT	 += -DUSE_PAPI
 EXTRALIB	 += -lpapi -lperfctr
 endif
 
+ifdef BUFFERSIZE
+CCOMMON_OPT	 += -DBUFFERSIZE=$(BUFFERSIZE)
+endif
+
 ifdef DYNAMIC_THREADS
 CCOMMON_OPT	 += -DDYNAMIC_THREADS
 endif
@@ -1342,11 +1385,9 @@ endif
 
 ifneq ($(ARCH), x86_64)
 ifneq ($(ARCH), x86)
-ifneq ($(CORE), LOONGSON3B)
 NO_AFFINITY = 1
 endif
 endif
-endif
 
 ifdef NO_AFFINITY
 ifeq ($(NO_AFFINITY), 0)
@@ -1438,6 +1479,10 @@ LAPACK_FFLAGS := $(FFLAGS)
 LAPACK_FPFLAGS := $(FPFLAGS)
 endif
 
+ifeq ($(F_COMPILER),NAG)
+LAPACK_FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
+endif
+
 LAPACK_CFLAGS = $(CFLAGS)
 LAPACK_CFLAGS += -DHAVE_LAPACK_CONFIG_H
 ifdef INTERFACE64

Makefile.x86_64

@@ -10,40 +10,46 @@ endif
 
 ifdef HAVE_SSE3
 CCOMMON_OPT += -msse3
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -msse3
 endif
+endif
 ifdef HAVE_SSSE3
 CCOMMON_OPT += -mssse3
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -mssse3
 endif
+endif
 ifdef HAVE_SSE4_1
 CCOMMON_OPT += -msse4.1
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -msse4.1
 endif
+endif
 ifndef OLDGCC
 ifdef HAVE_AVX
 CCOMMON_OPT += -mavx
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -mavx
 endif
 endif
+endif
 ifndef NO_AVX2
 ifdef HAVE_AVX2
 CCOMMON_OPT += -mavx2
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -mavx2
 endif
 endif
-ifndef OLDGCC
-ifdef HAVE_FMA3
-CCOMMON_OPT += -mfma
-FCOMMON_OPT += -mfma
-endif
 endif
 
 ifeq ($(CORE), SKYLAKEX)
 ifndef DYNAMIC_ARCH
 ifndef NO_AVX512
 CCOMMON_OPT += -march=skylake-avx512
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=skylake-avx512
+endif
 ifeq ($(OSNAME), CYGWIN_NT)
 CCOMMON_OPT += -fno-asynchronous-unwind-tables
 FCOMMON_OPT += -fno-asynchronous-unwind-tables
@@ -65,9 +71,11 @@ ifeq ($(C_COMPILER), GCC)
 # cooperlake support was added in 10.1
 ifeq ($(GCCVERSIONGTEQ10)$(GCCMINORVERSIONGTEQ1), 11)
 CCOMMON_OPT += -march=cooperlake
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=cooperlake
 endif
 endif
+endif
 ifeq ($(OSNAME), CYGWIN_NT)
 CCOMMON_OPT += -fno-asynchronous-unwind-tables
 FCOMMON_OPT += -fno-asynchronous-unwind-tables

README.md

@@ -13,10 +13,14 @@ Drone CI: [![Build Status](https://cloud.drone.io/api/badges/xianyi/OpenBLAS/sta
 
 ## Introduction
 
-OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
+OpenBLAS is an optimized BLAS (Basic Linear Algebra Subprograms) library based on GotoBLAS2 1.13 BSD version.
 
 Please read the documentation on the OpenBLAS wiki pages: <https://github.com/xianyi/OpenBLAS/wiki>.
 
+For a general introduction to the BLAS routines, please refer to the extensive documentation of their reference implementation hosted at netlib:
+<https://www.netlib.org/blas>. On that site you will likewise find documentation for the reference implementation of the higher-level library LAPACK - the **L**inear **A**lgebra **Pack**age that comes included with OpenBLAS. If you are looking for a general primer or refresher on Linear Algebra, the set of six
+20-minute lecture videos by Prof. Gilbert Strang on either MIT OpenCourseWare <https://ocw.mit.edu/resources/res-18-010-a-2020-vision-of-linear-algebra-spring-2020/> or Youtube <https://www.youtube.com/playlist?list=PLUl4u3cNGP61iQEFiWLE21EJCxwmWvvek> may be helpful.
+
 ## Binary Packages
 
 We provide official binary packages for the following platform:
@@ -208,7 +212,8 @@ Please note that it is not possible to combine support for different architectur
 - **Android**: Supported by the community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-build-OpenBLAS-for-Android>.
 - **AIX**: Supported on PPC up to POWER8
 - **Haiku**: Supported by the community. We don't actively test the library on this OS.
-- **SunOS**: Supported by the community. We don't actively test the library on this OS:
+- **SunOS**: Supported by the community. We don't actively test the library on this OS.
+- **Cortex-M**: Supported by the community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-use-OpenBLAS-on-Cortex-M>.
 
 ## Usage
 

appveyor.yml

@@ -30,10 +30,10 @@ environment:
     CONDA_INSTALL_LOCN: C:\\Miniconda36-x64
   matrix:
     - COMPILER: clang-cl
-      WITH_FORTRAN: yes
+      WITH_FORTRAN: ON
     - COMPILER: clang-cl
       DYNAMIC_ARCH: ON
-      WITH_FORTRAN: no
+      WITH_FORTRAN: OFF
     - COMPILER: cl
     - COMPILER: MinGW64-gcc-7.2.0-mingw
       DYNAMIC_ARCH: OFF
@@ -47,12 +47,7 @@ environment:
 install:
   - if [%COMPILER%]==[clang-cl] call %CONDA_INSTALL_LOCN%\Scripts\activate.bat
   - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
-  - if [%COMPILER%]==[clang-cl] conda install --yes --quiet clangdev cmake
-
-  - if [%WITH_FORTRAN%]==[no] conda install --yes --quiet ninja
-  - if [%WITH_FORTRAN%]==[yes] conda install --yes --quiet -c isuruf kitware-ninja
-  - if [%WITH_FORTRAN%]==[yes] conda install --yes --quiet flang
-
+  - if [%COMPILER%]==[clang-cl] conda install --yes --quiet clangdev cmake ninja flang=11.0.1
   - if [%COMPILER%]==[clang-cl] call "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvarsall.bat" x64
   - if [%COMPILER%]==[clang-cl] set "LIB=%CONDA_INSTALL_LOCN%\Library\lib;%LIB%"
   - if [%COMPILER%]==[clang-cl] set "CPATH=%CONDA_INSTALL_LOCN%\Library\include;%CPATH%"
@@ -68,15 +63,14 @@ before_build:
   - if [%COMPILER%]==[MinGW64-gcc-7.2.0-mingw] cmake -G "MinGW Makefiles" -DNOFORTRAN=1 ..
   - if [%COMPILER%]==[MinGW-gcc-6.3.0-32] cmake -G "MSYS Makefiles" -DNOFORTRAN=1 ..
   - if [%COMPILER%]==[MinGW-gcc-5.3.0] cmake -G "MSYS Makefiles" -DNOFORTRAN=1 ..
-  - if [%WITH_FORTRAN%]==[no] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DMSVC_STATIC_CRT=ON ..
-  - if [%WITH_FORTRAN%]==[yes] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
+  - if [%WITH_FORTRAN%]==[OFF] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DMSVC_STATIC_CRT=ON ..
+  - if [%WITH_FORTRAN%]==[ON] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
+  - if [%USE_OPENMP%]==[ON] cmake -DUSE_OPENMP=ON ..
   - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON -DDYNAMIC_LIST='CORE2;NEHALEM;SANDYBRIDGE;BULLDOZER;HASWELL' ..
 
 build_script:
   - cmake --build .
 
 test_script:
-  - echo Running Test
-  - cd utest
-  - openblas_utest
+  - ctest -j2
   

azure-pipelines.yml

@@ -68,4 +68,13 @@ jobs:
       dir
       openblas_utest.exe
   
+- job: OSX_OpenMP
+  pool:
+     vmImage: 'macOS-10.15'
+  steps:   
+  - script: |
+      brew update
+      make TARGET=CORE2 DYNAMIC_ARCH=1 USE_OPENMP=1 CC=gcc-10 FC=gfortran-10
       
+      
+       

benchmark/bench.h

@@ -3,6 +3,8 @@
 #include <time.h>
 #ifdef __CYGWIN32__
 #include <sys/time.h>
+#elif defined(__APPLE__)
+#include <mach/mach_time.h>
 #endif
 #include "common.h"
 
@@ -74,6 +76,9 @@ static void *huge_malloc(BLASLONG size){
 
 #if defined(__WIN32__) || defined(__WIN64__) || !defined(_POSIX_TIMERS)
   struct timeval start, stop;
+#elif defined(__APPLE__)
+ mach_timebase_info_data_t info;
+ uint64_t start = 0, stop = 0;
 #else
   struct timespec start = { 0, 0 }, stop = { 0, 0 };
 #endif
@@ -82,6 +87,9 @@ double getsec()
 {
 #if defined(__WIN32__) || defined(__WIN64__) || !defined(_POSIX_TIMERS)
     return (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+#elif defined(__APPLE__)
+    mach_timebase_info(&info);
+    return (double)(((stop - start) * info.numer)/info.denom) * 1.e-9;
 #else
     return (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_nsec - start.tv_nsec)) * 1.e-9;
 #endif
@@ -90,6 +98,8 @@ double getsec()
 void begin() {
 #if defined(__WIN32__) || defined(__WIN64__) || !defined(_POSIX_TIMERS)
     gettimeofday( &start, (struct timezone *)0);
+#elif defined(__APPLE__)
+    start = clock_gettime_nsec_np(CLOCK_UPTIME_RAW);
 #else
     clock_gettime(CLOCK_REALTIME, &start);
 #endif
@@ -98,7 +108,9 @@ void begin() {
 void end() {
 #if defined(__WIN32__) || defined(__WIN64__) || !defined(_POSIX_TIMERS)
     gettimeofday( &stop, (struct timezone *)0);
+#elif defined(__APPLE__)
+    stop = clock_gettime_nsec_np(CLOCK_UPTIME_RAW);
 #else
     clock_gettime(CLOCK_REALTIME, &stop);
 #endif
-}
+}

c_check

@@ -1,11 +1,11 @@
-#!/usr/bin/perl
+#!/usr/bin/env perl
 
 #use File::Basename;
 # use File::Temp qw(tempfile);
 
 # Checking cross compile
 $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
-$hostarch = `uname -m | sed -e s/i.86/x86/`;chop($hostarch);
+$hostarch = `uname -m | sed -e s/i.86/x86/`;
 $hostarch = `uname -p` if ($hostos eq "AIX" || $hostos eq "SunOS");
 chop($hostarch);
 $hostarch = "x86_64" if ($hostarch eq "amd64");
@@ -199,7 +199,7 @@ if (($architecture eq "mips") || ($architecture eq "mips64")) {
     } else {
 	$tmpf = new File::Temp( SUFFIX => '.c' , UNLINK => 1 );
 	$code = '"addvi.b $w0, $w1, 1"';
-	$msa_flags = "-mmsa -mfp64 -msched-weight -mload-store-pairs";
+	$msa_flags = "-mmsa -mfp64 -mload-store-pairs";
 	print $tmpf "#include <msa.h>\n\n";
 	print $tmpf "void main(void){ __asm__ volatile($code); }\n";
 

cblas.h

@@ -125,9 +125,14 @@ void cblas_zswap(OPENBLAS_CONST blasint n, void *x, OPENBLAS_CONST blasint incx,
 
 void cblas_srot(OPENBLAS_CONST blasint N, float *X, OPENBLAS_CONST blasint incX, float *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float c, OPENBLAS_CONST float s);
 void cblas_drot(OPENBLAS_CONST blasint N, double *X, OPENBLAS_CONST blasint incX, double *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double c, OPENBLAS_CONST double  s);
+void cblas_csrot(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float c, OPENBLAS_CONST float s);
+void cblas_zdrot(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double c, OPENBLAS_CONST double s);
 
 void cblas_srotg(float *a, float *b, float *c, float *s);
 void cblas_drotg(double *a, double *b, double *c, double *s);
+void cblas_crotg(void *a, void *b, float *c, void *s);
+void cblas_zrotg(void *a, void *b, double *c, void *s);
+
 
 void cblas_srotm(OPENBLAS_CONST blasint N, float *X, OPENBLAS_CONST blasint incX, float *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float *P);
 void cblas_drotm(OPENBLAS_CONST blasint N, double *X, OPENBLAS_CONST blasint incX, double *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double *P);

cmake/arch.cmake

@@ -45,6 +45,9 @@ endif ()
 if (DYNAMIC_ARCH)
   if (ARM64)
     set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
+    if (DYNAMIC_LIST)
+	    set(DYNAMIC_CORE ARMV8 ${DYNAMIC_LIST})
+    endif ()
   endif ()
   
   if (POWER)

cmake/lapacke.cmake

@@ -2499,6 +2499,5 @@ foreach (Utils_FILE ${Utils_SRC})
 endforeach ()
 
 set(lapacke_include_dir "${NETLIB_LAPACK_DIR}/LAPACKE/include")
-configure_file("${lapacke_include_dir}/lapacke_mangling_with_flags.h.in" "${lapacke_include_dir}/lapacke_mangling.h" COPYONLY)
 include_directories(${lapacke_include_dir})
 set_source_files_properties(${LAPACKE_SOURCES} PROPERTIES COMPILE_FLAGS "${LAPACK_CFLAGS}")

cmake/system.cmake

@@ -148,16 +148,20 @@ endif ()
 include("${PROJECT_SOURCE_DIR}/cmake/prebuild.cmake")
 if (DEFINED TARGET)
   if (${TARGET} STREQUAL COOPERLAKE AND NOT NO_AVX512)
-#    if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
       execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
-        if (${GCC_VERSION} VERSION_GREATER 10.1 OR ${GCC_VERSION} VERSION_EQUAL 10.1)
+        if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 10.09)
           set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=cooperlake")
         else()
           set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
         endif()
-#    elseif (${CMAKE_C_COMPILER_ID} STREQUAL "CLANG")
-#      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
-#    endif()    
+    elseif (${CMAKE_C_COMPILER_ID} STREQUAL "Clang" OR ${CMAKE_C_COMPILER_ID} STREQUAL "AppleClang")
+         if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 8.99)
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=cooperlake")
+        else()
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
+        endif()
+    endif()    
   endif()
   if (${TARGET} STREQUAL SKYLAKEX AND NOT NO_AVX512)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
@@ -233,6 +237,11 @@ if (BINARY64)
   endif ()
 endif ()
 
+if(EMBEDDED)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DOS_EMBEDDED")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -mthumb -mcpu=cortex-m4 -mfloat-abi=hard -mfpu=fpv4-sp-d16")
+endif()
+
 if (NEED_PIC)
   if (${CMAKE_C_COMPILER} STREQUAL "IBM")
     set(CCOMMON_OPT "${CCOMMON_OPT} -qpic=large")

cmake/utils.cmake

@@ -74,6 +74,9 @@ macro(ParseMakefileVars MAKEFILE_IN)
           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 ( ${CMAKE_MATCH_1} STREQUAL C_COMPILER)
+                 set (CMAKE_MATCH_1 CMAKE_C_COMPILER)
+            endif ()
             if (NOT ( ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2}))
 #              message (STATUS "condition is true")
               set (IfElse 1)

common.h

@@ -122,7 +122,7 @@ extern "C" {
 #define ATOM GOTO_ATOM
 #undef  GOTO_ATOM
 #endif
-#else
+#elif !defined(OS_EMBEDDED)
 #include <sys/mman.h>
 #ifndef NO_SYSV_IPC
 #include <sys/shm.h>
@@ -134,6 +134,9 @@ extern "C" {
 #if defined(SMP) || defined(USE_LOCKING)
 #include <pthread.h>
 #endif
+#else
+#include <time.h>
+#include <math.h>
 #endif
 
 #if defined(OS_SUNOS)
@@ -488,10 +491,12 @@ static inline unsigned long long rpcc(void){
   struct timespec ts;
   clock_gettime(CLOCK_MONOTONIC, &ts);
   return (unsigned long long)ts.tv_sec * 1000000000ull + ts.tv_nsec;
-#else
+#elif !defined(OS_EMBEDDED)
   struct timeval tv;
   gettimeofday(&tv,NULL);
   return (unsigned long long)tv.tv_sec * 1000000000ull + tv.tv_usec * 1000;
+#else
+  return 0;
 #endif
 }
 #define RPCC_DEFINED
@@ -521,6 +526,10 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #include "common_linux.h"
 #endif
 
+#ifdef OS_EMBEDDED
+#define DTB_DEFAULT_ENTRIES 64
+#endif
+
 #define MMAP_ACCESS (PROT_READ | PROT_WRITE)
 
 #ifdef __NetBSD__

common_arm64.h

@@ -39,7 +39,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define INLINE inline
 
-#ifdef F_INTERFACE_FLANG
+#if defined( F_INTERFACE_FLANG) || defined(F_INTERFACE_PGI)
 #define RETURN_BY_STACK
 #else
 #define RETURN_BY_COMPLEX

common_linux.h

@@ -75,18 +75,10 @@ static inline int my_mbind(void *addr, unsigned long len, int mode,
 // https://lsbbugs.linuxfoundation.org/show_bug.cgi?id=3482
         return 0;
 #else
-#if defined (LOONGSON3B)
-#if defined (__64BIT__)
-	return syscall(SYS_mbind, addr, len, mode, nodemask, maxnode, flags);
-#else
-	return 0; //NULL Implementation on Loongson 3B 32bit.
-#endif
-#else
 //Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34
 //	unsigned long null_nodemask=0;
 	return syscall(SYS_mbind, addr, len, mode, nodemask, maxnode, flags);
 #endif
-#endif
 }
 
 static inline int my_set_mempolicy(int mode, const unsigned long *addr, unsigned long flag) {

common_mips64.h

@@ -229,12 +229,7 @@ REALNAME: ;\
 
 #define BUFFER_SIZE     ( 32 << 21)
 
-#if defined(LOONGSON3A)
-#define PAGESIZE	(16UL << 10)
-#define FIXED_PAGESIZE	(16UL << 10)
-#endif
-
-#if defined(LOONGSON3B)
+#if defined(LOONGSON3R3) || defined(LOONGSON3R4)
 #define PAGESIZE	(16UL << 10)
 #define FIXED_PAGESIZE	(16UL << 10)
 #endif
@@ -250,7 +245,7 @@ REALNAME: ;\
 #define MAP_ANONYMOUS MAP_ANON
 #endif
 
-#if defined(LOONGSON3A) || defined(LOONGSON3B)
+#if defined(LOONGSON3R3) || defined(LOONGSON3R4)
 #define PREFETCHD_(x) ld $0, x
 #define PREFETCHD(x)  PREFETCHD_(x)
 #else

cpuid_mips64.c

@@ -70,19 +70,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* or implied, of The University of Texas at Austin.                 */
 /*********************************************************************/
 
-#define CPU_UNKNOWN     0
-#define CPU_SICORTEX    1
-#define CPU_LOONGSON3A  2
-#define CPU_LOONGSON3B  3
-#define CPU_I6400       4
-#define CPU_P6600       5
-#define CPU_I6500       6
+#define CPU_UNKNOWN      0
+#define CPU_SICORTEX     1
+#define CPU_LOONGSON3R3  2
+#define CPU_LOONGSON3R4  3
+#define CPU_I6400        4
+#define CPU_P6600        5
+#define CPU_I6500        6
 
 static char *cpuname[] = {
   "UNKNOWN",
   "SICORTEX",
-  "LOONGSON3A",
-  "LOONGSON3B",
+  "LOONGSON3R3",
+  "LOONGSON3R4",
   "I6400",
   "P6600",
   "I6500"
@@ -90,48 +90,13 @@ static char *cpuname[] = {
 
 int detect(void){
 
-#ifdef __linux
+#ifdef linux
   FILE *infile;
   char buffer[512], *p;
 
   p = (char *)NULL;
-  infile = fopen("/proc/cpuinfo", "r");
-  while (fgets(buffer, sizeof(buffer), infile)){
-    if (!strncmp("cpu", buffer, 3)){
-	p = strchr(buffer, ':') + 2;
-#if 0
-	fprintf(stderr, "%s\n", p);
-#endif
-	break;
-      }
-  }
-
-  fclose(infile);
-
-  if(p != NULL){
-  if (strstr(p, "Loongson-3A")){
-    return CPU_LOONGSON3A;
-  }else if(strstr(p, "Loongson-3B")){
-    return CPU_LOONGSON3B;
-  }else if (strstr(p, "Loongson-3")){
-    infile = fopen("/proc/cpuinfo", "r");
-    p = (char *)NULL;
-    while (fgets(buffer, sizeof(buffer), infile)){
-      if (!strncmp("system type", buffer, 11)){
-	p = strchr(buffer, ':') + 2;
-	break;
-      }
-    }
-    fclose(infile);
-    if (strstr(p, "loongson3a"))
-      return CPU_LOONGSON3A;
-  }else{
-    return CPU_SICORTEX;
-  }
-  }
   //Check model name for Loongson3
   infile = fopen("/proc/cpuinfo", "r");
-  p = (char *)NULL;
   while (fgets(buffer, sizeof(buffer), infile)){
     if (!strncmp("model name", buffer, 10)){
       p = strchr(buffer, ':') + 2;
@@ -140,14 +105,16 @@ int detect(void){
   }
   fclose(infile);
   if(p != NULL){
-  if (strstr(p, "Loongson-3A")){
-    return CPU_LOONGSON3A;
-  }else if(strstr(p, "Loongson-3B")){
-    return CPU_LOONGSON3B;
-  }
+  if (strstr(p, "Loongson-3A3000") || strstr(p, "Loongson-3B3000")){
+    return CPU_LOONGSON3R3;
+  }else if(strstr(p, "Loongson-3A4000") || strstr(p, "Loongson-3B4000")){
+    return CPU_LOONGSON3R4;
+  } else{
+    return CPU_SICORTEX;
   }
 #endif
     return CPU_UNKNOWN;
+  }
 }
 
 char *get_corename(void){
@@ -159,10 +126,10 @@ void get_architecture(void){
 }
 
 void get_subarchitecture(void){
-  if(detect()==CPU_LOONGSON3A) {
-    printf("LOONGSON3A");
-  }else if(detect()==CPU_LOONGSON3B){
-    printf("LOONGSON3B");
+  if(detect()==CPU_LOONGSON3R3) {
+    printf("LOONGSON3R3");
+  }else if(detect()==CPU_LOONGSON3R4){
+    printf("LOONGSON3R4");
   }else if(detect()==CPU_I6400){
     printf("I6400");
   }else if(detect()==CPU_P6600){
@@ -179,8 +146,8 @@ void get_subdirname(void){
 }
 
 void get_cpuconfig(void){
-  if(detect()==CPU_LOONGSON3A) {
-    printf("#define LOONGSON3A\n");
+  if(detect()==CPU_LOONGSON3R3) {
+    printf("#define LOONGSON3R3\n");
     printf("#define L1_DATA_SIZE 65536\n");
     printf("#define L1_DATA_LINESIZE 32\n");
     printf("#define L2_SIZE 512488\n");
@@ -188,8 +155,8 @@ void get_cpuconfig(void){
     printf("#define DTB_DEFAULT_ENTRIES 64\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 4\n");
-  }else if(detect()==CPU_LOONGSON3B){
-    printf("#define LOONGSON3B\n");
+  }else if(detect()==CPU_LOONGSON3R4){
+    printf("#define LOONGSON3R4\n");
     printf("#define L1_DATA_SIZE 65536\n");
     printf("#define L1_DATA_LINESIZE 32\n");
     printf("#define L2_SIZE 512488\n");
@@ -237,10 +204,10 @@ void get_cpuconfig(void){
 }
 
 void get_libname(void){
-  if(detect()==CPU_LOONGSON3A) {
-    printf("loongson3a\n");
-  }else if(detect()==CPU_LOONGSON3B) {
-    printf("loongson3b\n");
+  if(detect()==CPU_LOONGSON3R3) {
+    printf("loongson3r3\n");
+  }else if(detect()==CPU_LOONGSON3R4) {
+    printf("loongson3r4\n");
   }else if(detect()==CPU_I6400) {
     printf("i6400\n");
   }else if(detect()==CPU_P6600) {

cpuid_x86.c

@@ -1418,6 +1418,15 @@ int get_cpuname(void){
       case 9:
       case 8:      
         switch (model) {
+        case 12: // Tiger Lake
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+            return CPUTYPE_SANDYBRIDGE;
+          else
+          return CPUTYPE_NEHALEM;
 	case 14: // Kaby Lake and refreshes
           if(support_avx2())
             return CPUTYPE_HASWELL;
@@ -1436,6 +1445,15 @@ int get_cpuname(void){
         return CPUTYPE_SANDYBRIDGE;
           else
         return CPUTYPE_NEHALEM;
+    case 7: // Rocket Lake           
+	  if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	  return CPUTYPE_NEHALEM;
 	}
 	break;    
       }
@@ -2014,6 +2032,19 @@ int get_coretype(void){
   #endif
             else
               return CORE_NEHALEM;
+	  case 7:// Rocket Lake
+#ifndef NO_AVX512
+	  if(support_avx512())
+            return CORE_SKYLAKEX;
+#endif
+#ifndef NO_AVX2
+	  if(support_avx2())
+            return CORE_HASWELL;
+#endif
+	  if(support_avx())
+	    return CORE_SANDYBRIDGE;
+	  else
+	  return CORE_NEHALEM;
         }
       case 5:
         switch (model) {
@@ -2102,6 +2133,16 @@ int get_coretype(void){
         break;    	
       case 9:
       case 8:
+       if (model == 12) { // Tiger Lake
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+            return CPUTYPE_SANDYBRIDGE;
+          else
+          return CPUTYPE_NEHALEM;
+        }
         if (model == 14) { // Kaby Lake 
 	  if(support_avx())
 #ifndef NO_AVX2

ctest/CMakeLists.txt

@@ -5,9 +5,18 @@ enable_language(Fortran)
 
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DADD${BU} -DCBLAS")
 
+if(WIN32)
+FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.ps1
+"$ErrorActionPreference = \"Stop\"\n"
+"Get-Content $args[1] | & $args[0]\n"
+)
+set(test_helper powershell -ExecutionPolicy Bypass "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.ps1")
+else()
 FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh
 "$1 < $2\n"
 )
+set(test_helper sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh")
+endif()
 
 foreach(float_type ${FLOAT_TYPES})
   string(SUBSTRING ${float_type} 0 1 float_char_upper)
@@ -21,7 +30,7 @@ foreach(float_type ${FLOAT_TYPES})
     c_${float_char}blas1.c)
   target_link_libraries(x${float_char}cblat1 ${OpenBLAS_LIBNAME})
   add_test(NAME "x${float_char}cblat1"
-    COMMAND "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat1")
+    COMMAND $<TARGET_FILE:x${float_char}cblat1>)
 
   #level2
   add_executable(x${float_char}cblat2
@@ -33,7 +42,7 @@ foreach(float_type ${FLOAT_TYPES})
     constant.c)
   target_link_libraries(x${float_char}cblat2 ${OpenBLAS_LIBNAME})
   add_test(NAME "x${float_char}cblat2"
-    COMMAND sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh" "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat2" "${PROJECT_SOURCE_DIR}/ctest/${float_char}in2")
+    COMMAND ${test_helper} $<TARGET_FILE:x${float_char}cblat2> "${PROJECT_SOURCE_DIR}/ctest/${float_char}in2")
 
   #level3
   add_executable(x${float_char}cblat3
@@ -45,6 +54,6 @@ foreach(float_type ${FLOAT_TYPES})
     constant.c)
   target_link_libraries(x${float_char}cblat3 ${OpenBLAS_LIBNAME})
   add_test(NAME "x${float_char}cblat3"
-    COMMAND sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh" "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat3" "${PROJECT_SOURCE_DIR}/ctest/${float_char}in3")
+    COMMAND ${test_helper} $<TARGET_FILE:x${float_char}cblat3> "${PROJECT_SOURCE_DIR}/ctest/${float_char}in3")
 
 endforeach()

ctest/Makefile

@@ -212,6 +212,9 @@ ifeq ($(C_COMPILER), CLANG)
 CEXTRALIB = -lomp
 endif
 endif
+ifeq ($(F_COMPILER), NAG)
+CEXTRALIB = -lgomp
+endif
 endif
 
 ifeq ($(BUILD_SINGLE),1)

driver/level3/level3.c

@@ -339,8 +339,10 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 #else
         if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
         else
-        	if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
+/*
+		if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
         	else
+*/
           		if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 #endif
 

driver/level3/level3_thread.c

@@ -373,8 +373,10 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 #else
 	if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
 	else
+/*
           if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
           else
+*/
             if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 #endif
         /* Copy part of local region of B into workspace */

driver/others/Makefile

@@ -24,10 +24,14 @@ else
 ifeq ($(ARCH),zarch)
 COMMONOBJS += dynamic_zarch.$(SUFFIX)
 else
+ifeq ($(ARCH),mips64)
+COMMONOBJS += dynamic_mips64.$(SUFFIX)
+else
 COMMONOBJS	+=  dynamic.$(SUFFIX)
 endif
 endif
 endif
+endif
 else
 COMMONOBJS	+=  parameter.$(SUFFIX)
 endif
@@ -92,10 +96,14 @@ else
 ifeq ($(ARCH),zarch)
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic_zarch.$(SUFFIX)
 else
+ifeq ($(ARCH),mips64)
+HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic_mips64.$(SUFFIX)
+else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic.$(SUFFIX)
 endif
 endif
 endif
+endif
 else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) parameter.$(SUFFIX)
 endif

driver/others/blas_server.c

@@ -967,9 +967,11 @@ void goto_set_num_threads(int num_threads) {
   blas_cpu_number  = num_threads;
 
 #if defined(ARCH_MIPS64)
+#ifndef DYNAMIC_ARCH
   //set parameters for different number of threads.
   blas_set_parameter();
 #endif
+#endif
 
 }
 
@@ -1022,38 +1024,39 @@ int BLASFUNC(blas_thread_shutdown)(void){
 
   int i;
 
-  if (!blas_server_avail) return 0;
-
   LOCK_COMMAND(&server_lock);
 
-  for (i = 0; i < blas_num_threads - 1; i++) {
+  if (blas_server_avail) {
+
+    for (i = 0; i < blas_num_threads - 1; i++) {
 
 
-    pthread_mutex_lock (&thread_status[i].lock);
+      pthread_mutex_lock (&thread_status[i].lock);
 
-    atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)-1);
-    thread_status[i].status = THREAD_STATUS_WAKEUP;
-    pthread_cond_signal (&thread_status[i].wakeup);
+      atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)-1);
+      thread_status[i].status = THREAD_STATUS_WAKEUP;
+      pthread_cond_signal (&thread_status[i].wakeup);
 
-    pthread_mutex_unlock(&thread_status[i].lock);
+      pthread_mutex_unlock(&thread_status[i].lock);
 
-  }
+    }
 
-  for(i = 0; i < blas_num_threads - 1; i++){
-    pthread_join(blas_threads[i], NULL);
-  }
+    for(i = 0; i < blas_num_threads - 1; i++){
+      pthread_join(blas_threads[i], NULL);
+    }
 
-  for(i = 0; i < blas_num_threads - 1; i++){
-    pthread_mutex_destroy(&thread_status[i].lock);
-    pthread_cond_destroy (&thread_status[i].wakeup);
-  }
+    for(i = 0; i < blas_num_threads - 1; i++){
+      pthread_mutex_destroy(&thread_status[i].lock);
+      pthread_cond_destroy (&thread_status[i].wakeup);
+    }
 
 #ifdef NEED_STACKATTR
-  pthread_attr_destory(&attr);
+    pthread_attr_destroy(&attr);
 #endif
 
-  blas_server_avail = 0;
+    blas_server_avail = 0;
 
+  }
   UNLOCK_COMMAND(&server_lock);
 
   return 0;

driver/others/dynamic.c

@@ -644,6 +644,21 @@ static gotoblas_t *get_coretype(void){
         return NULL;  
       case 9:
       case 8:
+        if (model == 12) { // Tiger Lake
+          if (support_avx512()) 
+            return &gotoblas_SKYLAKEX;
+          if(support_avx2()){
+            openblas_warning(FALLBACK_VERBOSE, HASWELL_FALLBACK);
+            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;
+          }
+        }
 	if (model == 14 ) { // Kaby Lake, Coffee Lake
 	  if(support_avx2())
 	    return &gotoblas_HASWELL;
@@ -656,7 +671,7 @@ static gotoblas_t *get_coretype(void){
 	  }
 	}
       case 10:
-    if (model == 5 || model == 6) {
+        if (model == 5 || model == 6) {
 	  if(support_avx2())
 	    return &gotoblas_HASWELL;
 	  if(support_avx()) {
@@ -666,7 +681,20 @@ static gotoblas_t *get_coretype(void){
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
-    }
+        }
+        if (model == 7) {
+	  if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  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:

driver/others/dynamic_arm64.c

@@ -43,6 +43,63 @@
 #endif
 
 extern gotoblas_t  gotoblas_ARMV8;
+#ifdef DYNAMIC_LIST
+#ifdef DYN_CORTEXA53
+extern gotoblas_t  gotoblas_CORTEXA53;
+#else
+#define gotoblas_CORTEXA53 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEXA57
+extern gotoblas_t  gotoblas_CORTEXA57;
+#else
+#define gotoblas_CORTEXA57 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEXA72
+extern gotoblas_t  gotoblas_CORTEXA72;
+#else
+#define gotoblas_CORTEXA72 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEXA73
+extern gotoblas_t  gotoblas_CORTEXA73;
+#else
+#define gotoblas_CORTEXA73 gotoblas_ARMV8
+#endif
+#ifdef DYN_FALKOR
+extern gotoblas_t  gotoblas_FALKOR;
+#else
+#define gotoblas_FALKOR gotoblas_ARMV8
+#endif
+#ifdef DYN_TSV110
+extern gotoblas_t  gotoblas_TSV110;
+#else
+#define gotoblas_TSV110 gotoblas_ARMV8
+#endif
+#ifdef DYN_THUNDERX
+extern gotoblas_t  gotoblas_THUNDERX;
+#else
+#define gotoblas_THUNDERX gotoblas_ARMV8
+#endif
+#ifdef DYN_THUNDERX2T99
+extern gotoblas_t  gotoblas_THUNDERX2T99;
+#else
+#define gotoblas_THUNDERX2T99 gotoblas_ARMV8
+#endif
+#ifdef DYN_THUNDERX3T110
+extern gotoblas_t  gotoblas_THUNDERX3T110;
+#else
+#define gotoblas_THUNDERX3T110 gotoblas_ARMV8
+#endif
+#ifdef DYN_EMAG8180
+extern gotoblas_t  gotoblas_EMAG8180;
+#else
+#define gotoblas_EMAG8180 gotoblas_ARMV8
+#endif
+#ifdef DYN_NEOVERSEN1
+extern gotoblas_t  gotoblas_NEOVERSEN1;
+#else
+#define gotoblas_NEOVERSEN1 gotoblas_ARMV8
+#endif
+#else
 extern gotoblas_t  gotoblas_CORTEXA53;
 extern gotoblas_t  gotoblas_CORTEXA57;
 extern gotoblas_t  gotoblas_CORTEXA72;
@@ -54,6 +111,7 @@ extern gotoblas_t  gotoblas_TSV110;
 extern gotoblas_t  gotoblas_EMAG8180;
 extern gotoblas_t  gotoblas_NEOVERSEN1;
 extern gotoblas_t  gotoblas_THUNDERX3T110;
+#endif
 
 extern void openblas_warning(int verbose, const char * msg);
 
@@ -68,7 +126,7 @@ extern void openblas_warning(int verbose, const char * msg);
 #endif
 
 #define get_cpu_ftr(id, var) ({					\
-		__asm__("mrs %0, "#id : "=r" (var));		\
+		__asm__ ("mrs %0, "#id : "=r" (var));		\
 	})
 
 static char *corename[] = {

driver/others/dynamic_mips64.c

@@ -0,0 +1,230 @@
+/*****************************************************************************
+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 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 <sys/wait.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/resource.h>
+#include "common.h"
+
+extern gotoblas_t  gotoblas_LOONGSON3R3;
+extern gotoblas_t  gotoblas_LOONGSON3R4;
+
+extern void openblas_warning(int verbose, const char * msg);
+
+#define NUM_CORETYPES    2
+
+static char *corename[] = {
+  "loongson3r3",
+  "loongson3r4",
+  "UNKNOWN"
+};
+
+char *gotoblas_corename(void) {
+  if (gotoblas == &gotoblas_LOONGSON3R3)    return corename[0];
+  if (gotoblas == &gotoblas_LOONGSON3R4)    return corename[1];
+  return corename[NUM_CORETYPES];
+}
+
+static gotoblas_t *force_coretype(char *coretype) {
+  int i;
+  int found = -1;
+  char message[128];
+
+  for ( i=0 ; i < NUM_CORETYPES; i++)
+  {
+    if (!strncasecmp(coretype, corename[i], 20))
+    {
+        found = i;
+        break;
+    }
+  }
+
+  switch (found)
+  {
+    case  0: return (&gotoblas_LOONGSON3R3);
+    case  1: return (&gotoblas_LOONGSON3R4);
+  }
+  snprintf(message, 128, "Core not found: %s\n", coretype);
+  openblas_warning(1, message);
+  return NULL;
+}
+
+#define MMI_MASK    0x00000010
+#define MSA_MASK    0x00000020
+
+int fd[2];
+int support_cpucfg;
+
+static void handler(int signum)
+{
+    close(fd[1]);
+    exit(1);
+}
+
+/* Brief :  Function to check if cpucfg supported on loongson
+ * Return:  1   supported
+ *          0   not supported
+ */
+static int cpucfg_test(void) {
+    pid_t pid;
+    int status = 0;
+
+    support_cpucfg = 0;
+    pipe(fd);
+    pid = fork();
+    if (pid == 0) { /* Subprocess */
+        struct sigaction act;
+        close(fd[0]);
+        /* Set signal action for SIGILL. */
+        act.sa_handler = handler;
+        sigaction(SIGILL,&act,NULL);
+
+        /* Execute cpucfg in subprocess. */
+        __asm__ volatile(
+            ".insn              \n\t"
+            ".word (0xc8080118) \n\t"
+            :::
+        );
+        support_cpucfg = 1;
+        write(fd[1],&support_cpucfg,sizeof(support_cpucfg));
+        close(fd[1]);
+        exit(0);
+    } else if (pid > 0){ /* Parent process*/
+        close(fd[1]);
+        if ((waitpid(pid,&status,0) <= 0) ||
+            (read(fd[0],&support_cpucfg,sizeof(support_cpucfg)) <= 0))
+            support_cpucfg = 0;
+        close(fd[0]);
+    } else {
+        support_cpucfg = 0;
+    }
+
+    return support_cpucfg;
+}
+
+static gotoblas_t *get_coretype_from_cpucfg(void) {
+    int flag = 0;
+    __asm__ volatile(
+        ".insn                     \n\t"
+        "dli    $8,    0x01        \n\t"
+        ".word (0xc9084918)        \n\t"
+        "usw    $9,    0x00(%0)    \n\t"
+        :
+        : "r"(&flag)
+        : "memory"
+    );
+    if (flag & MSA_MASK)
+        return (&gotoblas_LOONGSON3R4);
+    if (flag & MMI_MASK)
+        return (&gotoblas_LOONGSON3R3);
+    return NULL;
+}
+
+static gotoblas_t *get_coretype_from_cpuinfo(void) {
+#ifdef linux
+  FILE *infile;
+  char buffer[512], *p;
+
+  p = (char *)NULL;
+  //Check model name for Loongson3
+  infile = fopen("/proc/cpuinfo", "r");
+  while (fgets(buffer, sizeof(buffer), infile)){
+    if (!strncmp("model name", buffer, 10)){
+      p = strchr(buffer, ':') + 2;
+      break;
+    }
+  }
+  fclose(infile);
+  if(p != NULL){
+   if (strstr(p, "Loongson-3A3000") || strstr(p, "Loongson-3B3000"))
+     return (&gotoblas_LOONGSON3R3);
+   else if(strstr(p, "Loongson-3A4000") || strstr(p, "Loongson-3B4000"))
+     return (&gotoblas_LOONGSON3R4);
+   else
+     return NULL;
+  }
+#endif
+    return NULL;
+}
+
+static gotoblas_t *get_coretype(void) {
+    int ret = 0;
+
+    ret = cpucfg_test();
+    if (ret == 1)
+        return get_coretype_from_cpucfg();
+    else
+        return get_coretype_from_cpuinfo();
+}
+
+void gotoblas_dynamic_init(void) {
+  char coremsg[128];
+  char coren[22];
+  char *p;
+
+  if (gotoblas) return;
+
+  p = getenv("OPENBLAS_CORETYPE");
+  if ( p )
+  {
+    gotoblas = force_coretype(p);
+  }
+  else
+  {
+    gotoblas = get_coretype();
+  }
+
+  if (gotoblas == NULL)
+  {
+    snprintf(coremsg, 128, "Falling back to loongson3r3 core\n");
+    openblas_warning(1, coremsg);
+    gotoblas = &gotoblas_LOONGSON3R3;
+  }
+
+  if (gotoblas && gotoblas->init) {
+    strncpy(coren, gotoblas_corename(), 20);
+    sprintf(coremsg, "Core: %s\n", coren);
+    openblas_warning(2, coremsg);
+    gotoblas -> init();
+  } else {
+    openblas_warning(0, "OpenBLAS : Architecture Initialization failed. No initialization function found.\n");
+    exit(1);
+  }
+
+}
+
+void gotoblas_dynamic_quit(void) {
+  gotoblas = NULL;
+}

driver/others/dynamic_power.c

@@ -27,7 +27,9 @@ static char *corename[] = {
 #define NUM_CORETYPES 4
 
 char *gotoblas_corename(void) {
+#ifndef C_PGI
 	if (gotoblas == &gotoblas_POWER6)	return corename[1];
+#endif
 	if (gotoblas == &gotoblas_POWER8)	return corename[2];
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	if (gotoblas == &gotoblas_POWER9)	return corename[3];
@@ -38,10 +40,164 @@ char *gotoblas_corename(void) {
 	return corename[0];
 }
 
+#if defined(__clang__)
+static int __builtin_cpu_supports(char* arg) 
+{
+	return 0;
+}
+#endif
+
+#if defined(C_PGI) || defined(__clang__)
+/*
+ * NV HPC compilers do not yet implement __builtin_cpu_is().
+ * Fake a version here for use in the CPU detection code below.
+ *
+ * Strategy here is to first check the CPU to see what it actually is,
+ * and then test the input to see if what the CPU actually is matches
+ * what was requested.
+ */
+
+#include <string.h>
+
+/*
+ *  Define POWER processor version table.
+ *
+ *  NOTE NV HPC SDK compilers only support POWER8 and POWER9 at this time
+ */
+
+#define CPU_UNKNOWN 0
+#define CPU_POWER5 5
+#define CPU_POWER6 6
+#define CPU_POWER8 8
+#define CPU_POWER9 9
+#define CPU_POWER10 10
+
+static  struct {
+    uint32_t    pvr_mask;
+    uint32_t    pvr_value;
+    const char* cpu_name;
+    uint32_t    cpu_type;
+} pvrPOWER [] = {
+
+    {   /* POWER6 in P5+ mode; 2.04-compliant processor */
+        .pvr_mask       = 0xffffffff,
+        .pvr_value      = 0x0f000001,
+        .cpu_name       = "POWER5+",
+        .cpu_type       = CPU_POWER5,
+    },
+
+    {   /* Power6 aka POWER6X*/
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x003e0000,
+        .cpu_name       = "POWER6 (raw)",
+        .cpu_type       = CPU_POWER6,
+    },
+
+    {   /* Power7 */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x003f0000,
+        .cpu_name       = "POWER7 (raw)",
+        .cpu_type       = CPU_POWER6,
+    },
+
+    {   /* Power7+ */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004A0000,
+        .cpu_name       = "POWER7+ (raw)",
+        .cpu_type       = CPU_POWER6,
+    },
+
+    {   /* Power8E */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004b0000,
+        .cpu_name       = "POWER8E (raw)",
+        .cpu_type       = CPU_POWER8,
+    },
+
+    {   /* Power8NVL */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004c0000,
+        .cpu_name       = "POWER8NVL (raw)",
+        .cpu_type       = CPU_POWER8,
+    },
+
+    {   /* Power8 */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004d0000,
+        .cpu_name       = "POWER8 (raw)",
+        .cpu_type       = CPU_POWER8,
+    },
+
+    {   /* Power9 DD2.0 */
+        .pvr_mask       = 0xffffefff,
+        .pvr_value      = 0x004e0200,
+        .cpu_name       = "POWER9 (raw)",
+        .cpu_type       = CPU_POWER9,
+    },
+
+    {   /* Power9 DD 2.1 */
+        .pvr_mask       = 0xffffefff,
+        .pvr_value      = 0x004e0201,
+        .cpu_name       = "POWER9 (raw)",
+        .cpu_type       = CPU_POWER9,
+    },
+
+    {   /* Power9 DD2.2 or later */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004e0000,
+        .cpu_name       = "POWER9 (raw)",
+        .cpu_type       = CPU_POWER9,
+    },
+
+    {   /* Power10 */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x00800000,
+        .cpu_name       = "POWER10 (raw)",
+        .cpu_type       = CPU_POWER10,
+    },
+
+    {   /* End of table, pvr_mask and pvr_value must be zero */
+        .pvr_mask       = 0x0,
+        .pvr_value      = 0x0,
+        .cpu_name       = "Unknown",
+        .cpu_type       = CPU_UNKNOWN,
+    },
+};
+
+static int __builtin_cpu_is(const char *cpu) {
+	int i;
+	uint32_t pvr;
+	uint32_t cpu_type;
+
+	asm("mfpvr    %0" : "=r"(pvr));
+
+	for (i = 0 ; i < sizeof pvrPOWER / sizeof *pvrPOWER ; ++i) {
+		if ((pvr & pvrPOWER[i].pvr_mask) == pvrPOWER[i].pvr_value) {
+			break;
+		}
+	}
+
+#if defined(DEBUG)
+	printf("%s: returning CPU=%s, cpu_type=%p\n", __func__,
+		pvrPOWER[i].cpu_name, pvrPOWER[i].cpu_type);
+#endif
+	cpu_type = pvrPOWER[i].cpu_type;
+
+	if (!strcmp(cpu, "power8"))
+		return cpu_type == CPU_POWER8;
+	if (!strcmp(cpu, "power9"))
+		return cpu_type == CPU_POWER9;
+	return 0;
+}
+
+#endif  /* C_PGI */
+
 static gotoblas_t *get_coretype(void) {
 
+#ifndef C_PGI
 	if (__builtin_cpu_is("power6") || __builtin_cpu_is("power6x"))
 		return &gotoblas_POWER6;
+#endif
 	if (__builtin_cpu_is("power8"))
 		return &gotoblas_POWER8;
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
@@ -52,6 +208,11 @@ static gotoblas_t *get_coretype(void) {
 	if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma"))
 		return &gotoblas_POWER10;
 #endif
+	/* Fall back to the POWER9 implementation if the toolchain is too old or the MMA feature is not set */
+#if (!defined __GNUC__) || ( __GNUC__ >= 11) || (__GNUC__ == 10 && __GNUC_MINOR__ >= 2)
+	if (__builtin_cpu_is("power10"))
+		return &gotoblas_POWER9;
+#endif	
 	return NULL;
 }
 
@@ -72,7 +233,9 @@ static gotoblas_t *force_coretype(char * coretype) {
 
 	switch (found)
 	{
+#ifndef C_PGI
 	case  1: return (&gotoblas_POWER6);
+#endif
 	case  2: return (&gotoblas_POWER8);
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	case  3: return (&gotoblas_POWER9);

driver/others/memory.c

@@ -222,11 +222,11 @@ int get_num_procs(void);
 #else
 int get_num_procs(void) {
   static int nums = 0;
+
+#if defined(__GLIBC_PREREQ)
   cpu_set_t cpuset,*cpusetp;
   size_t size;
   int ret;
-
-#if defined(__GLIBC_PREREQ)
 #if !__GLIBC_PREREQ(2, 7)
   int i;
 #if !__GLIBC_PREREQ(2, 6)
@@ -1241,7 +1241,7 @@ UNLOCK_COMMAND(&alloc_lock);
 
       func = &memoryalloc[0];
 
-      while ((func != NULL) && (map_address == (void *) -1)) {
+      while ((*func != NULL) && (map_address == (void *) -1)) {
 
         map_address = (*func)((void *)base_address);
 
@@ -1619,10 +1619,12 @@ static int on_process_term(void)
 #else
 #pragma data_seg(".CRT$XLB")
 #endif
-static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
+
 #ifdef _WIN64
+static const PIMAGE_TLS_CALLBACK dll_callback(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
 #pragma const_seg()
 #else
+static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
 #pragma data_seg()
 #endif
 
@@ -1631,10 +1633,12 @@ static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOI
 #else
 #pragma data_seg(".CRT$XTU")
 #endif
-static int(*p_process_term)(void) = on_process_term;
+
 #ifdef _WIN64
+static const int(*p_process_term)(void) = on_process_term;
 #pragma const_seg()
 #else
+static int(*p_process_term)(void) = on_process_term;
 #pragma data_seg()
 #endif
 #endif
@@ -1668,16 +1672,23 @@ void gotoblas_dummy_for_PGI(void) {
 #ifndef MEM_LARGE_PAGES
 #define MEM_LARGE_PAGES  0x20000000
 #endif
-#else
+#elif !defined(OS_EMBEDDED)
 #define ALLOC_MMAP
 #define ALLOC_MALLOC
+#else
+#define ALLOC_MALLOC
+
+inline int puts(const char *str) { return 0; }
+inline int printf(const char *format, ...) { return 0; }
+inline char *getenv(const char *name) { return ""; }
+inline int atoi(const char *str) { return 0; }
 #endif
 
 #include <stdlib.h>
 #include <stdio.h>
 #include <fcntl.h>
 
-#if !defined(OS_WINDOWS) || defined(OS_CYGWIN_NT)
+#if (!defined(OS_WINDOWS) || defined(OS_CYGWIN_NT)) && !defined(OS_EMBEDDED)
 #include <sys/mman.h>
 #ifndef NO_SYSV_IPC
 #include <sys/shm.h>

driver/others/parameter.c

@@ -717,7 +717,7 @@ void blas_set_parameter(void){
 
 #if defined(ARCH_MIPS64)
 void blas_set_parameter(void){
-#if defined(LOONGSON3A)
+#if defined(LOONGSON3R3) || defined(LOONGSON3R4)
 #ifdef SMP
   if(blas_num_threads == 1){
 #endif
@@ -731,20 +731,6 @@ void blas_set_parameter(void){
 #endif
 #endif
 
-#if defined(LOONGSON3B)
-#ifdef SMP
-  if(blas_num_threads == 1 || blas_num_threads == 2){
-#endif
-    //single thread
-    dgemm_r = 640;
-#ifdef SMP
-  }else{
-    //multi thread
-    dgemm_r = 160;
-  }
-#endif
-#endif
-
 }
 #endif
 

exports/gensymbol

@@ -1,4 +1,4 @@
-#!/usr/bin/perl
+#!/usr/bin/env perl
 
 # Changelog
 # 2017/09/03 staticfloat

f_check

@@ -1,4 +1,4 @@
-#!/usr/bin/perl
+#!/usr/bin/env perl
 
 $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
 
@@ -32,9 +32,9 @@ if ($compiler eq "") {
               "xlf95", "xlf90", "xlf",
               "ppuf77", "ppuf95", "ppuf90", "ppuxlf",
 	      "pathf90", "pathf95",
-	      "pgf95", "pgf90", "pgf77",
+	      "pgf95", "pgf90", "pgf77", "pgfortran", "nvfortran",
 	      "flang", "egfortran",
-              "ifort");
+              "ifort", "nagfor");
 
 OUTER:
     foreach $lists (@lists) {
@@ -64,7 +64,9 @@ if ($compiler eq "") {
     if (!$?) {
 
 	$data = `$compiler -O2 -S ftest.f > /dev/null 2>&1 && cat ftest.s && rm -f ftest.s`;
-
+	if ($data eq "") {
+		$data = `$compiler -O2 -S ftest.f > /dev/null 2>&1 && cat ftest.c && rm -f ftest.c`;
+	}
 	if ($data =~ /zhoge_/) {
 	    $bu       = "_";
 	}
@@ -76,6 +78,7 @@ if ($compiler eq "") {
 
 	} elsif ($data =~ /GNU/ || $data =~ /GCC/ ) {
 
+            $data =~ s/\(+.*?\)+//g;
 	    $data =~ /(\d+)\.(\d+).(\d+)/;
 	    $major = $1;
 	    $minor = $2;
@@ -87,7 +90,7 @@ if ($compiler eq "") {
 		if ($compiler =~ /flang/) {
 		    $vendor = FLANG;
 		    $openmp = "-fopenmp";
-	    } elsif ($compiler =~ /pgf/) {
+	    } elsif ($compiler =~ /pgf/ || $compiler =~ /nvf/) {
 		    $vendor = PGI;
 		    $openmp = "-mp";
 		} else {
@@ -123,7 +126,7 @@ if ($compiler eq "") {
 	    $openmp = "-mp";
 	}
 
-	if ($data =~ /PGF/) {
+	if ($data =~ /PGF/ || $data =~ /NVF/) {
 	    $vendor = PGI;
 	    $openmp = "-mp";
 	}
@@ -133,8 +136,16 @@ if ($compiler eq "") {
 	    $openmp = "-openmp";
 	}
 
+	if ($data =~ /NAG/) {
+	    $vendor = NAG;
+	    $openmp = "-openmp";
+	}
+
 	# for embedded underscore name, e.g. zho_ge, it may append 2 underscores.
 	$data = `$compiler -O2 -S ftest3.f > /dev/null 2>&1 && cat ftest3.s && rm -f ftest3.s`;
+	if ($data eq "") {
+		$data = `$compiler -O2 -S ftest3.f > /dev/null 2>&1 && cat ftest3.c && rm -f ftest3.c`;
+	}
 	if ($data =~ / zho_ge__/) {
 	    $need2bu       = 1;
 	}
@@ -177,7 +188,7 @@ if ($compiler eq "") {
 	    $openmp = "-mp";
 	}
 
-	if ($compiler =~ /pgf/) {
+	if ($compiler =~ /pgf/ || $compiler =~ /nvf/) {
 	    $vendor = PGI;
 	    $bu       = "_";
 	    $openmp = "-mp";
@@ -222,6 +233,12 @@ if ($compiler eq "") {
 	    $openmp = "-fopenmp";
 	}
 
+	if ($compiler =~ /nagfor/) {
+	    $vendor = NAG;
+	    $bu     = "_";
+	    $openmp = "-openmp";
+	}
+
 	if ($vendor eq "") {
 	    $nofortran = 1;
 	    $compiler = "gfortran";
@@ -275,14 +292,20 @@ if (!$?) {
 	if ($?) {
 	    $link = `$compiler $openmp -mabi=64 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
 	}
+       #For nagfor
+	if ($?) {
+	    $link = `$compiler $openmp -dryrun ftest2.f 2>&1 && rm -f a.out a.exe`;
+        }
 	$binary = "" if ($?);
     }
-
     if ($binary eq "") {
 	$link = `$compiler $openmp -v ftest2.f 2>&1 && rm -f a.out a.exe`;
     }
 }
 
+if ( $vendor eq "NAG") {
+	    $link = `$compiler $openmp -dryrun ftest2.f 2>&1 && rm -f a.out a.exe`;
+    }
 $linker_L = "";
 $linker_l = "";
 $linker_a = "";
@@ -330,12 +353,13 @@ if ($link ne "") {
 	    $flags =~ s/\@/\,/g;
 	    $linker_L .= "-Wl,". $flags . " " ;
 	}
-	if ($flags =~ /-lgomp/ && $CC =~ /clang/) {
+	if ($flags =~ /-lgomp/ && $ENV{"CC"} =~ /clang/) {
 	    $flags = "-lomp";
 	}
 
 	if (
 	    ($flags =~ /^\-l/)
+	    && ($flags !~ /ibrary/)
 	    && ($flags !~ /gfortranbegin/)
 	    && ($flags !~ /frtbegin/)
 	    && ($flags !~ /pathfstart/)
@@ -352,6 +376,16 @@ if ($link ne "") {
 	    $linker_l .= $flags . " ";
 	}
 
+	if ( $flags =~ /quickfit.o/ && $vendor == NAG) {
+	    $linker_l .= $flags . " ";
+	}
+	if ( $flags =~ /safefit.o/ && $vendor == NAG) {
+	    $linker_l .= $flags . " ";
+	}
+	if ( $flags =~ /thsafe.o/ && $vendor == NAG) {
+	    $linker_l .= $flags . " ";
+	}
+
 	$linker_a .= $flags . " " if $flags =~ /\.a$/;
     }
 

getarch.c

@@ -140,8 +140,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_PPC440FP2	*/
 /* #define FORCE_CELL		*/
 /* #define FORCE_SICORTEX	*/
-/* #define FORCE_LOONGSON3A	*/
-/* #define FORCE_LOONGSON3B	*/
+/* #define FORCE_LOONGSON3R3	*/
+/* #define FORCE_LOONGSON3R4	*/
 /* #define FORCE_I6400		*/
 /* #define FORCE_P6600		*/
 /* #define FORCE_P5600		*/
@@ -814,31 +814,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 
-#ifdef FORCE_LOONGSON3A
+#if defined FORCE_LOONGSON3R3 || defined FORCE_LOONGSON3A || defined FORCE_LOONGSON3B
 #define FORCE
 #define ARCHITECTURE    "MIPS"
-#define SUBARCHITECTURE "LOONGSON3A"
+#define SUBARCHITECTURE "LOONGSON3R3"
 #define SUBDIRNAME      "mips64"
-#define ARCHCONFIG   "-DLOONGSON3A " \
+#define ARCHCONFIG   "-DLOONGSON3R3 " \
        "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=32 " \
        "-DL2_SIZE=512488 -DL2_LINESIZE=32 " \
        "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 "
-#define LIBNAME   "loongson3a"
-#define CORENAME  "LOONGSON3A"
+#define LIBNAME   "loongson3r3"
+#define CORENAME  "LOONGSON3R3"
 #else
 #endif
 
-#ifdef FORCE_LOONGSON3B
+#ifdef FORCE_LOONGSON3R4
 #define FORCE
 #define ARCHITECTURE    "MIPS"
-#define SUBARCHITECTURE "LOONGSON3B"
+#define SUBARCHITECTURE "LOONGSON3R4"
 #define SUBDIRNAME      "mips64"
-#define ARCHCONFIG   "-DLOONGSON3B " \
+#define ARCHCONFIG   "-DLOONGSON3R4 " \
        "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=32 " \
        "-DL2_SIZE=512488 -DL2_LINESIZE=32 " \
        "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 "
-#define LIBNAME   "loongson3b"
-#define CORENAME  "LOONGSON3B"
+#define LIBNAME   "loongson3r4"
+#define CORENAME  "LOONGSON3R4"
 #else
 #endif
 
@@ -1375,6 +1375,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifdef __riscv
 #include "cpuid_riscv64.c"
+#define OPENBLAS_SUPPORTED
 #endif
 
 #ifdef __arm__

getarch_2nd.c

@@ -4,6 +4,14 @@
 #else
 #include "config_kernel.h"
 #endif
+#if (defined(__WIN32__) || defined(__WIN64__) || defined(__CYGWIN32__) || defined(__CYGWIN64__) || defined(_WIN32) || defined(_WIN64)) && defined(__64BIT__)
+typedef long long BLASLONG;
+typedef unsigned long long BLASULONG;
+#else
+typedef long BLASLONG;
+typedef unsigned long BLASULONG;
+#endif
+
 #include "param.h"
 
 int main(int argc, char **argv) {

interface/Makefile

@@ -316,7 +316,7 @@ CCBLAS1OBJS   = \
 	cblas_cscal.$(SUFFIX) cblas_csscal.$(SUFFIX) \
 	cblas_cswap.$(SUFFIX) cblas_scnrm2.$(SUFFIX) \
 	cblas_caxpby.$(SUFFIX) \
-	cblas_icmin.$(SUFFIX) cblas_icmax.$(SUFFIX) cblas_scsum.$(SUFFIX)
+	cblas_icmin.$(SUFFIX) cblas_icmax.$(SUFFIX) cblas_scsum.$(SUFFIX) cblas_csrot.$(SUFFIX) cblas_crotg.$(SUFFIX)
 
 CCBLAS2OBJS   = \
 	cblas_cgemv.$(SUFFIX) cblas_cgerc.$(SUFFIX) cblas_cgeru.$(SUFFIX) \
@@ -346,7 +346,7 @@ CZBLAS1OBJS   = \
 	cblas_zscal.$(SUFFIX) cblas_zdscal.$(SUFFIX) \
 	cblas_zswap.$(SUFFIX) cblas_dznrm2.$(SUFFIX) \
 	cblas_zaxpby.$(SUFFIX) \
-	cblas_izmin.$(SUFFIX) cblas_izmax.$(SUFFIX) cblas_dzsum.$(SUFFIX)
+	cblas_izmin.$(SUFFIX) cblas_izmax.$(SUFFIX) cblas_dzsum.$(SUFFIX) cblas_zdrot.$(SUFFIX) cblas_zrotg.$(SUFFIX)
 
 
 CZBLAS2OBJS   = \
@@ -1634,6 +1634,12 @@ cblas_srotg.$(SUFFIX) cblas_srotg.$(PSUFFIX): rotg.c
 cblas_drotg.$(SUFFIX) cblas_drotg.$(PSUFFIX): rotg.c
 	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
 
+cblas_crotg.$(SUFFIX) cblas_crotg.$(PSUFFIX): zrotg.c
+	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
+
+cblas_zrotg.$(SUFFIX) cblas_zrotg.$(PSUFFIX): zrotg.c
+	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
+
 cblas_srotm.$(SUFFIX) cblas_srotm.$(PSUFFIX): rotm.c
 	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
 
@@ -1664,6 +1670,12 @@ cblas_csscal.$(SUFFIX) cblas_csscal.$(PSUFFIX) : zscal.c
 cblas_zdscal.$(SUFFIX) cblas_zdscal.$(PSUFFIX) : zscal.c
 	$(CC) $(CFLAGS) -DCBLAS -c -DSSCAL $< -o $(@F)
 
+cblas_csrot.$(SUFFIX) cblas_csrot.$(PSUFFIX) : zrot.c
+	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+
+cblas_zdrot.$(SUFFIX) cblas_zdrot.$(PSUFFIX) : zrot.c
+	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+
 ifeq ($(BUILD_BFLOAT16),1)
 cblas_sbgemv.$(SUFFIX) cblas_sbgemv.$(PSUFFIX) : sbgemv.c
 	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)

interface/create

@@ -1,4 +1,4 @@
-#!/usr/bin/perl
+#!/usr/bin/env perl
 
 $count = 0;
 

interface/gemm.c

@@ -246,6 +246,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
 
 #ifdef SMP
   double MNK;
+#if defined(USE_SIMPLE_THREADED_LEVEL3) || !defined(NO_AFFINITY)
 #ifndef COMPLEX
 #ifdef XDOUBLE
   int mode  =  BLAS_XDOUBLE | BLAS_REAL;
@@ -264,6 +265,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
 #endif
 #endif
 #endif
+#endif
 
 #if defined(SMP) && !defined(NO_AFFINITY) && !defined(USE_SIMPLE_THREADED_LEVEL3)
   int nodes;
@@ -417,8 +419,10 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   sb = (XFLOAT *)(((BLASLONG)sa + ((GEMM_P * GEMM_Q * COMPSIZE * SIZE + GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B);
 
 #ifdef SMP
+#if defined(USE_SIMPLE_THREADED_LEVEL3) || !defined(NO_AFFINITY)
   mode |= (transa << BLAS_TRANSA_SHIFT);
   mode |= (transb << BLAS_TRANSB_SHIFT);
+#endif
 
   MNK = (double) args.m * (double) args.n * (double) args.k;
   if ( MNK <= (SMP_THRESHOLD_MIN  * (double) GEMM_MULTITHREAD_THRESHOLD)  )

interface/rotmg.c

@@ -107,7 +107,6 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 		dq1 =  dp1 * *dx1;
 		if(ABS(dq1) > ABS(dq2))
 		{
-			dflag = ZERO;
 			dh11  =  ONE;
 			dh22  =  ONE;
 			dh21 = -  dy1 / *dx1;

kernel/CMakeLists.txt

@@ -187,10 +187,11 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     endif ()
     # Makefile.L3
     set(USE_TRMM false)
-    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) OR (TARGET_CORE MATCHES COOPERLAKE))
+    string(TOUPPER ${TARGET_CORE} UC_TARGET_CORE)
+    if (ARM OR ARM64 OR (UC_TARGET_CORE MATCHES LONGSOON3B) OR (UC_TARGET_CORE MATCHES GENERIC) OR (UC_TARGET_CORE MATCHES HASWELL) OR (UC_TARGET_CORE MATCHES ZEN) OR (UC_TARGET_CORE MATCHES SKYLAKEX) OR (UC_TARGET_CORE MATCHES COOPERLAKE))
       set(USE_TRMM true)
     endif ()
-    if (ZARCH OR (TARGET_CORE MATCHES POWER8) OR (TARGET_CORE MATCHES POWER9) OR (TARGET_CORE MATCHES POWER10))
+    if (ZARCH OR (UC_TARGET_CORE MATCHES POWER8) OR (UC_TARGET_CORE MATCHES POWER9) OR (UC_TARGET_CORE MATCHES POWER10))
       set(USE_TRMM true)
     endif ()
 

kernel/Makefile

@@ -36,7 +36,7 @@ ifeq ($(TARGET_CORE), COOPERLAKE)
  ifeq ($(GCCVERSIONGTEQ10), 1) 
   override CFLAGS += -march=cooperlake
  else 
-  override CFLAGS += -march=skylake-avx512
+  override CFLAGS += -march=skylake-avx512 -mavx512f
  endif 
  ifeq ($(OSNAME), CYGWIN_NT)
   override CFLAGS += -fno-asynchronous-unwind-tables
@@ -47,7 +47,7 @@ ifeq ($(TARGET_CORE), COOPERLAKE)
   endif
  endif
 else ifeq ($(TARGET_CORE), SKYLAKEX)
- override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=skylake-avx512
+ override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=skylake-avx512 -mavx512f
  ifeq ($(OSNAME), CYGWIN_NT)
   override CFLAGS += -fno-asynchronous-unwind-tables
  endif
@@ -58,6 +58,8 @@ else ifeq ($(TARGET_CORE), SKYLAKEX)
  endif
 else ifeq ($(TARGET_CORE), HASWELL)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(AVX2OPT)
+else ifeq ($(TARGET_CORE), LOONGSON3R4)
+ override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(MSA_FLAGS)
 else
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
 endif
@@ -68,6 +70,9 @@ else
 TARGET_CORE = $(CORE)
 KDIR =
 TSUFFIX =
+ifeq ($(TARGET_CORE), LOONGSON3R4)
+  override CFLAGS += $(MSA_FLAGS)
+endif
 endif
 
 -include $(KERNELDIR)/KERNEL.$(TARGET_CORE)

kernel/Makefile.L1

@@ -1,3 +1,11 @@
+FMAFLAG=
+ifndef OLDGCC
+ifdef HAVE_FMA3
+FMAFLAG = -mfma
+endif
+endif
+
+
 ### AMAX ###
 
 ifndef SAMAXKERNEL
@@ -828,10 +836,10 @@ $(KDIR)xnrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xnrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KE
 	$(CC) $(CFLAGS) -DCOMPLEX -c -DXDOUBLE $< -o $@
 
 $(KDIR)srot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)srot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SROTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UCOMPLEX -UDOUBLE  $< -o $@
+	$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -UDOUBLE  $< -o $@
 
 $(KDIR)drot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)drot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DROTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UCOMPLEX -DDOUBLE  $< -o $@
+	$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DDOUBLE  $< -o $@
 
 $(KDIR)qrot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qrot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QROTKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -UCOMPLEX -DXDOUBLE $< -o $@

kernel/Makefile.L3

@@ -29,10 +29,6 @@ ifeq ($(ARCH), riscv64)
 USE_TRMM = 1
 endif
 
-ifeq ($(TARGET), LOONGSON3B)
-USE_TRMM = 1
-endif
-
 ifneq ($(DYNAMIC_ARCH), 1)
 ifeq ($(TARGET), GENERIC)
 USE_TRMM = 1

kernel/arm/omatcopy_rt.c

@@ -1,5 +1,5 @@
 /***************************************************************************
-Copyright (c) 2013, The OpenBLAS Project
+Copyright (c) 2021, 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
@@ -27,36 +27,208 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-/*****************************************************
- * 2014/06/09 Saar
- *
- * Order rowMajor
- * Trans
- *
-******************************************************/
-
 int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG ldb)
 {
-	BLASLONG i,j;
-	FLOAT *aptr,*bptr;
+    BLASLONG i, j;
+    FLOAT *a_offset, *a_offset1, *a_offset2, *a_offset3, *a_offset4;
+    FLOAT *b_offset, *b_offset1, *b_offset2, *b_offset3, *b_offset4;
 
-	if ( rows <= 0     )  return(0);
-	if ( cols <= 0     )  return(0);
+    if (rows <= 0)  return 0;
+    if (cols <= 0)  return 0;
 
-	aptr = a;
+    a_offset = a;
+    b_offset = b;
 
-	for ( i=0; i<rows ; i++ )
-	{
-		bptr = &b[i];
-		for(j=0; j<cols; j++)
-		{
-			bptr[j*ldb] = alpha * aptr[j];
-		}
-		aptr += lda;
-	}
+    i = (rows >> 2);
+    if (i > 0) {
+        do {
+            a_offset1 = a_offset;
+            a_offset2 = a_offset1 + lda;
+            a_offset3 = a_offset2 + lda;
+            a_offset4 = a_offset3 + lda;
+            a_offset += 4 * lda;
 
-	return(0);
+            b_offset1 = b_offset;
+            b_offset2 = b_offset1 + ldb;
+            b_offset3 = b_offset2 + ldb;
+            b_offset4 = b_offset3 + ldb;
+            b_offset += 4;
 
+            j = (cols >> 2);
+            if (j > 0) {
+                do {
+                /* Column 1 of MAT_B */
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha; // Row 1 of MAT_A
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+                *(b_offset3 + 0) = *(a_offset1 + 2)*alpha;
+                *(b_offset4 + 0) = *(a_offset1 + 3)*alpha;
+
+                /* Column 2 of MAT_B */
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha; // Row 2 of MAT_A
+                *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+                *(b_offset3 + 1) = *(a_offset2 + 2)*alpha;
+                *(b_offset4 + 1) = *(a_offset2 + 3)*alpha;
+
+                /* Column 3 of MAT_B */
+                *(b_offset1 + 2) = *(a_offset3 + 0)*alpha; // Row 3 of MAT_A
+                *(b_offset2 + 2) = *(a_offset3 + 1)*alpha;
+                *(b_offset3 + 2) = *(a_offset3 + 2)*alpha;
+                *(b_offset4 + 2) = *(a_offset3 + 3)*alpha;
+
+                /* Column 4 of MAT_B */
+                *(b_offset1 + 3) = *(a_offset4 + 0)*alpha; // Row 4 of MAT_A
+                *(b_offset2 + 3) = *(a_offset4 + 1)*alpha;
+                *(b_offset3 + 3) = *(a_offset4 + 2)*alpha;
+                *(b_offset4 + 3) = *(a_offset4 + 3)*alpha;
+
+                a_offset1 += 4;
+                a_offset2 += 4;
+                a_offset3 += 4;
+                a_offset4 += 4;
+                b_offset1 += ldb * 4;
+                b_offset2 += ldb * 4;
+                b_offset3 += ldb * 4;
+                b_offset4 += ldb * 4;
+                
+                j--;
+                } while (j > 0);
+            } // if(j > 0)
+
+
+            if (cols & 2) {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+                *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+
+                *(b_offset1 + 2) = *(a_offset3 + 0)*alpha;
+                *(b_offset2 + 2) = *(a_offset3 + 1)*alpha;
+
+                *(b_offset1 + 3) = *(a_offset4 + 0)*alpha;
+                *(b_offset2 + 3) = *(a_offset4 + 1)*alpha;
+
+                a_offset1 += 2;
+                a_offset2 += 2;
+                a_offset3 += 2;
+                a_offset4 += 2;
+
+                b_offset1 += ldb*2;
+
+            }
+
+            if (cols & 1) {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+
+                *(b_offset1 + 2) = *(a_offset3 + 0)*alpha;
+
+                *(b_offset1 + 3) = *(a_offset4 + 0)*alpha;
+            }
+
+            i--;
+        } while (i > 0);
+    }
+
+
+    if (rows & 2) {
+        a_offset1 = a_offset;
+        a_offset2 = a_offset1 + lda;
+        a_offset += 2 * lda;
+
+        b_offset1 = b_offset;
+        b_offset2 = b_offset1 + ldb;
+        b_offset3 = b_offset2 + ldb;
+        b_offset4 = b_offset3 + ldb;
+        b_offset += 2;
+
+        j = (cols >> 2);
+        if (j > 0){
+            do {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+                *(b_offset3 + 0) = *(a_offset1 + 2)*alpha;
+                *(b_offset4 + 0) = *(a_offset1 + 3)*alpha;
+
+                *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+                *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+                *(b_offset3 + 1) = *(a_offset2 + 2)*alpha;
+                *(b_offset4 + 1) = *(a_offset2 + 3)*alpha;
+                
+                a_offset1 += 4;
+                a_offset2 += 4;
+                b_offset1 += ldb * 4;
+                b_offset2 += ldb * 4;
+                b_offset3 += ldb * 4;
+                b_offset4 += ldb * 4;
+
+                j--;
+            } while (j > 0);
+        }
+
+
+        if (cols & 2){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+            *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+
+            *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+            *(b_offset2 + 1) = *(a_offset2 + 1)*alpha;
+
+            a_offset1 += 2;
+            a_offset2 += 2;
+            b_offset1 += ldb*2;
+
+        }
+
+
+        if (cols & 1){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+            *(b_offset1 + 1) = *(a_offset2 + 0)*alpha;
+        }
+    } // if (rows & 2)
+
+
+    if (rows & 1) {
+        a_offset1 = a_offset;
+        a_offset += lda;
+
+        b_offset1 = b_offset;
+        b_offset2 = b_offset1 + ldb;
+        b_offset3 = b_offset2 + ldb;
+        b_offset4 = b_offset3 + ldb;
+
+        j = (cols >> 2);
+        if (j > 0){
+            do {
+                *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+                *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+                *(b_offset3 + 0) = *(a_offset1 + 2)*alpha;
+                *(b_offset4 + 0) = *(a_offset1 + 3)*alpha;
+                
+                a_offset1 += 4;
+                b_offset1 += ldb * 4;
+                b_offset2 += ldb * 4;
+                b_offset3 += ldb * 4;
+                b_offset4 += ldb * 4;
+                
+                j--;
+            } while (j > 0);
+        }
+
+        if (cols & 2){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+            *(b_offset2 + 0) = *(a_offset1 + 1)*alpha;
+
+            a_offset1 += 2;
+            b_offset1 += ldb * 2;
+        }
+
+        if (cols & 1){
+            *(b_offset1 + 0) = *(a_offset1 + 0)*alpha;
+        }
+    }
+
+    return 0;
 }
 
-

kernel/arm/zdot.c

@@ -48,7 +48,7 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 
 	dot[0]=0.0;
 	dot[1]=0.0;
-#if !defined(__PPC__) && !defined(__SunOS)
+#if !defined(__PPC__) && !defined(__SunOS) && !defined(__PGI)
 	CREAL(result) = 0.0 ;
 	CIMAG(result) = 0.0 ;
 #else
@@ -73,7 +73,7 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 		i++ ;
 
 	}
-#if !defined(__PPC__)	&& !defined(__SunOS)
+#if !defined(__PPC__)	&& !defined(__SunOS) && !defined(__PGI)
         CREAL(result) = dot[0];
 	CIMAG(result) = dot[1];
 #else

kernel/arm64/KERNEL.ARMV8

@@ -97,9 +97,18 @@ CNRM2KERNEL    = znrm2.S
 ZNRM2KERNEL    = znrm2.S
 
 DDOTKERNEL     = dot.S
+ifneq ($(C_COMPILER), PGI)
 SDOTKERNEL     = ../generic/dot.c
+else
+SDOTKERNEL = dot.S
+endif
+ifneq ($(C_COMPILER), PGI)
 CDOTKERNEL     = zdot.S
 ZDOTKERNEL     = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
 DSDOTKERNEL    = dot.S
 
 DGEMM_BETA     = dgemm_beta.S

kernel/arm64/KERNEL.CORTEXA53

@@ -96,11 +96,20 @@ DNRM2KERNEL    = nrm2.S
 CNRM2KERNEL    = znrm2.S
 ZNRM2KERNEL    = znrm2.S
 
-DDOTKERNEL     = dot.S
-SDOTKERNEL     = ../generic/dot.c
-CDOTKERNEL     = zdot.S
-ZDOTKERNEL     = zdot.S
-DSDOTKERNEL    = dot.S
+ifneq ($(C_COMPILER), PGI)
+SDOTKERNEL   = ../generic/dot.c
+else
+SDOTKERNEL   = dot.S
+endif
+DDOTKERNEL   = dot.S
+ifneq ($(C_COMPILER), PGI)
+CDOTKERNEL   = zdot.S
+ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
+DSDOTKERNEL  = dot.S
 
 DGEMM_BETA     = dgemm_beta.S
 SGEMM_BETA     = sgemm_beta.S

kernel/arm64/KERNEL.CORTEXA57

@@ -70,10 +70,19 @@ DCOPYKERNEL  = copy.S
 CCOPYKERNEL  = copy.S
 ZCOPYKERNEL  = copy.S
 
+ifneq ($(C_COMPILER), PGI)
 SDOTKERNEL   = ../generic/dot.c
+else
+SDOTKERNEL   = dot.S
+endif
 DDOTKERNEL   = dot.S
+ifneq ($(C_COMPILER), PGI)
 CDOTKERNEL   = zdot.S
 ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
 DSDOTKERNEL  = dot.S
 
 SNRM2KERNEL  = nrm2.S

kernel/arm64/KERNEL.THUNDERX

@@ -47,8 +47,13 @@ ZCOPYKERNEL  = copy.S
 
 SDOTKERNEL   = dot_thunderx.c
 DDOTKERNEL   = ddot_thunderx.c
+ifneq ($(C_COMPILER), PGI)
 CDOTKERNEL   = zdot.S
 ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
 DSDOTKERNEL  = dot.S
 
 SNRM2KERNEL  = nrm2.S

kernel/arm64/KERNEL.TSV110

@@ -72,8 +72,13 @@ ZCOPYKERNEL  = copy.S
 
 SDOTKERNEL   = dot.S
 DDOTKERNEL   = dot.S
+ifneq ($(C_COMPILER), PGI)
 CDOTKERNEL   = zdot.S
 ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
 DSDOTKERNEL  = dot.S
 
 SNRM2KERNEL  = nrm2.S

kernel/arm64/dznrm2_thunderx2t99.c

@@ -58,6 +58,7 @@ extern int blas_level1_thread_with_return_value(int mode, BLASLONG m, BLASLONG n
 #define CUR_MAXINV	"d8"
 #define CUR_MAXINV_V	"v8.2d"
 #define CUR_MAX_V	"v8.2d"
+#define REGINF		"d9"
 
 static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 		         double *ssq, double *scale)
@@ -79,8 +80,10 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	ble	9f //nrm2_kernel_L999			\n"
 
 	"1: //nrm2_kernel_F_BEGIN:				\n"
+	"	mov	x6, #0x7FF0000000000000 //+Infinity	\n"
 	"	fmov	"REGZERO", xzr				\n"
 	"	fmov	"REGONE", #1.0				\n"
+	"	fmov	"REGINF", x6				\n"
 	"	lsl	"INC_X", "INC_X", #"INC_SHIFT"		\n"
 	"	mov	"J", "N"				\n"
 	"	cmp	"J", xzr				\n"
@@ -104,6 +107,8 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	ldr	d4, ["X"]				\n"
 	"	fabs	d4, d4					\n"
 	"	fmax	"CUR_MAX", "SCALE", d4			\n"
+	"	fcmp	"CUR_MAX", "REGINF"			\n"
+	"	beq	10f					\n"
 	"	fdiv	"SCALE", "SCALE", "CUR_MAX"		\n"
 	"	fmul	"SCALE", "SCALE", "SCALE"		\n"
 	"	fmul	"SSQ", "SSQ", "SCALE"			\n"
@@ -116,6 +121,8 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	ldr	d3, ["X", #8]				\n"
 	"	fabs	d3, d3					\n"
 	"	fmax	"CUR_MAX", "SCALE", d3			\n"
+	"	fcmp	"CUR_MAX", "REGINF"			\n"
+	"	beq	10f					\n"
 	"	fdiv	"SCALE", "SCALE", "CUR_MAX"		\n"
 	"	fmul	"SCALE", "SCALE", "SCALE"		\n"
 	"	fmul	"SSQ", "SSQ", "SCALE"			\n"
@@ -158,6 +165,8 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	fmaxp	v24.2d, v24.2d, v26.2d			\n"
 	"	fmaxp	v24.2d, v24.2d, v24.2d			\n"
 	"	fmax	"CUR_MAX", "SCALE", d24			\n"
+	"	fcmp	"CUR_MAX", "REGINF"			\n"
+	"	beq	10f					\n"
 	"	fdiv	"CUR_MAXINV", "REGONE", "CUR_MAX"	\n"
 	"	//dup	"CUR_MAX_V", v7.d[0]			\n"
 	"	fdiv	"SCALE", "SCALE", "CUR_MAX"		\n"
@@ -217,6 +226,8 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	fmaxp	v24.2d, v24.2d, v26.2d			\n"
 	"	fmaxp	v24.2d, v24.2d, v24.2d			\n"
 	"	fmax	"CUR_MAX", "SCALE", d24			\n"
+	"	fcmp	"CUR_MAX", "REGINF"			\n"
+	"	beq	10f					\n"
 	"	fdiv	"CUR_MAXINV", "REGONE", "CUR_MAX"	\n"
 	"	//dup	"CUR_MAX_V", v7.d[0]			\n"
 	"	fdiv	"SCALE", "SCALE", "CUR_MAX"		\n"
@@ -265,6 +276,8 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	ldr	d4, ["X"]				\n"
 	"	fabs	d4, d4					\n"
 	"	fmax	"CUR_MAX", "SCALE", d4			\n"
+	"	fcmp	"CUR_MAX", "REGINF"			\n"
+	"	beq	10f					\n"
 	"	fdiv	"SCALE", "SCALE", "CUR_MAX"		\n"
 	"	fmul	"SCALE", "SCALE", "SCALE"		\n"
 	"	fmul	"SSQ", "SSQ", "SCALE"			\n"
@@ -276,6 +289,8 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"	ldr	d3, ["X", #8]				\n"
 	"	fabs	d3, d3					\n"
 	"	fmax	"CUR_MAX", "SCALE", d3			\n"
+	"	fcmp	"CUR_MAX", "REGINF"			\n"
+	"	beq	10f					\n"
 	"	fdiv	"SCALE", "SCALE", "CUR_MAX"		\n"
 	"	fmul	"SCALE", "SCALE", "SCALE"		\n"
 	"	fmul	"SSQ", "SSQ", "SCALE"			\n"
@@ -291,6 +306,11 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	"9: //nrm2_kernel_L999:					\n"
 	"	str	"SSQ", [%[SSQ_]]			\n"
 	"	str	"SCALE", [%[SCALE_]]			\n"
+	"	b	11f					\n"
+	"10:							\n"
+	"	str	"REGINF", [%[SSQ_]]			\n"
+	"	str	"REGINF", [%[SCALE_]]			\n"
+	"11:							\n"
 
 	:
 	: [SSQ_]    "r"  (ssq),			//%0
@@ -300,7 +320,7 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	  [INCX_]   "r"  (inc_x)		//%4
 	: "cc",
 	  "memory",
-	  "x0", "x1", "x2", "x3", "x4", "x5",
+	  "x0", "x1", "x2", "x3", "x4", "x5", "x6",
 	  "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8"
 	);
 
@@ -359,6 +379,12 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 			cur_ssq = *ptr;
 			cur_scale = *(ptr + 1);
 
+			if (cur_ssq == INFINITY) {
+				ssq = INFINITY;
+				scale = INFINITY;
+				break;
+			}
+
 			if (cur_scale != 0) {
 				if (cur_scale > scale) {
 					scale = (scale / cur_scale);

kernel/mips/cgemm_kernel_8x4_msa.c

@@ -121,7 +121,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_KERNEL_8X1_MSA(OP0, OP1, OP2, OP3, OP4)                 \
 {                                                                     \
     LD_SP4_INC(pa0, 4, src_a0, src_a1, src_a2, src_a3);               \
-    src_bi = (v4f32) __msa_cast_to_vector_double(*((double *) pb0));  \
+    src_bi = (v4f32) COPY_DOUBLE_TO_VECTOR(*((double *) pb0));        \
     SPLATI_W2_SP(src_bi, 0, src_br, src_bi);                          \
                                                                       \
     PCKEVOD_W2_SP(src_a1, src_a0, src_a0r, src_a0i);                  \
@@ -200,7 +200,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_KERNEL_4X1_MSA(OP0, OP1, OP2, OP3, OP4)                 \
 {                                                                     \
     LD_SP2_INC(pa0, 4, src_a0, src_a1);                               \
-    src_bi = (v4f32) __msa_cast_to_vector_double(*((double *) pb0));  \
+    src_bi = (v4f32) COPY_DOUBLE_TO_VECTOR(*((double *) pb0));        \
     SPLATI_W2_SP(src_bi, 0, src_br, src_bi);                          \
                                                                       \
     PCKEVOD_W2_SP(src_a1, src_a0, src_a0r, src_a0i);                  \

kernel/mips/cgemv_n_msa.c

@@ -56,11 +56,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     #if !defined(XCONJ)
         #define OP0  +=
         #define OP1  -=
-        #define OP2  -=
+        #define OP2  +=
     #else
         #define OP0  -=
         #define OP1  -=
-        #define OP2  +=
+        #define OP2  -=
     #endif
 #endif
 

kernel/mips/cgemv_t_msa.c

@@ -32,14 +32,26 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #undef OP1
 #undef OP2
 
-#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
-    #define OP0  -=
-    #define OP1  +=
-    #define OP2  +=
+#if !defined(CONJ)
+    #if !defined(XCONJ)
+        #define OP0  -=
+        #define OP1  +=
+        #define OP2  +=
+    #else
+        #define OP0  +=
+        #define OP1  +=
+        #define OP2  -=
+    #endif
 #else
-    #define OP0  +=
-    #define OP1  +=
-    #define OP2  -=
+    #if !defined(XCONJ)
+        #define OP0  +=
+        #define OP1  -=
+        #define OP2  +=
+    #else
+        #define OP0  -=
+        #define OP1  -=
+        #define OP2  -=
+    #endif
 #endif
 
 #define CGEMV_T_8x4()                        \

kernel/mips/crot_msa.c

@@ -49,11 +49,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y,
     {
         if ((0 == c) && (0 == s))
         {
-            v4f32 zero = __msa_cast_to_vector_float(0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 0, 0.0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 1, 0.0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 2, 0.0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 3, 0.0);
+            v4f32 zero = {0.0, 0.0, 0.0, 0.0};
 
             /* process 2 elements */
             for (j = (n >> 1); j--;)

kernel/mips/cscal_msa.c

@@ -49,11 +49,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
     {
         if ((0.0 == da_r) && (0.0 == da_i))
         {
-            v4f32 zero_v = __msa_cast_to_vector_float(0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 0, 0.0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 1, 0.0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 2, 0.0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 3, 0.0);
+            v4f32 zero_v = {0.0, 0.0, 0.0, 0.0};
 
             for (i = (n >> 5); i--;)
             {

kernel/mips/dscal_msa.c

@@ -44,9 +44,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x,
     {
         if (0.0 == da)
         {
-            v2f64 zero_v = __msa_cast_to_vector_double(0);
-            zero_v = (v2f64) __msa_insert_d((v2i64) zero_v, 0, 0.0);
-            zero_v = (v2f64) __msa_insert_d((v2i64) zero_v, 1, 0.0);
+            v2f64 zero_v = {0.0, 0.0};
 
             for (i = (n >> 5); i--;)
             {

kernel/mips/dswap_msa.c

@@ -184,7 +184,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3,
             }
         }
     }
-    else
+    else if ((inc_x != 0) && (inc_y != 0))
     {
         for (i = (n >> 3); i--;)
         {
@@ -248,6 +248,32 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3,
             }
         }
     }
-
+    else
+    {
+        if (inc_x == inc_y)
+        {
+            if (n & 1)
+            {
+                x0 = *srcx;
+                *srcx  = *srcy;
+                *srcy  = x0;
+            }
+            else
+                return (0);
+        }
+        else
+        {
+            BLASLONG ix = 0, iy = 0;
+            while (i < n)
+            {
+                x0 = srcx[ix];
+                srcx[ix] = srcy[iy];
+                srcy[iy] = x0;
+                ix += inc_x;
+                iy += inc_y;
+                i++;
+            }
+        }
+    }
     return (0);
 }

kernel/mips/dtrsm_kernel_LN_8x4_msa.c

@@ -186,8 +186,7 @@ void dsolve_8x4_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     ILVRL_D2_DP(src_c14, src_c10, res_c12, res_c13);
     ILVRL_D2_DP(src_c15, src_c11, res_c14, res_c15);
 
-    src_a54 = __msa_cast_to_vector_double(*(a + 54));
-    src_a54 = (v2f64) __msa_splati_d((v2i64) src_a54, 0);
+    src_a54 = COPY_DOUBLE_TO_VECTOR(*(a + 54));
     src_a62 = LD_DP(a + 62);
     src_a63 = (v2f64) __msa_splati_d((v2i64) src_a62, 1);
     src_a62 = (v2f64) __msa_splati_d((v2i64) src_a62, 0);
@@ -200,8 +199,7 @@ void dsolve_8x4_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     src_a44 = LD_DP(a + 44);
     src_a45 = (v2f64) __msa_splati_d((v2i64) src_a44, 1);
     src_a44 = (v2f64) __msa_splati_d((v2i64) src_a44, 0);
-    src_a36 = __msa_cast_to_vector_double(*(a + 36));
-    src_a36 = (v2f64) __msa_splati_d((v2i64) src_a36, 0);
+    src_a36 = COPY_DOUBLE_TO_VECTOR(*(a + 36));
 
     res_c7 *= src_a63;
     res_c6 -= res_c7 * src_a62;
@@ -271,8 +269,7 @@ void dsolve_8x4_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     src_a26 = LD_DP(a + 26);
     src_a27 = (v2f64) __msa_splati_d((v2i64) src_a26, 1);
     src_a26 = (v2f64) __msa_splati_d((v2i64) src_a26, 0);
-    src_a18 = __msa_cast_to_vector_double(*(a + 18));
-    src_a18 = (v2f64) __msa_splati_d((v2i64) src_a18, 0);
+    src_a18 = COPY_DOUBLE_TO_VECTOR(*(a + 18));
 
     res_c3 -= res_c7 * src_a59;
     res_c2 -= res_c7 * src_a58;
@@ -358,8 +355,7 @@ void dsolve_8x4_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     src_a8 = LD_DP(a + 8);
     src_a9 = (v2f64) __msa_splati_d((v2i64) src_a8, 1);
     src_a8 = (v2f64) __msa_splati_d((v2i64) src_a8, 0);
-    src_a0 = __msa_cast_to_vector_double(*(a + 0));
-    src_a0 = (v2f64) __msa_splati_d((v2i64) src_a0, 0);
+    src_a0 = COPY_DOUBLE_TO_VECTOR(*(a + 0));
 
     res_c1 -= res_c2 * src_a17;
     res_c1 *= src_a9;
@@ -488,8 +484,7 @@ static void dsolve_8x2_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a52 = LD_DP(a - 12);
     src_a53 = (v2f64) __msa_splati_d((v2i64) src_a52, 1);
     src_a52 = (v2f64) __msa_splati_d((v2i64) src_a52, 0);
-    src_a54 = __msa_cast_to_vector_double(*(a - 10));
-    src_a54 = (v2f64) __msa_splati_d((v2i64) src_a54, 0);
+    src_a54 = COPY_DOUBLE_TO_VECTOR(*(a -10));
 
     src_a40 = LD_DP(a - 24);
     src_a41 = (v2f64) __msa_splati_d((v2i64) src_a40, 1);
@@ -526,8 +521,7 @@ static void dsolve_8x2_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a34 = LD_DP(a - 30);
     src_a35 = (v2f64) __msa_splati_d((v2i64) src_a34, 1);
     src_a34 = (v2f64) __msa_splati_d((v2i64) src_a34, 0);
-    src_a36 = __msa_cast_to_vector_double(*(a - 28));
-    src_a36 = (v2f64) __msa_splati_d((v2i64) src_a36, 0);
+    src_a36 = COPY_DOUBLE_TO_VECTOR(*(a -28));
 
     res_c4 *= src_a36;
     res_c3 -= res_c4 * src_a35;
@@ -544,10 +538,8 @@ static void dsolve_8x2_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a16 = LD_DP(a - 48);
     src_a17 = (v2f64) __msa_splati_d((v2i64) src_a16, 1);
     src_a16 = (v2f64) __msa_splati_d((v2i64) src_a16, 0);
-    src_a18 = __msa_cast_to_vector_double(*(a - 46));
-    src_a18 = (v2f64) __msa_splati_d((v2i64) src_a18, 0);
-    src_a0 = __msa_cast_to_vector_double(*(a - 64));
-    src_a0 = (v2f64) __msa_splati_d((v2i64) src_a0, 0);
+    src_a18 = COPY_DOUBLE_TO_VECTOR(*(a - 46));
+    src_a0 = COPY_DOUBLE_TO_VECTOR(*(a - 64));
     src_a8 = LD_DP(a - 56);
     src_a9 = (v2f64) __msa_splati_d((v2i64) src_a8, 1);
     src_a8 = (v2f64) __msa_splati_d((v2i64) src_a8, 0);
@@ -785,11 +777,8 @@ static void dsolve_4x4_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a10 = (v2f64) __msa_splati_d((v2i64) src_a9, 1);
     src_a9 = (v2f64) __msa_splati_d((v2i64) src_a9, 0);
 
-    src_a8 = __msa_cast_to_vector_double(*(a + 8));
-    src_a0 = __msa_cast_to_vector_double(*(a + 0));
-
-    src_a8 = (v2f64) __msa_splati_d((v2i64) src_a8, 0);
-    src_a0 = (v2f64) __msa_splati_d((v2i64) src_a0, 0);
+    src_a8 = COPY_DOUBLE_TO_VECTOR(*(a + 8));
+    src_a0 = COPY_DOUBLE_TO_VECTOR(*(a + 0));
 
     src_a4 = LD_DP(a + 4);
     src_a5 = (v2f64) __msa_splati_d((v2i64) src_a4, 1);
@@ -890,11 +879,8 @@ static void dsolve_4x2_ln_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a10 = (v2f64) __msa_splati_d((v2i64) src_a9, 1);
     src_a9 = (v2f64) __msa_splati_d((v2i64) src_a9, 0);
 
-    src_a8 = __msa_cast_to_vector_double(*(a + 8));
-    src_a0 = __msa_cast_to_vector_double(*(a + 0));
-
-    src_a8 = (v2f64) __msa_splati_d((v2i64) src_a8, 0);
-    src_a0 = (v2f64) __msa_splati_d((v2i64) src_a0, 0);
+    src_a8 = COPY_DOUBLE_TO_VECTOR(*(a + 8));
+    src_a0 = COPY_DOUBLE_TO_VECTOR(*(a + 0));
 
     src_a4 = LD_DP(a + 4);
     src_a5 = (v2f64) __msa_splati_d((v2i64) src_a4, 1);

kernel/mips/dtrsm_kernel_LT_8x4_msa.c

@@ -215,8 +215,7 @@ void dsolve_8x4_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     res_c14 -= res_c8 * src_a6;
     res_c15 -= res_c8 * src_a7;
 
-    src_a9 = __msa_cast_to_vector_double(*(a + 9));
-    src_a9 = (v2f64) __msa_splati_d((v2i64) src_a9, 0);
+    src_a9 = COPY_DOUBLE_TO_VECTOR(*(a + 9));
     src_a10 = LD_DP(a + 10);
     src_a11 = (v2f64) __msa_splati_d((v2i64) src_a10, 1);
     src_a10 = (v2f64) __msa_splati_d((v2i64) src_a10, 0);
@@ -280,8 +279,7 @@ void dsolve_8x4_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     res_c14 -= res_c10 * src_a22;
     res_c15 -= res_c10 * src_a23;
 
-    src_a27 = __msa_cast_to_vector_double(*(a + 27));
-    src_a27 = (v2f64) __msa_splati_d((v2i64) src_a27, 0);
+    src_a27 = COPY_DOUBLE_TO_VECTOR(*(a + 27));
     src_a28 = LD_DP(a + 28);
     src_a29 = (v2f64) __msa_splati_d((v2i64) src_a28, 1);
     src_a28 = (v2f64) __msa_splati_d((v2i64) src_a28, 0);
@@ -326,8 +324,7 @@ void dsolve_8x4_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     res_c14 -= res_c12 * src_a38;
     res_c15 -= res_c12 * src_a39;
 
-    src_a45 = __msa_cast_to_vector_double(*(a + 45));
-    src_a45 = (v2f64) __msa_splati_d((v2i64) src_a45, 0);
+    src_a45 = COPY_DOUBLE_TO_VECTOR(*(a + 45));
     src_a46 = LD_DP(a + 46);
     src_a47 = (v2f64) __msa_splati_d((v2i64) src_a46, 1);
     src_a46 = (v2f64) __msa_splati_d((v2i64) src_a46, 0);
@@ -353,8 +350,7 @@ void dsolve_8x4_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     ILVRL_D2_DP(res_c5, res_c4, src_c2, src_c6);
     ILVRL_D2_DP(res_c13, res_c12, src_c10, src_c14);
 
-    src_a63 = __msa_cast_to_vector_double(*(a + 63));
-    src_a63 = (v2f64) __msa_splati_d((v2i64) src_a63, 0);
+    src_a63 = COPY_DOUBLE_TO_VECTOR(*(a + 63));
     src_a54 = LD_DP(a + 54);
     src_a55 = (v2f64) __msa_splati_d((v2i64) src_a54, 1);
     src_a54 = (v2f64) __msa_splati_d((v2i64) src_a54, 0);
@@ -478,8 +474,7 @@ static void dsolve_8x2_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     res_c6 -= res_c0 * src_a6;
     res_c7 -= res_c0 * src_a7;
 
-    src_a9 = __msa_cast_to_vector_double(*(a + 9));
-    src_a9 = (v2f64) __msa_splati_d((v2i64) src_a9, 0);
+    src_a9 = COPY_DOUBLE_TO_VECTOR(*(a + 9));
     src_a10 = LD_DP(a + 10);
     src_a11 = (v2f64) __msa_splati_d((v2i64) src_a10, 1);
     src_a10 = (v2f64) __msa_splati_d((v2i64) src_a10, 0);
@@ -515,8 +510,7 @@ static void dsolve_8x2_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     res_c6 -= res_c2 * src_a22;
     res_c7 -= res_c2 * src_a23;
 
-    src_a27 = __msa_cast_to_vector_double(*(a + 27));
-    src_a27 = (v2f64) __msa_splati_d((v2i64) src_a27, 0);
+    src_a27 = COPY_DOUBLE_TO_VECTOR(*(a + 27));
     src_a28 = LD_DP(a + 28);
     src_a29 = (v2f64) __msa_splati_d((v2i64) src_a28, 1);
     src_a28 = (v2f64) __msa_splati_d((v2i64) src_a28, 0);
@@ -553,8 +547,7 @@ static void dsolve_8x2_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     res_c6 -= res_c4 * src_a38;
     res_c7 -= res_c4 * src_a39;
 
-    src_a45 = __msa_cast_to_vector_double(*(a + 45));
-    src_a45 = (v2f64) __msa_splati_d((v2i64) src_a45, 0);
+    src_a45 = COPY_DOUBLE_TO_VECTOR(*(a + 45));
     src_a46 = LD_DP(a + 46);
     src_a47 = (v2f64) __msa_splati_d((v2i64) src_a46, 1);
     src_a46 = (v2f64) __msa_splati_d((v2i64) src_a46, 0);
@@ -563,8 +556,7 @@ static void dsolve_8x2_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     res_c6 -= res_c5 * src_a46;
     res_c7 -= res_c5 * src_a47;
 
-    src_a63 = __msa_cast_to_vector_double(*(a + 63));
-    src_a63 = (v2f64) __msa_splati_d((v2i64) src_a63, 0);
+    src_a63 = COPY_DOUBLE_TO_VECTOR(*(a + 63));
     src_a54 = LD_DP(a + 54);
     src_a55 = (v2f64) __msa_splati_d((v2i64) src_a54, 1);
     src_a54 = (v2f64) __msa_splati_d((v2i64) src_a54, 0);
@@ -786,8 +778,7 @@ static void dsolve_4x4_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     res_c6 -= res_c4 * src_a2;
     res_c7 -= res_c4 * src_a3;
 
-    src_a5 = __msa_cast_to_vector_double(*(a + 5));
-    src_a5 = (v2f64) __msa_splati_d((v2i64) src_a5, 0);
+    src_a5 = COPY_DOUBLE_TO_VECTOR(*(a + 5));
     src_a6 = LD_DP(a + 6);
     src_a7 = (v2f64) __msa_splati_d((v2i64) src_a6, 1);
     src_a6 = (v2f64) __msa_splati_d((v2i64) src_a6, 0);
@@ -803,8 +794,7 @@ static void dsolve_4x4_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a10 = LD_DP(a + 10);
     src_a11 = (v2f64) __msa_splati_d((v2i64) src_a10, 1);
     src_a10 = (v2f64) __msa_splati_d((v2i64) src_a10, 0);
-    src_a15 = __msa_cast_to_vector_double(*(a + 15));
-    src_a15 = (v2f64) __msa_splati_d((v2i64) src_a15, 0);
+    src_a15 = COPY_DOUBLE_TO_VECTOR(*(a + 15));
 
     res_c2 *= src_a10;
     res_c3 -= res_c2 * src_a11;
@@ -881,8 +871,7 @@ static void dsolve_4x2_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     res_c2 -= res_c0 * src_a2;
     res_c3 -= res_c0 * src_a3;
 
-    src_a5 = __msa_cast_to_vector_double(*(a + 5));
-    src_a5 = (v2f64) __msa_splati_d((v2i64) src_a5, 0);
+    src_a5 = COPY_DOUBLE_TO_VECTOR(*(a + 5));
     src_a6 = LD_DP(a + 6);
     src_a7 = (v2f64) __msa_splati_d((v2i64) src_a6, 1);
     src_a6 = (v2f64) __msa_splati_d((v2i64) src_a6, 0);
@@ -894,8 +883,7 @@ static void dsolve_4x2_lt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_a10 = LD_DP(a + 10);
     src_a11 = (v2f64) __msa_splati_d((v2i64) src_a10, 1);
     src_a10 = (v2f64) __msa_splati_d((v2i64) src_a10, 0);
-    src_a15 = __msa_cast_to_vector_double(*(a + 15));
-    src_a15 = (v2f64) __msa_splati_d((v2i64) src_a15, 0);
+    src_a15 = COPY_DOUBLE_TO_VECTOR(*(a + 15));
 
     res_c2 *= src_a10;
     res_c3 -= res_c2 * src_a11;

kernel/mips/dtrsm_kernel_RN_8x4_msa.c

@@ -161,16 +161,14 @@ void dsolve_8x4_rn_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     src_b2 = LD_DP(b + 2);
     src_b3 = (v2f64) __msa_splati_d((v2i64) src_b2, 1);
     src_b2 = (v2f64) __msa_splati_d((v2i64) src_b2, 0);
-    src_b5 = __msa_cast_to_vector_double(*(b + 5));
-    src_b5 = (v2f64) __msa_splati_d((v2i64) src_b5, 0);
+    src_b5 = COPY_DOUBLE_TO_VECTOR(*(b + 5));
     src_b6 = LD_DP(b + 6);
     src_b7 = (v2f64) __msa_splati_d((v2i64) src_b6, 1);
     src_b6 = (v2f64) __msa_splati_d((v2i64) src_b6, 0);
     src_b10 = LD_DP(b + 10);
     src_b11 = (v2f64) __msa_splati_d((v2i64) src_b10, 1);
     src_b10 = (v2f64) __msa_splati_d((v2i64) src_b10, 0);
-    src_b15 = __msa_cast_to_vector_double(*(b + 15));
-    src_b15 = (v2f64) __msa_splati_d((v2i64) src_b15, 0);
+    src_b15 = COPY_DOUBLE_TO_VECTOR(*(b + 15));
 
     src_c0 *= src_b0;
     src_c1 *= src_b0;
@@ -294,8 +292,7 @@ static void dsolve_8x2_rn_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_b0 = LD_DP(b + 0);
     src_b1 = (v2f64) __msa_splati_d((v2i64) src_b0, 1);
     src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
-    src_b3 = __msa_cast_to_vector_double(*(b + 3));
-    src_b3 = (v2f64) __msa_splati_d((v2i64) src_b3, 0);
+    src_b3 = COPY_DOUBLE_TO_VECTOR(*(b + 3));
 
     src_c0 *= src_b0;
     src_c1 *= src_b0;
@@ -347,8 +344,7 @@ static void dsolve_8x1_rn_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG bk)
         }
     }
 
-    src_b0 = __msa_cast_to_vector_double(*b);
-    src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
+    src_b0 = COPY_DOUBLE_TO_VECTOR(*b);
 
     src_c0 *= src_b0;
     src_c1 *= src_b0;
@@ -407,16 +403,14 @@ static void dsolve_4x4_rn_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_b2 = LD_DP(b + 2);
     src_b3 = (v2f64) __msa_splati_d((v2i64) src_b2, 1);
     src_b2 = (v2f64) __msa_splati_d((v2i64) src_b2, 0);
-    src_b5 = __msa_cast_to_vector_double(*(b + 5));
-    src_b5 = (v2f64) __msa_splati_d((v2i64) src_b5, 0);
+    src_b5 = COPY_DOUBLE_TO_VECTOR(*(b + 5));
     src_b6 = LD_DP(b + 6);
     src_b7 = (v2f64) __msa_splati_d((v2i64) src_b6, 1);
     src_b6 = (v2f64) __msa_splati_d((v2i64) src_b6, 0);
     src_b10 = LD_DP(b + 10);
     src_b11 = (v2f64) __msa_splati_d((v2i64) src_b10, 1);
     src_b10 = (v2f64) __msa_splati_d((v2i64) src_b10, 0);
-    src_b15 = __msa_cast_to_vector_double(*(b + 15));
-    src_b15 = (v2f64) __msa_splati_d((v2i64) src_b15, 0);
+    src_b15 = COPY_DOUBLE_TO_VECTOR(*(b + 15));
 
     src_c0 *= src_b0;
     src_c1 *= src_b0;
@@ -490,8 +484,7 @@ static void dsolve_4x2_rn_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_b0 = LD_DP(b + 0);
     src_b1 = (v2f64) __msa_splati_d((v2i64) src_b0, 1);
     src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
-    src_b3 = __msa_cast_to_vector_double(*(b + 3));
-    src_b3 = (v2f64) __msa_splati_d((v2i64) src_b3, 0);
+    src_b3 = COPY_DOUBLE_TO_VECTOR(*(b + 3));
 
     src_c0 *= src_b0;
     src_c1 *= src_b0;

kernel/mips/dtrsm_kernel_RT_8x4_msa.c

@@ -168,11 +168,9 @@ void dsolve_8x4_rt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG bk)
     src_b8 = LD_DP(b + 8);
     src_b9 = (v2f64) __msa_splati_d((v2i64) src_b8, 1);
     src_b8 = (v2f64) __msa_splati_d((v2i64) src_b8, 0);
-    src_b10 = __msa_cast_to_vector_double(*(b + 10));
-    src_b10 = (v2f64) __msa_splati_d((v2i64) src_b10, 0);
+    src_b10 = COPY_DOUBLE_TO_VECTOR(*(b + 10));
 
-    src_b0 = __msa_cast_to_vector_double(*(b + 0));
-    src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
+    src_b0 = COPY_DOUBLE_TO_VECTOR(*(b + 0));
     src_b4 = LD_DP(b + 4);
     src_b5 = (v2f64) __msa_splati_d((v2i64) src_b4, 1);
     src_b4 = (v2f64) __msa_splati_d((v2i64) src_b4, 0);
@@ -298,8 +296,7 @@ static void dsolve_8x2_rt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     a -= 16;
     b -= 4;
 
-    src_b0 = __msa_cast_to_vector_double(*(b + 0));
-    src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
+    src_b0 = COPY_DOUBLE_TO_VECTOR(*(b + 0));
     src_b2 = LD_DP(b + 2);
     src_b3 = (v2f64) __msa_splati_d((v2i64) src_b2, 1);
     src_b2 = (v2f64) __msa_splati_d((v2i64) src_b2, 0);
@@ -377,8 +374,7 @@ static void dsolve_8x1_rt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG bk)
     a -= 8;
     b -= 1;
 
-    src_b0 = __msa_cast_to_vector_double(*b);
-    src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
+    src_b0 = COPY_DOUBLE_TO_VECTOR(*b);
 
     src_c0 *= src_b0;
     src_c1 *= src_b0;
@@ -445,11 +441,9 @@ static void dsolve_4x4_rt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     src_b8 = LD_DP(b + 8);
     src_b9 = (v2f64) __msa_splati_d((v2i64) src_b8, 1);
     src_b8 = (v2f64) __msa_splati_d((v2i64) src_b8, 0);
-    src_b10 = __msa_cast_to_vector_double(*(b + 10));
-    src_b10 = (v2f64) __msa_splati_d((v2i64) src_b10, 0);
+    src_b10 = COPY_DOUBLE_TO_VECTOR(*(b + 10));
 
-    src_b0 = __msa_cast_to_vector_double(*(b + 0));
-    src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
+    src_b0 = COPY_DOUBLE_TO_VECTOR(*(b + 0));
     src_b4 = LD_DP(b + 4);
     src_b5 = (v2f64) __msa_splati_d((v2i64) src_b4, 1);
     src_b4 = (v2f64) __msa_splati_d((v2i64) src_b4, 0);
@@ -527,8 +521,7 @@ static void dsolve_4x2_rt_msa(FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLO
     a -= 8;
     b -= 4;
 
-    src_b0 = __msa_cast_to_vector_double(*(b + 0));
-    src_b0 = (v2f64) __msa_splati_d((v2i64) src_b0, 0);
+    src_b0 = COPY_DOUBLE_TO_VECTOR(*(b + 0));
     src_b2 = LD_DP(b + 2);
     src_b3 = (v2f64) __msa_splati_d((v2i64) src_b2, 1);
     src_b2 = (v2f64) __msa_splati_d((v2i64) src_b2, 0);

kernel/mips/macros_msa.h

@@ -63,16 +63,12 @@ inline static void prefetch_load_lf(unsigned char *src)
 #define ST_DP(...) ST_D(v2f64, __VA_ARGS__)
 
 #define COPY_FLOAT_TO_VECTOR(a) ( {                \
-    v4f32  out;                                    \
-    out = __msa_cast_to_vector_float(a);           \
-    out = (v4f32) __msa_splati_w((v4i32) out, 0);  \
+    v4f32  out = {a, a, a, a};                     \
     out;                                           \
 } )
 
 #define COPY_DOUBLE_TO_VECTOR(a) ( {               \
-    v2f64  out;                                    \
-    out = __msa_cast_to_vector_double(a);          \
-    out = (v2f64) __msa_splati_d((v2i64) out, 0);  \
+    v2f64  out = {a, a};                           \
     out;                                           \
 } )
 

kernel/mips/srot_msa.c

@@ -48,11 +48,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y,
     {
         if ((0 == c) && (0 == s))
         {
-            v4f32 zero = __msa_cast_to_vector_float(0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 0, 0.0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 1, 0.0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 2, 0.0);
-            zero = (v4f32) __msa_insert_w((v4i32) zero, 3, 0.0);
+            v4f32 zero = {0.0, 0.0, 0.0, 0.0};
 
             /* process 4 floats */
             for (j = (n >> 2); j--;)

kernel/mips/sscal_msa.c

@@ -44,11 +44,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x,
     {
         if (0.0 == da)
         {
-            v4f32 zero_v = __msa_cast_to_vector_float(0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 0, 0.0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 1, 0.0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 2, 0.0);
-            zero_v = (v4f32) __msa_insert_w((v4i32) zero_v, 3, 0.0);
+            v4f32 zero_v = {0.0, 0.0, 0.0, 0.0};
 
             for (i = (n >> 6); i--;)
             {

kernel/mips/sswap_msa.c

@@ -198,7 +198,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3,
             }
         }
     }
-    else
+    else if ((inc_x != 0) && (inc_y != 0))
     {
         for (i = (n >> 3); i--;)
         {
@@ -262,6 +262,33 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3,
             }
         }
     }
+    else
+    {
+        if (inc_x == inc_y)
+        {
+            if (n & 1)
+            {
+                x0 = *srcx;
+                *srcx  = *srcy;
+                *srcy  = x0;
+            }
+            else
+                return (0);
+        }
+        else
+        {
+            BLASLONG ix = 0, iy = 0;
+            while (i < n)
+            {
+                x0 = srcx[ix];
+                srcx[ix] = srcy[iy];
+                srcy[iy] = x0;
+                ix += inc_x;
+                iy += inc_y;
+                i++;
+            }
+        }
+    }
 
     return (0);
 }

kernel/mips/zgemv_n_msa.c

@@ -56,11 +56,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     #if !defined(XCONJ)
         #define OP0  +=
         #define OP1  -=
-        #define OP2  -=
+        #define OP2  +=
     #else
         #define OP0  -=
         #define OP1  -=
-        #define OP2  +=
+        #define OP2  -=
     #endif
 #endif
 

kernel/mips/zgemv_t_msa.c

@@ -34,14 +34,26 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #undef OP3
 #undef OP4
 
-#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
-    #define OP0  -=
-    #define OP1  +=
-    #define OP2  +=
+#if !defined(CONJ)
+    #if !defined(XCONJ)
+        #define OP0  -=
+        #define OP1  +=
+        #define OP2  +=
+    #else
+        #define OP0  +=
+        #define OP1  +=
+        #define OP2  -=
+    #endif
 #else
-    #define OP0  +=
-    #define OP1  +=
-    #define OP2  -=
+    #if !defined(XCONJ)
+        #define OP0  +=
+        #define OP1  -=
+        #define OP2  +=
+    #else
+        #define OP0  -=
+        #define OP1  -=
+        #define OP2  -=
+    #endif
 #endif
 
 #define ZGEMV_T_8x1()                     \

kernel/mips/zscal_msa.c

@@ -49,9 +49,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
     {
         if ((0.0 == da_r) && (0.0 == da_i))
         {
-            v2f64 zero_v = __msa_cast_to_vector_double(0);
-            zero_v = (v2f64) __msa_insert_d((v2i64) zero_v, 0, 0.0);
-            zero_v = (v2f64) __msa_insert_d((v2i64) zero_v, 1, 0.0);
+            v2f64 zero_v = {0.0, 0.0};
 
             for (i = (n >> 4); i--;)
             {
@@ -475,9 +473,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
 
         if ((0.0 == da_r) && (0.0 == da_i))
         {
-            v2f64 zero_v = __msa_cast_to_vector_double(0);
-            zero_v = (v2f64) __msa_insert_d((v2i64) zero_v, 0, 0.0);
-            zero_v = (v2f64) __msa_insert_d((v2i64) zero_v, 1, 0.0);
+            v2f64 zero_v = {0.0, 0.0};
 
             for (i = (n >> 4); i--;)
             {

kernel/mips64/KERNEL.LOONGSON3B

@@ -1,64 +0,0 @@
-SAXPYKERNEL=axpy_loongson3a.S
-DAXPYKERNEL=daxpy_loongson3a_simd.S
-
-SGEMVNKERNEL = gemv_n_loongson3a.c
-SGEMVTKERNEL = gemv_t_loongson3a.c
-DGEMVNKERNEL = gemv_n_loongson3a.c
-DGEMVTKERNEL = gemv_t_loongson3a.c
-CGEMVNKERNEL = zgemv_n_loongson3a.c
-CGEMVTKERNEL = zgemv_t_loongson3a.c
-ZGEMVNKERNEL = zgemv_n_loongson3a.c
-ZGEMVTKERNEL = zgemv_t_loongson3a.c
-
-STRMMKERNEL	= ../generic/trmmkernel_2x2.c
-DTRMMKERNEL	= ../generic/trmmkernel_2x2.c
-CTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
-ZTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
-
-SGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
-SGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
-SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
-SGEMMONCOPYOBJ =  sgemm_oncopy.o
-SGEMMOTCOPYOBJ =  sgemm_otcopy.o
-
-DGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
-DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
-DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
-DGEMMONCOPYOBJ = dgemm_oncopy.o
-DGEMMOTCOPYOBJ = dgemm_otcopy.o
-
-CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
-CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
-CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
-CGEMMONCOPYOBJ =  cgemm_oncopy.o
-CGEMMOTCOPYOBJ =  cgemm_otcopy.o
-
-ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
-ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
-ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
-ZGEMMONCOPYOBJ =  zgemm_oncopy.o
-ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
-
-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
-
-
-
-

kernel/mips64/KERNEL.LOONGSON3R3

@@ -16,32 +16,32 @@ SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
 SGEMMITCOPY    =  ../generic/gemm_tcopy_8.c
 SGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
 SGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
-SGEMMINCOPYOBJ =  sgemm_incopy.o
-SGEMMITCOPYOBJ =  sgemm_itcopy.o
-SGEMMONCOPYOBJ =  sgemm_oncopy.o
-SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 DGEMMKERNEL    =  dgemm_kernel_loongson3a_4x4.S
 DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
 DGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
-DGEMMONCOPYOBJ = dgemm_oncopy.o
-DGEMMOTCOPYOBJ = dgemm_otcopy.o
+DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 CGEMMKERNEL    =  cgemm_kernel_loongson3a_4x2_ps.S
 CGEMMINCOPY    = ../generic/zgemm_ncopy_4.c
 CGEMMITCOPY    = ../generic/zgemm_tcopy_4.c
 CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
 CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
-CGEMMINCOPYOBJ =  cgemm_incopy.o
-CGEMMITCOPYOBJ =  cgemm_itcopy.o
-CGEMMONCOPYOBJ =  cgemm_oncopy.o
-CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 ZGEMMKERNEL    =  zgemm_kernel_loongson3a_2x2.S
 ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
 ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
-ZGEMMONCOPYOBJ =  zgemm_oncopy.o
-ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
 STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
 STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
@@ -64,6 +64,3 @@ ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
 ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
 
 DSDOTKERNEL     = ../mips/dot.c
-
-
-

kernel/mips64/KERNEL.LOONGSON3R4

@@ -0,0 +1,192 @@
+ifdef HAVE_MSA
+SAXPYKERNEL = ../mips/saxpy_msa.c
+DAXPYKERNEL = ../mips/daxpy_msa.c
+CAXPYKERNEL = ../mips/caxpy_msa.c
+ZAXPYKERNEL = ../mips/zaxpy_msa.c
+else
+SAXPYKERNEL = axpy_loongson3a.S
+DAXPYKERNEL = daxpy_loongson3a_simd.S
+endif
+
+ifdef HAVE_MSA
+SCOPYKERNEL  = ../mips/scopy_msa.c
+DCOPYKERNEL  = ../mips/dcopy_msa.c
+CCOPYKERNEL  = ../mips/ccopy_msa.c
+ZCOPYKERNEL  = ../mips/zcopy_msa.c
+endif
+
+ifdef HAVE_MSA
+SDOTKERNEL   = ../mips/sdot_msa.c
+DDOTKERNEL   = ../mips/ddot_msa.c
+CDOTKERNEL   = ../mips/cdot_msa.c
+ZDOTKERNEL   = ../mips/zdot_msa.c
+endif
+DSDOTKERNEL  = ../mips/dot.c
+
+ifdef HAVE_MSA
+SROTKERNEL   = ../mips/srot_msa.c
+DROTKERNEL   = ../mips/drot_msa.c
+CROTKERNEL   = ../mips/crot_msa.c
+ZROTKERNEL   = ../mips/zrot_msa.c
+endif
+
+ifdef HAVE_MSA
+SSCALKERNEL  = ../mips/sscal_msa.c
+DSCALKERNEL  = ../mips/dscal_msa.c
+CSCALKERNEL  = ../mips/cscal_msa.c
+ZSCALKERNEL  = ../mips/zscal_msa.c
+endif
+
+ifdef HAVE_MSA
+SGEMVNKERNEL = ../mips/sgemv_n_msa.c
+DGEMVNKERNEL = ../mips/dgemv_n_msa.c
+SGEMVTKERNEL = ../mips/sgemv_t_msa.c
+DGEMVTKERNEL = ../mips/dgemv_t_msa.c
+CGEMVNKERNEL = ../mips/cgemv_n_msa.c
+CGEMVTKERNEL = ../mips/cgemv_t_msa.c
+ZGEMVNKERNEL = ../mips/zgemv_n_msa.c
+ZGEMVTKERNEL = ../mips/zgemv_t_msa.c
+else
+SGEMVNKERNEL = gemv_n_loongson3a.c
+SGEMVTKERNEL = gemv_t_loongson3a.c
+DGEMVNKERNEL = gemv_n_loongson3a.c
+DGEMVTKERNEL = gemv_t_loongson3a.c
+CGEMVNKERNEL = zgemv_n_loongson3a.c
+CGEMVTKERNEL = zgemv_t_loongson3a.c
+ZGEMVNKERNEL = zgemv_n_loongson3a.c
+ZGEMVTKERNEL = zgemv_t_loongson3a.c
+endif
+
+ifdef HAVE_MSA
+SASUMKERNEL  = ../mips/sasum_msa.c
+DASUMKERNEL  = ../mips/dasum_msa.c
+CASUMKERNEL  = ../mips/casum_msa.c
+ZASUMKERNEL  = ../mips/zasum_msa.c
+endif
+
+ifdef HAVE_MSA
+SSWAPKERNEL  = ../mips/sswap_msa.c
+DSWAPKERNEL  = ../mips/dswap_msa.c
+CSWAPKERNEL  = ../mips/cswap_msa.c
+ZSWAPKERNEL  = ../mips/zswap_msa.c
+endif
+
+ifdef HAVE_MSA
+SGEMMKERNEL    = ../mips/sgemm_kernel_8x8_msa.c
+SGEMMONCOPY    = ../mips/sgemm_ncopy_8_msa.c
+SGEMMOTCOPY    = ../mips/sgemm_tcopy_8_msa.c
+SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+else
+SGEMMKERNEL    =  sgemm_kernel_8x4_ps.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
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifdef HAVE_MSA
+DGEMMKERNEL    = ../mips/dgemm_kernel_8x4_msa.c
+DGEMMINCOPY    = ../mips/dgemm_ncopy_8_msa.c
+DGEMMITCOPY    = ../mips/dgemm_tcopy_8_msa.c
+DGEMMONCOPY    = ../mips/dgemm_ncopy_4_msa.c
+DGEMMOTCOPY    = ../mips/dgemm_tcopy_4_msa.c
+DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+else
+DGEMMKERNEL    =  dgemm_kernel_loongson3a_4x4.S
+DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
+DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifdef HAVE_MSA
+CGEMMKERNEL    = ../mips/cgemm_kernel_8x4_msa.c
+CGEMMINCOPY    = ../mips/cgemm_ncopy_8_msa.c
+CGEMMITCOPY    = ../mips/cgemm_tcopy_8_msa.c
+CGEMMONCOPY    = ../mips/cgemm_ncopy_4_msa.c
+CGEMMOTCOPY    = ../mips/cgemm_tcopy_4_msa.c
+CGEMMINCOPYOBJ = cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ = cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ = cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ = cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+else
+CGEMMKERNEL    =  cgemm_kernel_loongson3a_4x2_ps.S
+CGEMMINCOPY    = ../generic/zgemm_ncopy_4.c
+CGEMMITCOPY    = ../generic/zgemm_tcopy_4.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifdef HAVE_MSA
+ZGEMMKERNEL    = ../mips/zgemm_kernel_4x4_msa.c
+ZGEMMONCOPY    = ../mips/zgemm_ncopy_4_msa.c
+ZGEMMOTCOPY    = ../mips/zgemm_tcopy_4_msa.c
+ZGEMMONCOPYOBJ = zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ = zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+else
+ZGEMMKERNEL    =  zgemm_kernel_loongson3a_2x2.S
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifdef HAVE_MSA
+STRSMKERNEL_LN = ../mips/strsm_kernel_LN_8x8_msa.c
+STRSMKERNEL_LT = ../mips/strsm_kernel_LT_8x8_msa.c
+STRSMKERNEL_RN = ../mips/strsm_kernel_RN_8x8_msa.c
+STRSMKERNEL_RT = ../mips/strsm_kernel_RT_8x8_msa.c
+else
+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
+endif
+
+ifdef HAVE_MSA
+DTRSMKERNEL_LN = ../mips/dtrsm_kernel_LN_8x4_msa.c
+DTRSMKERNEL_LT = ../mips/dtrsm_kernel_LT_8x4_msa.c
+DTRSMKERNEL_RN = ../mips/dtrsm_kernel_RN_8x4_msa.c
+DTRSMKERNEL_RT = ../mips/dtrsm_kernel_RT_8x4_msa.c
+else
+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
+endif
+
+ifdef HAVE_MSA
+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
+else
+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
+endif
+
+ifdef HAVE_MSA
+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
+else
+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
+endif

kernel/power/KERNEL.POWER10

@@ -63,15 +63,15 @@ 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
+STRSMKERNEL_LN	= trsm_kernel_LN_power10.c
+STRSMKERNEL_LT	= trsm_kernel_LT_power10.c
+STRSMKERNEL_RN	= trsm_kernel_RN_power10.c
+STRSMKERNEL_RT	= trsm_kernel_RT_power10.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
+DTRSMKERNEL_LN	= trsm_kernel_LN_power10.c
+DTRSMKERNEL_LT	= trsm_kernel_LT_power10.c
+DTRSMKERNEL_RN	= trsm_kernel_RN_power10.c
+DTRSMKERNEL_RT	= trsm_kernel_RT_power10.c
 
 CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
@@ -154,11 +154,7 @@ ZCOPYKERNEL  = zcopy_power10.c
 SDOTKERNEL   =  sdot_power10.c
 DDOTKERNEL   =  ddot_power10.c
 DSDOTKERNEL  =  sdot_power10.c
-ifneq ($(GCCVERSIONGTEQ9),1)
-CDOTKERNEL   =  cdot_power9.S
-else
 CDOTKERNEL   =  cdot.c
-endif
 ZDOTKERNEL   =  zdot.c
 #
 SNRM2KERNEL  = ../arm/nrm2.c
@@ -173,8 +169,13 @@ ZROTKERNEL   = zrot.c
 #
 SSCALKERNEL  = sscal.c
 DSCALKERNEL  = dscal.c
+ifeq ($(C_COMPILER), PGI)
+CSCALKERNEL  = ../arm/zscal.c
+ZSCALKERNEL  = ../arm/zscal.c
+else
 CSCALKERNEL  = zscal.c
 ZSCALKERNEL  = zscal.c
+endif
 #
 SSWAPKERNEL  = sswap.c
 DSWAPKERNEL  = dswap.c

kernel/power/KERNEL.POWER8

@@ -242,8 +242,13 @@ ZROTKERNEL   = zrot.c
 #
 SSCALKERNEL  = sscal.c
 DSCALKERNEL  = dscal.c
+ifeq ($(C_COMPILER), PGI)
+CSCALKERNEL  = ../arm/zscal.c
+ZSCALKERNEL  = ../arm/zscal.c
+else
 CSCALKERNEL  = zscal.c
 ZSCALKERNEL  = zscal.c
+endif
 #
 SSWAPKERNEL  = sswap.c
 DSWAPKERNEL  = dswap.c

kernel/power/KERNEL.POWER9

@@ -52,15 +52,15 @@ 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
+STRSMKERNEL_LN	= trsm_kernel_LN_power10.c
+STRSMKERNEL_LT	= trsm_kernel_LT_power10.c
+STRSMKERNEL_RN	= trsm_kernel_RN_power10.c
+STRSMKERNEL_RT	= trsm_kernel_RT_power10.c
 
-DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LN	= trsm_kernel_LN_power10.c
 DTRSMKERNEL_LT	= dtrsm_kernel_LT_16x4_power8.S
-DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
-DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+DTRSMKERNEL_RN	= trsm_kernel_RN_power10.c
+DTRSMKERNEL_RT	= trsm_kernel_RT_power10.c
 
 CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
@@ -166,8 +166,13 @@ ZROTKERNEL   = zrot.c
 #
 SSCALKERNEL  = sscal.c
 DSCALKERNEL  = dscal.c
+ifeq ($(C_COMPILER), PGI)
+CSCALKERNEL  = ../arm/zscal.c
+ZSCALKERNEL  = ../arm/zscal.c
+else
 CSCALKERNEL  = zscal.c
 ZSCALKERNEL  = zscal.c
+endif
 #
 SSWAPKERNEL  = sswap.c
 DSWAPKERNEL  = dswap.c

kernel/power/caxpy_microk_power10.c

@@ -112,10 +112,14 @@ static void caxpy_kernel_8 (long n, float *x, float *y,
        "xvmaddasp	38, 58, 33	\n\t"
        "xvmaddasp	39, 59, 33	\n\t"
 
-       "stxvp		48, 0(%4)	\n\t"
-       "stxvp		50, 32(%4)	\n\t"
-       "stxvp		34, 64(%4)	\n\t"
-       "stxvp		38, 96(%4)	\n\t"
+       "stxv		49, 0(%4)	\n\t"
+       "stxv		48, 16(%4)	\n\t"
+       "stxv		51, 32(%4)	\n\t"
+       "stxv		50, 48(%4)	\n\t"
+       "stxv		35, 64(%4)	\n\t"
+       "stxv		34, 80(%4)	\n\t"
+       "stxv		39, 96(%4)	\n\t"
+       "stxv		38, 112(%4)	\n\t"
 
        "addi		%4, %4, 128	\n\t"
        "xxperm 52, 40, %x10 \n\t"       // exchange real and imag part
@@ -159,10 +163,14 @@ static void caxpy_kernel_8 (long n, float *x, float *y,
        "xvmaddasp	38, 58, 33	\n\t"
        "xvmaddasp	39, 59, 33	\n\t"
 
-       "stxvp		48, 0(%4)	\n\t"
-       "stxvp		50, 32(%4)	\n\t"
-       "stxvp		34, 64(%4)	\n\t"
-       "stxvp		38, 96(%4)	\n\t"
+       "stxv		49, 0(%4)	\n\t"
+       "stxv		48, 16(%4)	\n\t"
+       "stxv		51, 32(%4)	\n\t"
+       "stxv		50, 48(%4)	\n\t"
+       "stxv		35, 64(%4)	\n\t"
+       "stxv		34, 80(%4)	\n\t"
+       "stxv		39, 96(%4)	\n\t"
+       "stxv		38, 112(%4)	\n\t"
 
      "#n=%1 x=%5=%2 y=%0=%3 alpha=(%7,%8) mvecp=%6=%9 ytmp=%4\n"
      :

kernel/power/ccopy_microk_power10.c

@@ -0,0 +1,115 @@
+/***************************************************************************
+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.
+*****************************************************************************/
+
+#define HAVE_KERNEL 1
+
+static void copy_kernel (BLASLONG n, FLOAT *x, FLOAT *y)
+{
+  __asm__
+    (
+       "lxvp		32, 0(%2)	\n\t"
+       "lxvp		34, 32(%2)	\n\t"
+       "lxvp		36, 64(%2)	\n\t"
+       "lxvp		38, 96(%2)	\n\t"
+       "lxvp		40, 128(%2)	\n\t"
+       "lxvp		42, 160(%2)	\n\t"
+       "lxvp		44, 192(%2)	\n\t"
+       "lxvp		46, 224(%2)	\n\t"
+
+       "addi		%2, %2, 256	\n\t"
+       "addic.		%1, %1, -32	\n\t"
+       "ble		two%=		\n\t"
+
+       ".align	5		\n"
+     "one%=:				\n\t"
+
+       "stxv		33, 0(%3)	\n\t"
+       "stxv		32, 16(%3)	\n\t"
+       "stxv		35, 32(%3)	\n\t"
+       "stxv		34, 48(%3)	\n\t"
+       "stxv		37, 64(%3)	\n\t"
+       "stxv		36, 80(%3)	\n\t"
+       "stxv		39, 96(%3)	\n\t"
+       "stxv		38, 112(%3)	\n\t"
+       "lxvp		32, 0(%2)	\n\t"
+       "lxvp		34, 32(%2)	\n\t"
+       "lxvp		36, 64(%2)	\n\t"
+       "lxvp		38, 96(%2)	\n\t"
+
+       "stxv		41, 128(%3)	\n\t"
+       "stxv		40, 144(%3)	\n\t"
+       "stxv		43, 160(%3)	\n\t"
+       "stxv		42, 176(%3)	\n\t"
+       "stxv		45, 192(%3)	\n\t"
+       "stxv		44, 208(%3)	\n\t"
+       "stxv		47, 224(%3)	\n\t"
+       "stxv		46, 240(%3)	\n\t"
+       "lxvp		40, 128(%2)	\n\t"
+       "lxvp		42, 160(%2)	\n\t"
+       "lxvp		44, 192(%2)	\n\t"
+       "lxvp		46, 224(%2)	\n\t"
+
+
+       "addi		%3, %3, 256	\n\t"
+       "addi		%2, %2, 256	\n\t"
+
+       "addic.		%1, %1, -32	\n\t"
+       "bgt		one%=		\n"
+
+     "two%=:				\n\t"
+
+       "stxv		33, 0(%3)	\n\t"
+       "stxv		32, 16(%3)	\n\t"
+       "stxv		35, 32(%3)	\n\t"
+       "stxv		34, 48(%3)	\n\t"
+       "stxv		37, 64(%3)	\n\t"
+       "stxv		36, 80(%3)	\n\t"
+       "stxv		39, 96(%3)	\n\t"
+       "stxv		38, 112(%3)	\n\t"
+       "stxv		41, 128(%3)	\n\t"
+       "stxv		40, 144(%3)	\n\t"
+       "stxv		43, 160(%3)	\n\t"
+       "stxv		42, 176(%3)	\n\t"
+       "stxv		45, 192(%3)	\n\t"
+       "stxv		44, 208(%3)	\n\t"
+       "stxv		47, 224(%3)	\n\t"
+       "stxv		46, 240(%3)	\n\t"
+
+     "#n=%1 x=%4=%2 y=%0=%3"
+     :
+       "=m" (*y),
+       "+r" (n),	// 1
+       "+b" (x),	// 2
+       "+b" (y) 	// 3
+     :
+       "m" (*x)
+     :
+       "cr0",
+       "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
+       "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47"
+     );
+}

kernel/power/ccopy_power10.c

@@ -28,7 +28,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if defined(__VEC__) || defined(__ALTIVEC__)
-#include "copy_microk_power10.c"
+#include "ccopy_microk_power10.c"
 #endif
 
 #ifndef HAVE_KERNEL
@@ -86,7 +86,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
 	if ( (inc_x == 1) && (inc_y == 1 ))
 	{
 
-		BLASLONG n1 = n & -64;
+		BLASLONG n1 = n & -32;
 		if ( n1 > 0 )
 		{
 			copy_kernel(n1, x, y);

kernel/power/cdot.c

@@ -28,6 +28,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 
 #include "common.h"
+#if defined(POWER10)
+#include "cdot_microk_power10.c"
+#else
 #ifndef HAVE_KERNEL_8
 #include <altivec.h> 
 
@@ -99,6 +102,7 @@ static void cdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, float *dot)
  
 }
 #endif
+#endif
  
 
 OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) {
@@ -116,7 +120,11 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 
     if ((inc_x == 1) && (inc_y == 1)) {
 
+#if defined(POWER10)
+        BLASLONG n1 = n & -16;
+#else
         BLASLONG n1 = n & -8;
+#endif
         BLASLONG j=0; 
 
         if (n1){

kernel/power/cdot_microk_power10.c

@@ -0,0 +1,177 @@
+/***************************************************************************
+Copyright (c) 2021, 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 HAVE_KERNEL_8 1
+
+static void cdot_kernel_8 (long n, float *x, float *y, float *dot)
+{
+  __vector unsigned char mask = { 11,10,9,8,15,14,13,12,3,2,1,0,7,6,5,4};
+  __asm__
+    (
+       "dcbt		0, %2		\n\t"
+       "dcbt		0, %3		\n\t"
+
+       "xxlxor		32, 32,	32	\n\t"
+       "xxlxor		33, 33,	33	\n\t"
+       "xxlxor		34, 34,	34	\n\t"
+       "xxlxor		35, 35,	35	\n\t"
+       "xxlxor		36, 36,	36	\n\t"
+       "xxlxor		37, 37,	37	\n\t"
+       "xxlxor		38, 38,	38	\n\t"
+       "xxlxor		39, 39,	39	\n\t"
+
+       "lxvp            40, 0(%2)       \n\t"
+       "lxvp            42, 32(%2)      \n\t"
+       "lxvp            44, 64(%2)      \n\t"
+       "lxvp            46, 96(%2)      \n\t"
+       "lxvp            48, 0(%3)       \n\t"
+       "lxvp            50, 32(%3)      \n\t"
+       "lxvp            52, 64(%3)      \n\t"
+       "lxvp            54, 96(%3)      \n\t"
+
+       "xxperm          56, 48, %x7     \n\t"
+       "xxperm          57, 49, %x7     \n\t"
+       "xxperm          58, 50, %x7     \n\t"
+       "xxperm          59, 51, %x7     \n\t"
+
+       "xxperm          60, 52, %x7     \n\t"
+       "xxperm          61, 53, %x7     \n\t"
+       "xxperm          62, 54, %x7     \n\t"
+       "xxperm          63, 55, %x7     \n\t"
+
+       "addi		%2, %2, 128	\n\t"
+       "addi		%3, %3, 128	\n\t"
+
+       "addic.		%1, %1, -16	\n\t"
+       "ble		two%=		\n\t"
+
+       ".align	5		\n"
+     "one%=:				\n\t"
+
+       "xvmaddasp	32, 40, 48	\n\t"	// x0_r * y0_r , x0_i * y0_i
+       "xvmaddasp	34, 41, 49	\n\t"	// x1_r * y1_r , x1_i * y1_i
+       "lxvp            48, 0(%3)       \n\t"
+
+       "xvmaddasp	36, 42, 50	\n\t"	// x2_r * y2_r , x2_i * y2_i
+       "xvmaddasp	38, 43, 51	\n\t"	// x3_r * y3_r , x3_i * y3_i
+       "lxvp            50, 32(%3)      \n\t"
+
+       "xvmaddasp	33, 40, 56 	\n\t"	// x0_r * y0_i , x0_i * y0_r
+       "xvmaddasp	35, 41, 57 	\n\t"	// x1_r * y1_i , x1_i * y1_r
+       "lxvp            40, 0(%2)       \n\t"
+
+       "xvmaddasp	37, 42, 58	\n\t"	// x2_r * y2_i , x2_i * y2_r
+       "xvmaddasp	39, 43, 59	\n\t"	// x3_r * y3_i , x3_i * y3_r
+       "lxvp            42, 32(%2)      \n\t"
+
+       "xxperm          56, 48, %x7     \n\t"
+       "xxperm          57, 49, %x7     \n\t"
+       "xxperm          58, 50, %x7     \n\t"
+       "xxperm          59, 51, %x7     \n\t"
+
+       "xvmaddasp	32, 44, 52	\n\t"	// x0_r * y0_r , x0_i * y0_i
+       "xvmaddasp	34, 45, 53	\n\t"	// x1_r * y1_r , x1_i * y1_i
+       "lxvp            52, 64(%3)      \n\t"
+
+       "xvmaddasp	36, 46, 54 	\n\t"	// x2_r * y2_r , x2_i * y2_i
+       "xvmaddasp	38, 47, 55	\n\t"	// x3_r * y3_r , x3_i * y3_i
+       "lxvp            54, 96(%3)      \n\t"
+
+       "xvmaddasp	33, 44, 60	\n\t"	// x0_r * y0_i , x0_i * y0_r
+       "xvmaddasp	35, 45, 61	\n\t"	// x1_r * y1_i , x1_i * y1_r
+       "lxvp            44, 64(%2)      \n\t"
+       "xvmaddasp	37, 46, 62	\n\t"	// x2_r * y2_i , x2_i * y2_r
+       "xvmaddasp	39, 47, 63	\n\t"	// x3_r * y3_i , x3_i * y3_r
+       "lxvp            46, 96(%2)      \n\t"
+
+       "xxperm          60, 52, %x7     \n\t"
+       "xxperm          61, 53, %x7     \n\t"
+       "xxperm          62, 54, %x7     \n\t"
+       "xxperm          63, 55, %x7     \n\t"
+
+       "addi		%2, %2, 128 	\n\t"
+       "addi		%3, %3, 128	\n\t"
+
+       "addic.		%1, %1, -16	\n\t"
+       "bgt		one%=		\n"
+
+     "two%=:				\n\t"
+
+       "xvmaddasp	32, 40, 48	\n\t"	// x0_r * y0_r , x0_i * y0_i
+       "xvmaddasp	34, 41, 49	\n\t"	// x1_r * y1_r , x1_i * y1_i
+       "xvmaddasp	36, 42, 50	\n\t"	// x2_r * y2_r , x2_i * y2_i
+       "xvmaddasp	38, 43, 51	\n\t"	// x3_r * y3_r , x3_i * y3_i
+
+       "xvmaddasp	33, 40, 56	\n\t"	// x0_r * y0_i , x0_i * y0_r
+       "xvmaddasp	35, 41, 57	\n\t"	// x1_r * y1_i , x1_i * y1_r
+       "xvmaddasp	37, 42, 58	\n\t"	// x2_r * y2_i , x2_i * y2_r
+       "xvmaddasp	39, 43, 59	\n\t"	// x3_r * y3_i , x3_i * y3_r
+
+       "xvmaddasp	32, 44, 52	\n\t"	// x0_r * y0_r , x0_i * y0_i
+       "xvmaddasp	34, 45, 53	\n\t"	// x1_r * y1_r , x1_i * y1_i
+       "xvmaddasp	36, 46, 54	\n\t"	// x2_r * y2_r , x2_i * y2_i
+       "xvmaddasp	38, 47, 55	\n\t"	// x3_r * y3_r , x3_i * y3_i
+
+       "xvmaddasp	33, 44, 60	\n\t"	// x0_r * y0_i , x0_i * y0_r
+       "xvmaddasp	35, 45, 61	\n\t"	// x1_r * y1_i , x1_i * y1_r
+       "xvmaddasp	37, 46, 62	\n\t"	// x2_r * y2_i , x2_i * y2_r
+       "xvmaddasp	39, 47, 63	\n\t"	// x3_r * y3_i , x3_i * y3_r
+
+       "xvaddsp		32, 32, 34	\n\t"
+       "xvaddsp		36, 36, 38	\n\t"
+
+       "xvaddsp		33, 33, 35	\n\t"
+       "xvaddsp		37, 37, 39	\n\t"
+
+       "xvaddsp		35, 32, 36	\n\t"
+       "xvaddsp		34, 33, 37	\n\t"
+       "xxswapd		32, 35		\n\t"
+       "xxswapd		33, 34		\n\t"
+       "xvaddsp		35, 35, 32	\n\t"
+       "xvaddsp		34, 34, 33	\n\t"
+       "xxpermdi	34, 34, 35, 2	\n\t"
+       "stxv		34, 0(%6)       \n\t"
+
+     "#n=%1 x=%4=%2 y=%5=%3 dot=%0=%6"
+     :
+       "=m" (*dot),
+       "+r" (n),	// 1
+       "+b" (x),	// 2
+       "+b" (y)		// 3
+     :
+       "m" (*x),
+       "m" (*y),
+       "b" (dot),	// 6
+       "wa" (mask)
+     :
+       "cr0",
+       "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
+       "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47",
+       "vs48","vs49","vs50","vs51","vs52","vs53","vs54","vs55",
+       "vs56","vs57","vs58","vs59","vs60","vs61","vs62","vs63"
+     );
+}

kernel/power/copy_microk_power10.c

@@ -62,38 +62,39 @@ static void copy_kernel (BLASLONG n, FLOAT *x, FLOAT *y)
      "one%=:				\n\t"
 
        "stxvp		32, 0(%3)	\n\t"
-       "lxvp		32, 0(%2)	\n\t"
        "stxvp		34, 32(%3)	\n\t"
-       "lxvp		34, 32(%2)	\n\t"
        "stxvp		36, 64(%3)	\n\t"
-       "lxvp		36, 64(%2)	\n\t"
        "stxvp		38, 96(%3)	\n\t"
+       "lxvp		32, 0(%2)	\n\t"
+       "lxvp		34, 32(%2)	\n\t"
+       "lxvp		36, 64(%2)	\n\t"
        "lxvp		38, 96(%2)	\n\t"
 
        "stxvp		40, 128(%3)	\n\t"
-       "lxvp		40, 128(%2)	\n\t"
        "stxvp		42, 160(%3)	\n\t"
-       "lxvp		42, 160(%2)	\n\t"
        "stxvp		44, 192(%3)	\n\t"
-       "lxvp		44, 192(%2)	\n\t"
        "stxvp		46, 224(%3)	\n\t"
+       "lxvp		40, 128(%2)	\n\t"
+       "lxvp		42, 160(%2)	\n\t"
+       "lxvp		44, 192(%2)	\n\t"
        "lxvp		46, 224(%2)	\n\t"
 
        "stxvp		48, 256(%3)	\n\t"
-       "lxvp		48, 256(%2)	\n\t"
        "stxvp		50, 288(%3)	\n\t"
-       "lxvp		50, 288(%2)	\n\t"
        "stxvp		52, 320(%3)	\n\t"
-       "lxvp		52, 320(%2)	\n\t"
        "stxvp		54, 352(%3)	\n\t"
+       "lxvp		48, 256(%2)	\n\t"
+       "lxvp		50, 288(%2)	\n\t"
+       "lxvp		52, 320(%2)	\n\t"
        "lxvp		54, 352(%2)	\n\t"
+
        "stxvp		56, 384(%3)	\n\t"
-       "lxvp		56, 384(%2)	\n\t"
        "stxvp		58, 416(%3)	\n\t"
-       "lxvp		58, 416(%2)	\n\t"
        "stxvp		60, 448(%3)	\n\t"
-       "lxvp		60, 448(%2)	\n\t"
        "stxvp		62, 480(%3)	\n\t"
+       "lxvp		56, 384(%2)	\n\t"
+       "lxvp		58, 416(%2)	\n\t"
+       "lxvp		60, 448(%2)	\n\t"
        "lxvp		62, 480(%2)	\n\t"
 
        "addi		%3, %3, 512	\n\t"

kernel/power/cscal_microk_power10.c

@@ -0,0 +1,176 @@
+/***************************************************************************
+Copyright (c) 2021, 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 HAVE_KERNEL_8 1
+
+static void zscal_kernel_8 (long n, float *x, float alpha_r, float alpha_i)
+{
+  __vector float t0 = {-alpha_i, alpha_i, -alpha_i, alpha_i};
+  __vector unsigned char mask = { 11,10,9,8,15,14,13,12,3,2,1,0,7,6,5,4};
+  __asm__
+    (
+       "dcbt		0, %2		\n\t"
+       "xscvdpspn	32, %x3    \n\t"
+       "xxspltw		32, 32, 0	\n\t"
+
+       "lxvp		40, 0(%2)	\n\t"
+       "lxvp		42, 32(%2)	\n\t"
+       "lxvp		44, 64(%2)	\n\t"
+       "lxvp		46, 96(%2)	\n\t"
+
+       "addic.		%1, %1, -16	\n\t"
+       "ble		two%=		\n\t"
+
+       ".align	5		\n"
+     "one%=:				\n\t"
+
+       "xvmulsp		48, 40, 32	\n\t"	// x0_r * alpha_r, x0_i * alpha_r
+       "xvmulsp		49, 41, 32	\n\t"
+       "xvmulsp		50, 42, 32	\n\t"
+       "xvmulsp		51, 43, 32	\n\t"
+       "xvmulsp		52, 44, 32	\n\t"
+       "xvmulsp		53, 45, 32	\n\t"
+       "xvmulsp		54, 46, 32	\n\t"
+       "xvmulsp		55, 47, 32	\n\t"
+
+       "xxperm 34, 40, %x5 \n\t"
+       "xxperm 35, 41, %x5 \n\t"
+       "xxperm 36, 42, %x5 \n\t"
+       "xxperm 37, 43, %x5 \n\t"
+       "xxperm 38, 44, %x5 \n\t"
+       "xxperm 39, 45, %x5 \n\t"
+       "xxperm 56, 46, %x5 \n\t"
+       "xxperm 57, 47, %x5 \n\t"
+
+       "xvmulsp		34, 34, %x4	\n\t"	// x0_i * -alpha_i, x0_r * alpha_i
+       "xvmulsp		35, 35, %x4	\n\t"
+
+       "lxvp		40, 128(%2)	\n\t"
+
+       "xvmulsp		36, 36, %x4	\n\t"
+       "xvmulsp		37, 37, %x4	\n\t"
+
+       "lxvp		42, 160(%2)	\n\t"
+
+       "xvmulsp		38, 38, %x4	\n\t"
+       "xvmulsp		39, 39, %x4	\n\t"
+
+       "lxvp		44, 192(%2)	\n\t"
+
+       "xvmulsp		56, 56, %x4	\n\t"
+       "xvmulsp		57, 57, %x4	\n\t"
+
+       "lxvp		46, 224(%2)	\n\t"
+
+       "xvaddsp		48, 48, 34	\n\t"
+       "xvaddsp		49, 49, 35	\n\t"
+       "xvaddsp		50, 50, 36	\n\t"
+       "xvaddsp		51, 51, 37	\n\t"
+
+       "stxvp		48, 0(%2)	\n\t"
+
+       "xvaddsp		52, 52, 38	\n\t"
+       "xvaddsp		53, 53, 39	\n\t"
+
+       "stxvp		50, 32(%2)	\n\t"
+
+       "xvaddsp		54, 54, 56	\n\t"
+       "xvaddsp		55, 55, 57	\n\t"
+
+       "stxvp		52, 64(%2)	\n\t"
+       "stxvp		54, 96(%2)	\n\t"
+
+       "addi		%2, %2, 128	\n\t"
+
+       "addic.		%1, %1, -16	\n\t"
+       "bgt		one%=		\n"
+
+     "two%=:				\n\t"
+
+       "xvmulsp		48, 40, 32	\n\t"	// x0_r * alpha_r, x0_i * alpha_r
+       "xvmulsp		49, 41, 32	\n\t"
+       "xvmulsp		50, 42, 32	\n\t"
+       "xvmulsp		51, 43, 32	\n\t"
+       "xvmulsp		52, 44, 32	\n\t"
+       "xvmulsp		53, 45, 32	\n\t"
+       "xvmulsp		54, 46, 32	\n\t"
+       "xvmulsp		55, 47, 32	\n\t"
+
+       "xxperm 34, 40, %x5 \n\t"
+       "xxperm 35, 41, %x5 \n\t"
+       "xxperm 36, 42, %x5 \n\t"
+       "xxperm 37, 43, %x5 \n\t"
+       "xxperm 38, 44, %x5 \n\t"
+       "xxperm 39, 45, %x5 \n\t"
+       "xxperm 56, 46, %x5 \n\t"
+       "xxperm 57, 47, %x5 \n\t"
+
+
+       "xvmulsp		34, 34, %x4	\n\t"	// x0_i * -alpha_i, x0_r * alpha_i
+       "xvmulsp		35, 35, %x4	\n\t"
+       "xvmulsp		36, 36, %x4	\n\t"
+       "xvmulsp		37, 37, %x4	\n\t"
+       "xvmulsp		38, 38, %x4	\n\t"
+       "xvmulsp		39, 39, %x4	\n\t"
+       "xvmulsp		56, 56, %x4	\n\t"
+       "xvmulsp		57, 57, %x4	\n\t"
+
+       "xvaddsp		48, 48, 34	\n\t"
+       "xvaddsp		49, 49, 35	\n\t"
+       "xvaddsp		50, 50, 36	\n\t"
+       "xvaddsp		51, 51, 37	\n\t"
+
+       "stxvp		48, 0(%2)	\n\t"
+
+       "xvaddsp		52, 52, 38	\n\t"
+       "xvaddsp		53, 53, 39	\n\t"
+
+       "stxvp		50, 32(%2)	\n\t"
+
+       "xvaddsp		54, 54, 56	\n\t"
+       "xvaddsp		55, 55, 57	\n\t"
+
+       "stxvp		52, 64(%2)	\n\t"
+       "stxvp		54, 96(%2)	\n\t"
+
+     "#n=%1 x=%0=%2 alpha=(%3,%4)\n"
+     :
+       "+m" (*x),
+       "+r" (n),	// 1
+       "+b" (x)		// 2
+     :
+       "f" (alpha_r),	// 3 
+       "wa" (t0),	// 4 
+       "wa" (mask)	// 5
+     :
+       "cr0",
+       "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
+       "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47",
+       "vs48","vs49","vs50","vs51","vs52","vs53","vs54","vs55",
+       "vs56","vs57"
+     );
+}

kernel/power/cswap.c

@@ -36,9 +36,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(POWER8)  || defined(POWER9) || defined(POWER10)
 #if defined(__VEC__) || defined(__ALTIVEC__)
+#if defined(POWER8)  || defined(POWER9)
 #include "cswap_microk_power8.c"
+#elif defined(POWER10)
+#include "cswap_microk_power10.c"
 #endif
 #endif
 

kernel/power/cswap_microk_power10.c

@@ -0,0 +1,127 @@
+/***************************************************************************
+Copyright (c) 2021, 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.
+*****************************************************************************/
+
+#if defined(DOUBLE)
+#define HAVE_KERNEL_16 1
+static void zswap_kernel_16 (long n, double *x, double *y)
+#else
+#define HAVE_KERNEL_32 1
+static void cswap_kernel_32 (long n, float *x, float *y)
+#endif
+{
+  __asm__
+    (
+       ".align	5		\n"
+     "one%=:				\n\t"
+       "lxvp            32, 0(%4)       \n\t"
+       "lxvp            34, 32(%4)      \n\t"
+       "lxvp            36, 64(%4)      \n\t"
+       "lxvp            38, 96(%4)      \n\t"
+
+       "lxvp            40, 128(%4)     \n\t"
+       "lxvp            42, 160(%4)     \n\t"
+       "lxvp            44, 192(%4)     \n\t"
+       "lxvp            46, 224(%4)     \n\t"
+
+       "lxvp            48, 0(%3)       \n\t"
+       "lxvp            50, 32(%3)      \n\t"
+       "lxvp            52, 64(%3)      \n\t"
+       "lxvp            54, 96(%3)      \n\t"
+
+       "lxvp            56, 128(%3)     \n\t"
+       "lxvp            58, 160(%3)     \n\t"
+       "lxvp            60, 192(%3)     \n\t"
+       "lxvp            62, 224(%3)     \n\t"
+
+
+       "stxv		33, 0(%3)	\n\t"
+       "stxv		32, 16(%3)	\n\t"
+       "stxv		35, 32(%3)	\n\t"
+       "stxv		34, 48(%3)	\n\t"
+       "stxv		37, 64(%3)	\n\t"
+       "stxv		36, 80(%3)	\n\t"
+       "stxv		39, 96(%3)	\n\t"
+       "stxv		38, 112(%3)	\n\t"
+
+       "addi		%3, %3, 128	\n\t"
+
+       "stxv		41, 0(%3)	\n\t"
+       "stxv		40, 16(%3)	\n\t"
+       "stxv		43, 32(%3)	\n\t"
+       "stxv		42, 48(%3)	\n\t"
+       "stxv		45, 64(%3)	\n\t"
+       "stxv		44, 80(%3)	\n\t"
+       "stxv		47, 96(%3)	\n\t"
+       "stxv		46, 112(%3)	\n\t"
+
+       "addi		%3, %3, 128	\n\t"
+
+       "stxv		49, 0(%4)	\n\t"
+       "stxv		48, 16(%4)	\n\t"
+       "stxv		51, 32(%4)	\n\t"
+       "stxv		50, 48(%4)	\n\t"
+       "stxv		53, 64(%4)	\n\t"
+       "stxv		52, 80(%4)	\n\t"
+       "stxv		55, 96(%4)	\n\t"
+       "stxv		54, 112(%4)	\n\t"
+
+       "addi		%4, %4, 128	\n\t"
+
+       "stxv		57, 0(%4)	\n\t"
+       "stxv		56, 16(%4)	\n\t"
+       "stxv		59, 32(%4)	\n\t"
+       "stxv		58, 48(%4)	\n\t"
+       "stxv		61, 64(%4)	\n\t"
+       "stxv		60, 80(%4)	\n\t"
+       "stxv		63, 96(%4)	\n\t"
+       "stxv		62, 112(%4)	\n\t"
+
+       "addi		%4, %4, 128	\n\t"
+
+#if defined(DOUBLE)
+       "addic.		%2, %2, -16	\n\t"
+#else
+       "addic.		%2, %2, -32	\n\t"
+#endif
+       "bgt		one%=		\n"
+
+     "#n=%2 x=%0=%3 y=%1=%4"
+     :
+       "+m" (*x),
+       "+m" (*y),
+       "+r" (n),	// 2
+       "+b" (x),	// 3
+       "+b" (y)		// 4
+     :
+     :
+       "cr0",
+       "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
+       "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47",
+       "vs48","vs49","vs50","vs51","vs52","vs53","vs54","vs55",
+       "vs56","vs57","vs58","vs59","vs60","vs61","vs62","vs63"
+     );
+}

kernel/power/dasum.c

@@ -46,9 +46,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #if defined(__VEC__) || defined(__ALTIVEC__)
+#if defined(POWER8) || defined(POWER9)
 #include "dasum_microk_power8.c"
+#elif defined(POWER10)
+#include "dasum_microk_power10.c"
 #endif
 #endif
 
@@ -110,6 +112,21 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	if ( inc_x == 1 )
 	{
 
+#if defined(POWER10)
+		if ( n >= 16 )
+		{
+			BLASLONG align = ((32 - ((uintptr_t)x & (uintptr_t)0x1F)) >> 3) & 0x3;
+			for (i = 0; i < align; i++) {
+				sumf += ABS(x[i]);
+			}
+		}
+		n1 = (n-i) & -16;
+		if ( n1 > 0 )
+		{
+			sumf += dasum_kernel_16(n1, &x[i]);
+			i+=n1;
+		}
+#else
 		n1 = n & -16;
 		if ( n1 > 0 )
 		{
@@ -117,6 +134,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 			sumf = dasum_kernel_16(n1, x);
 			i=n1;
 		}
+#endif
 
 		while(i < n)
 		{

kernel/power/dasum_microk_power10.c

@@ -0,0 +1,152 @@
+/***************************************************************************
+Copyright (c) 2021, 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 HAVE_KERNEL_16 1
+
+static double dasum_kernel_16 (long n, double *x)
+{
+  double sum;
+  __vector double t0;
+  __vector double t1;
+  __vector double t2;
+  __vector double t3;
+
+  __asm__
+    (
+       "dcbt		0, %2		\n\t"
+
+       "xxlxor		32, 32,	32	\n\t"
+       "xxlxor		33, 33,	33	\n\t"
+       "xxlxor		34, 34,	34	\n\t"
+       "xxlxor		35, 35,	35	\n\t"
+       "xxlxor		36, 36,	36	\n\t"
+       "xxlxor		37, 37,	37	\n\t"
+       "xxlxor		38, 38,	38	\n\t"
+       "xxlxor		39, 39,	39	\n\t"
+
+       "lxvp            40, 0(%2)       \n\t"
+       "lxvp            42, 32(%2)      \n\t"
+       "lxvp            44, 64(%2)      \n\t"
+       "lxvp            46, 96(%2)      \n\t"
+
+       "addi		%2, %2, 128	\n\t"
+
+       "addic.		%1, %1, -16	\n\t"
+       "ble		two%=		\n\t"
+
+       ".align	5		\n"
+     "one%=:				\n\t"
+
+       "xvabsdp		48, 40		\n\t"
+       "xvabsdp		49, 41		\n\t"
+       "xvabsdp		50, 42		\n\t"
+       "xvabsdp		51, 43		\n\t"
+       "lxvp            40, 0(%2)       \n\t"
+
+
+       "xvabsdp		%x3, 44		\n\t"
+       "xvabsdp		%x4, 45		\n\t"
+       "lxvp            42, 32(%2)      \n\t"
+
+
+       "xvabsdp		%x5, 46		\n\t"
+       "xvabsdp		%x6, 47		\n\t"
+       "lxvp            44, 64(%2)      \n\t"
+
+
+       "xvadddp		32, 32, 48	\n\t"
+       "xvadddp		33, 33, 49	\n\t"
+
+       "lxvp            46, 96(%2)      \n\t"
+
+       "xvadddp		34, 34, 50	\n\t"
+       "xvadddp		35, 35, 51	\n\t"
+       "addi		%2, %2, 128	\n\t"
+       "xvadddp		36, 36, %x3	\n\t"
+       "xvadddp		37, 37, %x4	\n\t"
+       "addic.		%1, %1, -16	\n\t"
+       "xvadddp		38, 38, %x5	\n\t"
+       "xvadddp		39, 39, %x6	\n\t"
+
+       "bgt		one%=		\n"
+
+     "two%=:				\n\t"
+
+       "xvabsdp		48, 40		\n\t"
+       "xvabsdp		49, 41		\n\t"
+       "xvabsdp		50, 42		\n\t"
+       "xvabsdp		51, 43		\n\t"
+       "xvabsdp		%x3, 44		\n\t"
+       "xvabsdp		%x4, 45		\n\t"
+       "xvabsdp		%x5, 46		\n\t"
+       "xvabsdp		%x6, 47		\n\t"
+
+       "xvadddp		32, 32, 48	\n\t"
+       "xvadddp		33, 33, 49	\n\t"
+       "xvadddp		34, 34, 50	\n\t"
+       "xvadddp		35, 35, 51	\n\t"
+       "xvadddp		36, 36, %x3	\n\t"
+       "xvadddp		37, 37, %x4	\n\t"
+       "xvadddp		38, 38, %x5	\n\t"
+       "xvadddp		39, 39, %x6	\n\t"
+
+       "xvadddp		32, 32, 33	\n\t"
+       "xvadddp		34, 34, 35	\n\t"
+       "xvadddp		36, 36, 37	\n\t"
+       "xvadddp		38, 38, 39	\n\t"
+
+       "xvadddp		32, 32, 34	\n\t"
+       "xvadddp		36, 36, 38	\n\t"
+
+       "xvadddp		32, 32, 36	\n\t"
+
+       XXSWAPD_S(33,32)
+       "xsadddp		%x0, 32, 33	\n"
+
+     "#n=%1 x=%3=%2 sum=%0\n"
+     "#t0=%x3 t1=%x4 t2=%x5 t3=%x6"
+     :
+       "=d" (sum),	// 0
+       "+r" (n),	// 1
+       "+b" (x),	// 2
+       "=wa" (t0),	// 3
+       "=wa" (t1),	// 4
+       "=wa" (t2),	// 5
+       "=wa" (t3)	// 6
+     :
+       "m" (*x)
+     :
+       "cr0",
+       "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
+       "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47",
+       "vs48","vs49","vs50","vs51"
+     );
+
+  return sum;
+}
+
+

kernel/power/daxpy_power10.c

@@ -66,12 +66,19 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 	if ( (inc_x == 1) && (inc_y == 1) )
 	{
 
-		BLASLONG n1 = n & -16;
+                if ( n >= 16 )
+                {
+                       BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 3) & 0x3;
+                        for (i = 0; i < align; i++) {
+                          y[i] += da * x[i] ;
+                        }
+                }
+                BLASLONG n1 = (n-i) & -16;
+                if ( n1 )
+                      daxpy_kernel_8(n1, &x[i], &y[i], da);
 
-		if ( n1 )
-			daxpy_kernel_8(n1, x, y, da);
+                i += n1;
 
-		i = n1;
 		while(i < n)
 		{
 

kernel/power/dcopy_power10.c

@@ -85,12 +85,18 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
 
 	if ( (inc_x == 1) && (inc_y == 1 ))
 	{
-
-		BLASLONG n1 = n & -64;
-		if ( n1 > 0 )
+		if ( n >= 64 )
 		{
-			copy_kernel(n1, x, y);
-			i=n1;
+			BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 3) & 0x3;
+			for (i = 0; i < align; i++) {
+				y[i] = x[i] ;
+			}
+		}
+		BLASLONG n1 = (n-i) & -64;
+		if ( n1 )
+		{
+			copy_kernel(n1, &x[i], &y[i]);
+			i += n1;
 		}
 
 		while(i < n)

kernel/power/dgemm_kernel_power10.c

@@ -29,7 +29,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 typedef __vector unsigned char  vec_t;
 typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
-typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
+#if !__has_builtin(__builtin_vsx_assemble_pair)
+#define __builtin_vsx_assemble_pair __builtin_mma_assemble_pair
+#endif
+
+#if !__has_builtin(__builtin_vsx_disassemble_pair)
+#define __builtin_vsx_disassemble_pair __builtin_mma_disassemble_pair
+#endif
 
 #ifdef TRMMKERNEL
 #define SAVE_ACC(ACC, J)  \
@@ -186,8 +192,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  vec_t *rb = (vec_t *) & BO[0];
 	  __vector_pair rowB, rowB1;
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	  __builtin_mma_assemble_pair (&rowB1, rb[3], rb[2]);
+	  __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_vsx_assemble_pair (&rowB1, rb[3], rb[2]);
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB1, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[1]);
@@ -200,8 +206,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	    {
 	      rowA = (vec_t *) & AO[l << 3];
 	      rb = (vec_t *) & BO[l << 3];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      __builtin_mma_assemble_pair (&rowB1, rb[3], rb[2]);
+	      __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_vsx_assemble_pair (&rowB1, rb[3], rb[2]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB1, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[1]);
@@ -242,8 +248,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __vector_pair rowB, rowB1;
 	  vec_t *rb = (vec_t *) & BO[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	  __builtin_mma_assemble_pair (&rowB1, rb[3], rb[2]);
+	  __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_vsx_assemble_pair (&rowB1, rb[3], rb[2]);
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB1, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[1]);
@@ -252,8 +258,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	    {
 	      rowA = (vec_t *) & AO[l << 2];
 	      rb = (vec_t *) & BO[l << 3];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      __builtin_mma_assemble_pair (&rowB1, rb[3], rb[2]);
+	      __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_vsx_assemble_pair (&rowB1, rb[3], rb[2]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB1, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc2, rowB, rowA[1]);
@@ -286,16 +292,16 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __vector_pair rowB, rowB1;
 	  vec_t *rb = (vec_t *) & BO[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	  __builtin_mma_assemble_pair (&rowB1, rb[3], rb[2]);
+	  __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_vsx_assemble_pair (&rowB1, rb[3], rb[2]);
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB1, rowA[0]);
 	  for (l = 1; l < temp; l++)
 	    {
 	      rowA = (vec_t *) & AO[l << 1];
 	      rb = (vec_t *) & BO[l << 3];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
-	      __builtin_mma_assemble_pair (&rowB1, rb[3], rb[2]);
+	      __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_vsx_assemble_pair (&rowB1, rb[3], rb[2]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB1, rowA[0]);
 	    }
@@ -398,7 +404,7 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __vector_pair rowB;
 	  vec_t *rb = (vec_t *) & BO[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
 	  __builtin_mma_xvf64ger (&acc2, rowB, rowA[2]);
@@ -407,7 +413,7 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	    {
 	      rowA = (vec_t *) & AO[l << 3];
 	      rb = (vec_t *) & BO[l << 2];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_vsx_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]);
@@ -440,14 +446,14 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __vector_pair rowB;
 	  vec_t *rb = (vec_t *) & BO[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
 	  for (l = 1; l < temp; l++)
 	    {
 	      rowA = (vec_t *) & AO[l << 2];
 	      rb = (vec_t *) & BO[l << 2];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
 	    }
@@ -476,13 +482,13 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __vector_pair rowB;
 	  vec_t *rb = (vec_t *) & BO[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  for (l = 1; l < temp; l++)
 	    {
 	      rowA = (vec_t *) & AO[l << 1];
 	      rb = (vec_t *) & BO[l << 2];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      __builtin_vsx_assemble_pair (&rowB, rb[1], rb[0]);
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	    }
 	  SAVE_ACC (&acc0, 0);
@@ -562,11 +568,9 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1, acc2, acc3;
 	  BLASLONG l = 0;
-	  FLOAT t[4] = { 0, 0, 0, 0 };
-	  t[0] = BO[0], t[1] = BO[1];
 	  __vector_pair rowB;
-	  vec_t *rb = (vec_t *) & t[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_vsx_assemble_pair (&rowB, rb[0], rb[0]);
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
@@ -574,9 +578,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  __builtin_mma_xvf64ger (&acc3, rowB, rowA[3]);
 	  for (l = 1; l < temp; l++)
 	    {
-	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      rb = (vec_t *) & t[0];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      rb = (vec_t *) & BO[l << 1];
+	      __builtin_vsx_assemble_pair (&rowB, rb[0], rb[0]);
 	      rowA = (vec_t *) & AO[l << 3];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
@@ -607,19 +610,16 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t result[4];
 	  __vector_quad acc0, acc1;
 	  BLASLONG l = 0;
-	  FLOAT t[4] = { 0, 0, 0, 0 };
-	  t[0] = BO[0], t[1] = BO[1];
 	  __vector_pair rowB;
-	  vec_t *rb = (vec_t *) & t[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_vsx_assemble_pair (&rowB, rb[0], rb[0]);
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  __builtin_mma_xvf64ger (&acc1, rowB, rowA[1]);
 	  for (l = 1; l < temp; l++)
 	    {
-	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      rb = (vec_t *) & t[0];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      rb = (vec_t *) & BO[l << 1];
+	      __builtin_vsx_assemble_pair (&rowB, rb[0], rb[0]);
 	      rowA = (vec_t *) & AO[l << 2];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	      __builtin_mma_xvf64gerpp (&acc1, rowB, rowA[1]);
@@ -646,18 +646,15 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT * A, FLOAT * B,
 	  v4sf_t result[4];
 	  __vector_quad acc0;
 	  BLASLONG l = 0;
-	  FLOAT t[4] = { 0, 0, 0, 0 };
-	  t[0] = BO[0], t[1] = BO[1];
 	  __vector_pair rowB;
-	  vec_t *rb = (vec_t *) & t[0];
-	  __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	  vec_t *rb = (vec_t *) & BO[0];
+	  __builtin_vsx_assemble_pair (&rowB, rb[0], rb[0]);
 	  vec_t *rowA = (vec_t *) & AO[0];
 	  __builtin_mma_xvf64ger (&acc0, rowB, rowA[0]);
 	  for (l = 1; l < temp; l++)
 	    {
-	      t[0] = BO[l << 1], t[1] = BO[(l << 1) + 1];
-	      rb = (vec_t *) & t[0];
-	      __builtin_mma_assemble_pair (&rowB, rb[1], rb[0]);
+	      rb = (vec_t *) & BO[l << 1];
+	      __builtin_vsx_assemble_pair (&rowB, rb[0], rb[0]);
 	      rowA = (vec_t *) & AO[l << 1];
 	      __builtin_mma_xvf64gerpp (&acc0, rowB, rowA[0]);
 	    }

kernel/power/drot.c

@@ -39,9 +39,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #pragma GCC optimize "O1"
 
-#if defined(POWER8) || defined(POWER9) || defined(POWER10)
 #if defined(__VEC__) || defined(__ALTIVEC__)
+#if defined(POWER8) || defined(POWER9)
 #include "drot_microk_power8.c"
+#elif defined(POWER10)
+#include "drot_microk_power10.c"
 #endif
 #endif
 
@@ -115,12 +117,30 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
 	if ( (inc_x == 1) && (inc_y == 1) )
 	{
 
+#if defined(POWER10)
+		if ( n >= 16 )
+		{
+			BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 3) & 0x3;
+			for (i = 0; i < align; i++) {
+				temp  = c*x[i] + s*y[i] ;
+				y[i]  = c*y[i] - s*x[i] ;
+				x[i]  = temp ;
+			}
+		}
+		BLASLONG n1 = (n-i) & -16;
+		if ( n1 > 0 )
+		{
+			drot_kernel_16(n1,&x[i], &y[i], c, s);
+			i+=n1;
+		}
+#else
 		BLASLONG n1 = n & -16;
 		if ( n1 > 0 )
 		{
 			drot_kernel_16(n1, x1, y1, c, s);
 			i=n1;
 		}
+#endif
 
 		while(i < n)
 		{

kernel/power/drot_microk_power10.c

@@ -0,0 +1,148 @@
+/***************************************************************************
+Copyright (c) 2021, 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 HAVE_KERNEL_16 1
+
+static void drot_kernel_16 (long n, double *x, double *y, double c, double s)
+{
+  __asm__
+    (
+       XXSPLTD_S(36,%x5,0)	// load c to both dwords
+       XXSPLTD_S(37,%x6,0)	// load s to both dwords
+       "lxvp            32, 0(%3)       \n\t"   // load x
+       "lxvp            34, 32(%3)      \n\t"
+       "lxvp            48, 0(%4)       \n\t"   // load y
+       "lxvp            50, 32(%4)      \n\t"
+
+       "addic.		%2, %2, -8	\n\t"
+       "ble		two%=		\n\t"
+
+       ".align	5		\n"
+     "one%=:				\n\t"
+
+       "xvmuldp		40, 32, 36	\n\t"	// c * x
+       "xvmuldp		41, 33, 36	\n\t"
+       "xvmuldp		42, 34, 36	\n\t"
+       "xvmuldp		43, 35, 36	\n\t"
+
+       "xvmuldp		44, 32, 37	\n\t"	// s * x
+       "xvmuldp		45, 33, 37	\n\t"
+       "xvmuldp		46, 34, 37	\n\t"
+       "xvmuldp		47, 35, 37	\n\t"
+
+       "lxvp            32, 64(%3)       \n\t"   // load x
+       "lxvp            34, 96(%3)      \n\t"
+       "xvmuldp		52, 48, 36	\n\t"	// c * y
+       "xvmuldp		53, 49, 36	\n\t"
+       "xvmuldp		54, 50, 36	\n\t"
+       "xvmuldp		55, 51, 36	\n\t"
+
+       "xvmuldp		38, 48, 37	\n\t"	// s * y
+       "xvmuldp		39, 49, 37	\n\t"
+       "xvmuldp		56, 50, 37	\n\t"
+       "xvmuldp		57, 51, 37	\n\t"
+
+       "lxvp            48, 64(%4)       \n\t"   // load y
+       "lxvp            50, 96(%4)      \n\t"
+
+       "xvadddp		40, 40, 38	\n\t"	// c * x + s * y
+       "xvadddp		41, 41, 39	\n\t"	// c * x + s * y
+       "xvadddp		42, 42, 56	\n\t"	// c * x + s * y
+       "xvadddp		43, 43, 57	\n\t"	// c * x + s * y
+
+       "stxvp           40, 0(%3)       \n\t"   // store x
+       "stxvp           42, 32(%3)      \n\t"
+
+       "xvsubdp         52, 52, 44      \n\t"   // c * y - s * x
+       "xvsubdp         53, 53, 45      \n\t"   // c * y - s * x
+       "xvsubdp         54, 54, 46      \n\t"   // c * y - s * x
+       "xvsubdp         55, 55, 47      \n\t"   // c * y - s * x
+
+       "stxvp           52, 0(%4)       \n\t"   // store y
+       "stxvp           54, 32(%4)      \n\t"
+
+       "addi		%3, %3, 64	\n\t"
+       "addi		%4, %4, 64	\n\t"
+
+       "addic.		%2, %2, -8	\n\t"
+       "bgt		one%=		\n"
+
+     "two%=:				\n\t"
+
+       "xvmuldp		40, 32, 36	\n\t"	// c * x
+       "xvmuldp		41, 33, 36	\n\t"
+       "xvmuldp		42, 34, 36	\n\t"
+       "xvmuldp		43, 35, 36	\n\t"
+
+       "xvmuldp         52, 48, 36      \n\t"   // c * y
+       "xvmuldp         53, 49, 36      \n\t"
+       "xvmuldp         54, 50, 36      \n\t"
+       "xvmuldp         55, 51, 36      \n\t"
+
+       "xvmuldp         44, 32, 37      \n\t"   // s * x
+       "xvmuldp         45, 33, 37      \n\t"
+       "xvmuldp         46, 34, 37      \n\t"
+       "xvmuldp         47, 35, 37      \n\t"
+
+       "xvmuldp         38, 48, 37     \n\t"   // s * y
+       "xvmuldp         39, 49, 37     \n\t"
+       "xvmuldp         56, 50, 37     \n\t"
+       "xvmuldp         57, 51, 37     \n\t"
+
+       "xvadddp         40, 40, 38     \n\t"   // c * x + s * y
+       "xvadddp         41, 41, 39     \n\t"   // c * x + s * y
+       "xvadddp         42, 42, 56     \n\t"   // c * x + s * y
+       "xvadddp         43, 43, 57     \n\t"   // c * x + s * y
+
+       "stxvp           40, 0(%3)       \n\t"   // store x
+       "stxvp           42, 32(%3)      \n\t"
+       "xvsubdp         52, 52, 44      \n\t"   // c * y - s * x
+       "xvsubdp         53, 53, 45      \n\t"   // c * y - s * x
+       "xvsubdp         54, 54, 46      \n\t"   // c * y - s * x
+       "xvsubdp         55, 55, 47      \n\t"   // c * y - s * x
+
+       "stxvp           52, 0(%4)       \n\t"   // store y
+       "stxvp           54, 32(%4)      \n\t"
+
+     "#n=%2 x=%0=%3 y=%1=%4 c=%5 s=%6\n"
+     :
+       "+m" (*x),
+       "+m" (*y),
+       "+r" (n),	// 2
+       "+b" (x),	// 3
+       "+b" (y) 	// 4
+     :
+       "d" (c),		// 5 
+       "d" (s)		// 6 
+     :
+       "cr0",
+       "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
+       "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47",
+       "vs48","vs49","vs50","vs51","vs52","vs53","vs54","vs55",
+       "vs56","vs57"
+     );
+}