Merge pull request #2228 from martin-frbg/issue2227

Add Intel Goldmont Plus CPUID

.drone.yml

@@ -0,0 +1,143 @@
+---
+kind: pipeline
+name: arm64_gcc_make
+
+platform:
+  os: linux
+  arch: arm64
+
+steps:
+- name: Build and Test
+  image: ubuntu:19.04
+  environment:
+    CC: gcc
+    COMMON_FLAGS: 'DYNAMIC_ARCH=1 TARGET=ARMV8 NUM_THREADS=32'
+  commands:
+    - echo "MAKE_FLAGS:= $COMMON_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC gfortran perl
+    - $CC --version
+    - make QUIET_MAKE=1 $COMMON_FLAGS
+    - make -C test $COMMON_FLAGS
+    - make -C ctest $COMMON_FLAGS
+    - make -C utest $COMMON_FLAGS
+
+---
+kind: pipeline
+name: arm32_gcc_make
+
+platform:
+  os: linux
+  arch: arm
+
+steps:
+- name: Build and Test
+  image: ubuntu:19.04
+  environment:
+    CC: gcc
+    COMMON_FLAGS: 'DYNAMIC_ARCH=1 TARGET=ARMV6 NUM_THREADS=32'
+  commands:
+    - echo "MAKE_FLAGS:= $COMMON_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC gfortran perl
+    - $CC --version
+    - make QUIET_MAKE=1 $COMMON_FLAGS
+    - make -C test $COMMON_FLAGS
+    - make -C ctest $COMMON_FLAGS
+    - make -C utest $COMMON_FLAGS
+
+---
+kind: pipeline
+name: arm64_clang_make
+
+platform:
+  os: linux
+  arch: arm64
+
+steps:
+- name: Build and Test
+  image: ubuntu:18.04
+  environment:
+    CC: clang
+    COMMON_FLAGS: 'DYNAMIC_ARCH=1 TARGET=ARMV8 NUM_THREADS=32'
+  commands:
+    - echo "MAKE_FLAGS:= $COMMON_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC gfortran perl
+    - $CC --version
+    - make QUIET_MAKE=1 $COMMON_FLAGS
+    - make -C test $COMMON_FLAGS
+    - make -C ctest $COMMON_FLAGS
+    - make -C utest $COMMON_FLAGS
+
+---
+kind: pipeline
+name: arm32_clang_cmake
+
+platform:
+  os: linux
+  arch: arm
+
+steps:
+- name: Build and Test
+  image: ubuntu:18.04
+  environment:
+    CC: clang
+    CMAKE_FLAGS: '-DDYNAMIC_ARCH=1 -DTARGET=ARMV6 -DNUM_THREADS=32 -DNOFORTRAN=ON -DBUILD_WITHOUT_LAPACK=ON'
+  commands:
+    - echo "CMAKE_FLAGS:= $CMAKE_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC g++ perl cmake
+    - $CC --version
+    - mkdir build && cd build
+    - cmake $CMAKE_FLAGS ..
+    - make -j
+    - ctest
+
+---
+kind: pipeline
+name: arm64_gcc_cmake
+
+platform:
+  os: linux
+  arch: arm64
+
+steps:
+- name: Build and Test
+  image: ubuntu:18.04
+  environment:
+    CC: gcc
+    CMAKE_FLAGS: '-DDYNAMIC_ARCH=1 -DTARGET=ARMV8 -DNUM_THREADS=32 -DNOFORTRAN=ON -DBUILD_WITHOUT_LAPACK=ON'
+  commands:
+    - echo "CMAKE_FLAGS:= $CMAKE_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC g++ perl cmake
+    - $CC --version
+    - mkdir build && cd build
+    - cmake $CMAKE_FLAGS ..
+    - make -j
+    - ctest
+
+---
+kind: pipeline
+name: arm64_clang_cmake
+
+platform:
+  os: linux
+  arch: arm64
+
+steps:
+- name: Build and Test
+  image: ubuntu:18.04
+  environment:
+    CC: clang
+    CMAKE_FLAGS: '-DDYNAMIC_ARCH=1 -DTARGET=ARMV8 -DNUM_THREADS=32 -DNOFORTRAN=ON -DBUILD_WITHOUT_LAPACK=ON'
+  commands:
+    - echo "CMAKE_FLAGS:= $CMAKE_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC g++ perl cmake
+    - $CC --version
+    - mkdir build && cd build
+    - cmake $CMAKE_FLAGS ..
+    - make -j
+    - ctest

.travis.yml

@@ -25,6 +25,15 @@ matrix:
         - TARGET_BOX=LINUX64
         - BTYPE="BINARY=64"
 
+    - <<: *test-ubuntu
+      os: linux-ppc64le
+      before_script:
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=POWER8 NUM_THREADS=32"
+      env:
+        # for matrix annotation only
+        - TARGET_BOX=PPC64LE_LINUX
+        - BTYPE="BINARY=64 USE_OPENMP=1"
+
     - <<: *test-ubuntu
       env:
         - TARGET_BOX=LINUX64
@@ -164,42 +173,6 @@ matrix:
       env:
         - BTYPE="BINARY=32"
 
-    - &emulated-arm
-      dist: trusty
-      sudo: required
-      services: docker
-      env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=gcc
-      name: "Emulated Build for ARMV6 with gcc"
-      before_install: sudo docker run --rm --privileged multiarch/qemu-user-static:register --reset
-      script: |
-        echo "FROM openblas/alpine:${IMAGE_ARCH}
-        COPY . /tmp/openblas
-        RUN mkdir /tmp/openblas/build                             &&  \
-            cd /tmp/openblas/build                                &&  \
-            CC=${COMPILER} cmake -D DYNAMIC_ARCH=OFF                  \
-                                 -D TARGET=${TARGET_ARCH}             \
-                                 -D BUILD_SHARED_LIBS=ON              \
-                                 -D BUILD_WITHOUT_LAPACK=ON           \
-                                 -D BUILD_WITHOUT_CBLAS=ON            \
-                                 -D CMAKE_BUILD_TYPE=Release ../  &&  \
-            cmake --build ." > Dockerfile
-        docker build .
-    - <<: *emulated-arm
-      env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=clang
-      name: "Emulated Build for ARMV6 with clang"
-    - <<: *emulated-arm
-      env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=gcc
-      name: "Emulated Build for ARMV8 with gcc"
-    - <<: *emulated-arm
-      env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=clang
-      name: "Emulated Build for ARMV8 with clang"
-
-  allow_failures:
-    - env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=gcc
-    - env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=clang
-    - env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=gcc
-    - env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=clang
-
 # whitelist
 branches:
   only:

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 6)
+set(OpenBLAS_PATCH_VERSION 8.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -20,9 +20,14 @@ if(MSVC)
 option(BUILD_WITHOUT_LAPACK "Do not build LAPACK and LAPACKE (Only BLAS or CBLAS)" ON)
 endif()
 option(BUILD_WITHOUT_CBLAS "Do not build the C interface (CBLAS) to the BLAS functions" OFF)
-option(DYNAMIC_ARCH "Include support for multiple CPU targets, with automatic selection at runtime (x86/x86_64 only)" OFF)
-option(DYNAMIC_OLDER "Include specific support for older cpu models (Penryn,Dunnington,Atom,Nano,Opteron) with DYNAMIC_ARCH" OFF)
+option(DYNAMIC_ARCH "Include support for multiple CPU targets, with automatic selection at runtime (x86/x86_64, aarch64 or ppc only)" OFF)
+option(DYNAMIC_OLDER "Include specific support for older x86 cpu models (Penryn,Dunnington,Atom,Nano,Opteron) with DYNAMIC_ARCH" OFF)
 option(BUILD_RELAPACK "Build with ReLAPACK (recursive implementation of several LAPACK functions on top of standard LAPACK)" OFF)
+if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
+option(NO_AFFINITY "Disable support for CPU affinity masks to avoid binding processes from e.g. R or numpy/scipy to a single core" ON)
+else()
+set(NO_AFFINITY 1)
+endif()
 
 # Add a prefix or suffix to all exported symbol names in the shared library.
 # Avoids conflicts with other BLAS libraries, especially when using
@@ -206,7 +211,8 @@ if (USE_THREAD)
   target_link_libraries(${OpenBLAS_LIBNAME} ${CMAKE_THREAD_LIBS_INIT})
 endif()
 
-if (MSVC OR NOT NOFORTRAN)
+#if (MSVC OR NOT NOFORTRAN)
+if (NOT NO_CBLAS)
   # Broken without fortran on unix
   add_subdirectory(utest)
 endif()

CONTRIBUTORS.md

@@ -167,4 +167,7 @@ In chronological order:
   * [2017-02-26] ztrmm kernel for IBM z13
   * [2017-03-13] strmm and ctrmm kernel for IBM z13
   * [2017-09-01] initial Blas Level-1,2 (double precision) for IBM z13
-
+  * [2018-03-07] added missing Blas Level 1-2  (double precision) simd codes
+  * [2019-02-01] added missing Blas Level-1,2 (single precision)  simd codes
+  * [2019-03-14] power9 dgemm/dtrmm kernel
+  * [2019-04-29] power9 sgemm/strmm kernel 

Changelog.txt

@@ -1,4 +1,46 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.7
+11-Aug 2019
+
+common:
+        * having the gmake special variables TARGET_ARCH or TARGET_MACH
+          defined no longer causes build failures in ctest or utest
+        * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
+          has the same effect as setting them to 1
+        * a new test program was added to allow checking the library for
+          thread safety
+        * a new option USE_LOCKING was added to ensure thread safety when
+          OpenBLAS itself is built without multithreading but will be 
+          called from multiple threads.
+        * a build failure on Linux with glibc versions earlier than 2.5
+          was fixed
+        * a runtime error with CPU enumeration (and NO_AFFINITY not set)
+          on glibc 2.6 was fixed
+        * NO_AFFINITY was added to the CMAKE options (and defaults to being
+          active on Linux, as in the gmake builds)
+
+x86_64:
+        * the build-time logic for detection of AVX512 availability in
+          the processor and compiler was fixed
+        * gmake builds on OSX now set the internal name of the library to 
+          libopenblas.0.dylib (consistent with CMAKE)
+        * the Haswell DGEMM kernel received a significant speedup through
+          improved prefetch and load instructions
+        * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
+          increased by avoiding vpermpd instructions
+        * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
+          to fix remaining errors in DGEMM, DSYMM and DTRMM
+
+## POWER:
+        * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
+        * added optimized kernels for POWER9 SGEMM and STRMM
+
+## ARMV7:
+        * fixed the softfp implementations of xAMAX and IxAMAX
+        * removed the predefined -march= flags on both ARMV5 and ARMV6 as
+          they were appropriate for only a subset of platforms
+
 ====================================================================
 Version 0.3.6
 29-Apr-2019

Makefile

@@ -34,7 +34,7 @@ endif
 
 LAPACK_NOOPT := $(filter-out -O0 -O1 -O2 -O3 -Ofast,$(LAPACK_FFLAGS))
 
-SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench
+SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench cpp_thread_test
 
 .PHONY : all libs netlib $(RELA) test ctest shared install
 .NOTPARALLEL : all libs $(RELA) prof lapack-test install blas-test
@@ -109,6 +109,7 @@ endif
 ifeq ($(OSNAME), Darwin)
 	@$(MAKE) -C exports dyn
 	@ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib
+	@ln -fs $(LIBDYNNAME) $(LIBPREFIX).$(MAJOR_VERSION).dylib
 endif
 ifeq ($(OSNAME), WINNT)
 	@$(MAKE) -C exports dll
@@ -123,10 +124,13 @@ ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	touch $(LIBNAME)
 ifndef NO_FBLAS
 	$(MAKE) -C test all
-	$(MAKE) -C utest all
 endif
+	$(MAKE) -C utest all
 ifndef NO_CBLAS
 	$(MAKE) -C ctest all
+ifeq ($(CPP_THREAD_SAFETY_TEST), 1)
+	$(MAKE) -C cpp_thread_test all
+endif
 endif
 endif
 

Makefile.arm

@@ -1,7 +1,7 @@
 ifeq ($(CORE), $(filter $(CORE),ARMV7 CORTEXA9 CORTEXA15))
 ifeq ($(OSNAME), Android)
-CCOMMON_OPT += -mfpu=neon  -march=armv7-a
-FCOMMON_OPT += -mfpu=neon  -march=armv7-a
+CCOMMON_OPT += -mfpu=neon
+FCOMMON_OPT += -mfpu=neon
 else
 CCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
 FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
@@ -9,11 +9,6 @@ endif
 endif
 
 ifeq ($(CORE), ARMV6)
-CCOMMON_OPT += -mfpu=vfp -march=armv6
-FCOMMON_OPT += -mfpu=vfp -march=armv6
-endif
-
-ifeq ($(CORE), ARMV5)
-CCOMMON_OPT += -march=armv5
-FCOMMON_OPT += -march=armv5
+CCOMMON_OPT += -mfpu=vfp
+FCOMMON_OPT += -mfpu=vfp
 endif

Makefile.install

@@ -83,7 +83,8 @@ ifeq ($(OSNAME), Darwin)
 	@-cp $(LIBDYNNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@-install_name_tool -id "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME)" "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
-	ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib
+	ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib ; \
+	ln -fs $(LIBDYNNAME) $(LIBPREFIX).$(MAJOR_VERSION).dylib
 endif
 ifeq ($(OSNAME), WINNT)
 	@-cp $(LIBDLLNAME) "$(DESTDIR)$(OPENBLAS_BINARY_DIR)"

Makefile.power

@@ -29,6 +29,10 @@ FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -malign-power -fno-fas
 endif
 endif
 
+# workaround for C->FORTRAN ABI violation in LAPACKE
+ifeq ($(F_COMPILER), GFORTRAN)
+FCOMMON_OPT += -fno-optimize-sibling-calls
+endif
 
 FLAMEPATH	= $(HOME)/flame/lib
 

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.6
+VERSION = 0.3.8.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -58,6 +58,12 @@ VERSION = 0.3.6
 # For force setting for multi  threaded, specify USE_THREAD = 1
 # USE_THREAD = 0
 
+# If you want to build a single-threaded OpenBLAS, but expect to call this
+# from several concurrent threads in some other program, comment this in for
+# thread safety. (This is done automatically for USE_THREAD=1 , and should not
+# be necessary when USE_OPENMP=1)
+# USE_LOCKING = 1
+
 # If you're going to use this library with OpenMP, please comment it in.
 # This flag is always set for POWER8. Don't set USE_OPENMP = 0 if you're targeting POWER8.
 # USE_OPENMP = 1
@@ -157,6 +163,10 @@ NO_AFFINITY = 1
 # Don't use Haswell optimizations if binutils is too old (e.g. RHEL6)
 # NO_AVX2 = 1
 
+# Don't use SkylakeX optimizations if binutils or compiler are too old (the build
+# system will try to determine this automatically)
+# NO_AVX512 = 1
+
 # Don't use parallel make.
 # NO_PARALLEL_MAKE = 1
 
@@ -181,17 +191,17 @@ NO_AFFINITY = 1
 # time out to improve performance. This number should be from 4 to 30
 # which corresponds to (1 << n) cycles. For example, if you set to 26,
 # thread will be running for (1 << 26) cycles(about 25ms on 3.0GHz
-# system). Also you can control this mumber by THREAD_TIMEOUT
+# system). Also you can control this number by THREAD_TIMEOUT
 # CCOMMON_OPT	+= -DTHREAD_TIMEOUT=26
 
-# Using special device driver for mapping physically contigous memory
+# Using special device driver for mapping physically contiguous memory
 # to the user space. If bigphysarea is enabled, it will use it.
 # DEVICEDRIVER_ALLOCATION = 1
 
 # If you need to synchronize FP CSR between threads (for x86/x86_64 only).
 # CONSISTENT_FPCSR = 1
 
-# If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute
+# If any gemm argument m, n or k is less or equal this threshold, gemm will be execute
 # with single thread. (Actually in recent versions this is a factor proportional to the
 # number of floating point operations necessary for the given problem size, no longer
 # an individual dimension). You can use this setting to avoid the overhead of multi-
@@ -239,6 +249,21 @@ COMMON_PROF = -pg
 # SYMBOLPREFIX=
 # SYMBOLSUFFIX=
 
+# Run a C++ based thread safety tester after the build is done.
+# This is mostly intended as a developer feature to spot regressions, but users and
+# package maintainers can enable this if they have doubts about the thread safety of
+# the library, given the configuration in this file.
+# By default, the thread safety tester launches 52 concurrent calculations at the same
+# time.
+#
+# Please note that the test uses ~1300 MiB of RAM for the DGEMM test.
+#
+# The test requires CBLAS to be built, a C++11 capable compiler and the presence of
+# an OpenMP implementation. If you are cross-compiling this test will probably not
+# work at all.
+#
+# CPP_THREAD_SAFETY_TEST = 1
+
 #
 #  End of user configuration
 #

Makefile.system

@@ -9,6 +9,11 @@ ifndef TOPDIR
 TOPDIR = .
 endif
 
+# If ARCH is not set, we use the host system's architecture.
+ifndef ARCH
+ARCH := $(shell uname -m)
+endif
+
 # Catch conflicting usage of ARCH in some BSD environments
 ifeq ($(ARCH), amd64)
 override ARCH=x86_64
@@ -137,7 +142,12 @@ endif
 endif
 
 
-
+# On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
+ifeq ($(ARCH), x86_64)
+ifeq ($(findstring pgcc,$(HOSTCC)),)
+GETARCH_FLAGS += -march=native
+endif
+endif
 
 ifdef INTERFACE64
 ifneq ($(INTERFACE64), 0)
@@ -237,6 +247,10 @@ SMP = 1
 endif
 endif
 
+ifeq ($(SMP), 1)
+USE_LOCKING = 
+endif
+
 ifndef NEED_PIC
 NEED_PIC = 1
 endif
@@ -253,9 +267,10 @@ OBJCOPY = $(CROSS_SUFFIX)objcopy
 OBJCONV = $(CROSS_SUFFIX)objconv
 
 
-# For detect fortran failed, only build BLAS.
+# When fortran support was either not detected or actively deselected, only build BLAS.
 ifeq ($(NOFORTRAN), 1)
 NO_LAPACK = 1
+override FEXTRALIB = 
 endif
 
 #
@@ -388,6 +403,12 @@ ifneq ($(MAX_STACK_ALLOC), 0)
 CCOMMON_OPT	+= -DMAX_STACK_ALLOC=$(MAX_STACK_ALLOC)
 endif
 
+ifdef USE_LOCKING
+ifneq ($(USE_LOCKING), 0)
+CCOMMON_OPT	+= -DUSE_LOCKING
+endif
+endif
+
 #
 #  Architecture dependent settings
 #
@@ -744,6 +765,8 @@ CCOMMON_OPT += -DF_INTERFACE_GFORT
 FCOMMON_OPT += -Wall
 # make single-threaded LAPACK calls thread-safe #1847
 FCOMMON_OPT += -frecursive
+# work around ABI problem with passing single-character arguments
+FCOMMON_OPT += -fno-optimize-sibling-calls
 #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
 ifneq ($(NO_LAPACK), 1)
 EXTRALIB += -lgfortran
@@ -1049,7 +1072,7 @@ ifdef USE_SIMPLE_THREADED_LEVEL3
 CCOMMON_OPT	+= -DUSE_SIMPLE_THREADED_LEVEL3
 endif
 
-ifdef USE_TLS
+ifeq ($(USE_TLS), 1)
 CCOMMON_OPT += -DUSE_TLS
 endif
 
@@ -1102,8 +1125,12 @@ endif
 endif
 
 ifdef NO_AFFINITY
+ifeq ($(NO_AFFINITY), 0)
+override undefine NO_AFFINITY
+else
 CCOMMON_OPT	+= -DNO_AFFINITY
 endif
+endif
 
 ifdef FUNCTION_PROFILE
 CCOMMON_OPT	+= -DFUNCTION_PROFILE

Makefile.x86_64

@@ -28,11 +28,15 @@ endif
 ifeq ($(CORE), HASWELL)
 ifndef DYNAMIC_ARCH
 ifndef NO_AVX2
+ifeq ($(C_COMPILER), GCC)
 CCOMMON_OPT += -mavx2
+endif
+ifeq ($(F_COMPILER), GFORTRAN)
 FCOMMON_OPT += -mavx2
 endif
 endif
 endif
+endif
 
 
 

README.md

@@ -6,11 +6,13 @@ Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.svg?branch=dev
 
 AppVeyor: [![Build status](https://ci.appveyor.com/api/projects/status/09sohd35n8nkkx64/branch/develop?svg=true)](https://ci.appveyor.com/project/xianyi/openblas/branch/develop)
 
+[![Build Status](https://dev.azure.com/xianyi/OpenBLAS/_apis/build/status/xianyi.OpenBLAS?branchName=develop)](https://dev.azure.com/xianyi/OpenBLAS/_build/latest?definitionId=1&branchName=develop)
+
 ## Introduction
 
 OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
 
-Please read the documentation on the OpenBLAS wiki pages: <http://github.com/xianyi/OpenBLAS/wiki>.
+Please read the documentation on the OpenBLAS wiki pages: <https://github.com/xianyi/OpenBLAS/wiki>.
 
 ## Binary Packages
 
@@ -22,7 +24,7 @@ You can download them from [file hosting on sourceforge.net](https://sourceforge
 
 ## Installation from Source
 
-Download from project homepage, http://xianyi.github.com/OpenBLAS/, or check out the code
+Download from project homepage, https://xianyi.github.com/OpenBLAS/, or check out the code
 using Git from https://github.com/xianyi/OpenBLAS.git.
 
 ### Dependencies
@@ -63,9 +65,7 @@ A debug version can be built using `make DEBUG=1`.
 
 ### Compile with MASS support on Power CPU (optional)
 
-The [IBM MASS](http://www-01.ibm.com/software/awdtools/mass/linux/mass-linux.html) library
-consists of a set of mathematical functions for C, C++, and Fortran applications that are
-are tuned for optimum performance on POWER architectures.
+The [IBM MASS](https://www.ibm.com/support/home/product/W511326D80541V01/other_software/mathematical_acceleration_subsystem) library consists of a set of mathematical functions for C, C++, and Fortran applications that are tuned for optimum performance on POWER architectures.
 OpenBLAS with MASS requires a 64-bit, little-endian OS on POWER.
 The library can be installed as shown:
 
@@ -115,6 +115,7 @@ Please read `GotoBLAS_01Readme.txt`.
 - **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thanks to Werner Saar)
 - **AMD PILEDRIVER**: Uses Bulldozer codes with some optimizations.
 - **AMD STEAMROLLER**: Uses Bulldozer codes with some optimizations.
+- **AMD ZEN**: Uses Haswell codes with some optimizations.
 
 #### MIPS64
 
@@ -133,11 +134,13 @@ Please read `GotoBLAS_01Readme.txt`.
 
 #### PPC/PPC64
 
-- **POWER8**: Optmized Level-3 BLAS and some Level-1, only with `USE_OPENMP=1`
+- **POWER8**: Optimized BLAS, only for PPC64LE (Little Endian), only with `USE_OPENMP=1`
+- **POWER9**: Optimized Level-3 BLAS (real) and some Level-1,2. PPC64LE with OpenMP only. 
 
 #### IBM zEnterprise System
 
 - **Z13**: Optimized Level-3 BLAS and Level-1,2 (double precision)
+- **Z14**: Optimized Level-3 BLAS and Level-1,2 (single precision)
 
 ### Supported OS
 

appveyor.yml

@@ -35,7 +35,14 @@ environment:
       DYNAMIC_ARCH: ON
       WITH_FORTRAN: no
     - COMPILER: cl
-
+    - COMPILER: MinGW64-gcc-7.2.0-mingw
+      DYNAMIC_ARCH: OFF
+      WITH_FORTRAN: ignore
+    - COMPILER: MinGW64-gcc-7.2.0
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
+      COMPILER: MinGW-gcc-5.3.0
+      WITH_FORTRAN: ignore
+    
 install:
   - if [%COMPILER%]==[clang-cl] call %CONDA_INSTALL_LOCN%\Scripts\activate.bat
   - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
@@ -52,7 +59,14 @@ install:
 before_build:
   - ps: if (-Not (Test-Path .\build)) { mkdir build }
   - cd build
+  - set PATH=%PATH:C:\Program Files\Git\usr\bin;=%
+  - if [%COMPILER%]==[MinGW-gcc-5.3.0] set PATH=C:\MinGW\bin;C:\msys64\usr\bin;C:\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev1\mingw64\bin;%PATH%
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0-mingw] set PATH=C:\MinGW\bin;C:\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev1\mingw64\bin;%PATH%
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0] set PATH=C:\msys64\usr\bin;C:\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev1\mingw64\bin;%PATH%
   - if [%COMPILER%]==[cl] cmake -G "Visual Studio 15 2017 Win64" ..
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0-mingw] cmake -G "MinGW Makefiles" -DNOFORTRAN=1 ..
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0] cmake -G "MSYS Makefiles"  -DBINARY=32 -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 [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON -DDYNAMIC_LIST='CORE2;NEHALEM;SANDYBRIDGE;BULLDOZER;HASWELL' ..
@@ -64,3 +78,4 @@ test_script:
   - echo Running Test
   - cd utest
   - openblas_utest
+  

azure-pipelines.yml

@@ -0,0 +1,51 @@
+trigger:
+  # start a new build for every push
+  batch: False
+  branches:
+    include:
+      - develop
+
+jobs:
+# manylinux1 is useful to test because the
+# standard Docker container uses an old version
+# of gcc / glibc
+- job: manylinux1_gcc
+  pool:
+    vmImage: 'ubuntu-16.04'
+  steps:
+  - script: |
+      echo "FROM quay.io/pypa/manylinux1_x86_64
+        COPY . /tmp/openblas
+        RUN cd /tmp/openblas                                      &&  \
+            COMMON_FLAGS='DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32' && \
+            BTYPE='BINARY=64' CC=gcc && \
+            make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE && \
+            make -C test $COMMON_FLAGS $BTYPE && \
+            make -C ctest $COMMON_FLAGS $BTYPE && \
+            make -C utest $COMMON_FLAGS $BTYPE" > Dockerfile
+      docker build .
+    displayName: Run manylinux1 docker build
+- job: Intel_SDE_skx
+  pool:
+    vmImage: 'ubuntu-16.04'
+  steps:
+  - script: |
+      # at the time of writing the available Azure Ubuntu vm image
+      # does not support AVX512VL, so use more recent LTS version
+      echo "FROM ubuntu:bionic
+      COPY . /tmp/openblas
+      RUN apt-get -y update && apt-get -y install \\
+          cmake \\
+          gfortran \\
+          make \\
+          wget
+      RUN mkdir /tmp/SDE && cd /tmp/SDE && \\
+          mkdir sde-external-8.35.0-2019-03-11-lin && \\
+          wget --quiet -O sde-external-8.35.0-2019-03-11-lin.tar.bz2 https://www.dropbox.com/s/fopsnzj67572sj5/sde-external-8.35.0-2019-03-11-lin.tar.bz2?dl=0 && \\
+          tar -xjvf sde-external-8.35.0-2019-03-11-lin.tar.bz2 -C /tmp/SDE/sde-external-8.35.0-2019-03-11-lin --strip-components=1
+      RUN cd /tmp/openblas && CC=gcc make QUIET_MAKE=1 DYNAMIC_ARCH=1 NUM_THREADS=32 BINARY=64
+      CMD cd /tmp/openblas && echo 0 > /proc/sys/kernel/yama/ptrace_scope && CC=gcc OPENBLAS_VERBOSE=2 /tmp/SDE/sde-external-8.35.0-2019-03-11-lin/sde64 -cpuid_in /tmp/SDE/sde-external-8.35.0-2019-03-11-lin/misc/cpuid/skx/cpuid.def -- make -C utest DYNAMIC_ARCH=1 NUM_THREADS=32 BINARY=64" > Dockerfile
+      docker build -t intel_sde .
+      # we need a privileged docker run for sde process attachment
+      docker run --privileged intel_sde
+    displayName: 'Run AVX512 SkylakeX docker build / test'

benchmark/gemm.c

@@ -207,7 +207,7 @@ int main(int argc, char *argv[]){
   for (i = 0; i < m * n * COMPSIZE; i++) {
     c[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
   }
- 
+
   fprintf(stderr, "          SIZE                   Flops             Time\n");
 
   for (i = from; i <= to; i += step) {

c_check

@@ -240,7 +240,7 @@ if (($architecture eq "x86") || ($architecture eq "x86_64")) {
 	} else {
 	    $no_avx512 = 0;
 	}
-	unlink("tmpf.o");
+	unlink("$tmpf.o");
     }
 }
 

cmake/arch.cmake

@@ -73,14 +73,16 @@ if (DYNAMIC_ARCH)
     endif ()
     if (NOT NO_AVX512)
       set(DYNAMIC_CORE ${DYNAMIC_CORE} SKYLAKEX)
-    endif ()
+      string(REGEX REPLACE "-march=native" "" CMAKE_C_FLAGS ${CMAKE_C_FLAGS})
+   endif ()
     if (DYNAMIC_LIST)
 	set(DYNAMIC_CORE PRESCOTT ${DYNAMIC_LIST})
     endif ()
   endif ()
 
   if (NOT DYNAMIC_CORE)
-    unset(DYNAMIC_ARCH)
+    message (STATUS "DYNAMIC_ARCH is not supported on this architecture, removing from options")
+    unset(DYNAMIC_ARCH CACHE)
   endif ()
 endif ()
 

cmake/fc.cmake

@@ -44,7 +44,10 @@ endif ()
 
 if (${F_COMPILER} STREQUAL "GFORTRAN")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_GFORT")
+  # ensure reentrancy of lapack codes
   set(FCOMMON_OPT "${FCOMMON_OPT} -Wall -frecursive")
+  # work around ABI violation in passing string arguments from C
+  set(FCOMMON_OPT "${FCOMMON_OPT} -fno-optimize-sibling-calls")
   #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
   if (NOT NO_LAPACK)
     set(EXTRALIB "{EXTRALIB} -lgfortran")

cmake/kernel.cmake

@@ -1,7 +1,7 @@
 # helper functions for the kernel CMakeLists.txt
 
 
-# Set the default filenames for L1 objects. Most of these will be overriden by the appropriate KERNEL file.
+# Set the default filenames for L1 objects. Most of these will be overridden by the appropriate KERNEL file.
 macro(SetDefaultL1)
   set(SAMAXKERNEL amax.S)
   set(DAMAXKERNEL amax.S)

cmake/prebuild.cmake

@@ -59,6 +59,9 @@ set(FU "")
 if (APPLE OR (MSVC AND NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang"))
   set(FU "_")
 endif()
+if(MINGW AND NOT MINGW64)
+  set(FU "_")
+endif()
 
 set(COMPILER_ID ${CMAKE_C_COMPILER_ID})
 if (${COMPILER_ID} STREQUAL "GNU")
@@ -82,6 +85,11 @@ endif ()
 # f_check
 if (NOT NOFORTRAN)
   include("${PROJECT_SOURCE_DIR}/cmake/f_check.cmake")
+else ()
+ file(APPEND ${TARGET_CONF_TEMP}
+   "#define BUNDERSCORE _\n"
+   "#define NEEDBUNDERSCORE 1\n")
+ set(BU "_")
 endif ()
 
 # Cannot run getarch on target if we are cross-compiling

cmake/system.cmake

@@ -65,6 +65,18 @@ if (DEFINED TARGET)
   set(GETARCH_FLAGS "-DFORCE_${TARGET}")
 endif ()
 
+# On x86_64 build getarch with march=native. This is required to detect AVX512 support in getarch.
+if (X86_64 AND NOT ${CMAKE_C_COMPILER_ID} STREQUAL "PGI")
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -march=native")
+endif ()
+
+# On x86 no AVX support is available
+if (X86 OR X86_64)
+if ((DEFINED BINARY AND BINARY EQUAL 32) OR ("$CMAKE_SIZEOF_VOID_P}" EQUAL "4"))
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX -DNO_AVX2 -DNO_AVX512")
+endif ()
+endif ()
+
 if (INTERFACE64)
   message(STATUS "Using 64-bit integers.")
   set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DUSE64BITINT")
@@ -136,10 +148,16 @@ endif ()
 
 if (USE_THREAD)
   message(STATUS "Multi-threading enabled with ${NUM_THREADS} threads.")
+else()
+  if (${USE_LOCKING})
+    set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_LOCKING")
+  endif ()
 endif ()
 
 include("${PROJECT_SOURCE_DIR}/cmake/prebuild.cmake")
-
+if (DEFINED BINARY)
+  message(STATUS "Compiling a ${BINARY}-bit binary.")
+endif ()
 if (NOT DEFINED NEED_PIC)
   set(NEED_PIC 1)
 endif ()
@@ -156,6 +174,9 @@ include("${PROJECT_SOURCE_DIR}/cmake/cc.cmake")
 if (NOT NOFORTRAN)
   # Fortran Compiler dependent settings
   include("${PROJECT_SOURCE_DIR}/cmake/fc.cmake")
+else ()
+set(NO_LAPACK 1)
+set(NO_LAPACKE 1)
 endif ()
 
 if (BINARY64)
@@ -181,9 +202,14 @@ if (NEED_PIC)
 endif ()
 
 if (DYNAMIC_ARCH)
-  set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
-  if (DYNAMIC_OLDER)
-    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
+  if (X86 OR X86_64 OR ARM64 OR PPC)
+    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
+    if (DYNAMIC_OLDER)
+      set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
+    endif ()
+  else ()
+    unset (DYNAMIC_ARCH)
+    message (STATUS "DYNAMIC_ARCH is not supported on the target architecture, removing")
   endif ()
 endif ()
 
@@ -283,7 +309,7 @@ endif ()
 
 set(KERNELDIR	"${PROJECT_SOURCE_DIR}/kernel/${ARCH}")
 
-# TODO: nead to convert these Makefiles
+# TODO: need to convert these Makefiles
 # include ${PROJECT_SOURCE_DIR}/cmake/${ARCH}.cmake
 
 if (${CORE} STREQUAL "PPC440")

cmake/system_check.cmake

@@ -15,7 +15,7 @@ if (${HOST_OS} STREQUAL "LINUX")
     EXECUTE_PROCESS( COMMAND uname -o COMMAND tr -d '\n' OUTPUT_VARIABLE OPERATING_SYSTEM)
       if(${OPERATING_SYSTEM} MATCHES "Android")
         set(HOST_OS ANDROID)
-      endif(${OPERATING_SYSTEM} MATCHES "Android")
+      endif()
 endif()
 
 

cmake/utils.cmake

@@ -89,7 +89,7 @@ function(AllCombinations list_in absent_codes_in)
   set(CODES_OUT ${CODES_OUT} PARENT_SCOPE)
 endfunction ()
 
-# generates object files for each of the sources, using the BLAS naming scheme to pass the funciton name as a preprocessor definition
+# generates object files for each of the sources, using the BLAS naming scheme to pass the function name as a preprocessor definition
 # @param sources_in the source files to build from
 # @param defines_in (optional) preprocessor definitions that will be applied to all objects
 # @param name_in (optional) if this is set this name will be used instead of the filename. Use a * to indicate where the float character should go, if no star the character will be prepended.

common.h

@@ -131,7 +131,7 @@ extern "C" {
 #include <time.h>
 #include <unistd.h>
 #include <math.h>
-#ifdef SMP
+#if defined(SMP) || defined(USE_LOCKING)
 #include <pthread.h>
 #endif
 #endif
@@ -200,7 +200,7 @@ extern "C" {
 #error "You can't specify both LOCK operation!"
 #endif
 
-#ifdef SMP
+#if defined(SMP) || defined(USE_LOCKING)
 #define USE_PTHREAD_LOCK
 #undef	USE_PTHREAD_SPINLOCK
 #endif

common_power.h

@@ -241,7 +241,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define HAVE_PREFETCH
 #endif
 
-#if defined(POWER3) || defined(POWER6) || defined(PPCG4) || defined(CELL) || defined(POWER8) || defined(POWER9) || ( defined(PPC970) && defined(OS_DARWIN) )
+#if defined(POWER3) || defined(POWER6) || defined(PPCG4) || defined(CELL) || defined(POWER8) || defined(POWER9) || ( defined(PPC970) && ( defined(OS_DARWIN) || defined(OS_FREEBSD) ) )
 #define DCBT_ARG	0
 #else
 #define DCBT_ARG	8
@@ -499,7 +499,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 
 #if defined(ASSEMBLER) && !defined(NEEDPARAM)
 
-#ifdef OS_LINUX
+#if defined(OS_LINUX) || defined(OS_FREEBSD)
 #ifndef __64BIT__
 #define PROLOGUE \
 	.section .text;\
@@ -784,7 +784,7 @@ Lmcount$lazy_ptr:
 
 #define HALT		mfspr	r0, 1023
 
-#ifdef OS_LINUX
+#if defined(OS_LINUX) || defined(OS_FREEBSD)
 #if defined(PPC440) || defined(PPC440FP2)
 #undef  MAX_CPU_NUMBER
 #define MAX_CPU_NUMBER 1
@@ -829,7 +829,7 @@ Lmcount$lazy_ptr:
 #define MAP_ANONYMOUS MAP_ANON
 #endif
 
-#ifdef OS_LINUX
+#if defined(OS_LINUX) || defined(OS_FREEBSD)
 #ifndef __64BIT__
 #define FRAMESLOT(X) (((X) * 4) + 8)
 #else

common_stackalloc.h

@@ -45,7 +45,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  * SIZE must be carefully chosen to be:
  * - as small as possible to maximize the number of stack allocation
  * - large enough to support all architectures and kernel
- * Chosing a too small SIZE will lead to a stack smashing.
+ * Choosing a SIZE too small will lead to a stack smashing.
  */
 #define STACK_ALLOC(SIZE, TYPE, BUFFER)                                        \
   /* make it volatile because some function (ex: dgemv_n.S) */                 \

common_x86.h

@@ -214,7 +214,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #endif
 
 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
-//Enable some optimazation for barcelona.
+//Enable some optimization for barcelona.
 #define BARCELONA_OPTIMIZATION
 #endif
 

common_x86_64.h

@@ -129,12 +129,13 @@ static __inline void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
   *ecx=cpuinfo[2];
   *edx=cpuinfo[3];
 #else
-        __asm__ __volatile__("cpuid"
+        __asm__ __volatile__("mov $0, %%ecx;"
+			     "cpuid"
 			     : "=a" (*eax),
 			     "=b" (*ebx),
 			     "=c" (*ecx),
 			     "=d" (*edx)
-			     : "0" (op), "c"(0));
+			     : "0" (op));
 #endif
 }
 
@@ -276,7 +277,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #ifdef ASSEMBLER
 
 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
-//Enable some optimazation for barcelona.
+//Enable some optimization for barcelona.
 #define BARCELONA_OPTIMIZATION
 #endif
 

cpp_thread_test/Makefile

@@ -0,0 +1,14 @@
+include ../Makefile.rule
+
+all :: dgemv_tester dgemm_tester
+
+dgemv_tester :
+	$(CXX) $(COMMON_OPT) -Wall -Wextra -Wshadow -fopenmp -std=c++11 dgemv_thread_safety.cpp ../libopenblas.a -lpthread -o dgemv_tester
+	./dgemv_tester
+
+dgemm_tester : dgemv_tester
+	$(CXX) $(COMMON_OPT) -Wall -Wextra -Wshadow -fopenmp -std=c++11 dgemm_thread_safety.cpp ../libopenblas.a -lpthread -o dgemm_tester
+	./dgemm_tester
+
+clean ::
+	rm -f dgemv_tester dgemm_tester

cpp_thread_test/cpp_thread_safety_common.h

@@ -0,0 +1,55 @@
+inline void pauser(){
+    /// a portable way to pause a program
+    std::string dummy;
+    std::cout << "Press enter to continue...";
+    std::getline(std::cin, dummy);
+}
+
+void FillMatrices(std::vector<std::vector<double>>& matBlock, std::mt19937_64& PRNG, std::uniform_real_distribution<double>& rngdist, const blasint randomMatSize, const uint32_t numConcurrentThreads, const uint32_t numMat){
+	for(uint32_t i=0; i<numMat; i++){
+		for(uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize*randomMatSize); j++){
+			matBlock[i][j] = rngdist(PRNG);
+		}
+	}
+	for(uint32_t i=numMat; i<(numConcurrentThreads*numMat); i+=numMat){
+		for(uint32_t j=0; j<numMat; j++){
+			matBlock[i+j] = matBlock[j];
+		}
+	}
+}
+
+void FillVectors(std::vector<std::vector<double>>& vecBlock, std::mt19937_64& PRNG, std::uniform_real_distribution<double>& rngdist, const blasint randomMatSize, const uint32_t numConcurrentThreads, const uint32_t numVec){
+	for(uint32_t i=0; i<numVec; i++){
+		for(uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize); j++){
+			vecBlock[i][j] = rngdist(PRNG);
+		}
+	}
+	for(uint32_t i=numVec; i<(numConcurrentThreads*numVec); i+=numVec){
+		for(uint32_t j=0; j<numVec; j++){
+			vecBlock[i+j] = vecBlock[j];
+		}
+	}
+}
+
+std::mt19937_64 InitPRNG(){
+	std::random_device rd;
+	std::mt19937_64 PRNG(rd()); //seed PRNG using /dev/urandom or similar OS provided RNG
+	std::uniform_real_distribution<double> rngdist{-1.0, 1.0};
+	//make sure the internal state of the PRNG is properly mixed by generating 10M random numbers
+	//PRNGs often have unreliable distribution uniformity and other statistical properties before their internal state is sufficiently mixed
+	for (uint32_t i=0;i<10000000;i++) rngdist(PRNG);
+	return PRNG;
+}
+
+void PrintMatrices(const std::vector<std::vector<double>>& matBlock, const blasint randomMatSize, const uint32_t numConcurrentThreads, const uint32_t numMat){
+	for (uint32_t i=0;i<numConcurrentThreads*numMat;i++){
+		std::cout<<i<<std::endl;
+		for (uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize); j++){
+			for (uint32_t k = 0; k < static_cast<uint32_t>(randomMatSize); k++){
+				std::cout<<matBlock[i][j*randomMatSize + k]<<"  ";
+			}
+			std::cout<<std::endl;
+		}
+		std::cout<<std::endl;
+	}
+}

cpp_thread_test/dgemm_thread_safety.cpp

@@ -0,0 +1,92 @@
+#include <iostream>
+#include <vector>
+#include <random>
+#include <future>
+#include <omp.h>
+#include "../cblas.h"
+#include "cpp_thread_safety_common.h"
+
+void launch_cblas_dgemm(double* A, double* B, double* C, const blasint randomMatSize){
+	cblas_dgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, randomMatSize, randomMatSize, randomMatSize, 1.0, A, randomMatSize, B, randomMatSize, 0.1, C, randomMatSize);
+}
+
+int main(int argc, char* argv[]){
+	blasint randomMatSize = 1024; //dimension of the random square matrices used
+	uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested
+	uint32_t numTestRounds = 16; //number of testing rounds before success exit
+	
+	if (argc > 4){
+		std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl;
+		abort();
+	}
+	
+	if(argc == 4){
+		std::vector<std::string> cliArgs;
+		for (int i = 1; i < argc; i++){
+			cliArgs.push_back(argv[i]);
+			std::cout<<argv[i]<<std::endl;
+		}
+		randomMatSize = std::stoul(cliArgs[0]);
+		numConcurrentThreads = std::stoul(cliArgs[1]);
+		numTestRounds = std::stoul(cliArgs[2]);
+	}
+	
+	std::uniform_real_distribution<double> rngdist{-1.0, 1.0};
+	std::vector<std::vector<double>> matBlock(numConcurrentThreads*3);
+	std::vector<std::future<void>> futureBlock(numConcurrentThreads);
+	
+	std::cout<<"*----------------------------*\n";
+	std::cout<<"| DGEMM thread safety tester |\n";
+	std::cout<<"*----------------------------*\n";
+	std::cout<<"Size of random matrices(N=M=K): "<<randomMatSize<<'\n';
+	std::cout<<"Number of concurrent calls into OpenBLAS : "<<numConcurrentThreads<<'\n';
+	std::cout<<"Number of testing rounds : "<<numTestRounds<<'\n';
+	std::cout<<"This test will need "<<(static_cast<uint64_t>(randomMatSize*randomMatSize)*numConcurrentThreads*3*8)/static_cast<double>(1024*1024)<<" MiB of RAM\n"<<std::endl;
+	
+	std::cout<<"Initializing random number generator..."<<std::flush;
+	std::mt19937_64 PRNG = InitPRNG();
+	std::cout<<"done\n";
+	
+	std::cout<<"Preparing to test CBLAS DGEMM thread safety\n";
+	std::cout<<"Allocating matrices..."<<std::flush;
+	for(uint32_t i=0; i<(numConcurrentThreads*3); i++){
+		matBlock[i].resize(randomMatSize*randomMatSize);
+	}
+	std::cout<<"done\n";
+	//pauser();
+	std::cout<<"Filling matrices with random numbers..."<<std::flush;
+	FillMatrices(matBlock, PRNG, rngdist, randomMatSize, numConcurrentThreads, 3);
+	//PrintMatrices(matBlock, randomMatSize, numConcurrentThreads, 3);
+	std::cout<<"done\n";
+	std::cout<<"Testing CBLAS DGEMM thread safety\n";
+	omp_set_num_threads(numConcurrentThreads);
+	for(uint32_t R=0; R<numTestRounds; R++){
+		std::cout<<"DGEMM round #"<<R<<std::endl;
+		std::cout<<"Launching "<<numConcurrentThreads<<" threads simultaneously using OpenMP..."<<std::flush;
+		#pragma omp parallel for default(none) shared(futureBlock, matBlock, randomMatSize, numConcurrentThreads)
+		for(uint32_t i=0; i<numConcurrentThreads; i++){
+			futureBlock[i] = std::async(std::launch::async, launch_cblas_dgemm, &matBlock[i*3][0], &matBlock[i*3+1][0], &matBlock[i*3+2][0], randomMatSize);
+			//launch_cblas_dgemm( &matBlock[i][0], &matBlock[i+1][0], &matBlock[i+2][0]);
+		}
+		std::cout<<"done\n";
+		std::cout<<"Waiting for threads to finish..."<<std::flush;
+		for(uint32_t i=0; i<numConcurrentThreads; i++){
+			futureBlock[i].get();
+		}
+		std::cout<<"done\n";
+		//PrintMatrices(matBlock, randomMatSize, numConcurrentThreads, 3);
+		std::cout<<"Comparing results from different threads..."<<std::flush;
+		for(uint32_t i=3; i<(numConcurrentThreads*3); i+=3){ //i is the index of matrix A, for a given thread
+			for(uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize*randomMatSize); j++){
+				if (std::abs(matBlock[i+2][j] - matBlock[2][j]) > 1.0E-13){ //i+2 is the index of matrix C, for a given thread
+					std::cout<<"ERROR: one of the threads returned a different result! Index : "<<i+2<<std::endl;
+					std::cout<<"CBLAS DGEMM thread safety test FAILED!"<<std::endl;
+					return -1;
+				}
+			}
+		}
+		std::cout<<"OK!\n"<<std::endl;
+	}
+	std::cout<<"CBLAS DGEMM thread safety test PASSED!\n"<<std::endl;
+	return 0;
+}

cpp_thread_test/dgemv_thread_safety.cpp

@@ -0,0 +1,101 @@
+#include <iostream>
+#include <vector>
+#include <random>
+#include <future>
+#include <omp.h>
+#include "../cblas.h"
+#include "cpp_thread_safety_common.h"
+
+void launch_cblas_dgemv(double* A, double* x, double* y, const blasint randomMatSize){
+	const blasint inc = 1;
+	cblas_dgemv(CblasColMajor, CblasNoTrans, randomMatSize, randomMatSize, 1.0, A, randomMatSize, x, inc, 0.1, y, inc);
+}
+
+int main(int argc, char* argv[]){
+	blasint randomMatSize = 1024; //dimension of the random square matrices and vectors being used
+	uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested
+	uint32_t numTestRounds = 16; //number of testing rounds before success exit
+	
+	if (argc > 4){
+		std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl;
+		abort();
+	}
+	if(argc == 4){
+		std::vector<std::string> cliArgs;
+		for (int i = 1; i < argc; i++){
+			cliArgs.push_back(argv[i]);
+			std::cout<<argv[i]<<std::endl;
+		}
+		randomMatSize = std::stoul(cliArgs.at(0));
+		numConcurrentThreads = std::stoul(cliArgs.at(1));
+		numTestRounds = std::stoul(cliArgs.at(2));
+	}
+	
+	std::uniform_real_distribution<double> rngdist{-1.0, 1.0};
+	std::vector<std::vector<double>> matBlock(numConcurrentThreads);
+	std::vector<std::vector<double>> vecBlock(numConcurrentThreads*2);
+	std::vector<std::future<void>> futureBlock(numConcurrentThreads);
+	
+	std::cout<<"*----------------------------*\n";
+	std::cout<<"| DGEMV thread safety tester |\n";
+	std::cout<<"*----------------------------*\n";
+	std::cout<<"Size of random matrices and vectors(N=M): "<<randomMatSize<<'\n';
+	std::cout<<"Number of concurrent calls into OpenBLAS : "<<numConcurrentThreads<<'\n';
+	std::cout<<"Number of testing rounds : "<<numTestRounds<<'\n';
+	std::cout<<"This test will need "<<((static_cast<uint64_t>(randomMatSize*randomMatSize)*numConcurrentThreads*8)+(static_cast<uint64_t>(randomMatSize)*numConcurrentThreads*8*2))/static_cast<double>(1024*1024)<<" MiB of RAM\n"<<std::endl;
+	
+	std::cout<<"Initializing random number generator..."<<std::flush;
+	std::mt19937_64 PRNG = InitPRNG();
+	std::cout<<"done\n";
+	
+	std::cout<<"Preparing to test CBLAS DGEMV thread safety\n";
+	std::cout<<"Allocating matrices..."<<std::flush;
+	for(uint32_t i=0; i<numConcurrentThreads; i++){
+		matBlock.at(i).resize(randomMatSize*randomMatSize);
+	}
+	std::cout<<"done\n";
+	std::cout<<"Allocating vectors..."<<std::flush;
+	for(uint32_t i=0; i<(numConcurrentThreads*2); i++){
+		vecBlock.at(i).resize(randomMatSize);
+	}
+	std::cout<<"done\n";
+	//pauser();
+	
+	std::cout<<"Filling matrices with random numbers..."<<std::flush;
+	FillMatrices(matBlock, PRNG, rngdist, randomMatSize, numConcurrentThreads, 1);
+	//PrintMatrices(matBlock, randomMatSize, numConcurrentThreads);
+	std::cout<<"done\n";
+	std::cout<<"Filling vectors with random numbers..."<<std::flush;
+	FillVectors(vecBlock, PRNG, rngdist, randomMatSize, numConcurrentThreads, 2);
+	std::cout<<"done\n";
+	
+	std::cout<<"Testing CBLAS DGEMV thread safety"<<std::endl;
+	omp_set_num_threads(numConcurrentThreads);
+	for(uint32_t R=0; R<numTestRounds; R++){
+		std::cout<<"DGEMV round #"<<R<<std::endl;
+		std::cout<<"Launching "<<numConcurrentThreads<<" threads simultaneously using OpenMP..."<<std::flush;
+		#pragma omp parallel for default(none) shared(futureBlock, matBlock, vecBlock, randomMatSize, numConcurrentThreads)
+		for(uint32_t i=0; i<numConcurrentThreads; i++){
+			futureBlock[i] = std::async(std::launch::async, launch_cblas_dgemv, &matBlock[i][0], &vecBlock[i*2][0], &vecBlock[i*2+1][0], randomMatSize);
+		}
+		std::cout<<"done\n";
+		std::cout<<"Waiting for threads to finish..."<<std::flush;
+		for(uint32_t i=0; i<numConcurrentThreads; i++){
+			futureBlock[i].get();
+		}
+		std::cout<<"done\n";
+		std::cout<<"Comparing results from different threads..."<<std::flush;
+		for(uint32_t i=2; i<(numConcurrentThreads*2); i+=2){ //i is the index of vector x, for a given thread
+			for(uint32_t j = 0; j < static_cast<uint32_t>(randomMatSize); j++){
+				if (std::abs(vecBlock[i+1][j] - vecBlock[1][j]) > 1.0E-13){ //i+1 is the index of vector y, for a given thread
+					std::cout<<"ERROR: one of the threads returned a different result! Index : "<<i+1<<std::endl;
+					std::cout<<"CBLAS DGEMV thread safety test FAILED!"<<std::endl;
+					return -1;
+				}
+			}
+		}
+		std::cout<<"OK!\n"<<std::endl;
+	}
+	std::cout<<"CBLAS DGEMV thread safety test PASSED!\n"<<std::endl;
+	return 0;
+}

cpuid_arm64.c

@@ -94,7 +94,7 @@ int get_feature(char *search)
 	if( p == NULL ) return 0;
 
 	t = strtok(p," ");
-	while( t = strtok(NULL," "))
+	while( (t = strtok(NULL," ")))
 	{
 		if (!strcmp(t, search))   { return(1); }
 	}
@@ -344,7 +344,7 @@ void get_features(void)
 	if( p == NULL ) return;
 
 	t = strtok(p," ");
-	while( t = strtok(NULL," "))
+	while( (t = strtok(NULL," ")))
 	{
 	}
 

cpuid_x86.c

@@ -1211,7 +1211,7 @@ int get_cpuname(void){
 	  return CPUTYPE_CORE2;
 	}
 	break;
-      case 1:
+      case 1:  // family 6 exmodel 1
 	switch (model) {
 	case  6:
 	  return CPUTYPE_CORE2;
@@ -1228,7 +1228,7 @@ int get_cpuname(void){
 	  return CPUTYPE_DUNNINGTON;
 	}
 	break;
-      case  2:
+      case  2: // family 6 exmodel 2
 	switch (model) {
 	case 5:
 	  //Intel Core (Clarkdale) / Core (Arrandale)
@@ -1257,7 +1257,7 @@ int get_cpuname(void){
 	  return CPUTYPE_NEHALEM;
 	}
 	break;
-      case 3:
+      case 3: // family 6 exmodel 3
 	switch (model) {
 	case  7:
 	    // Bay Trail	
@@ -1287,7 +1287,7 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
 	}
 	break;
-      case 4:
+      case 4: // family 6 exmodel 4
         switch (model) {
         case 5:
 	case 6:
@@ -1321,7 +1321,7 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
         }
         break;
-      case 5:
+      case 5:  // family 6 exmodel 5
         switch (model) {
 	case 6:
 	  //Broadwell
@@ -1364,7 +1364,7 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
 	}
 	break;
-      case 6:
+      case 6:  // family 6 exmodel 6
         switch (model) {
         case 6: // Cannon Lake
           if(support_avx512())
@@ -1376,7 +1376,22 @@ int get_cpuname(void){
 	  else
 	  return CPUTYPE_NEHALEM;
         }
-      break;  
+      break;
+      case 7: // family 6 exmodel 7
+        switch (model) {
+        case 10: // Goldmont Plus
+          return CPUTYPE_NEHALEM;
+        case 14: // Ice Lake
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	  return CPUTYPE_NEHALEM;
+        }
+      break;
       case 9:
       case 8:      
         switch (model) {

ctest/Makefile

@@ -6,6 +6,8 @@ TOPDIR = ..
 include $(TOPDIR)/Makefile.system
 
 override CFLAGS += -DADD$(BU) -DCBLAS
+override TARGET_ARCH=
+override TARGET_MACH=
 
 LIB = $(TOPDIR)/$(LIBNAME)
 

ctest/c_cblat1.f

@@ -577,7 +577,7 @@
       SUBROUTINE STEST1(SCOMP1,STRUE1,SSIZE,SFAC)
 *     ************************* STEST1 *****************************
 *
-*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMODATE THE FORTRAN
+*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMMODATE THE FORTRAN
 *     REQUIREMENT THAT WHEN A DUMMY ARGUMENT IS AN ARRAY, THE
 *     ACTUAL ARGUMENT MUST ALSO BE AN ARRAY OR AN ARRAY ELEMENT.
 *

ctest/c_dblat1.f

@@ -653,7 +653,7 @@
       SUBROUTINE STEST1(SCOMP1,STRUE1,SSIZE,SFAC)
 *     ************************* STEST1 *****************************
 *
-*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMODATE THE FORTRAN
+*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMMODATE THE FORTRAN
 *     REQUIREMENT THAT WHEN A DUMMY ARGUMENT IS AN ARRAY, THE
 *     ACTUAL ARGUMENT MUST ALSO BE AN ARRAY OR AN ARRAY ELEMENT.
 *

ctest/c_sblat1.f

@@ -653,7 +653,7 @@
       SUBROUTINE STEST1(SCOMP1,STRUE1,SSIZE,SFAC)
 *     ************************* STEST1 *****************************
 *
-*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMODATE THE FORTRAN
+*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMMODATE THE FORTRAN
 *     REQUIREMENT THAT WHEN A DUMMY ARGUMENT IS AN ARRAY, THE
 *     ACTUAL ARGUMENT MUST ALSO BE AN ARRAY OR AN ARRAY ELEMENT.
 *

ctest/c_zblat1.f

@@ -577,7 +577,7 @@
       SUBROUTINE STEST1(SCOMP1,STRUE1,SSIZE,SFAC)
 *     ************************* STEST1 *****************************
 *
-*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMODATE THE FORTRAN
+*     THIS IS AN INTERFACE SUBROUTINE TO ACCOMMODATE THE FORTRAN
 *     REQUIREMENT THAT WHEN A DUMMY ARGUMENT IS AN ARRAY, THE
 *     ACTUAL ARGUMENT MUST ALSO BE AN ARRAY OR AN ARRAY ELEMENT.
 *

driver/others/blas_server.c

@@ -109,7 +109,7 @@ extern unsigned int openblas_thread_timeout();
 /* equal to "OMP_NUM_THREADS - 1" and thread only wakes up when     */
 /* jobs is queued.                                                  */
 
-/* We need this grobal for cheking if initialization is finished.   */
+/* We need this global for checking if initialization is finished.  */
 int blas_server_avail   __attribute__((aligned(ATTRIBUTE_SIZE))) = 0;
 
 /* Local Variables */
@@ -150,8 +150,8 @@ static unsigned int thread_timeout = (1U << (THREAD_TIMEOUT));
 
 #ifdef MONITOR
 
-/* Monitor is a function to see thread's status for every seconds. */
-/* Usually it turns off and it's for debugging.                    */
+/* Monitor is a function to see thread's status for every second. */
+/* Usually it turns off and it's for debugging.                   */
 
 static pthread_t      monitor_thread;
 static int main_status[MAX_CPU_NUMBER];

driver/others/blas_server_win32.c

@@ -50,7 +50,7 @@
 
 /* This is a thread implementation for Win32 lazy implementation */
 
-/* Thread server common infomation */
+/* Thread server common information */
 typedef struct{
   CRITICAL_SECTION lock;
   HANDLE filled;
@@ -61,7 +61,7 @@ typedef struct{
 
 } blas_pool_t;
 
-/* We need this global for cheking if initialization is finished.   */
+/* We need this global for checking if initialization is finished.   */
 int blas_server_avail = 0;
 
 /* Local Variables */

driver/others/dynamic.c

@@ -585,9 +585,27 @@ static gotoblas_t *get_coretype(void){
 	  }
         }
         return NULL;  
+      case 7:
+        if (model == 14) {
+	// Ice 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;
+          }
+        }
+        return NULL;  
       case 9:
       case 8:
-	if (model == 14 ) { // Kaby Lake
+	if (model == 14 ) { // Kaby Lake, Coffee Lake
 	  if(support_avx2())
 	    return &gotoblas_HASWELL;
 	  if(support_avx()) {

driver/others/init.c

@@ -765,7 +765,7 @@ int gotoblas_set_affinity(int pos) {
 
   int mynode = 1;
 
-  /* if number of threads is larger than inital condition */
+  /* if number of threads is larger than initial condition */
   if (pos < 0) {
       sched_setaffinity(0, sizeof(cpu_orig_mask), &cpu_orig_mask[0]);
       return 0;
@@ -857,7 +857,14 @@ void gotoblas_affinity_init(void) {
   common -> shmid = pshmid;
 
   if (common -> magic != SH_MAGIC) {
+
+#if defined(__GLIBC_PREREQ)
+#if __GLIBC_PREREQ(2, 7)
     cpu_set_t *cpusetp;
+#else
+    cpu_set_t cpuset;
+#endif
+#endif    
     int nums;
     int ret;
 
@@ -890,7 +897,7 @@ void gotoblas_affinity_init(void) {
     }
     CPU_FREE(cpusetp);
 #else
-    ret = sched_getaffinity(0,sizeof(cpu_set_t), cpusetp);
+    ret = sched_getaffinity(0,sizeof(cpu_set_t), &cpuset);
     if (ret!=0) {
         common->num_procs = nums;
     } else {
@@ -898,11 +905,11 @@ void gotoblas_affinity_init(void) {
     int i;
     int n = 0;
     for (i=0;i<nums;i++)
-        if (CPU_ISSET(i,cpusetp)) n++;
+        if (CPU_ISSET(i,&cpuset)) n++;
     common->num_procs = n;
     }
 #else
-    common->num_procs = CPU_COUNT(sizeof(cpu_set_t),cpusetp);
+    common->num_procs = CPU_COUNT(&cpuset);
     }
 #endif
 

driver/others/memory.c

@@ -229,7 +229,7 @@ int get_num_procs(void) {
   n=0;
   #if !__GLIBC_PREREQ(2, 6)
   for (i=0;i<nums;i++)
-     if (CPU_ISSET(i,cpuset)) n++;
+     if (CPU_ISSET(i,&cpuset)) n++;
   nums=n;
   #else
   nums = CPU_COUNT(sizeof(cpuset),&cpuset);
@@ -1622,6 +1622,7 @@ void gotoblas_dummy_for_PGI(void) {
   gotoblas_init();
   gotoblas_quit();
 
+#if __PGIC__ < 19
 #if 0
   asm ("\t.section\t.ctors,\"aw\",@progbits; .align 8; .quad gotoblas_init; .section .text");
   asm ("\t.section\t.dtors,\"aw\",@progbits; .align 8; .quad gotoblas_quit; .section .text");
@@ -1629,6 +1630,7 @@ void gotoblas_dummy_for_PGI(void) {
   asm (".section .init,\"ax\"; call gotoblas_init@PLT; .section .text");
   asm (".section .fini,\"ax\"; call gotoblas_quit@PLT; .section .text");
 #endif
+#endif
 }
 #endif
 
@@ -1772,7 +1774,7 @@ int get_num_procs(void) {
   n=0;
   #if !__GLIBC_PREREQ(2, 6)
   for (i=0;i<nums;i++)
-     if (CPU_ISSET(i,cpuset)) n++;
+     if (CPU_ISSET(i,&cpuset)) n++;
   nums=n;
   #else
   nums = CPU_COUNT(sizeof(cpuset),&cpuset);
@@ -2039,8 +2041,12 @@ static BLASULONG  alloc_lock = 0UL;
 
 static void alloc_mmap_free(struct release_t *release){
 
+if (!release->address) return;
+
   if (munmap(release -> address, BUFFER_SIZE)) {
-    printf("OpenBLAS : munmap failed\n");
+      int errsv=errno;
+       perror("OpenBLAS : munmap failed:");
+       printf("error code=%d,\trelease->address=%lx\n",errsv,release->address);
   }
 }
 
@@ -2062,15 +2068,21 @@ static void *alloc_mmap(void *address){
   }
 
   if (map_address != (void *)-1) {
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     LOCK_COMMAND(&alloc_lock);
 #endif    
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_mmap_free;
     release_pos ++;
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     UNLOCK_COMMAND(&alloc_lock);
 #endif    
+  } else {
+#ifdef DEBUG  
+        int errsv=errno;
+       perror("OpenBLAS : mmap failed:");
+       printf("error code=%d,\tmap_address=%lx\n",errsv,map_address);
+#endif
   }
 
 #ifdef OS_LINUX
@@ -2214,13 +2226,13 @@ static void *alloc_mmap(void *address){
 #endif
 
   if (map_address != (void *)-1) {
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     LOCK_COMMAND(&alloc_lock);
 #endif
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_mmap_free;
     release_pos ++;
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     UNLOCK_COMMAND(&alloc_lock);
 #endif
   }
@@ -2701,7 +2713,7 @@ void *blas_memory_alloc(int procpos){
 
   position = 0;
 
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   LOCK_COMMAND(&alloc_lock);
 #endif
   do {
@@ -2718,7 +2730,7 @@ void *blas_memory_alloc(int procpos){
     position ++;
 
   } while (position < NUM_BUFFERS);
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   UNLOCK_COMMAND(&alloc_lock);	
 #endif
   goto error;
@@ -2730,7 +2742,7 @@ void *blas_memory_alloc(int procpos){
 #endif
 
   memory[position].used = 1;
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   UNLOCK_COMMAND(&alloc_lock);
 #else
   blas_unlock(&memory[position].lock);	
@@ -2751,7 +2763,7 @@ void *blas_memory_alloc(int procpos){
 
 #ifdef ALLOC_DEVICEDRIVER
 	if ((*func ==  alloc_devicedirver) && (map_address == (void *)-1)) {
-	    fprintf(stderr, "OpenBLAS Warning ... Physically contigous allocation was failed.\n");
+	    fprintf(stderr, "OpenBLAS Warning ... Physically contiguous allocation was failed.\n");
 	}
 #endif
 
@@ -2779,11 +2791,11 @@ void *blas_memory_alloc(int procpos){
 
     } while ((BLASLONG)map_address == -1);
 
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     LOCK_COMMAND(&alloc_lock);
 #endif    
     memory[position].addr = map_address;
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     UNLOCK_COMMAND(&alloc_lock);
 #endif
 
@@ -2839,7 +2851,7 @@ void blas_memory_free(void *free_area){
 #endif
 
   position = 0;
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   LOCK_COMMAND(&alloc_lock);
 #endif
   while ((position < NUM_BUFFERS) && (memory[position].addr != free_area))
@@ -2855,7 +2867,7 @@ void blas_memory_free(void *free_area){
   WMB;
 
   memory[position].used = 0;
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   UNLOCK_COMMAND(&alloc_lock);
 #endif
 
@@ -2872,7 +2884,7 @@ void blas_memory_free(void *free_area){
   for (position = 0; position < NUM_BUFFERS; position++)
     printf("%4ld  %p : %d\n", position, memory[position].addr, memory[position].used);
 #endif
-#if defined(SMP) && !defined(USE_OPENMP)
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   UNLOCK_COMMAND(&alloc_lock);
 #endif
   return;
@@ -2924,7 +2936,7 @@ void blas_shutdown(void){
 
 #if defined(OS_LINUX) && !defined(NO_WARMUP)
 
-#ifdef SMP
+#if defined(SMP) || defined(USE_LOCKING)
 #if   defined(USE_PTHREAD_LOCK)
 static pthread_mutex_t    init_lock = PTHREAD_MUTEX_INITIALIZER;
 #elif defined(USE_PTHREAD_SPINLOCK)
@@ -2949,7 +2961,7 @@ static void _touch_memory(blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n,
     if (hot_alloc != 2) {
 #endif
 
-#ifdef SMP
+#if defined(SMP) || defined(USE_LOCKING)
   LOCK_COMMAND(&init_lock);
 #endif
 
@@ -2959,7 +2971,7 @@ static void _touch_memory(blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n,
     size    -= PAGESIZE;
   }
 
-#ifdef SMP
+#if defined(SMP) || defined(USE_LOCKING)
   UNLOCK_COMMAND(&init_lock);
 #endif
 
@@ -3192,7 +3204,7 @@ void gotoblas_dummy_for_PGI(void) {
 
   gotoblas_init();
   gotoblas_quit();
-
+#if __PGIC__ < 19
 #if 0
   asm ("\t.section\t.ctors,\"aw\",@progbits; .align 8; .quad gotoblas_init; .section .text");
   asm ("\t.section\t.dtors,\"aw\",@progbits; .align 8; .quad gotoblas_quit; .section .text");
@@ -3200,6 +3212,7 @@ void gotoblas_dummy_for_PGI(void) {
   asm (".section .init,\"ax\"; call gotoblas_init@PLT; .section .text");
   asm (".section .fini,\"ax\"; call gotoblas_quit@PLT; .section .text");
 #endif
+#endif	
 }
 #endif
 

dynamic.c

@@ -0,0 +1,897 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#include "common.h"
+
+#ifdef _MSC_VER
+#define strncasecmp _strnicmp
+#define strcasecmp _stricmp
+#endif
+
+#ifdef ARCH_X86
+#define EXTERN extern
+#else
+#define EXTERN
+#endif
+
+#ifdef DYNAMIC_LIST
+extern gotoblas_t gotoblas_PRESCOTT;
+
+#ifdef DYN_ATHLON
+extern gotoblas_t gotoblas_ATHLON;
+#else
+#define gotoblas_ATHLON gotoblas_PRESCOTT
+#endif
+#ifdef DYN_KATMAI
+extern gotoblas_t gotoblas_KATMAI;
+#else
+#define gotoblas_KATMAI gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BANIAS
+extern gotoblas_t gotoblas_BANIAS;
+#else
+#define gotoblas_BANIAS gotoblas_PRESCOTT
+#endif
+#ifdef DYN_COPPERMINE
+extern gotoblas_t gotoblas_COPPERMINE;
+#else
+#define gotoblas_COPPERMINE gotoblas_PRESCOTT
+#endif
+#ifdef DYN_NORTHWOOD
+extern gotoblas_t gotoblas_NORTHWOOD;
+#else
+#define gotoblas_NORTHWOOD gotoblas_PRESCOTT
+#endif
+#ifdef DYN_CORE2
+extern gotoblas_t gotoblas_CORE2;
+#else
+#define gotoblas_CORE2 gotoblas_PRESCOTT
+#endif
+#ifdef DYN_NEHALEM
+extern gotoblas_t gotoblas_NEHALEM;
+#else
+#define gotoblas_NEHALEM gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BARCELONA
+extern gotoblas_t gotoblas_BARCELONA;
+#elif defined(DYN_NEHALEM)
+#define gotoblas_BARCELONA gotoblas_NEHALEM
+#else
+#define gotoblas_BARCELONA gotoblas_PRESCOTT
+#endif
+#ifdef DYN_ATOM
+extern gotoblas_t gotoblas_ATOM;
+elif defined(DYN_NEHALEM)
+#define gotoblas_ATOM gotoblas_NEHALEM
+#else
+#define gotoblas_ATOM gotoblas_PRESCOTT
+#endif
+#ifdef DYN_NANO
+extern gotoblas_t gotoblas_NANO;
+#else
+#define gotoblas_NANO gotoblas_PRESCOTT
+#endif
+#ifdef DYN_PENRYN
+extern gotoblas_t gotoblas_PENRYN;
+#else
+#define gotoblas_PENRYN gotoblas_PRESCOTT
+#endif
+#ifdef DYN_DUNNINGTON
+extern gotoblas_t gotoblas_DUNNINGTON;
+#else
+#define gotoblas_DUNNINGTON gotoblas_PRESCOTT
+#endif
+#ifdef DYN_OPTERON
+extern gotoblas_t gotoblas_OPTERON;
+#else
+#define gotoblas_OPTERON gotoblas_PRESCOTT
+#endif
+#ifdef DYN_OPTERON_SSE3
+extern gotoblas_t gotoblas_OPTERON_SSE3;
+#else
+#define gotoblas_OPTERON_SSE3 gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BOBCAT
+extern gotoblas_t gotoblas_BOBCAT;
+#elif defined(DYN_NEHALEM)
+#define gotoblas_BOBCAT gotoblas_NEHALEM
+#else
+#define gotoblas_BOBCAT gotoblas_PRESCOTT
+#endif
+#ifdef DYN_SANDYBRIDGE
+extern gotoblas_t gotoblas_SANDYBRIDGE;
+#elif defined(DYN_NEHALEM)
+#define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
+#else
+#define gotoblas_SANDYBRIDGE gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BULLDOZER
+extern gotoblas_t gotoblas_BULLDOZER;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_BULLDOZER gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_BULLDOZER gotoblas_NEHALEM
+#else
+#define gotoblas_BULLDOZER gotoblas_PRESCOTT
+#endif
+#ifdef DYN_PILEDRIVER
+extern gotoblas_t gotoblas_PILEDRIVER;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_PILEDRIVER gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_PILEDRIVER gotoblas_NEHALEM
+#else
+#define gotoblas_PILEDRIVER gotoblas_PRESCOTT
+#endif
+#ifdef DYN_STEAMROLLER
+extern gotoblas_t gotoblas_STEAMROLLER;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_STEAMROLLER gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_STEAMROLLER gotoblas_NEHALEM
+#else
+#define gotoblas_STEAMROLLER gotoblas_PRESCOTT
+#endif
+#ifdef DYN_EXCAVATOR
+extern gotoblas_t gotoblas_EXCAVATOR;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_EXCAVATOR gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_EXCAVATOR gotoblas_NEHALEM
+#else
+#define gotoblas_EXCAVATOR gotoblas_PRESCOTT
+#endif
+#ifdef DYN_HASWELL
+extern gotoblas_t gotoblas_HASWELL;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_HASWELL gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_HASWELL gotoblas_NEHALEM
+#else
+#define gotoblas_HASWELL gotoblas_PRESCOTT
+#endif
+#ifdef DYN_ZEN
+extern gotoblas_t gotoblas_ZEN;
+#elif defined(DYN_HASWELL)
+#define gotoblas_ZEN gotoblas_HASWELL
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_ZEN gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_ZEN gotoblas_NEHALEM
+#else
+#define gotoblas_ZEN gotoblas_PRESCOTT
+#endif
+#ifdef DYN_SKYLAKEX
+extern gotoblas_t gotoblas_SKYLAKEX;
+#elif defined(DYN_HASWELL)
+#define gotoblas_SKYLAKEX gotoblas_HASWELL
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_SKYLAKEX gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_SKYLAKEX gotoblas_NEHALEM
+#else
+#define gotoblas_SKYLAKEX gotoblas_PRESCOTT
+#endif
+
+
+#else // not DYNAMIC_LIST
+EXTERN gotoblas_t  gotoblas_KATMAI;
+EXTERN gotoblas_t  gotoblas_COPPERMINE;
+EXTERN gotoblas_t  gotoblas_NORTHWOOD;
+EXTERN gotoblas_t  gotoblas_BANIAS;
+EXTERN gotoblas_t  gotoblas_ATHLON;
+
+extern gotoblas_t  gotoblas_PRESCOTT;
+extern gotoblas_t  gotoblas_CORE2;
+extern gotoblas_t  gotoblas_NEHALEM;
+extern gotoblas_t  gotoblas_BARCELONA;
+#ifdef DYNAMIC_OLDER
+extern gotoblas_t  gotoblas_ATOM;
+extern gotoblas_t  gotoblas_NANO;
+extern gotoblas_t  gotoblas_PENRYN;
+extern gotoblas_t  gotoblas_DUNNINGTON;
+extern gotoblas_t  gotoblas_OPTERON;
+extern gotoblas_t  gotoblas_OPTERON_SSE3;
+extern gotoblas_t  gotoblas_BOBCAT;
+#else
+#define gotoblas_ATOM gotoblas_NEHALEM
+#define gotoblas_NANO gotoblas_NEHALEM
+#define gotoblas_PENRYN gotoblas_CORE2
+#define gotoblas_DUNNINGTON gotoblas_CORE2
+#define gotoblas_OPTERON gotoblas_CORE2
+#define gotoblas_OPTERON_SSE3 gotoblas_CORE2
+#define gotoblas_BOBCAT gotoblas_CORE2
+#endif
+
+#ifndef NO_AVX
+extern gotoblas_t  gotoblas_SANDYBRIDGE;
+extern gotoblas_t  gotoblas_BULLDOZER;
+extern gotoblas_t  gotoblas_PILEDRIVER;
+extern gotoblas_t  gotoblas_STEAMROLLER;
+extern gotoblas_t  gotoblas_EXCAVATOR;
+#ifdef NO_AVX2
+#define gotoblas_HASWELL gotoblas_SANDYBRIDGE
+#define gotoblas_SKYLAKEX gotoblas_SANDYBRIDGE
+#define gotoblas_ZEN gotoblas_SANDYBRIDGE
+#else
+extern gotoblas_t  gotoblas_HASWELL;
+extern gotoblas_t  gotoblas_ZEN;
+#ifndef NO_AVX512
+extern gotoblas_t  gotoblas_SKYLAKEX;
+#else
+#define gotoblas_SKYLAKEX gotoblas_HASWELL
+#endif
+#endif
+#else
+//Use NEHALEM kernels for sandy bridge
+#define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
+#define gotoblas_HASWELL gotoblas_NEHALEM
+#define gotoblas_SKYLAKEX gotoblas_NEHALEM
+#define gotoblas_BULLDOZER gotoblas_BARCELONA
+#define gotoblas_PILEDRIVER gotoblas_BARCELONA
+#define gotoblas_STEAMROLLER gotoblas_BARCELONA
+#define gotoblas_EXCAVATOR gotoblas_BARCELONA
+#define gotoblas_ZEN gotoblas_BARCELONA
+#endif
+
+#endif // DYNAMIC_LIST
+
+#define VENDOR_INTEL      1
+#define VENDOR_AMD        2
+#define VENDOR_CENTAUR    3
+#define VENDOR_HYGON	  4
+#define VENDOR_UNKNOWN   99
+
+#define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
+
+#ifndef NO_AVX
+static inline void xgetbv(int op, int * eax, int * edx){
+  //Use binary code for xgetbv
+  __asm__ __volatile__
+    (".byte 0x0f, 0x01, 0xd0": "=a" (*eax), "=d" (*edx) : "c" (op) : "cc");
+}
+#endif
+
+int support_avx(){
+#ifndef NO_AVX
+  int eax, ebx, ecx, edx;
+  int ret=0;
+
+  cpuid(1, &eax, &ebx, &ecx, &edx);
+  if ((ecx & (1 << 28)) != 0 && (ecx & (1 << 27)) != 0 && (ecx & (1 << 26)) != 0){
+    xgetbv(0, &eax, &edx);
+    if((eax & 6) == 6){
+      ret=1;  //OS support AVX
+    }
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
+int support_avx2(){
+#ifndef NO_AVX2
+  int eax, ebx, ecx=0, edx;
+  int ret=0;
+
+  if (!support_avx())
+    return 0;
+  cpuid(7, &eax, &ebx, &ecx, &edx);
+  if((ebx & (1<<7)) != 0)
+      ret=1;  //OS supports AVX2
+  return ret;
+#else
+  return 0;
+#endif
+}
+
+int support_avx512(){
+#if !defined(NO_AVX) && !defined(NO_AVX512)
+  int eax, ebx, ecx, edx;
+  int ret=0;
+
+  if (!support_avx())
+    return 0;
+  cpuid(7, &eax, &ebx, &ecx, &edx);
+  if((ebx & (1<<7)) != 1){
+      ret=0;  //OS does not even support AVX2
+  }
+  if((ebx & (1<<31)) != 0){
+    xgetbv(0, &eax, &edx);
+    if((eax & 0xe0) == 0xe0)
+      ret=1;  //OS supports AVX512VL
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
+extern void openblas_warning(int verbose, const char * msg);
+#define FALLBACK_VERBOSE 1
+#define NEHALEM_FALLBACK "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n"
+#define SANDYBRIDGE_FALLBACK "OpenBLAS : Your OS does not support AVX2 instructions. OpenBLAS is using Sandybridge kernels as a fallback, which may give poorer performance.\n"
+#define HASWELL_FALLBACK "OpenBLAS : Your OS does not support AVX512VL instructions. OpenBLAS is using Haswell kernels as a fallback, which may give poorer performance.\n"
+#define BARCELONA_FALLBACK "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Barcelona kernels as a fallback, which may give poorer performance.\n"
+
+static int get_vendor(void){
+  int eax, ebx, ecx, edx;
+
+  union
+  {
+        char vchar[16];
+        int  vint[4];
+  } vendor;
+
+  cpuid(0, &eax, &ebx, &ecx, &edx);
+
+  *(&vendor.vint[0]) = ebx;
+  *(&vendor.vint[1]) = edx;
+  *(&vendor.vint[2]) = ecx;
+
+  vendor.vchar[12] = '\0';
+
+  if (!strcmp(vendor.vchar, "GenuineIntel")) return VENDOR_INTEL;
+  if (!strcmp(vendor.vchar, "AuthenticAMD")) return VENDOR_AMD;
+  if (!strcmp(vendor.vchar, "CentaurHauls")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor.vchar, "HygonGenuine")) return VENDOR_HYGON;
+
+  if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
+
+  return VENDOR_UNKNOWN;
+}
+
+static gotoblas_t *get_coretype(void){
+
+  int eax, ebx, ecx, edx;
+  int family, exfamily, model, vendor, exmodel;
+
+  cpuid(1, &eax, &ebx, &ecx, &edx);
+
+  family   = BITMASK(eax,  8, 0x0f);
+  exfamily = BITMASK(eax, 20, 0xff);
+  model    = BITMASK(eax,  4, 0x0f);
+  exmodel  = BITMASK(eax, 16, 0x0f);
+
+  vendor = get_vendor();
+
+  if (vendor == VENDOR_INTEL){
+    switch (family) {
+    case 0x6:
+      switch (exmodel) {
+      case 0:
+	if (model <= 0x7) return &gotoblas_KATMAI;
+	if ((model == 0x8) || (model == 0xa) || (model == 0xb)) return &gotoblas_COPPERMINE;
+	if ((model == 0x9) || (model == 0xd)) return &gotoblas_BANIAS;
+	if (model == 14) return &gotoblas_BANIAS;
+	if (model == 15) return &gotoblas_CORE2;
+	return NULL;
+
+      case 1:
+	if (model == 6) return &gotoblas_CORE2;
+	if (model == 7) return &gotoblas_PENRYN;
+	if (model == 13) return &gotoblas_DUNNINGTON;
+	if ((model == 10) || (model == 11) || (model == 14) || (model == 15)) return &gotoblas_NEHALEM;
+	if (model == 12) return &gotoblas_ATOM;
+	return NULL;
+
+      case 2:
+	//Intel Core (Clarkdale) / Core (Arrandale)
+	// Pentium (Clarkdale) / Pentium Mobile (Arrandale)
+	// Xeon (Clarkdale), 32nm
+	if (model ==  5) return &gotoblas_NEHALEM;
+
+	//Intel Xeon Processor 5600 (Westmere-EP)
+	//Xeon Processor E7 (Westmere-EX)
+	//Xeon E7540
+	if (model == 12 || model == 14 || model == 15) return &gotoblas_NEHALEM;
+
+	//Intel Core i5-2000 /i7-2000 (Sandy Bridge)
+	//Intel Core i7-3000 / Xeon E5
+	if (model == 10 || model == 13) {
+	  if(support_avx())
+	    return &gotoblas_SANDYBRIDGE;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	return NULL;
+      case 3:
+	//Intel Sandy Bridge 22nm (Ivy Bridge?)
+	if (model == 10 || model == 14) {
+	  if(support_avx())
+	    return &gotoblas_SANDYBRIDGE;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	//Intel Haswell
+	if (model == 12 || model == 15) {
+	  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.
+	  }
+	}
+	//Intel Broadwell
+	if (model == 13) {
+	  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.
+	  }
+	}
+	if (model == 7) return &gotoblas_ATOM; //Bay Trail	
+	return NULL;
+      case 4:
+		//Intel Haswell
+	if (model == 5 || model == 6) {
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	//Intel Broadwell
+	if (model == 7 || model == 15) {
+	  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.
+	  }
+	}
+	//Intel Skylake
+	if (model == 14) {
+	  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.
+	  }
+	}
+	//Intel Braswell / Avoton
+	if (model == 12 || model == 13) { 
+	  return &gotoblas_NEHALEM;
+	}	
+	return NULL;
+      case 5:
+	//Intel Broadwell
+	if (model == 6) {
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	if (model == 5) {	
+	// Intel Skylake X
+          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;
+          }
+	}
+	//Intel Skylake
+	if (model == 14) {
+	  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.
+	  }
+	}
+	//Intel Phi Knights Landing
+	if (model == 7) {
+	  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; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	//Apollo Lake or Denverton
+	if (model == 12 || model == 15) { 
+	  return &gotoblas_NEHALEM;
+	}	
+	return NULL;
+      case 6:
+        if (model == 6) {
+          // Cannon Lake
+	  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;
+	  }
+        }
+        return NULL;  
+      case 7:
+        if (model == 10) // Goldmont plus 
+           return &gotoblas_NEHALEM;
+        if (model == 14) {
+	// Ice 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;
+          }
+        }
+        return NULL;  
+      case 9:
+      case 8:
+	if (model == 14 ) { // Kaby Lake, Coffee Lake
+	  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:
+      if (model <= 0x2) return &gotoblas_NORTHWOOD;
+      return &gotoblas_PRESCOTT;
+    }
+  }
+
+  if (vendor == VENDOR_AMD || vendor == VENDOR_HYGON){
+    if (family <= 0xe) {
+        // Verify that CPU has 3dnow and 3dnowext before claiming it is Athlon
+        cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
+        if ( (eax & 0xffff)  >= 0x01) {
+            cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
+            if ((edx & (1 << 30)) == 0 || (edx & (1 << 31)) == 0)
+              return NULL;
+          }
+        else
+          return NULL;
+
+        return &gotoblas_ATHLON;
+      }
+    if (family == 0xf){
+      if ((exfamily == 0) || (exfamily == 2)) {
+	if (ecx & (1 <<  0)) return &gotoblas_OPTERON_SSE3;
+	else return &gotoblas_OPTERON;
+      }  else if (exfamily == 5) {
+	return &gotoblas_BOBCAT;
+      } else if (exfamily == 6) {
+	if(model == 1){
+	  //AMD Bulldozer Opteron 6200 / Opteron 4200 / AMD FX-Series
+	  if(support_avx())
+	    return &gotoblas_BULLDOZER;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	  }
+	}else if(model == 2 || model == 3){
+	  //AMD Bulldozer Opteron 6300 / Opteron 4300 / Opteron 3300
+	  if(support_avx())
+	    return &gotoblas_PILEDRIVER;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	  }
+	}else if(model == 5){
+	  if(support_avx())
+	    return &gotoblas_EXCAVATOR;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	  }
+	}else if(model == 0 || model == 8){
+	  if (exmodel == 1) {
+	    //AMD Trinity
+	    if(support_avx())
+	      return &gotoblas_PILEDRIVER;
+	    else{
+	      openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	      return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	    }
+	   }else if (exmodel == 3) {
+	    //AMD STEAMROLLER
+	    if(support_avx())
+	      return &gotoblas_STEAMROLLER;
+	    else{
+	      openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	      return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	    }
+	  }else if (exmodel == 6) {
+	    if(support_avx())
+	      return &gotoblas_EXCAVATOR;
+	    else{
+	      openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	      return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	    }
+
+	  }
+	}
+      } else if (exfamily == 8) {
+	if (model == 1 || model == 8) {
+	  if(support_avx())
+	    return &gotoblas_ZEN;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+      } else if (exfamily == 9) {
+	  if(support_avx())
+	    return &gotoblas_ZEN;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+        }
+      }else {
+	return &gotoblas_BARCELONA;
+      }
+    }
+  }
+
+  if (vendor == VENDOR_CENTAUR) {
+    switch (family) {
+    case 0x6:
+      return &gotoblas_NANO;
+    }
+  }
+
+  return NULL;
+}
+
+static char *corename[] = {
+    "Unknown",
+    "Katmai",
+    "Coppermine",
+    "Northwood",
+    "Prescott",
+    "Banias",
+    "Atom",
+    "Core2",
+    "Penryn",
+    "Dunnington",
+    "Nehalem",
+    "Athlon",
+    "Opteron",
+    "Opteron_SSE3",
+    "Barcelona",
+    "Nano",
+    "Sandybridge",
+    "Bobcat",
+    "Bulldozer",
+    "Piledriver",
+    "Haswell",
+    "Steamroller",
+    "Excavator",
+    "Zen",
+    "SkylakeX"	
+};
+
+char *gotoblas_corename(void) {
+
+  if (gotoblas == &gotoblas_KATMAI)       return corename[ 1];
+  if (gotoblas == &gotoblas_COPPERMINE)   return corename[ 2];
+  if (gotoblas == &gotoblas_NORTHWOOD)    return corename[ 3];
+  if (gotoblas == &gotoblas_PRESCOTT)     return corename[ 4];
+  if (gotoblas == &gotoblas_BANIAS)       return corename[ 5];
+  if (gotoblas == &gotoblas_ATOM)         return corename[ 6];
+  if (gotoblas == &gotoblas_CORE2)        return corename[ 7];
+  if (gotoblas == &gotoblas_PENRYN)       return corename[ 8];
+  if (gotoblas == &gotoblas_DUNNINGTON)   return corename[ 9];
+  if (gotoblas == &gotoblas_NEHALEM)      return corename[10];
+  if (gotoblas == &gotoblas_ATHLON)       return corename[11];
+  if (gotoblas == &gotoblas_OPTERON_SSE3) return corename[12];
+  if (gotoblas == &gotoblas_OPTERON)      return corename[13];
+  if (gotoblas == &gotoblas_BARCELONA)    return corename[14];
+  if (gotoblas == &gotoblas_NANO)         return corename[15];
+  if (gotoblas == &gotoblas_SANDYBRIDGE)  return corename[16];
+  if (gotoblas == &gotoblas_BOBCAT)       return corename[17];
+  if (gotoblas == &gotoblas_BULLDOZER)    return corename[18];
+  if (gotoblas == &gotoblas_PILEDRIVER)   return corename[19];
+  if (gotoblas == &gotoblas_HASWELL)      return corename[20];
+  if (gotoblas == &gotoblas_STEAMROLLER)  return corename[21];
+  if (gotoblas == &gotoblas_EXCAVATOR)    return corename[22];
+  if (gotoblas == &gotoblas_ZEN)          return corename[23];
+  if (gotoblas == &gotoblas_SKYLAKEX)     return corename[24];
+  return corename[0];
+}
+
+
+static gotoblas_t *force_coretype(char *coretype){
+
+	int i ;
+	int found = -1;
+	char message[128];
+	//char mname[20];
+
+	for ( i=1 ; i <= 24; i++)
+	{
+		if (!strncasecmp(coretype,corename[i],20))
+		{
+			found = i;
+			break;
+		}
+	}
+	if (found < 0)
+	{
+	        //strncpy(mname,coretype,20);
+	        snprintf(message, 128, "Core not found: %s\n",coretype);
+    		openblas_warning(1, message);
+		return(NULL);
+	}
+
+	switch (found)
+	{
+		case 24: return (&gotoblas_SKYLAKEX);	
+		case 23: return (&gotoblas_ZEN);
+		case 22: return (&gotoblas_EXCAVATOR);
+		case 21: return (&gotoblas_STEAMROLLER);
+		case 20: return (&gotoblas_HASWELL);
+		case 19: return (&gotoblas_PILEDRIVER);
+		case 18: return (&gotoblas_BULLDOZER);
+		case 17: return (&gotoblas_BOBCAT);
+		case 16: return (&gotoblas_SANDYBRIDGE);
+		case 15: return (&gotoblas_NANO);
+		case 14: return (&gotoblas_BARCELONA);
+		case 13: return (&gotoblas_OPTERON);
+		case 12: return (&gotoblas_OPTERON_SSE3);
+		case 11: return (&gotoblas_ATHLON);
+		case 10: return (&gotoblas_NEHALEM);
+		case  9: return (&gotoblas_DUNNINGTON);
+		case  8: return (&gotoblas_PENRYN);
+		case  7: return (&gotoblas_CORE2);
+		case  6: return (&gotoblas_ATOM);
+		case  5: return (&gotoblas_BANIAS);
+		case  4: return (&gotoblas_PRESCOTT);
+		case  3: return (&gotoblas_NORTHWOOD);
+		case  2: return (&gotoblas_COPPERMINE);
+		case  1: return (&gotoblas_KATMAI);
+	}
+	return(NULL);
+
+}
+
+
+
+
+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();
+  }
+
+#ifdef ARCH_X86
+  if (gotoblas == NULL) gotoblas = &gotoblas_KATMAI;
+#else
+  if (gotoblas == NULL) gotoblas = &gotoblas_PRESCOTT;
+  /* sanity check, if 64bit pointer we can't have a 32 bit cpu */
+  if (sizeof(void*) == 8) {
+      if (gotoblas == &gotoblas_KATMAI ||
+          gotoblas == &gotoblas_COPPERMINE ||
+          gotoblas == &gotoblas_NORTHWOOD ||
+          gotoblas == &gotoblas_BANIAS ||
+          gotoblas == &gotoblas_ATHLON)
+          gotoblas = &gotoblas_PRESCOTT;
+  }
+#endif
+
+  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;
+
+}

exports/Makefile

@@ -105,6 +105,10 @@ $(LIBPREFIX).def : gensymbol
 libgoto_hpl.def : gensymbol
 	perl ./gensymbol win2khpl $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
 
+ifeq ($(OSNAME), Darwin)
+INTERNALNAME = $(LIBPREFIX).$(MAJOR_VERSION).dylib
+endif
+
 ifeq (, $(SYMBOLPREFIX)$(SYMBOLSUFFIX))
 $(LIBDYNNAME) : ../$(LIBNAME) osx.def
 else
@@ -114,9 +118,9 @@ $(LIBDYNNAME) : ../$(LIBNAME).osx.renamed osx.def
 endif
 ifneq (,$(filter 1 2,$(NOFORTRAN)))
 #only build without Fortran
-	$(CC) $(CFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 else
-	$(FC) $(FFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(FC) $(FFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 endif
 
 dllinit.$(SUFFIX) : dllinit.c

f_check

@@ -125,7 +125,7 @@ if ($compiler eq "") {
 	    $openmp = "-openmp";
 	}
 
-	# for embeded underscore name, e.g. zho_ge, it may append 2 underscores.
+	# 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 =~ / zho_ge__/) {
 	    $need2bu       = 1;

interface/CMakeLists.txt

@@ -24,7 +24,7 @@ set(BLAS1_MANGLED_SOURCES
   axpby.c
 )
 
-# TODO: USE_NETLIB_GEMV shoudl switch gemv.c to netlib/*gemv.f
+# TODO: USE_NETLIB_GEMV should switch gemv.c to netlib/*gemv.f
 # these all have 'z' sources for complex versions
 set(BLAS2_SOURCES
   gemv.c ger.c

interface/axpy.c

@@ -91,7 +91,7 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint inc
   //disable multi-thread when incx==0 or incy==0
   //In that case, the threads would be dependent.
   //
-  //Temporarily work-around the low performance issue with small imput size &
+  //Temporarily work-around the low performance issue with small input size &
   //multithreads.
   if (incx == 0 || incy == 0 || n <= MULTI_THREAD_MINIMAL)
 	  nthreads = 1;

interface/zaxpy.c

@@ -99,7 +99,7 @@ void CNAME(blasint n, FLOAT *ALPHA, FLOAT *x, blasint incx, FLOAT *y, blasint in
   //disable multi-thread when incx==0 or incy==0
   //In that case, the threads would be dependent.
   //
-  //Temporarily work-around the low performance issue with small imput size &
+  //Temporarily work-around the low performance issue with small input size &
   //multithreads.
   if (incx == 0 || incy == 0 || n <= MULTI_THREAD_MINIMAL)
 	  nthreads = 1;

kernel/arm/KERNEL.ARMV6

@@ -1,30 +1,30 @@
 include $(KERNELDIR)/KERNEL.ARMV5
 
-SAMAXKERNEL  = iamax_vfp.S
-DAMAXKERNEL  = iamax_vfp.S
-CAMAXKERNEL  = iamax_vfp.S
-ZAMAXKERNEL  = iamax_vfp.S
+SAMAXKERNEL  = amax_vfp.S
+DAMAXKERNEL  = amax_vfp.S
+#CAMAXKERNEL  = amax_vfp.S
+#ZAMAXKERNEL  = amax_vfp.S
 
-SAMINKERNEL  = iamax_vfp.S
-DAMINKERNEL  = iamax_vfp.S
-CAMINKERNEL  = iamax_vfp.S
-ZAMINKERNEL  = iamax_vfp.S
+SAMINKERNEL  = amax_vfp.S
+DAMINKERNEL  = amax_vfp.S
+#CAMINKERNEL  = amax_vfp.S
+#ZAMINKERNEL  = amax_vfp.S
 
-SMAXKERNEL   = iamax_vfp.S
-DMAXKERNEL   = iamax_vfp.S
+SMAXKERNEL   = amax_vfp.S
+DMAXKERNEL   = amax_vfp.S
 
-SMINKERNEL   = iamax_vfp.S
-DMINKERNEL   = iamax_vfp.S
+SMINKERNEL   = amax_vfp.S
+DMINKERNEL   = amax_vfp.S
 
 ISAMAXKERNEL = iamax_vfp.S
 IDAMAXKERNEL = iamax_vfp.S
-ICAMAXKERNEL = iamax_vfp.S
-IZAMAXKERNEL = iamax_vfp.S
+#ICAMAXKERNEL = iamax_vfp.S
+#IZAMAXKERNEL = iamax_vfp.S
 
 ISAMINKERNEL = iamax_vfp.S
 IDAMINKERNEL = iamax_vfp.S
-ICAMINKERNEL = iamax_vfp.S
-IZAMINKERNEL = iamax_vfp.S
+#ICAMINKERNEL = iamax_vfp.S
+#IZAMINKERNEL = iamax_vfp.S
 
 ISMAXKERNEL  = iamax_vfp.S
 IDMAXKERNEL  = iamax_vfp.S

kernel/arm/amax_vfp.S

@@ -0,0 +1,445 @@
+/***************************************************************************
+Copyright (c) 2013, 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.
+*****************************************************************************/
+
+/**************************************************************************************
+* 2013/11/14 Saar
+* 	 BLASTEST 		: OK
+* 	 CTEST			: OK
+* 	 TEST			: OK
+*
+**************************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define STACKSIZE 256
+
+#define	N	r0
+#define	X	r1
+#define	INC_X	r2
+
+#define I	r12
+
+#define X_PRE	512
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+
+#if	defined(USE_ABS)
+
+#if	defined(DOUBLE)
+
+#define	VABS(x0,x1)	vabs.f64	x0, x1
+
+#else
+
+#define	VABS(x0,x1)	vabs.f32	x0, x1
+
+#endif
+
+#else
+
+#define VABS(x0,x1)	nop
+
+#endif
+
+/*****************************************************************************************/
+
+#if	defined(USE_MIN)
+
+#define	MOVCOND		movlt
+
+#if	defined(DOUBLE)
+
+#define	VMOVCOND	vmovlt.f64
+
+#else
+
+#define	VMOVCOND	vmovlt.f32
+
+#endif
+
+#else
+
+#define	MOVCOND		movgt
+
+#if	defined(DOUBLE)
+
+#define	VMOVCOND	vmovgt.f64
+
+#else
+
+#define	VMOVCOND	vmovgt.f32
+
+#endif
+
+
+#endif
+
+
+/*****************************************************************************************/
+
+
+
+#if	!defined(COMPLEX)
+
+#if	defined(DOUBLE)
+
+.macro INIT_F
+
+	vldmia.f64	X!, { d0 }
+	VABS(   d0,  d0 )
+
+.endm
+
+.macro KERNEL_F1
+
+	vldmia.f64	X!, { d4 }
+	VABS(   d4,  d4 )
+	vcmpe.f64  	d4,  d0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	d0,  d4
+
+.endm
+
+.macro INIT_S
+
+	vldmia.f64	X, { d0 }
+	VABS(   d0,  d0 )
+	add	X, X, INC_X
+
+.endm
+
+
+.macro KERNEL_S1
+
+	vldmia.f64	X, { d4 }
+	VABS(   d4,  d4 )
+	vcmpe.f64  	d4,  d0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	d0,  d4
+	add	X, X, INC_X
+
+.endm
+
+#else
+
+.macro INIT_F
+
+	vldmia.f32	X!, { s0 }
+	VABS(   s0,  s0 )
+
+.endm
+
+.macro KERNEL_F1
+
+	vldmia.f32	X!, { s4 }
+	VABS(   s4,  s4 )
+	vcmpe.f32  	s4,  s0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	s0,  s4
+
+.endm
+
+.macro INIT_S
+
+	vldmia.f32	X, { s0 }
+	VABS(   s0,  s0 )
+	add	X, X, INC_X
+
+.endm
+
+
+.macro KERNEL_S1
+
+	vldmia.f32	X, { s4 }
+	VABS(   s4,  s4 )
+	vcmpe.f32  	s4,  s0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	s0,  s4
+	add	X, X, INC_X
+
+.endm
+
+
+
+
+#endif
+
+#else
+
+#if	defined(DOUBLE)
+
+.macro INIT_F
+
+	vldmia.f64	X!, { d0 -d1 }
+	vabs.f64   d0,  d0
+	vabs.f64   d1,  d1
+	vadd.f64   d0  , d0,  d1
+.endm
+
+
+.macro KERNEL_F1
+
+	vldmia.f64	X!, { d4 - d5 }
+	vabs.f64   d4,  d4
+	vabs.f64   d5,  d5
+	vadd.f64   d4  , d4,  d5
+	vcmpe.f64  	d4,  d0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	d0,  d4
+
+.endm
+
+.macro INIT_S
+
+	vldmia.f64	X, { d0 -d1 }
+	vabs.f64   d0,  d0
+	vabs.f64   d1,  d1
+	vadd.f64   d0  , d0,  d1
+	add	X, X, INC_X
+
+.endm
+
+
+
+.macro KERNEL_S1
+
+	vldmia.f64	X, { d4 - d5 }
+	vabs.f64   d4,  d4
+	vabs.f64   d5,  d5
+	vadd.f64   d4  , d4,  d5
+	vcmpe.f64  	d4,  d0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	d0,  d4
+	add	X, X, INC_X
+
+.endm
+
+#else
+
+.macro INIT_F
+
+	vldmia.f32	X!, { s0 -s1 }
+	vabs.f32   s0,  s0
+	vabs.f32   s1,  s1
+	vadd.f32   s0  , s0,  s1
+
+.endm
+
+
+.macro KERNEL_F1
+
+	vldmia.f32	X!, { s4 - s5 }
+	vabs.f32   s4,  s4
+	vabs.f32   s5,  s5
+	vadd.f32   s4  , s4,  s5
+	vcmpe.f32  	s4,  s0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	s0,  s4
+
+.endm
+
+.macro INIT_S
+
+	vldmia.f32	X, { s0 -s1 }
+	vabs.f32   s0,  s0
+	vabs.f32   s1,  s1
+	vadd.f32   s0  , s0,  s1
+	add	X, X, INC_X
+
+.endm
+
+
+
+.macro KERNEL_S1
+
+	vldmia.f32	X, { s4 - s5 }
+	vabs.f32   s4,  s4
+	vabs.f32   s5,  s5
+	vadd.f32   s4  , s4,  s5
+	vcmpe.f32  	s4,  s0
+	vmrs		APSR_nzcv, fpscr
+	VMOVCOND	s0,  s4
+	add	X, X, INC_X
+
+.endm
+
+
+
+
+#endif
+
+#endif
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+
+        movs    r12, #0                                          // clear floating point register
+        vmov    s0, r12
+#if     defined(DOUBLE)
+        vcvt.f64.f32    d0, s0
+#endif
+
+
+	cmp	N, #0
+	ble	amax_kernel_L999
+
+	cmp	INC_X, #0
+	beq	amax_kernel_L999
+
+
+	cmp	INC_X, #1
+	bne	amax_kernel_S_BEGIN
+
+
+amax_kernel_F_BEGIN:
+
+	INIT_F
+
+	subs	N, N , #1
+	ble	amax_kernel_L999
+
+	asrs	I, N, #2					// I = N / 4
+	ble	amax_kernel_F1
+
+	.align 5
+
+amax_kernel_F4:
+
+	pld	[ X, #X_PRE ]
+	KERNEL_F1
+	KERNEL_F1
+#if defined(COMPLEX) && defined(DOUBLE)
+	pld	[ X, #X_PRE ]
+#endif
+	KERNEL_F1
+	KERNEL_F1
+
+	subs	I, I, #1
+	ble	amax_kernel_F1
+
+
+#if defined(COMPLEX) || defined(DOUBLE)
+	pld	[ X, #X_PRE ]
+#endif
+	KERNEL_F1
+	KERNEL_F1
+#if defined(COMPLEX) && defined(DOUBLE)
+	pld	[ X, #X_PRE ]
+#endif
+	KERNEL_F1
+	KERNEL_F1
+
+	subs	I, I, #1
+	bne	amax_kernel_F4
+
+amax_kernel_F1:
+
+	ands	I, N, #3
+	ble	amax_kernel_L999
+
+amax_kernel_F10:
+
+	KERNEL_F1
+
+	subs    I, I, #1
+        bne     amax_kernel_F10
+
+	b	amax_kernel_L999
+
+amax_kernel_S_BEGIN:
+
+#if defined(COMPLEX)
+
+#if defined(DOUBLE)
+	lsl	INC_X, INC_X, #4				// INC_X * SIZE * 2
+#else
+	lsl	INC_X, INC_X, #3				// INC_X * SIZE * 2
+#endif
+
+#else
+
+#if defined(DOUBLE)
+	lsl	INC_X, INC_X, #3				// INC_X * SIZE
+#else
+	lsl	INC_X, INC_X, #2				// INC_X * SIZE
+#endif
+
+#endif
+
+	INIT_S
+
+	subs	N, N , #1
+	ble	amax_kernel_L999
+
+	asrs	I, N, #2					// I = N / 4
+	ble	amax_kernel_S1
+
+	.align 5
+
+amax_kernel_S4:
+
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+
+	subs	I, I, #1
+	bne	amax_kernel_S4
+
+amax_kernel_S1:
+
+	ands	I, N, #3
+	ble	amax_kernel_L999
+
+amax_kernel_S10:
+
+	KERNEL_S1
+
+	subs    I, I, #1
+        bne     amax_kernel_S10
+
+
+amax_kernel_L999:
+#if !defined(__ARM_PCS_VFP)
+#if defined(DOUBLE)
+	vmov	r0, r1, d0
+#else	
+	vmov    r0, s0
+#endif
+#endif
+	bx	lr
+
+	EPILOGUE
+

kernel/power/KERNEL.POWER9

@@ -3,12 +3,12 @@
 #CGEMM_BETA = ../generic/zgemm_beta.c
 #ZGEMM_BETA = ../generic/zgemm_beta.c
 
-STRMMKERNEL	= strmm_kernel_16x8_power8.S
+STRMMKERNEL	= sgemm_kernel_power9.S
 DTRMMKERNEL	= dgemm_kernel_power9.S
-CTRMMKERNEL	= ctrmm_kernel_8x4_power8.S
-ZTRMMKERNEL	= ztrmm_kernel_8x2_power8.S
+CTRMMKERNEL	= cgemm_kernel_power9.S
+ZTRMMKERNEL	= zgemm_kernel_power9.S
 
-SGEMMKERNEL    =  sgemm_kernel_16x8_power8.S
+SGEMMKERNEL    =  sgemm_kernel_power9.S
 SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
@@ -28,9 +28,9 @@ DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-CGEMMKERNEL    = cgemm_kernel_8x4_power8.S
+CGEMMKERNEL    = cgemm_kernel_power9.S
 CGEMMINCOPY    = ../generic/zgemm_ncopy_8.c
-CGEMMITCOPY    = cgemm_tcopy_8_power8.S
+CGEMMITCOPY    = ../generic/zgemm_tcopy_8.c
 CGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
 CGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
@@ -38,7 +38,7 @@ CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
 CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
 
-ZGEMMKERNEL    = zgemm_kernel_8x2_power8.S
+ZGEMMKERNEL    = zgemm_kernel_power9.S
 ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
 ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
 ZGEMMINCOPY    = ../generic/zgemm_ncopy_8.c

kernel/power/axpy.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define N	r3
 #define X	r6

kernel/power/axpy_ppc440.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define N	r3
 #define X	r6

kernel/power/cgemm_kernel_8x4_power8.S

@@ -97,7 +97,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -265,7 +265,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	stfs	f2,  ALPHA_I_SP
 	// stw	r0,  FZERO
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifdef __64BIT__
 	ld	LDC, FRAMESLOT(0) + 0(FRAMEPOINTER)
 #endif
@@ -286,7 +286,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #ifdef TRMMKERNEL
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(1) + 0(FRAMEPOINTER)
 #endif
 

kernel/power/cgemm_kernel_power9.S

@@ -0,0 +1,293 @@
+/***************************************************************************
+Copyright (c) 2013-2019, 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.
+*****************************************************************************/
+
+/**************************************************************************************
+* Abdelrauf(quickwritereader@gmail.com)
+* 	 BLASTEST 		: OK
+* 	 CTEST			: OK
+* 	 TEST			: OK
+*	 LAPACK-TEST		: OK
+**************************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+#include "def_vsx.h"
+
+ 
+#define LOAD	ld
+#define STACKSIZE  (512 )  
+#define FLINK_SAVE (STACKSIZE+16) /* 16($r12) */  
+#define	M	r3
+#define	N	r4
+#define	K	r5
+
+
+#define A	r8
+#define	B	r9
+#define	C	r10
+#define	LDC	r6
+#define OFFSET	r7
+
+
+#define alpha_r vs19
+#define alpha_i vs20
+#define save_permute_1 vs21
+#define permute_mask vs22
+#define o0	0
+ 
+
+#define T1	r11
+#define T2	r12
+#define T3	r14
+#define T4	r15
+#define T5	r16
+#define T6	r17
+#define L	r18
+#define T7	r19
+#define T8	r20
+#define TEMP_REG	r21
+#define	I	r22
+#define J	r23
+#define AO	r24
+#define	BO	r25
+#define	CO 	r26
+#define T9	r27
+#define	T10	r28
+#define	PRE	r29
+
+#define T12	r30
+#define T13	r31
+
+#include "cgemm_macros_power9.S"
+
+.equ    perm_const1, 0x0405060700010203
+.equ    perm_const2, 0x0c0d0e0f08090a0b
+.equ save_permute_12, 0x0c0d0e0f1c1d1e1f
+.equ save_permute_11, 0x0405060714151617
+
+
+
+#ifndef NEEDPARAM
+
+	PROLOGUE
+	PROFCODE
+
+
+	addi	SP, SP, -STACKSIZE
+	mflr r0
+
+
+	stfd	f14,    0(SP)
+	stfd	f15,    8(SP)
+	stfd	f16,   16(SP)
+	stfd	f17,   24(SP)
+
+	stfd	f18,   32(SP)
+	stfd	f19,   40(SP)
+	stfd	f20,   48(SP)
+	stfd	f21,   56(SP)
+
+	stfd	f22,   64(SP)
+	stfd	f23,   72(SP)
+	stfd	f24,   80(SP)
+	stfd	f25,   88(SP)
+
+	stfd	f26,   96(SP)
+	stfd	f27,  104(SP)
+	stfd	f28,  112(SP)
+	stfd	f29,  120(SP)
+
+	stfd	f30,  128(SP)
+	stfd	f31,  136(SP)
+
+
+	std	r31,  144(SP)
+	std	r30,  152(SP)
+	std	r29,  160(SP)
+	std	r28,  168(SP)
+	std	r27,  176(SP)
+	std	r26,  184(SP)
+	std	r25,  192(SP)
+	std	r24,  200(SP)
+	std	r23,  208(SP)
+	std	r22,  216(SP)
+	std	r21,  224(SP)
+	std	r20,  232(SP)
+	std	r19,  240(SP)
+	std	r18,  248(SP)
+	std	r17,  256(SP)
+	std	r16,  264(SP)
+	std	r15,  272(SP)
+	std	r14,  280(SP)
+ 
+ 
+  stxv    vs52,  288(SP)
+  stxv    vs53,  304(SP)
+  stxv    vs54,  320(SP)
+  stxv    vs55,  336(SP)
+  stxv    vs56,  352(SP)
+  stxv    vs57,  368(SP)
+  stxv    vs58,  384(SP)
+  stxv    vs59,  400(SP)
+  stxv    vs60,  416(SP)
+  stxv    vs61,  432(SP)
+  stxv    vs62,  448(SP)
+  stxv    vs63,  464(SP)
+  std     r0,   FLINK_SAVE(SP)
+ 
+
+
+	ld	LDC, FRAMESLOT(0) + STACKSIZE(SP)
+
+
+
+#ifdef TRMMKERNEL
+	ld	OFFSET,  FRAMESLOT(1) + STACKSIZE(SP)
+#endif
+   slwi    LDC, LDC, ZBASE_SHIFT
+
+ 
+ 
+	/*alpha is stored in f1. convert to single and splat*/
+    xscvdpspn alpha_r,vs1 
+    xscvdpspn alpha_i,vs2 
+	xxspltw   alpha_r,alpha_r,0 
+	xxspltw   alpha_i,alpha_i,0 
+/*load reverse permute mask for big endian
+  uint128 = 0xc0d0e0f08090a0b0405060700010203
+*/ 
+		
+	lis T2, perm_const2@highest
+	lis T1, perm_const1@highest
+	lis T3, save_permute_12@highest
+	lis T4, save_permute_11@highest
+
+	
+	ori T2, T2, perm_const2@higher
+	ori T1, T1, perm_const1@higher
+	ori T3, T3, save_permute_12@higher
+	ori T4, T4, save_permute_11@higher
+
+	
+	rldicr T2, T2, 32, 31
+	rldicr T1, T1, 32, 31
+	rldicr T3, T3, 32, 31
+	rldicr T4, T4, 32, 31 
+
+	oris T2, T2, perm_const2@h
+	oris T1, T1, perm_const1@h
+	oris T3, T3, save_permute_12@h
+	oris T4, T4, save_permute_11@h
+
+	
+	ori T2, T2, perm_const2@l  
+	ori T1, T1, perm_const1@l
+	ori T3, T3, save_permute_12@l  
+	ori T4, T4, save_permute_11@l
+
+	
+  li r0,0
+  li PRE,512
+
+#if defined(CC) || defined(CR) || defined(RC) || defined(RR) 
+/*negate for this case as we will use addition -1*(a+b) */
+  xvnegsp alpha_r,alpha_r
+  xvnegsp alpha_i,alpha_i
+#endif
+
+	mtvsrdd permute_mask,T2,T1
+	mtvsrdd save_permute_1,T3,T4 	
+
+     /*mask is reverse permute so we have to make it inner permute */
+ 	xxpermdi	permute_mask,	permute_mask,	permute_mask,2 
+
+#include "cgemm_logic_power9.S"
+
+.L999: 
+	lfd	f14,    0(SP)
+	lfd	f15,    8(SP)
+	lfd	f16,   16(SP)
+	lfd	f17,   24(SP)
+
+	lfd	f18,   32(SP)
+	lfd	f19,   40(SP)
+	lfd	f20,   48(SP)
+	lfd	f21,   56(SP)
+
+	lfd	f22,   64(SP)
+	lfd	f23,   72(SP)
+	lfd	f24,   80(SP)
+	lfd	f25,   88(SP)
+
+	lfd	f26,   96(SP)
+	lfd	f27,  104(SP)
+	lfd	f28,  112(SP)
+	lfd	f29,  120(SP)
+
+	lfd	f30,  128(SP)
+	lfd	f31,  136(SP)
+
+	ld	r31,  144(SP)
+	ld	r30,  152(SP)
+	ld	r29,  160(SP)
+	ld	r28,  168(SP)
+	ld	r27,  176(SP)
+	ld	r26,  184(SP)
+	ld	r25,  192(SP)
+	ld	r24,  200(SP)
+	ld	r23,  208(SP)
+	ld	r22,  216(SP)
+	ld	r21,  224(SP)
+	ld	r20,  232(SP)
+	ld	r19,  240(SP)
+	ld	r18,  248(SP)
+	ld	r17,  256(SP)
+	ld	r16,  264(SP)
+	ld	r15,  272(SP)
+	ld	r14,  280(SP)
+
+	ld    r0, 	 FLINK_SAVE(SP)	
+ 
+    lxv    vs52,  288(SP)
+    lxv    vs53,  304(SP)
+    lxv    vs54,  320(SP)
+    lxv    vs55,  336(SP)
+    lxv    vs56,  352(SP)
+    lxv    vs57,  368(SP)
+    lxv    vs58,  384(SP) 
+    lxv    vs59,  400(SP)
+	mtlr r0
+    lxv    vs60,  416(SP)
+    lxv    vs61,  432(SP) 
+    lxv    vs62,  448(SP)
+    lxv    vs63,  464(SP)
+
+	addi	SP, SP, STACKSIZE 
+	blr
+
+
+	EPILOGUE
+#endif

kernel/power/ctrmm_kernel_8x4_power8.S

@@ -98,7 +98,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -264,7 +264,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	stfs	f2,  ALPHA_I_SP
 	// stw	r0,  FZERO
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifdef __64BIT__
 	ld	LDC, FRAMESLOT(0) + STACKSIZE(SP)
 #endif
@@ -285,7 +285,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #ifdef TRMMKERNEL
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(1) + STACKSIZE(SP)
 #endif
 

kernel/power/dgemm_kernel_16x4_power8.S

@@ -97,7 +97,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -271,7 +271,7 @@ li r11,0
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/dgemm_kernel_power9.S

@@ -135,18 +135,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	std	r14,  280(SP)
  
  
-    stxv    v20,  288(SP)
-    stxv    v21,  304(SP)
-    stxv    v22,  320(SP)
-    stxv    v23,  336(SP)
-    stxv    v24,  352(SP)
-    stxv    v25,  368(SP)
-    stxv    v26,  384(SP)
-    stxv    v27,  400(SP)
-    stxv    v28,  416(SP)
-    stxv    v29,  432(SP)
-    stxv    v30,  448(SP)
-    stxv    v31,  464(SP)
+    stxv    vs52,  288(SP)
+    stxv    vs53,  304(SP)
+    stxv    vs54,  320(SP)
+    stxv    vs55,  336(SP)
+    stxv    vs56,  352(SP)
+    stxv    vs57,  368(SP)
+    stxv    vs58,  384(SP) 
+    stxv    vs59,  400(SP) 
+    stxv    vs60,  416(SP)
+    stxv    vs61,  432(SP) 
+    stxv    vs62,  448(SP)
+    stxv    vs63,  464(SP)
 
 
 	stfd	f1,  ALPHA_SP
@@ -229,18 +229,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	ld	r15,  272(SP)
 	ld	r14,  280(SP)
  
-    lxv    v20,  288(SP)
-    lxv    v21,  304(SP)
-    lxv    v22,  320(SP)
-    lxv    v23,  336(SP)
-    lxv    v24,  352(SP)
-    lxv    v25,  368(SP)
-    lxv    v26,  384(SP)
-    lxv    v27,  400(SP)
-    lxv    v28,  416(SP)
-    lxv    v29,  432(SP)
-    lxv    v30,  448(SP)
-    lxv    v31,  464(SP)
+    lxv    vs52,  288(SP)
+    lxv    vs53,  304(SP)
+    lxv    vs54,  320(SP)
+    lxv    vs55,  336(SP)
+    lxv    vs56,  352(SP)
+    lxv    vs57,  368(SP)
+    lxv    vs58,  384(SP) 
+    lxv    vs59,  400(SP) 
+    lxv    vs60,  416(SP)
+    lxv    vs61,  432(SP) 
+    lxv    vs62,  448(SP)
+    lxv    vs63,  464(SP)
 
 	addi	SP, SP, STACKSIZE
 	blr

kernel/power/dtrmm_kernel_16x4_power8.S

@@ -96,7 +96,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -257,8 +257,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         stvx    v31, r11, r0
         li r11,0
 
-	stw	r31,  144(SP)
-
 	stfd	f1,  ALPHA_SP
 	stw	r0,  FZERO
 
@@ -271,7 +269,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/dtrsm_kernel_LT_16x4_power8.S

@@ -61,7 +61,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -217,7 +217,7 @@ li r11,0
 #endif
 
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/gemm_beta.S

@@ -62,7 +62,7 @@
 	stfd	f31,   16(SP)
 	stw	r0,    24(SP)
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	LDC,    FRAMESLOT(0) + STACKSIZE(SP)
 #else

kernel/power/gemm_kernel.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -186,7 +186,7 @@
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,   FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -228,7 +228,7 @@
 
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/gemm_kernel_altivec.S

@@ -58,7 +58,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7

kernel/power/gemm_kernel_altivec_cell.S

@@ -58,7 +58,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7

kernel/power/gemm_kernel_altivec_g4.S

@@ -58,7 +58,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7

kernel/power/gemm_kernel_cell.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -192,7 +192,7 @@
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -226,7 +226,7 @@
 	li	PREC,   4 * SIZE
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/gemm_kernel_g4.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -184,7 +184,7 @@
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/gemm_kernel_hummer.S

@@ -46,7 +46,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #define A	r6
 #define	B	r7
 #define	C	r8

kernel/power/gemm_kernel_power3.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -187,7 +187,7 @@
 	li	PREC,   4 * SIZE
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/gemm_kernel_power6.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -183,7 +183,7 @@
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/gemm_kernel_ppc440.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -183,7 +183,7 @@
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/gemv_n.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define	N	r4
@@ -252,7 +252,7 @@
 	stw	r27,   196(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	INCY,    FRAMESLOT(0) + STACKSIZE(SP)
 	lwz	BUFFER,  FRAMESLOT(1) + STACKSIZE(SP)

kernel/power/gemv_n_ppc440.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define	N	r4
@@ -199,7 +199,7 @@
 	stw	r23,   180(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	INCY,    FRAMESLOT(0) + STACKSIZE(SP)
 	lwz	BUFFER,  FRAMESLOT(1) + STACKSIZE(SP)

kernel/power/gemv_t.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define	N	r4
@@ -260,7 +260,7 @@
 	stw	r29,   220(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	INCY,	 FRAMESLOT(0) + STACKSIZE(SP)
 	lwz	BUFFER,  FRAMESLOT(1) + STACKSIZE(SP)

kernel/power/gemv_t_ppc440.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define	N	r4
@@ -190,7 +190,7 @@
 	stw	r22,   192(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	INCY,	 FRAMESLOT(0) + STACKSIZE(SP)
 	lwz	BUFFER,  FRAMESLOT(1) + STACKSIZE(SP)

kernel/power/ger.S

@@ -47,7 +47,7 @@
 #endif
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define	N	r4
@@ -224,7 +224,7 @@
 	stw	r27,   196(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	LDA,     FRAMESLOT(0) + STACKSIZE(SP)
 	lwz	BUFFER,  FRAMESLOT(1) + STACKSIZE(SP)

kernel/power/icamax.c

@@ -75,7 +75,7 @@ static inline __attribute__((always_inline)) __vector float mvec_mergeo(__vector
 static BLASLONG   ciamax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *maxf) { 
 
     BLASLONG index;
-    BLASLONG i;
+    BLASLONG i=0;
 #if  defined(USE_MASK_PERMUTATIONS)    
     register __vector unsigned int static_index0 = {0,1,2,3};
 #else

kernel/power/icamin.c

@@ -50,7 +50,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 static BLASLONG   ciamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *minf) { 
 
     BLASLONG index;
-    BLASLONG i;
+    BLASLONG i=0;
     register __vector unsigned int static_index0 = {0,1,2,3};
     register __vector unsigned int temp0 = {4,4,4, 4}; //temporary vector register
     register __vector unsigned int temp1=  temp0<<1;  //{8,8,8,8}

kernel/power/scal.S

@@ -43,7 +43,7 @@
 #define XX	r4
 #define PREA	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define X r6
 #define INCX r7

kernel/power/scal_ppc440.S

@@ -43,7 +43,7 @@
 #define XX	r4
 #define PRE	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define X r6
 #define INCX r7

kernel/power/sgemm_kernel_16x8_power8.S

@@ -95,7 +95,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -273,7 +273,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	slwi	LDC, LDC, 2
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + 0(FRAMEPOINTER)
 #endif
 

kernel/power/sgemm_kernel_power9.S

@@ -0,0 +1,272 @@
+/***************************************************************************
+Copyright (c) 2013-2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+ 
+#define ASSEMBLER
+#include "common.h"
+#include "def_vsx.h"
+
+ 
+#define LOAD	ld
+#define STACKSIZE  (512 )  
+#define FLINK_SAVE (STACKSIZE+16) /* 16($r12) */  
+#define	M	r3
+#define	N	r4
+#define	K	r5
+
+ 
+#define A	r7
+#define	B	r8
+#define	C	r9
+#define	LDC	r10
+#define OFFSET	r6
+ 
+ 
+
+#define alpha_r vs20
+#define save_permute_1 vs21
+#define save_permute_2 vs22
+#define permute_mask vs23
+#define o0	0
+ 
+
+#define T1	r11
+#define T2	r12
+#define T3	r14
+#define T4	r15
+#define T5	r16
+#define T6	r17
+#define L	r18
+#define T7	r19
+#define T8	r20
+#define TEMP_REG	r21
+#define	I	r22
+#define J	r23
+#define AO	r24
+#define	BO	r25
+#define	CO 	r26
+#define T9	r27
+#define	T10	r28
+#define	T11	r29
+
+#define T12	r30
+#define T13	r31
+
+#include "sgemm_macros_power9.S"
+
+.equ    perm_const1, 0x0405060700010203
+.equ    perm_const2, 0x0c0d0e0f08090a0b
+.equ save_permute_11, 0x1415161718191a1b
+.equ save_permute_12, 0x0405060708090a0b
+.equ save_permute_21, 0x101112131c1d1e1f
+.equ save_permute_22, 0x000102030c0d0e0f 
+
+
+#ifndef NEEDPARAM
+
+	PROLOGUE
+	PROFCODE
+
+	addi	SP, SP, -STACKSIZE
+	mflr r0
+
+
+	stfd	f14,    0(SP)
+	stfd	f15,    8(SP)
+	stfd	f16,   16(SP)
+	stfd	f17,   24(SP)
+
+	stfd	f18,   32(SP)
+	stfd	f19,   40(SP)
+	stfd	f20,   48(SP)
+	stfd	f21,   56(SP)
+
+	stfd	f22,   64(SP)
+	stfd	f23,   72(SP)
+	stfd	f24,   80(SP)
+	stfd	f25,   88(SP)
+
+	stfd	f26,   96(SP)
+	stfd	f27,  104(SP)
+	stfd	f28,  112(SP)
+	stfd	f29,  120(SP)
+
+	stfd	f30,  128(SP)
+	stfd	f31,  136(SP)
+
+ 
+	std	r31,  144(SP)
+	std	r30,  152(SP)
+	std	r29,  160(SP)
+	std	r28,  168(SP)
+	std	r27,  176(SP)
+	std	r26,  184(SP)
+	std	r25,  192(SP)
+	std	r24,  200(SP)
+	std	r23,  208(SP)
+	std	r22,  216(SP)
+	std	r21,  224(SP)
+	std	r20,  232(SP)
+	std	r19,  240(SP)
+	std	r18,  248(SP)
+	std	r17,  256(SP)
+	std	r16,  264(SP)
+	std	r15,  272(SP)
+	std	r14,  280(SP)
+ 
+ 
+  stxv    vs52,  288(SP)
+  stxv    vs53,  304(SP)
+  stxv    vs54,  320(SP)
+  stxv    vs55,  336(SP)
+  stxv    vs56,  352(SP)
+  stxv    vs57,  368(SP)
+  stxv    vs58,  384(SP)
+  stxv    vs59,  400(SP)
+  stxv    vs60,  416(SP)
+  stxv    vs61,  432(SP)
+  stxv    vs62,  448(SP)
+  stxv    vs63,  464(SP)
+  std     r0,   FLINK_SAVE(SP)
+ 
+
+#if defined(TRMMKERNEL) 
+	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
+#endif
+   slwi    LDC, LDC, 2
+
+ 
+ 
+	/*alpha is stored in f1. convert to single and splat*/
+  xscvdpspn alpha_r,vs1 
+	xxspltw   alpha_r,alpha_r,0 
+ 
+/*load reverse permute mask for big endian
+  uint128 = 0xc0d0e0f08090a0b0405060700010203
+*/ 
+		
+	lis T2, perm_const2@highest
+	lis T1, perm_const1@highest
+	lis T3, save_permute_12@highest
+	lis T4, save_permute_11@highest
+	lis T5, save_permute_22@highest
+	lis T6, save_permute_21@highest
+	ori T2, T2, perm_const2@higher
+	ori T1, T1, perm_const1@higher
+	ori T3, T3, save_permute_12@higher
+	ori T4, T4, save_permute_11@higher
+	ori T5, T5, save_permute_22@higher
+	ori T6, T6, save_permute_21@higher
+	rldicr T2, T2, 32, 31
+	rldicr T1, T1, 32, 31
+	rldicr T3, T3, 32, 31
+	rldicr T4, T4, 32, 31
+	rldicr T5, T5, 32, 31
+	rldicr T6, T6, 32, 31
+	oris T2, T2, perm_const2@h
+	oris T1, T1, perm_const1@h
+	oris T3, T3, save_permute_12@h
+	oris T4, T4, save_permute_11@h
+	oris T5, T5, save_permute_22@h
+	oris T6, T6, save_permute_21@h
+	ori T2, T2, perm_const2@l  
+	ori T1, T1, perm_const1@l
+	ori T3, T3, save_permute_12@l  
+	ori T4, T4, save_permute_11@l
+	ori T5, T5, save_permute_22@l 
+	ori T6, T6, save_permute_21@l
+  li r0,0
+	mtvsrdd permute_mask,T2,T1
+	mtvsrdd save_permute_1,T3,T4	
+	mtvsrdd save_permute_2,T5,T6	
+
+#include "sgemm_logic_power9.S"
+
+.L999: 
+	lfd	f14,    0(SP)
+	lfd	f15,    8(SP)
+	lfd	f16,   16(SP)
+	lfd	f17,   24(SP)
+
+	lfd	f18,   32(SP)
+	lfd	f19,   40(SP)
+	lfd	f20,   48(SP)
+	lfd	f21,   56(SP)
+
+	lfd	f22,   64(SP)
+	lfd	f23,   72(SP)
+	lfd	f24,   80(SP)
+	lfd	f25,   88(SP)
+
+	lfd	f26,   96(SP)
+	lfd	f27,  104(SP)
+	lfd	f28,  112(SP)
+	lfd	f29,  120(SP)
+
+	lfd	f30,  128(SP)
+	lfd	f31,  136(SP)
+
+	ld	r31,  144(SP)
+	ld	r30,  152(SP)
+	ld	r29,  160(SP)
+	ld	r28,  168(SP)
+	ld	r27,  176(SP)
+	ld	r26,  184(SP)
+	ld	r25,  192(SP)
+	ld	r24,  200(SP)
+	ld	r23,  208(SP)
+	ld	r22,  216(SP)
+	ld	r21,  224(SP)
+	ld	r20,  232(SP)
+	ld	r19,  240(SP)
+	ld	r18,  248(SP)
+	ld	r17,  256(SP)
+	ld	r16,  264(SP)
+	ld	r15,  272(SP)
+	ld	r14,  280(SP)
+
+	ld    r0, 	 FLINK_SAVE(SP)	
+ 
+    lxv    vs52,  288(SP)
+    lxv    vs53,  304(SP)
+    lxv    vs54,  320(SP)
+    lxv    vs55,  336(SP)
+    lxv    vs56,  352(SP)
+    lxv    vs57,  368(SP)
+    lxv    vs58,  384(SP) 
+    lxv    vs59,  400(SP)
+	mtlr r0
+    lxv    vs60,  416(SP)
+    lxv    vs61,  432(SP) 
+    lxv    vs62,  448(SP)
+    lxv    vs63,  464(SP)
+
+	addi	SP, SP, STACKSIZE 
+	blr
+
+
+	EPILOGUE
+#endif

kernel/power/strmm_kernel_16x8_power8.S

@@ -96,7 +96,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -271,7 +271,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	slwi	LDC, LDC, BASE_SHIFT
 
 #if defined(TRMMKERNEL)
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 

kernel/power/swap.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define N	r3
 #define X	r6

kernel/power/symv_L.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define N	r4
@@ -248,7 +248,7 @@
 	stw	r27,   196(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	BUFFER,  FRAMESLOT(0) + STACKSIZE(SP)
 #else

kernel/power/symv_U.S

@@ -39,7 +39,7 @@
 #define ASSEMBLER
 #include "common.h"
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define M	r3
 #define IS	r4
@@ -247,7 +247,7 @@
 	stw	r27,   196(SP)
 #endif
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	lwz	BUFFER,  FRAMESLOT(0) + STACKSIZE(SP)
 #else

kernel/power/trsm_kernel_LN.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -180,7 +180,7 @@
 
 	slwi	LDC, LDC, BASE_SHIFT
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -236,7 +236,7 @@
 
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/trsm_kernel_LT.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -180,7 +180,7 @@
 
 	slwi	LDC, LDC, BASE_SHIFT
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -257,7 +257,7 @@
 
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/trsm_kernel_RT.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -180,7 +180,7 @@
 
 	slwi	LDC, LDC, BASE_SHIFT
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -254,7 +254,7 @@
 
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/trsm_kernel_cell_LN.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -180,7 +180,7 @@
 
 	slwi	LDC, LDC, BASE_SHIFT
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -231,7 +231,7 @@
 	li	PREC,  -4 * SIZE
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/trsm_kernel_cell_LT.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -180,7 +180,7 @@
 
 	slwi	LDC, LDC, BASE_SHIFT
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -257,7 +257,7 @@
 
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/trsm_kernel_cell_RT.S

@@ -59,7 +59,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 #define A	r6
 #define	B	r7
@@ -180,7 +180,7 @@
 
 	slwi	LDC, LDC, BASE_SHIFT
 
-#if defined(linux) && defined(__64BIT__)
+#if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
 	ld	OFFSET,  FRAMESLOT(0) + STACKSIZE(SP)
 #endif
 
@@ -231,7 +231,7 @@
 	li	PREC,  -4 * SIZE
 #else
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #ifndef __64BIT__
 	mr	PREA,  r10
 	lwz	PREB,  FRAMESLOT(0) + STACKSIZE(SP)

kernel/power/trsm_kernel_hummer_LN.S

@@ -46,7 +46,7 @@
 #define	N	r4
 #define	K	r5
 
-#ifdef linux
+#if defined(linux) || defined(__FreeBSD__)
 #define A	r6
 #define	B	r7
 #define	C	r8