Merge branch 'develop' into risc-v

.drone.yml

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

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

@@ -43,7 +43,7 @@ jobs:
       - name: Update Homebrew
         if: github.event_name != 'pull_request'
         run: brew update || true
-
+          
       - name: Install prerequisites
         run: brew install --fetch-HEAD --HEAD --only-dependencies --keep-tmp openblas
 

.gitignore

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

.travis.yml

@@ -1,33 +1,38 @@
 # XXX: Precise is already deprecated, new default is Trusty.
 # https://blog.travis-ci.com/2017-07-11-trusty-as-default-linux-is-coming
-dist: precise
+dist: focal
 sudo: true
 language: c
 
 matrix:
   include:
     - &test-ubuntu
-      os: linux
+#      os: linux
       compiler: gcc
       addons:
         apt:
           packages:
             - gfortran
+#      before_script: &common-before
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+#      script:
+#        - 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
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64"
+#
+#    - <<: *test-ubuntu
+      os: linux-ppc64le
       before_script: &common-before
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=POWER8 NUM_THREADS=32"
       script:
         - 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
-      env:
-        - 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
@@ -55,38 +60,38 @@ matrix:
         - TARGET_BOX=IBMZ_LINUX
         - BTYPE="BINARY=64 USE_OPENMP=0 CC=clang"
 
-    - <<: *test-ubuntu
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 USE_OPENMP=1"
-
-    - <<: *test-ubuntu
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 INTERFACE64=1"
-
-    - <<: *test-ubuntu
-      compiler: clang
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 CC=clang"
-
-    - <<: *test-ubuntu
-      compiler: clang
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 INTERFACE64=1 CC=clang"
-
-    - <<: *test-ubuntu
-      addons:
-        apt:
-          packages:
-            - gcc-multilib
-            - gfortran-multilib
-      env:
-        - TARGET_BOX=LINUX32
-        - BTYPE="BINARY=32"
-
+#    - <<: *test-ubuntu
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 USE_OPENMP=1"
+#
+#    - <<: *test-ubuntu
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 INTERFACE64=1"
+#
+#    - <<: *test-ubuntu
+#      compiler: clang
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 CC=clang"
+#
+#    - <<: *test-ubuntu
+#      compiler: clang
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 INTERFACE64=1 CC=clang"
+#
+#    - <<: *test-ubuntu
+#      addons:
+#        apt:
+#          packages:
+#            - gcc-multilib
+#            - gfortran-multilib
+#      env:
+#        - TARGET_BOX=LINUX32
+#        - BTYPE="BINARY=32"
+#
     - os: linux
       arch: ppc64le
       dist: bionic
@@ -121,47 +126,47 @@ matrix:
         # for matrix annotation only
         - TARGET_BOX=PPC64LE_LINUX_P9
 
-    - os: linux
-      compiler: gcc
-      addons:
-        apt:
-          packages:
-            - binutils-mingw-w64-x86-64
-            - gcc-mingw-w64-x86-64
-            - gfortran-mingw-w64-x86-64
-      before_script: *common-before
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=WIN64
-        - BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
-
+#    - os: linux
+#      compiler: gcc
+#      addons:
+#        apt:
+#          packages:
+#            - binutils-mingw-w64-x86-64
+#            - gcc-mingw-w64-x86-64
+#            - gfortran-mingw-w64-x86-64
+#      before_script: *common-before
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - TARGET_BOX=WIN64
+#        - BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
+#
     # Build & test on Alpine Linux inside chroot, i.e. on system with musl libc.
     # These jobs needs sudo, so Travis runs them on VM-based infrastructure
     # which is slower than container-based infrastructure used for jobs
     # that don't require sudo.
-    - &test-alpine
-      os: linux
-      dist: trusty
-      sudo: true
-      language: minimal
-      before_install:
-        - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.9.0/alpine-chroot-install' \
-          && echo 'e5dfbbdc0c4b3363b99334510976c86bfa6cb251  alpine-chroot-install' | sha1sum -c || exit 1"
-        - alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
-      install:
-        - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
-      before_script: *common-before
-      script:
-        # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
-        - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-              CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
-        - alpine make -C test $COMMON_FLAGS $BTYPE
-        - alpine make -C ctest $COMMON_FLAGS $BTYPE
-        - alpine make -C utest $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64"
+ #   - &test-alpine
+ #     os: linux
+ #     dist: trusty
+ #     sudo: true
+ #     language: minimal
+ #     before_install:
+ #       - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.9.0/alpine-chroot-install' \
+ #         && echo 'e5dfbbdc0c4b3363b99334510976c86bfa6cb251  alpine-chroot-install' | sha1sum -c || exit 1"
+ #       - alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
+ #     install:
+ #       - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
+ #     before_script: *common-before
+ #     script:
+ #       # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
+ #       - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+ #             CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
+ #       - alpine make -C test $COMMON_FLAGS $BTYPE
+ #       - alpine make -C ctest $COMMON_FLAGS $BTYPE
+ #       - alpine make -C utest $COMMON_FLAGS $BTYPE
+ #     env:
+ #       - TARGET_BOX=LINUX64_MUSL
+ #       - BTYPE="BINARY=64"
 
     # XXX: This job segfaults in TESTS OF THE COMPLEX LEVEL 3 BLAS,
     # but only on Travis CI, cannot reproduce it elsewhere.
@@ -171,89 +176,98 @@ matrix:
     #    - TARGET_BOX=LINUX64_MUSL
     #    - BTYPE="BINARY=64 USE_OPENMP=1"
 
-    - <<: *test-alpine
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 INTERFACE64=1"
+#    - <<: *test-alpine
+#      env:
+#        - TARGET_BOX=LINUX64_MUSL
+#        - BTYPE="BINARY=64 INTERFACE64=1"
+#
+#    # Build with the same flags as Alpine do in OpenBLAS package.
+#    - <<: *test-alpine
+#      env:
+#        - TARGET_BOX=LINUX64_MUSL
+#        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=CORE2"
 
-    # Build with the same flags as Alpine do in OpenBLAS package.
-    - <<: *test-alpine
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=CORE2"
+#    - &test-cmake
+#      os: linux
+#      compiler: clang
+#      addons:
+#        apt:
+#          packages:
+#            - gfortran
+#            - cmake
+#      dist: trusty
+#      sudo: true
+#      before_script:
+#        - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
+#      script:
+#        - mkdir build
+#        - CONFIG=Release
+#        - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
+#        - cmake --build build --config $CONFIG -- -j2
+#      env:
+#        - CMAKE=1
+#    - <<: *test-cmake
+#      env:
+#        - CMAKE=1 CMAKE_ARGS="-DNOFORTRAN=1"
+#    - <<: *test-cmake
+#      compiler: gcc
+#      env:
+#        - CMAKE=1
 
-    - &test-cmake
-      os: linux
-      compiler: clang
-      addons:
-        apt:
-          packages:
-            - gfortran
-            - cmake
-      dist: trusty
-      sudo: true
-      before_script:
-        - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
-      script:
-        - mkdir build
-        - CONFIG=Release
-        - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
-        - cmake --build build --config $CONFIG -- -j2
-      env:
-        - CMAKE=1
-    - <<: *test-cmake
-      env:
-        - CMAKE=1 CMAKE_ARGS="-DNOFORTRAN=1"
-    - <<: *test-cmake
-      compiler: gcc
-      env:
-        - CMAKE=1
-
-    - &test-macos
-      os: osx
-      osx_image: xcode11.5
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-9"
+#    - &test-macos
+#      os: osx
+#      osx_image: xcode11.5
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-9"
+#
+#    - <<: *test-macos
+#      osx_image: xcode12
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#        - brew update
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - BTYPE="TARGET=HASWELL USE_OPENMP=1 BINARY=64 INTERFACE64=1 CC=gcc-10 FC=gfortran-10"
+#
+#    - <<: *test-macos
+#      osx_image: xcode12
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#        - brew update
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-10"      
 
-    - <<: *test-macos
-      osx_image: xcode12
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-        - brew update
-        - brew install gcc@10
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-10"
-        
   #  - <<: *test-macos
   #    osx_image: xcode10
   #    env:
   #      - BTYPE="TARGET=NEHALEM BINARY=32 NOFORTRAN=1"
 
-    - <<: *test-macos
-      osx_image: xcode11.5
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-        - brew update
-      env:
+#    - <<: *test-macos
+#      osx_image: xcode11.5
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#        - brew update
+#      env:
 #        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
 #        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch arm64 -miphoneos-version-min=10.0"
-        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
-        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch arm64 -miphoneos-version-min=10.0"
-        - BTYPE="TARGET=ARMV8 BINARY=64 HOSTCC=clang NOFORTRAN=1"
-    - <<: *test-macos
-      osx_image: xcode11.5
-      env:
-#        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
-#        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch armv7 -miphoneos-version-min=5.1"
-        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
-        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch armv7 -miphoneos-version-min=5.1"
-        - BTYPE="TARGET=ARMV7 HOSTCC=clang NOFORTRAN=1"
+#        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+#        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch arm64 -miphoneos-version-min=10.0"
+#        - BTYPE="TARGET=ARMV8 BINARY=64 HOSTCC=clang NOFORTRAN=1"
+#    - <<: *test-macos
+#      osx_image: xcode11.5
+#      env:
+##        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+##        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch armv7 -miphoneos-version-min=5.1"
+#        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+#        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch armv7 -miphoneos-version-min=5.1"
+#        - BTYPE="TARGET=ARMV7 HOSTCC=clang NOFORTRAN=1"
 
     - &test-graviton2
       os: linux

CMakeLists.txt

@@ -3,10 +3,13 @@
 ##
 
 cmake_minimum_required(VERSION 2.8.5)
+
 project(OpenBLAS C ASM)
+
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 12.dev)
+set(OpenBLAS_PATCH_VERSION 20)
+
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -14,54 +17,74 @@ include(GNUInstallDirs)
 
 include(CMakePackageConfigHelpers)
 
+if(MSVC AND NOT DEFINED NOFORTRAN)
+  set(NOFORTRAN ON)
+endif()
 
 #######
 if(MSVC)
-option(BUILD_WITHOUT_LAPACK "Do not build LAPACK and LAPACKE (Only BLAS or CBLAS)" ON)
+  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, aarch64 or ppc only)" OFF)
+
 option(DYNAMIC_OLDER "Include specific support for older x86 cpu models (Penryn,Dunnington,Atom,Nano,Opteron) with DYNAMIC_ARCH" OFF)
+
 option(BUILD_RELAPACK "Build with ReLAPACK (recursive implementation of several LAPACK functions on top of standard LAPACK)" OFF)
+
 option(USE_LOCKING "Use locks even in single-threaded builds to make them callable from multiple threads" OFF)
+
 if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
-option(NO_AFFINITY "Disable support for CPU affinity masks to avoid binding processes from e.g. R or numpy/scipy to a single core" ON)
+  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)
+  set(NO_AFFINITY 1)
 endif()
+
 option(CPP_THREAD_SAFETY_TEST "Run a massively parallel DGEMM test to confirm thread safety of the library (requires OpenMP and about 1.3GB of RAM)" OFF)
+
 option(CPP_THREAD_SAFETY_GEMV "Run a massively parallel DGEMV test to confirm thread safety of the library (requires OpenMP)" OFF)
+option(BUILD_STATIC_LIBS "Build static library" OFF)
+if(NOT BUILD_STATIC_LIBS AND NOT BUILD_SHARED_LIBS)
+  set(BUILD_STATIC_LIBS ON CACHE BOOL "Build static library" FORCE)
+endif()
+if((BUILD_STATIC_LIBS AND BUILD_SHARED_LIBS) AND MSVC)
+  message(WARNING "Could not enable both BUILD_STATIC_LIBS and BUILD_SHARED_LIBS with MSVC, Disable BUILD_SHARED_LIBS")
+  set(BUILD_SHARED_LIBS OFF CACHE BOOL "Build static library" FORCE)
+endif()
 
 # Add a prefix or suffix to all exported symbol names in the shared library.
 # Avoids conflicts with other BLAS libraries, especially when using
 # 64 bit integer interfaces in OpenBLAS.
-
 set(SYMBOLPREFIX "" CACHE STRING  "Add a prefix to all exported symbol names in the shared library to avoid conflicts with other BLAS libraries" )
+
 set(SYMBOLSUFFIX "" CACHE STRING  "Add a suffix to all exported symbol names in the shared library, e.g. _64 for INTERFACE64 builds" )
+
 #######
 if(BUILD_WITHOUT_LAPACK)
-set(NO_LAPACK 1)
-set(NO_LAPACKE 1)
+  set(NO_LAPACK 1)
+  set(NO_LAPACKE 1)
 endif()
 
 if(BUILD_WITHOUT_CBLAS)
-set(NO_CBLAS 1)
+  set(NO_CBLAS 1)
 endif()
 
 #######
 
 if(MSVC AND MSVC_STATIC_CRT)
-    set(CompilerFlags
-            CMAKE_CXX_FLAGS
-            CMAKE_CXX_FLAGS_DEBUG
-            CMAKE_CXX_FLAGS_RELEASE
-            CMAKE_C_FLAGS
-            CMAKE_C_FLAGS_DEBUG
-            CMAKE_C_FLAGS_RELEASE
-            )
-    foreach(CompilerFlag ${CompilerFlags})
-      string(REPLACE "/MD" "/MT" ${CompilerFlag} "${${CompilerFlag}}")
-    endforeach()
+  set(CompilerFlags
+        CMAKE_CXX_FLAGS
+        CMAKE_CXX_FLAGS_DEBUG
+        CMAKE_CXX_FLAGS_RELEASE
+        CMAKE_C_FLAGS
+        CMAKE_C_FLAGS_DEBUG
+        CMAKE_C_FLAGS_RELEASE
+        )
+  foreach(CompilerFlag ${CompilerFlags})
+    string(REPLACE "/MD" "/MT" ${CompilerFlag} "${${CompilerFlag}}")
+  endforeach()
 endif()
 
 message(WARNING "CMake support is experimental. It does not yet support all build options and may not produce the same Makefiles that OpenBLAS ships with.")
@@ -95,7 +118,7 @@ endif ()
 # set which float types we want to build for
 if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_COMPLEX AND NOT DEFINED BUILD_COMPLEX16)
   # if none are defined, build for all
-#  set(BUILD_BFLOAT16 true)
+  # set(BUILD_BFLOAT16 true)
   set(BUILD_SINGLE true)
   set(BUILD_DOUBLE true)
   set(BUILD_COMPLEX true)
@@ -129,7 +152,7 @@ endif ()
 
 if (BUILD_BFLOAT16)
   message(STATUS "Building Half Precision")
-  list(APPEND FLOAT_TYPES "BFLOAT16") # defines nothing
+  #  list(APPEND FLOAT_TYPES "BFLOAT16") # defines nothing
 endif ()
 
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
@@ -140,9 +163,10 @@ endif ()
 set( CMAKE_LIBRARY_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
 set( CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
 if(MSVC)
-set( CMAKE_LIBRARY_OUTPUT_DIRECTORY_DEBUG ${PROJECT_BINARY_DIR}/lib/Debug)
-set( CMAKE_ARCHIVE_OUTPUT_DIRECTORY_RELEASE ${PROJECT_BINARY_DIR}/lib/Release)
+  set( CMAKE_LIBRARY_OUTPUT_DIRECTORY_DEBUG ${PROJECT_BINARY_DIR}/lib/Debug)
+  set( CMAKE_ARCHIVE_OUTPUT_DIRECTORY_RELEASE ${PROJECT_BINARY_DIR}/lib/Release)
 endif ()
+
 # get obj vars into format that add_library likes: $<TARGET_OBJS:objlib> (see http://www.cmake.org/cmake/help/v3.0/command/add_library.html)
 set(TARGET_OBJS "")
 foreach (SUBDIR ${SUBDIRS})
@@ -180,12 +204,63 @@ if (${DYNAMIC_ARCH})
 endif ()
 
 # add objects to the openblas lib
-add_library(${OpenBLAS_LIBNAME} ${LA_SOURCES} ${LAPACKE_SOURCES} ${RELA_SOURCES} ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
-target_include_directories(${OpenBLAS_LIBNAME} INTERFACE $<INSTALL_INTERFACE:include/openblas${SUFFIX64}>)
+if(NOT NO_LAPACK)
+  add_library(LAPACK OBJECT ${LA_SOURCES})
+  list(APPEND TARGET_OBJS "$<TARGET_OBJECTS:LAPACK>")
+endif()
+if(NOT NO_LAPACKE)
+  add_library(LAPACKE OBJECT ${LAPACKE_SOURCES})
+  list(APPEND TARGET_OBJS "$<TARGET_OBJECTS:LAPACKE>")
+endif()
+if(BUILD_RELAPACK)
+  add_library(RELAPACK OBJECT ${RELA_SOURCES})
+  list(APPEND TARGET_OBJS "$<TARGET_OBJECTS:RELAPACK>")
+endif()
+set(OpenBLAS_LIBS "")
+if(BUILD_STATIC_LIBS)
+  add_library(${OpenBLAS_LIBNAME}_static STATIC ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
+  target_include_directories(${OpenBLAS_LIBNAME}_static INTERFACE $<INSTALL_INTERFACE:include/openblas${SUFFIX64}>)
+  list(APPEND OpenBLAS_LIBS ${OpenBLAS_LIBNAME}_static)
+endif()
+if(BUILD_SHARED_LIBS)
+  add_library(${OpenBLAS_LIBNAME}_shared SHARED ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
+  target_include_directories(${OpenBLAS_LIBNAME}_shared INTERFACE $<INSTALL_INTERFACE:include/openblas${SUFFIX64}>)
+  list(APPEND OpenBLAS_LIBS ${OpenBLAS_LIBNAME}_shared)
+endif()
+if(BUILD_STATIC_LIBS)
+  add_library(${OpenBLAS_LIBNAME} ALIAS ${OpenBLAS_LIBNAME}_static)
+else()
+  add_library(${OpenBLAS_LIBNAME} ALIAS ${OpenBLAS_LIBNAME}_shared)
+endif()
+
+set_target_properties(${OpenBLAS_LIBS} PROPERTIES OUTPUT_NAME ${OpenBLAS_LIBNAME})
 
 # Android needs to explicitly link against libm
 if(ANDROID)
-  target_link_libraries(${OpenBLAS_LIBNAME} m)
+  if(BUILD_STATIC_LIBS)
+    target_link_libraries(${OpenBLAS_LIBNAME}_static m)
+  endif()
+  if(BUILD_SHARED_LIBS)
+    target_link_libraries(${OpenBLAS_LIBNAME}_shared m)
+  endif()
+endif()
+
+if (APPLE AND DYNAMIC_ARCH AND BUILD_SHARED_LIBS)
+  set (CMAKE_C_USE_RESPONSE_FILE_FOR_OBJECTS 1)
+  if (NOT NOFORTRAN)  
+  set (CMAKE_Fortran_USE_RESPONSE_FILE_FOR_OBJECTS 1)
+  set (CMAKE_Fortran_CREATE_SHARED_LIBRARY
+ "sh -c 'cat ${CMAKE_BINARY_DIR}/CMakeFiles/openblas_shared.dir/objects*.rsp | xargs -n 1024 ar -ru libopenblas.a && exit 0' "
+ "sh -c 'ar -ru libopenblas.a ${CMAKE_BINARY_DIR}/driver/others/CMakeFiles/driver_others.dir/xerbla.c.o && exit 0' "
+ "sh -c 'echo \"\" | ${CMAKE_Fortran_COMPILER} -o dummy.o -c -x f95-cpp-input - '"
+ "sh -c '${CMAKE_Fortran_COMPILER} -fpic -shared -Wl,-all_load -Wl,-force_load,libopenblas.a -Wl,-noall_load dummy.o -o ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libopenblas.${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.dylib'"
+ "sh -c 'ls -l ${CMAKE_BINARY_DIR}/lib'")
+  else ()
+  set (CMAKE_C_CREATE_SHARED_LIBRARY
+   "sh -c 'cat ${CMAKE_BINARY_DIR}/CMakeFiles/openblas_shared.dir/objects*.rsp | xargs -n 1024 ar -ru libopenblas.a && exit 0' "
+   "sh -c 'ar -ru libopenblas.a ${CMAKE_BINARY_DIR}/driver/others/CMakeFiles/driver_others.dir/xerbla.c.o && exit 0' "
+   "sh -c '${CMAKE_C_COMPILER} -fpic -shared -Wl,-all_load -Wl,-force_load,libopenblas.a -Wl,-noall_load -o ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libopenblas.${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.dylib'")
+  endif ()
 endif()
 
 # Handle MSVC exports
@@ -194,21 +269,21 @@ if(MSVC AND BUILD_SHARED_LIBS)
     include("${PROJECT_SOURCE_DIR}/cmake/export.cmake")
   else()
     # Creates verbose .def file (51KB vs 18KB)
-    set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES WINDOWS_EXPORT_ALL_SYMBOLS true)
+    set_target_properties(${OpenBLAS_LIBNAME}_shared PROPERTIES WINDOWS_EXPORT_ALL_SYMBOLS true)
   endif()
 endif()
 
 # Set output for libopenblas
-set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
-set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES LIBRARY_OUTPUT_NAME_DEBUG "${OpenBLAS_LIBNAME}_d")
-set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES EXPORT_NAME "OpenBLAS")
+set_target_properties( ${OpenBLAS_LIBS} PROPERTIES RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
+set_target_properties( ${OpenBLAS_LIBS} PROPERTIES LIBRARY_OUTPUT_NAME_DEBUG "${OpenBLAS_LIBNAME}_d")
+set_target_properties( ${OpenBLAS_LIBS} PROPERTIES EXPORT_NAME "OpenBLAS")
 
 foreach (OUTPUTCONFIG ${CMAKE_CONFIGURATION_TYPES})
   string( TOUPPER ${OUTPUTCONFIG} OUTPUTCONFIG )
 
-  set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
-  set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
-  set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES ARCHIVE_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
+  set_target_properties( ${OpenBLAS_LIBS} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
+  set_target_properties( ${OpenBLAS_LIBS} PROPERTIES LIBRARY_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
+  set_target_properties( ${OpenBLAS_LIBS} PROPERTIES ARCHIVE_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
 endforeach()
 
 enable_testing()
@@ -217,10 +292,17 @@ if (USE_THREAD)
   # Add threading library to linker
   find_package(Threads)
   if (THREADS_HAVE_PTHREAD_ARG)
-    set_property(TARGET ${OpenBLAS_LIBNAME} PROPERTY COMPILE_OPTIONS "-pthread")
-    set_property(TARGET ${OpenBLAS_LIBNAME} PROPERTY INTERFACE_COMPILE_OPTIONS "-pthread")
+    set_target_properties(${OpenBLAS_LIBS} PROPERTIES
+      COMPILE_OPTIONS "-pthread"
+      INTERFACE_COMPILE_OPTIONS "-pthread"
+    )
+  endif()
+  if(BUILD_STATIC_LIBS)
+    target_link_libraries(${OpenBLAS_LIBNAME}_static ${CMAKE_THREAD_LIBS_INIT})
+  endif()
+  if(BUILD_SHARED_LIBS)
+    target_link_libraries(${OpenBLAS_LIBNAME}_shared ${CMAKE_THREAD_LIBS_INIT})
   endif()
-  target_link_libraries(${OpenBLAS_LIBNAME} ${CMAKE_THREAD_LIBS_INIT})
 endif()
 
 #if (MSVC OR NOT NOFORTRAN)
@@ -229,104 +311,116 @@ if (NOT NO_CBLAS)
   add_subdirectory(utest)
 endif()
 
-if (NOT MSVC AND NOT NOFORTRAN)
+if (NOT NOFORTRAN)
   # Build test and ctest
   add_subdirectory(test)
   if(NOT NO_CBLAS)
     add_subdirectory(ctest)
   endif()
   add_subdirectory(lapack-netlib/TESTING)
-	  if (CPP_THREAD_SAFETY_TEST OR CPP_THREAD_SAFETY_GEMV)
-	  	add_subdirectory(cpp_thread_test)
-    	  endif()
+  if (CPP_THREAD_SAFETY_TEST OR CPP_THREAD_SAFETY_GEMV)
+    add_subdirectory(cpp_thread_test)
+  endif()
 endif()
 
-set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
+set_target_properties(${OpenBLAS_LIBS} PROPERTIES
   VERSION ${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}
   SOVERSION ${OpenBLAS_MAJOR_VERSION}
 )
 
 if (BUILD_SHARED_LIBS AND BUILD_RELAPACK)
   if (NOT MSVC)
-    target_link_libraries(${OpenBLAS_LIBNAME} "-Wl,-allow-multiple-definition")
+    target_link_libraries(${OpenBLAS_LIBNAME}_shared "-Wl,-allow-multiple-definition")
   else()
-   set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /FORCE:MULTIPLE")
+    set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /FORCE:MULTIPLE")
   endif()
 endif()
 
 if (BUILD_SHARED_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFFIX} STREQUAL "")
-if (NOT DEFINED ARCH)
-  set(ARCH_IN "x86_64")
-else()
-  set(ARCH_IN ${ARCH})
-endif()
+  if (NOT DEFINED ARCH)
+    set(ARCH_IN "x86_64")
+  else()
+    set(ARCH_IN ${ARCH})
+  endif()
 
-if (${CORE} STREQUAL "generic")
-  set(ARCH_IN "GENERIC")
-endif ()
+  if (${CORE} STREQUAL "generic")
+    set(ARCH_IN "GENERIC")
+  endif ()
 
-if (NOT DEFINED EXPRECISION)
-  set(EXPRECISION_IN 0)
-else()
-  set(EXPRECISION_IN ${EXPRECISION})
-endif()
+  if (NOT DEFINED EXPRECISION)
+    set(EXPRECISION_IN 0)
+  else()
+    set(EXPRECISION_IN ${EXPRECISION})
+  endif()
 
-if (NOT DEFINED NO_CBLAS)
-  set(NO_CBLAS_IN 0)
-else()
-  set(NO_CBLAS_IN ${NO_CBLAS})
-endif()
+  if (NOT DEFINED NO_CBLAS)
+    set(NO_CBLAS_IN 0)
+  else()
+    set(NO_CBLAS_IN ${NO_CBLAS})
+  endif()
 
-if (NOT DEFINED NO_LAPACK)
-  set(NO_LAPACK_IN 0)
-else()
-  set(NO_LAPACK_IN ${NO_LAPACK})
-endif()
+  if (NOT DEFINED NO_LAPACK)
+    set(NO_LAPACK_IN 0)
+  else()
+    set(NO_LAPACK_IN ${NO_LAPACK})
+  endif()
 
-if (NOT DEFINED NO_LAPACKE)
-  set(NO_LAPACKE_IN 0)
-else()
-  set(NO_LAPACKE_IN ${NO_LAPACKE})
-endif()
+  if (NOT DEFINED NO_LAPACKE)
+    set(NO_LAPACKE_IN 0)
+  else()
+    set(NO_LAPACKE_IN ${NO_LAPACKE})
+  endif()
 
-if (NOT DEFINED NEED2UNDERSCORES)
-  set(NEED2UNDERSCORES_IN 0)
-else()
-  set(NEED2UNDERSCORES_IN ${NEED2UNDERSCORES})
-endif()
+  if (NOT DEFINED NEED2UNDERSCORES)
+    set(NEED2UNDERSCORES_IN 0)
+  else()
+    set(NEED2UNDERSCORES_IN ${NEED2UNDERSCORES})
+  endif()
 
-if (NOT DEFINED ONLY_CBLAS)
-  set(ONLY_CBLAS_IN 0)
-else()
-  set(ONLY_CBLAS_IN ${ONLY_CBLAS})
-endif()
+  if (NOT DEFINED ONLY_CBLAS)
+    set(ONLY_CBLAS_IN 0)
+  else()
+    set(ONLY_CBLAS_IN ${ONLY_CBLAS})
+  endif()
 
-if (NOT DEFINED BU)
-  set(BU _)
-endif()
+  if (NOT DEFINED BU)
+    set(BU _)
+  endif()
 
-if (NOT ${SYMBOLPREFIX} STREQUAL "")
-message(STATUS "adding prefix ${SYMBOLPREFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
-endif()
-if (NOT ${SYMBOLSUFFIX} STREQUAL "")
-message(STATUS "adding suffix ${SYMBOLSUFFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
-endif()
-	add_custom_command(TARGET ${OpenBLAS_LIBNAME} POST_BUILD
-  	COMMAND perl  ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BUILD_LAPACK_DEPRECATED}" > ${PROJECT_BINARY_DIR}/objcopy.def
-        COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
-        COMMENT "renaming symbols"
-        )
+  if (NOT ${SYMBOLPREFIX} STREQUAL "")
+    message(STATUS "adding prefix ${SYMBOLPREFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
+  endif()
+  if (NOT ${SYMBOLSUFFIX} STREQUAL "")
+    message(STATUS "adding suffix ${SYMBOLSUFFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
+  endif()
+
+  add_custom_command(TARGET ${OpenBLAS_LIBNAME}_shared POST_BUILD
+    COMMAND perl  ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BUILD_LAPACK_DEPRECATED}" > ${PROJECT_BINARY_DIR}/objcopy.def
+    COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
+    COMMENT "renaming symbols"
+    )
 endif()
 
 
 # Install project
 
 # Install libraries
-install(TARGETS ${OpenBLAS_LIBNAME}
-	EXPORT "OpenBLAS${SUFFIX64}Targets"
-	RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
-	ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
-  LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
+if(BUILD_SHARED_LIBS AND BUILD_STATIC_LIBS)
+  install(TARGETS ${OpenBLAS_LIBNAME}_shared
+    EXPORT "OpenBLAS${SUFFIX64}Targets"
+    RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
+    ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
+  install(TARGETS ${OpenBLAS_LIBNAME}_static
+    ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
+else()
+  install(TARGETS ${OpenBLAS_LIBS}
+    EXPORT "OpenBLAS${SUFFIX64}Targets"
+    RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
+    ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
+endif()
 
 # Install headers
 set(CMAKE_INSTALL_INCLUDEDIR ${CMAKE_INSTALL_INCLUDEDIR}/openblas${SUFFIX64})
@@ -362,36 +456,41 @@ if(NOT NOFORTRAN)
 endif()
 
 if(NOT NO_CBLAS)
-	message (STATUS "Generating cblas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
-	set(CBLAS_H ${CMAKE_BINARY_DIR}/generated/cblas.h)
-	file(READ ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h CBLAS_H_CONTENTS)
-	string(REPLACE "common" "openblas_config" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
-	if (NOT ${SYMBOLPREFIX} STREQUAL "")
-	string(REPLACE " cblas" " ${SYMBOLPREFIX}cblas" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
-	string(REPLACE " openblas" " ${SYMBOLPREFIX}openblas" CBLAS_H_CONTENTS_NEW	"${CBLAS_H_CONTENTS}")
-	string (REPLACE " ${SYMBOLPREFIX}openblas_complex" " openblas_complex" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
-	string(REPLACE " goto" " ${SYMBOLPREFIX}goto" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
-	endif()
-	if (NOT ${SYMBOLSUFFIX} STREQUAL "")
-	string(REGEX REPLACE "(cblas[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
-	string(REGEX REPLACE "(openblas[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
-	string(REGEX REPLACE "(openblas_complex[^ ]*)${SYMBOLSUFFIX}" "\\1" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
-	string(REGEX REPLACE "(goto[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS_NEW	"${CBLAS_H_CONTENTS}")
-	endif()
-	file(WRITE ${CBLAS_H} "${CBLAS_H_CONTENTS_NEW}")
-	install (FILES ${CBLAS_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
+  message (STATUS "Generating cblas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
+  set(CBLAS_H ${CMAKE_BINARY_DIR}/generated/cblas.h)
+  file(READ ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h CBLAS_H_CONTENTS)
+  string(REPLACE "common" "openblas_config" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
+  if (NOT ${SYMBOLPREFIX} STREQUAL "")
+    string(REPLACE " cblas" " ${SYMBOLPREFIX}cblas" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+    string(REPLACE " openblas" " ${SYMBOLPREFIX}openblas" CBLAS_H_CONTENTS_NEW	"${CBLAS_H_CONTENTS}")
+    string (REPLACE " ${SYMBOLPREFIX}openblas_complex" " openblas_complex" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+    string(REPLACE " goto" " ${SYMBOLPREFIX}goto" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
+  endif()
+  if (NOT ${SYMBOLSUFFIX} STREQUAL "")
+    string(REGEX REPLACE "(cblas[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+    string(REGEX REPLACE "(openblas[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
+    string(REGEX REPLACE "(openblas_complex[^ ]*)${SYMBOLSUFFIX}" "\\1" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+    string(REGEX REPLACE "(goto[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS_NEW	"${CBLAS_H_CONTENTS}")
+  endif()
+  file(WRITE ${CBLAS_H} "${CBLAS_H_CONTENTS_NEW}")
+  install (FILES ${CBLAS_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 endif()
 
 if(NOT NO_LAPACKE)
-	message (STATUS "Copying LAPACKE header files to ${CMAKE_INSTALL_INCLUDEDIR}")
-	add_dependencies( ${OpenBLAS_LIBNAME} genlapacke)
-	FILE(GLOB_RECURSE INCLUDE_FILES "${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/*.h")
-	install (FILES ${INCLUDE_FILES} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
-
-	ADD_CUSTOM_TARGET(genlapacke
-	COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/include/lapacke_mangling_with_flags.h.in "${CMAKE_BINARY_DIR}/lapacke_mangling.h"
-	)
-	install (FILES ${CMAKE_BINARY_DIR}/lapacke_mangling.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/openblas${SUFFIX64})
+  message (STATUS "Copying LAPACKE header files to ${CMAKE_INSTALL_INCLUDEDIR}")
+  if(BUILD_STATIC_LIBS)
+    add_dependencies( ${OpenBLAS_LIBNAME}_static genlapacke)
+  endif()
+  if(BUILD_SHARED_LIBS)
+    add_dependencies( ${OpenBLAS_LIBNAME}_shared genlapacke)
+  endif()
+  FILE(GLOB_RECURSE INCLUDE_FILES "${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/*.h")
+  install (FILES ${INCLUDE_FILES} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
+
+  ADD_CUSTOM_TARGET(genlapacke
+  COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/include/lapacke_mangling_with_flags.h.in "${CMAKE_BINARY_DIR}/lapacke_mangling.h"
+  )
+  install (FILES ${CMAKE_BINARY_DIR}/lapacke_mangling.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/openblas${SUFFIX64})
 endif()
 
 # Install pkg-config files
@@ -416,4 +515,3 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${PN}ConfigVersion.cmake
 install(EXPORT "${PN}${SUFFIX64}Targets"
         NAMESPACE "${PN}${SUFFIX64}::"
         DESTINATION ${CMAKECONFIG_INSTALL_DIR})
-

CONTRIBUTORS.md

@@ -194,3 +194,16 @@ In chronological order:
 
 * PingTouGe Semiconductor Co., Ltd.
   * [2020-10] Add RISC-V Vector (0.7.1) support. Optimize BLAS kernels for Xuantie C910
+
+* River Dillon <oss@outerpassage.net>
+  * [2021-07-10] fix compilation with musl libc
+
+* Bine Brank <https://github.com/binebrank>
+  * [2021-10-27] Add vector-length-agnostic DGEMM kernels for Arm SVE
+  * [2021-11-20] Vector-length-agnostic Arm SVE copy routines for DGEMM, DTRMM, DSYMM
+  * [2021-11-12] SVE kernels for SGEMM, STRMM and corresponding SVE copy functions
+  * [2022-01-06] SVE kernels for CGEMM, ZGEMM, CTRMM, ZTRMM and corresponding SVE copy functions
+  * [2022-01-18] SVE kernels and copy functions for TRSM
+
+* Ilya Kurdyukov <https://github.com/ilyakurdyukov>
+  * [2021-02-21] Add basic support for the Elbrus E2000 architecture

Changelog.txt

@@ -1,4 +1,340 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.20
+ 20-Feb-2022
+
+general:
+ - some code cleanup, with added casts etc.
+ - fixed obtaining the cpu count with OpenMP and OMP_PROC_BIND unset
+ - fixed pivot index calculation by ?LASWP for negative increments other than one
+ - fixed input argument check in LAPACK ? GEQRT2
+ - improved the check for a Fortran compiler in CMAKE builds
+ - disabled building OpenBLAS' optimized versions of LAPACK complex SPMV,SPR,SYMV,SYR with NO_LAPACK=1
+ - fixed building of LAPACK on certain distributed filesystems with parallel gmake
+ - fixed building the shared library on MacOS with classic flang
+
+x86_64:
+ - fixed cross-compilation with CMAKE for CORE2 target
+ - fixed miscompilation of AVX512 code in DYNAMIC_ARCH builds
+ - added support for the "incidental" AVX512 hardware in Alder Lake when enabled in BIOS
+
+E2K:
+ - add new architecture (Russian Elbrus E2000 family)
+
+SPARC:
+ - fix IMIN/IMAX
+
+ARMV8:
+ - added SVE-enabled CGEMM and ZGEMM kernels for ARMV8SVE and A64FX
+ - added support for Neoverse N2 and V1 cpus
+
+MIPS,MIPS64:
+ - fixed autodetection of MSA capability
+
+LOONGARCH64:
+ - added an optimized DGEMM kernel
+
+====================================================================
+Version 0.3.19
+ 19-Dec-2021
+ 
+ general:
+ - reverted unsafe TRSV/ZRSV optimizations introduced in 0.3.16
+ - fixed a potential thread race in the thread buffer reallocation routines
+   that were introduced in 0.3.18
+ - fixed miscounting of thread pool size on Linux with OMP_PROC_BIND=TRUE
+ - fixed CBLAS interfaces for CSROT/ZSROT and CROTG/ZROTG
+ - made automatic library suffix for CMAKE builds with INTERFACE64 available
+   to CBLAS-only builds
+
+x86_64:
+ - DYNAMIC_ARCH builds now fall back to the cpu with most similar capabilities
+   when an unknown CPUID is encountered, instead of defaulting to Prescott
+ - added cpu detection for Intel Alder Lake
+ - added cpu detection for Intel Sapphire Rapids
+ - added an optimized SBGEMM kernel for Sapphire Rapids
+ - fixed DYNAMIC_ARCH builds on OSX with CMAKE
+ - worked around DYNAMIC_ARCH builds made on Sandybridge failing on SkylakeX
+ - fixed missing thread initialization for static builds on Windows/MSVC
+ - fixed an excessive read in ZSYMV
+
+POWER:
+ - added support for POWER10 in big-endian mode
+ - added support for building with CMAKE
+ - added optimized SGEMM and DGEMM kernels for small matrix sizes
+
+ARMV8:
+ - added basic support and cputype detection for Fujitsu A64FX
+ - added a generic ARMV8SVE target  
+ - added SVE-enabled SGEMM and DGEMM kernels for ARMV8SVE and A64FX
+ - added optimized CGEMM and ZGEMM kernels for Cortex A53 and A55 cpus
+ - fixed cpuid detection for Apple M1 and improved performance
+ - improved compiler flag setting in CMAKE builds
+
+RISCV64:
+ - fixed improper initialization in CSCAL/ZSCAL for strided access patterns
+
+MIPS:
+ - added a GENERIC target for MIPS32
+ - added support for cross-compiling to MIPS32 on x86_64 using CMAKE
+
+MIPS64:
+ - fixed misdetection of MSA capability
+
+====================================================================
+Version 0.3.18
+ 02-Oct-2021
+
+general:
+ - when the build-time number of preconfigured threads is exceeded
+   at runtime (typically by an external program calling BLAS functions
+   from a larger number of threads in parallel), OpenBLAS will now 
+   allocate an auxiliary control structure for up to 512 additional
+   threads instead of aborting
+ - added support for Loongson's LoongArch64 cpu architecture
+ - fixed building OpenBLAS with CMAKE and -DBUILD_BFLOAT16=ON
+ - added support for building OpenBLAS as a CMAKE subproject
+ - added support for building for Windows/ARM64 targets with clang
+ - improved support for building with the IBM xlf compiler
+ - imported Reference-LAPACK PR 625 (out-of-bounds reads in ?LARRV)
+ - imported Reference-LAPACK PR 597 for testsuite compatibility with
+   LLVM's libomp
+
+x86_64:
+ - added SkylakeX S/DGEMM kernels for small problem sizes (M*N*K<=1000000)
+ - added optimized SBGEMM for Intel Cooper Lake
+ - reinstated the performance patch for AVX512 SGEMV_T with a proper fix
+ - added a workaround for a gcc11 tree-vectorizer bug that caused spurious
+   failures in the test programs for complex BLAS3 when compiling at -O3
+   (the default for cmake "release" builds)
+ - added support for runtime cpu count detection under Haiku OS
+ - worked around a long-standing miscompilation issue of the Haswell DGEMV_T
+   kernel with gcc that could produce NaN output in some corner cases
+
+POWER:
+ - improved performance of DASUM on POWER10
+
+ARMV8:
+ - fixed crashes (use of reserved register x18) on Apple M1 under OSX
+ - fixed building with gcc releases earlier than 5.1
+
+MIPS:
+ - fixed building under BSD
+
+MIPS64:
+ - fixed building under BSD
+
+====================================================================
+Version 0.3.17
+ 15-Jul-2021
+
+common:
+ - reverted the optimization of SGEMV_N/DGEMV_N for small input sizes
+   and consecutive arguments as it led to stack overflows on x86_64
+   with some operating systems (notably OSX and Windows)
+ 
+ x86_64:
+  - reverted the performance patch for SGEMV_T on AVX512 as it caused
+    wrong results in some applications
+    
+ SPARC:
+  - fixed compilation with compilers other than gcc
+====================================================================
+Version 0.3.16
+ 11-Jul-2021
+
+common:
+ - drastically reduced the stack size requirements for running the LAPACK
+   testsuite (Reference-LAPACK PR 553)
+ - fixed spurious test failures in the LAPACK testsuite (Reference-LAPACK
+   PR 564)
+ - expressly setting DYNAMIC_ARCH=0 no longer enables dynamic_arch mode 
+ - improved performance of xGER, xSPR, xSPR2, xSYR, xSYR2, xTRSV, SGEMV_N 
+   and DGEMV_N, for small input sizes and consecutive arguments
+ - improved performance of xGETRF, xPORTF and xPOTRI for small input sizes
+   by disabling multithreading
+ - fixed installing with BSD versions of the "install" utility
+
+RISCV:
+ - fixed the implementation of xIMIN
+ - improved the performance of DSDOT
+ - fixed linking of the tests on C910V with current vendor gcc
+
+POWER:
+- fixed SBGEMM computation for some odd value inputs
+- fixed compilation for PPCG4, PPC970, POWER3, POWER4 and POWER5
+
+x86_64:
+ - improved performance of SGEMV_N and SGEMV_T for small N on AVX512-capable cpus
+ - worked around a miscompilation of ZGEMM/ZTRMM on Sandybridge with old gcc
+   versions
+ - fixed compilation with MS Visual Studio versions older than 2017
+ - fixed macro name collision with winnt.h from the latest Win10 SDK
+ - added cpu type autodetection for Intel Ice Lake SP
+ - fixed cpu type autodetection for Intel Tiger Lake
+ - added cpu type autodetection for recent Centaur/Zhaoxin models
+ - fixed compilation with musl libc
+
+ARM64:
+- fixed compilation with gcc/gfortran on the Apple M1
+- fixed linking of the tests on FreeBSD
+- fixed missing restore of a register in the recently rewritten DNRM2 kernel
+  for ThunderX2 and Neoverse N1 that could cause spurious failures in e.g.
+  DGEEV
+- added compiler optimization flags for the EMAG8180
+- added initial support for Cortex A55
+
+ARM:
+- fixed linking of the tests on FreeBSD
+
+====================================================================
+Version 0.3.15
+  2-May-2021
+
+common:
+ - imported improvements and bugfixes from Reference-LAPACK 3.9.1
+ - imported LAPACKE interface fixes from Reference-LAPACK PRs 534 + 537
+ - fixed a problem in the cpu detection of 0.3.14 that prevented cross-compilation
+ - fixed a sequence problem in the generation of softlinks to the library in GMAKE
+
+RISC V:
+ - fixed compilation on RISCV (missing entry in getarch)
+ - fixed a potential division by zero in CROTG and ZROTG
+
+POWER:
+ - fixed LAPACK testsuite failures seen with the NVIDIA HPC compiler
+ - improved CGEMM, DGEMM and ZGEMM performance on POWER10
+ - added an optimized ZGEMV kernel for POWER10
+ - fixed a potential division by zero in CROTG and ZROTG
+
+x86_64:
+ - added support for Intel Control-flow Enforcement Technology (CET)
+ - reverted the DOMATCOPY_RT code to the generic C version
+ - fixed a bug in the AVX512 SGEMM kernel introduced in 0.3.14
+ - fixed misapplication of -msse flag to non-SSE cpus in DYNAMIC_ARCH
+ - added support for compilation of the benchmarks on older OSX versions
+ - fix propagation of the NO_AVX512 option in CMAKE builds
+ - fix compilation of the AVX512 SGEMM kernel with clang-cl on Windows
+ - fixed compilation of the CTESTs with INTERFACE64=1 (random faults on OSX)
+ - corrected the Haswell DROT kernel to require AVX2/FMA3 rather than AVX512
+
+ARM:
+ - fixed a potential division by zero in CROTG and ZROTG
+ - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
+
+ARM64:
+ - fixed spurious reads outside the array in the SGEMM tcopy macro
+ - fixed a potential division by zero in CROTG and ZROTG
+ - fixed a segmentation fault in DYNAMIC_ARCH builds (reappeared in 0.3.14)
+
+MIPS
+ - fixed a potential division by zero in CROTG and ZROTG
+ - fixed a potential overflow in IMATCOPY/ZIMATCOPY and the CTESTs
+
+MIPS64:
+ - fixed a potential division by zero in CROTG and ZROTG
+
+SPARC:
+ - fixed a potential division by zero in CROTG and ZROTG
+
+====================================================================
+Version 0.3.14
+ 17-Mar-2021
+ 
+ common:
+	* Fixed a race condition on thread shutdown in non-OpenMP builds
+	* Fixed custom BUFFERSIZE option getting ignored in gmake builds
+	* Fixed CMAKE compilation of the TRMM kernels for GENERIC platforms
+	* Added CBLAS interfaces for CROTG, ZROTG, CSROT and ZDROT
+	* Improved performance of OMATCOPY_RT across all platforms
+	* Changed perl scripts to use env instead of a hardcoded /usr/bin/perl
+	* Fixed potential misreading of the GCC compiler version in the build scripts
+	* Fixed convergence problems in LAPACK complex GGEV/GGES (Reference-LAPACK #477)
+	* Reduced the stacksize requirements for running the LAPACK testsuite (Reference-LAPACK #335)
+
+ RISCV:
+ 	* Fixed compilation on RISCV (missing entry in getarch)
+	
+ POWER:
+ 	* Fixed compilation for DYNAMIC_ARCH with clang and with old gcc versions
+	* Added support for compilation on FreeBSD/ppc64le
+ 	* Added optimized POWER10 kernels for SSCAL, DSCAL, CSCAL, ZSCAL
+	* Added optimized POWER10 kernels for SROT, DROT, CDOT, SASUM, DASUM
+	* Improved SSWAP, DSWAP, CSWAP, ZSWAP performance on POWER10
+	* Improved SCOPY and CCOPY performance on POWER10
+	* Improved SGEMM and DGEMM performance on POWER10
+	* Added support for compilation with the NVIDIA HPC compiler
+
+ x86_64:
+	* Added an optimized bfloat16 GEMM kernel for Cooperlake
+	* Added CPUID autodetection for Intel Rocket Lake and Tiger Lake cpus
+	* Improved the performance of SASUM,DASUM,SROT,DROT on AMD Ryzen cpus
+	* Added support for compilation with the NAG Fortran compiler
+	* Fixed recognition of the AMD AOCC compiler
+	* Fixed compilation for DYNAMIC_ARCH with clang on Windows
+	* Added support for running the BLAS/CBLAS tests on Windows
+	* Fixed signatures of the tls callback functions for Windows x64
+	* Fixed various issues with fma intrinsics support handling 
+
+ ARM:
+	* Added support for embedded Cortex M targets via a new option EMBEDDED
+
+ ARMV8:
+ 	* Fixed the THUNDERX2T99 and NEOVERSEN1 DNRM2/ZNRM2 kernels for inputs with Inf
+	* Added support for the DYNAMIC_LIST option
+	* Added support for compilation with the NVIDIA HPC compiler
+ 	* Added support for compiling with the NAG Fortran compiler
+
+====================================================================
+Version 0.3.13
+ 12-Dec-2020
+ 
+ common:
+	* Added a generic bfloat16 SBGEMV kernel
+	* Fixed a potentially severe memory leak after fork in OpenMP builds
+	  that was introduced in 0.3.12
+	* Added detection of the Fujitsu Fortran compiler
+	* Added detection of the (e)gfortran compiler on OpenBSD
+	* Added support for overriding the default name of the library independently
+	  from symbol suffixing in the gmake builds (already supported in cmake)
+
+RISCV:
+	* Added a RISC V port optimized for C910V
+
+POWER:
+	* Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
+	* Improved DGEMM performance on POWER10
+	* Improved STRSM and DTRSM performance on POWER9 and POWER10
+	* Fixed segmemtation faults in DYNAMIC_ARCH builds
+ 	* Fixed compilation with the PGI compiler
+
+x86:
+	* Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
+	
+x86_64:
+	* Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
+	* Improved the performance of SASUM and DASUM kernels through parallelization
+	* Improved the performance of SROT and DROT kernels
+	* Improved the performance of multithreaded xSYRK
+	* Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
+	  (where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
+	  wrong results)
+	* Fixed miscompilations by old gcc 4.6
+	* Fixed misdetection of AVX2 capability in some Sandybridge cpus
+	* Fixed lockups in builds combining DYNAMIC_ARCH with TARGET=GENERIC on OpenBSD
+
+ARM64:
+	* Fixed segmemtation faults in DYNAMIC_ARCH builds
+
+MIPS:
+	* Improved kernels for Loongson 3R3 ("3A") and 3R4 ("3B") models, including MSA
+	* Fixed bugs in the MSA kernels for CGEMM, CTRMM, CGEMV and ZGEMV
+	* Added handling of zero increments in the MSA kernels for SSWAP and DSWAP
+	* Added DYNAMIC_ARCH support for MIPS64 (currently Loongson3R3/3R4 only)
+
+SPARC:
+	* Fixed building 32 and 64 bit SPARC kernels with the SolarisStudio compilers
+
 ====================================================================
 Version 0.3.12
  24-Oct-2020

Makefile

@@ -32,7 +32,7 @@ export NOFORTRAN
 export NO_LAPACK
 endif
 
-LAPACK_NOOPT := $(filter-out -O0 -O1 -O2 -O3 -Ofast,$(LAPACK_FFLAGS))
+LAPACK_NOOPT := $(filter-out -O0 -O1 -O2 -O3 -Ofast -O -Og -Os,$(LAPACK_FFLAGS))
 
 SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench cpp_thread_test
 
@@ -59,6 +59,9 @@ endif
 	@$(CC) --version > /dev/null 2>&1;\
 	if [ $$? -eq 0 ]; then \
 	   cverinfo=`$(CC) --version | sed -n '1p'`; \
+	   if [ -z "$${cverinfo}" ]; then \
+	   cverinfo=`$(CC) --version | sed -n '2p'`; \
+	   fi; \
 	   echo "  C compiler       ... $(C_COMPILER)  (cmd & version : $${cverinfo})";\
 	else  \
 	   echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))";\
@@ -67,6 +70,9 @@ ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	@$(FC) --version > /dev/null 2>&1;\
 	if [ $$? -eq 0 ]; then \
 	   fverinfo=`$(FC) --version | sed -n '1p'`; \
+	   if [ -z "$${fverinfo}" ]; then \
+	   fverinfo=`$(FC) --version | sed -n '2p'`; \
+	   fi; \
 	   echo "  Fortran compiler ... $(F_COMPILER)  (cmd & version : $${fverinfo})";\
 	else \
 	   echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))";\
@@ -161,7 +167,6 @@ ifeq ($(NO_SHARED), 1)
 	$(error OpenBLAS: neither static nor shared are enabled.)
 endif
 endif
-	@-ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 	@for d in $(SUBDIRS) ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d $(@F) || exit 1 ; \
@@ -190,6 +195,7 @@ endif
 ifdef USE_THREAD
 	@echo USE_THREAD=$(USE_THREAD) >>  Makefile.conf_last
 endif
+	@-ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 	@touch lib.grd
 
 prof : prof_blas prof_lapack
@@ -263,7 +269,7 @@ prof_lapack : lapack_prebuild
 lapack_prebuild :
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	-@echo "FC          = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "override FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS_DRV  = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS_NOOPT       = -O0 $(LAPACK_NOOPT)" >> $(NETLIB_LAPACK_DIR)/make.inc

Makefile.arm64

@@ -1,80 +1,234 @@
+ifneq ($(C_COMPILER), PGI)
+
+ifeq ($(C_COMPILER), CLANG)
+ISCLANG=1
+endif
+ifneq (1, $(filter 1,$(GCCVERSIONGT4) $(ISCLANG)))
+CCOMMON_OPT += -march=armv8-a
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a
+endif
+
+
+else 
+
 
 ifeq ($(CORE), ARMV8)
 CCOMMON_OPT += -march=armv8-a
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a
 endif
+endif
+
+ifeq ($(CORE), ARMV8SVE)
+CCOMMON_OPT += -march=armv8-a+sve
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a+sve
+endif
+endif
 
 ifeq ($(CORE), CORTEXA53)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
 endif
+endif
 
 ifeq ($(CORE), CORTEXA57)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a57
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a57
 endif
+endif
 
 ifeq ($(CORE), CORTEXA72)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
 endif
+endif
 
 ifeq ($(CORE), CORTEXA73)
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a73
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a73
 endif
+endif
 
 # Use a72 tunings because Neoverse-N1 is only available
 # in GCC>=9
 ifeq ($(CORE), NEOVERSEN1)
-ifeq ($(GCCVERSIONGTEQ7), 1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
 ifeq ($(GCCVERSIONGTEQ9), 1)
 CCOMMON_OPT += -march=armv8.2-a -mtune=neoverse-n1
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=neoverse-n1
+endif
+else
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
+endif
+endif
+else
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+endif
+endif
+endif
+
+# Use a72 tunings because Neoverse-V1 is only available
+# in GCC>=9.4
+ifeq ($(CORE), NEOVERSEV1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
+ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq (1, $(filter 1,$(GCCMINORVERSIONGTEQ4) $(GCCVERSIONGTEQ10)))
+CCOMMON_OPT += -march=armv8.4-a -mtune=neoverse-v1
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.4-a -mtune=neoverse-v1
+endif
+else
+CCOMMON_OPT += -march=armv8.4-a -mtune=native
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.4-a -mtune=native
+endif
+endif
+else
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
+endif
+endif
+else
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+endif
+endif
+endif
+
+# Use a72 tunings because Neoverse-N2 is only available
+# in GCC>=9.4
+ifeq ($(CORE), NEOVERSEN2)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
+ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq (1, $(filter 1,$(GCCMINORVERSIONGTEQ4) $(GCCVERSIONGTEQ10)))
+CCOMMON_OPT += -march=armv8.5-a -mtune=neoverse-n2
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.5-a -mtune=neoverse-n2
+endif
+else
+CCOMMON_OPT += -march=armv8.5-a -mtune=native
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.5-a -mtune=native
+endif
+endif
 else
 CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a72
 endif
+endif
 else
 CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
 endif
 endif
+endif
+
+# Use a53 tunings because a55 is only available in GCC>=8.1
+ifeq ($(CORE), CORTEXA55)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
+ifeq ($(GCCVERSIONGTEQ8), 1)
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
+endif
+else
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a53
+endif
+endif
+else
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+endif
+endif
+endif
 
 ifeq ($(CORE), THUNDERX)
 CCOMMON_OPT += -march=armv8-a -mtune=thunderx
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=thunderx
 endif
+endif
 
 ifeq ($(CORE), FALKOR)
 CCOMMON_OPT += -march=armv8-a -mtune=falkor
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=falkor
 endif
+endif
 
 ifeq ($(CORE), THUNDERX2T99)
 CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
+endif
 
 ifeq ($(CORE), THUNDERX3T110)
 ifeq ($(GCCVERSIONGTEQ10), 1)
 CCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+endif
 else
 CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
 endif
+endif
 
 ifeq ($(CORE), VORTEX)
 CCOMMON_OPT += -march=armv8.3-a
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.3-a
 endif
+endif
 
-ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ9) $(ISCLANG)))
 ifeq ($(CORE), TSV110)
 CCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
 endif
 endif
+endif
 
+ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq ($(CORE), EMAG8180)
+CCOMMON_OPT += -march=armv8-a -mtune=emag
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=emag
+endif
+endif
+endif
+
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ11) $(ISCLANG)))
+ifeq ($(CORE), A64FX)
+CCOMMON_OPT += -march=armv8.2-a+sve -mtune=a64fx
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a+sve -mtune=a64fx
+endif
+endif
+endif
+
+endif
+
+endif

Makefile.e2k

@@ -0,0 +1 @@
+COPT	= -Wall -O2 # -DGEMMTEST

Makefile.install

@@ -74,17 +74,17 @@ endif
 ifneq ($(OSNAME), AIX)
 ifndef NO_LAPACKE
 	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapack.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapack.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_utils.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapack.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapack.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_utils.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h"
 endif
 
 #for install static library
 ifneq ($(NO_STATIC),1)
 	@echo Copying the static library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
-	@install -pm644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@install -m644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 endif
@@ -92,7 +92,7 @@ endif
 ifneq ($(NO_SHARED),1)
 	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
 ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku FreeBSD DragonFly))
-	@install -pm755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@install -m755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)

Makefile.loongarch64

@@ -0,0 +1,3 @@
+ifdef BINARY64
+else
+endif

Makefile.power

@@ -10,9 +10,15 @@ USE_OPENMP = 1
 endif
 
 ifeq ($(CORE), POWER10)
+ifneq ($(C_COMPILER), PGI)
 CCOMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
+ifeq ($(F_COMPILER), IBM)
+FCOMMON_OPT += -O2 -qrecur -qnosave
+else
 FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -fno-fast-math
 endif
+endif
+endif
 
 ifeq ($(CORE), POWER9)
 ifneq ($(C_COMPILER), PGI)
@@ -31,7 +37,11 @@ else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
+ifeq ($(F_COMPILER), IBM)
+FCOMMON_OPT += -O2 -qrecur -qnosave
+else
 FCOMMON_OPT += -O2 -frecursive -fno-fast-math
+endif
 ifeq ($(C_COMPILER), GCC)
 ifneq ($(GCCVERSIONGT4), 1)
 $(warning your compiler is too old to fully support POWER9, getting a newer version of gcc is recommended)
@@ -55,7 +65,11 @@ CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
 ifeq ($(OSNAME), AIX)
+ifeq ($(F_COMPILER), IBM)
+FCOMMON_OPT += -O2 -qrecur -qnosave
+else
 FCOMMON_OPT += -O1 -frecursive -mcpu=power8 -mtune=power8  -fno-fast-math 
+endif
 else
 FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8  -fno-fast-math 
 endif

Makefile.prebuild

@@ -3,6 +3,10 @@
 export BINARY
 export USE_OPENMP
 
+ifdef DYNAMIC_ARCH
+override HOST_CFLAGS += -DDYNAMIC_ARCH
+endif
+
 ifdef TARGET_CORE
 TARGET_MAKE = Makefile_kernel.conf
 TARGET_CONF = config_kernel.h

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.12.dev
+VERSION = 0.3.20
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library

Makefile.system

@@ -9,11 +9,10 @@ ifndef TOPDIR
 TOPDIR = . 
 endif
 
- # If ARCH is not set, we use the host system's architecture for getarch compile options.
-ifndef ARCH
+# we need to use the host system's architecture for getarch compile options even especially when cross-compiling
 HOSTARCH := $(shell uname -m)
-else
-HOSTARCH = $(ARCH)
+ifeq ($(HOSTARCH), amd64)
+HOSTARCH=x86_64
 endif
 
 # Catch conflicting usage of ARCH in some BSD environments
@@ -21,6 +20,8 @@ ifeq ($(ARCH), amd64)
 override ARCH=x86_64
 else ifeq ($(ARCH), powerpc64)
 override ARCH=power
+else ifeq ($(ARCH), powerpc64le)
+override ARCH=power
 else ifeq ($(ARCH), powerpc)
 override ARCH=power
 else ifeq ($(ARCH), i386)
@@ -31,6 +32,10 @@ else ifeq ($(ARCH), armv7)
 override ARCH=arm
 else ifeq ($(ARCH), aarch64)
 override ARCH=arm64
+else ifeq ($(ARCH), mipsel)
+override ARCH=mips
+else ifeq ($(ARCH), mips64el)
+override ARCH=mips64
 else ifeq ($(ARCH), zarch)
 override ARCH=zarch
 endif
@@ -96,7 +101,7 @@ GETARCH_FLAGS += -DUSER_TARGET
 ifeq ($(TARGET), GENERIC)
 ifeq ($(DYNAMIC_ARCH), 1)
 override NO_EXPRECISION=1
-export NO_EXPRECiSION
+export NO_EXPRECISION
 endif
 endif
 endif
@@ -113,6 +118,9 @@ endif
 ifeq ($(TARGET), COOPERLAKE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
+ifeq ($(TARGET), SAPPHIRERAPIDS)
+GETARCH_FLAGS := -DFORCE_NEHALEM
+endif
 ifeq ($(TARGET), SANDYBRIDGE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
@@ -137,8 +145,13 @@ endif
 ifeq ($(TARGET), POWER8)
 GETARCH_FLAGS := -DFORCE_POWER6
 endif
+ifeq ($(TARGET), POWER9)
+GETARCH_FLAGS := -DFORCE_POWER6
+endif
+ifeq ($(TARGET), POWER10)
+GETARCH_FLAGS := -DFORCE_POWER6
+endif
 endif
-
 
 #TARGET_CORE will override TARGET which is used in DYNAMIC_ARCH=1.
 #
@@ -158,6 +171,9 @@ endif
 ifeq ($(TARGET_CORE), COOPERLAKE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
+ifeq ($(TARGET_CORE), SAPPHIRERAPIDS)
+GETARCH_FLAGS := -DFORCE_NEHALEM
+endif
 ifeq ($(TARGET_CORE), SANDYBRIDGE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
@@ -181,7 +197,7 @@ endif
 
 # On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
 ifeq ($(HOSTARCH), x86_64)
-ifeq ($(findstring pgcc,$(HOSTCC)),)
+ifeq ($(findstring pgcc,$(HOSTCC))$(findstring nvc,$(HOSTCC)),)
 GETARCH_FLAGS += -march=native
 endif
 endif
@@ -242,12 +258,26 @@ else
 ONLY_CBLAS = 0
 endif
 
+#For small matrix optimization
+ifeq ($(ARCH), x86_64)
+SMALL_MATRIX_OPT = 1
+else ifeq ($(CORE), POWER10)
+SMALL_MATRIX_OPT = 1
+endif
+ifeq ($(SMALL_MATRIX_OPT), 1)
+CCOMMON_OPT += -DSMALL_MATRIX_OPT
+endif
+
 # This operation is expensive, so execution should be once.
 ifndef GOTOBLAS_MAKEFILE
 export GOTOBLAS_MAKEFILE = 1
 
+# Determine if the assembler is GNU Assembler
+HAVE_GAS := $(shell $(AS) -v < /dev/null 2>&1 | grep GNU 2>&1 >/dev/null ; echo $$?)
+GETARCH_FLAGS += -DHAVE_GAS=$(HAVE_GAS)
+
 # Generating Makefile.conf and config.h
-DUMMY := $(shell $(MAKE) -C $(TOPDIR) -f Makefile.prebuild CC="$(CC)" FC="$(FC)" HOSTCC="$(HOSTCC)" HOST_CFLAGS="$(GETARCH_FLAGS)" CFLAGS="$(CFLAGS)" BINARY=$(BINARY) USE_OPENMP=$(USE_OPENMP) TARGET_CORE=$(TARGET_CORE) ONLY_CBLAS=$(ONLY_CBLAS) TARGET=$(TARGET) all)
+DUMMY := $(shell $(MAKE) -C $(TOPDIR) -f Makefile.prebuild CC="$(CC)" FC="$(FC)" HOSTCC="$(HOSTCC)" HOST_CFLAGS="$(GETARCH_FLAGS)" CFLAGS="$(CFLAGS)" BINARY=$(BINARY) USE_OPENMP=$(USE_OPENMP) DYNAMIC_ARCH=$(DYNAMIC_ARCH) TARGET_CORE=$(TARGET_CORE) ONLY_CBLAS=$(ONLY_CBLAS) TARGET=$(TARGET) all)
 
 ifndef TARGET_CORE
 include $(TOPDIR)/Makefile.conf
@@ -293,7 +323,7 @@ else
 SMP = 1
 endif
 else
-ifeq ($(NUM_THREAD), 1)
+ifeq ($(NUM_THREADS), 1)
 SMP =
 else
 SMP = 1
@@ -331,6 +361,7 @@ GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
 GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
 GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
 GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
+GCCVERSIONGTEQ8 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 8)
 GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
 GCCVERSIONGTEQ11 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 11)
 GCCVERSIONGTEQ10 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 10)
@@ -343,6 +374,7 @@ else
 endif
 GCCMINORVERSIONGTEQ1 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 1)
 GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 2)
+GCCMINORVERSIONGTEQ4 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 4)
 GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 7)
 endif
 
@@ -378,6 +410,12 @@ ifeq ($(OSNAME), AIX)
 EXTRALIB	+= -lm
 endif
 
+ifeq ($(OSNAME), FreeBSD)
+ifeq ($(ARCH), $(filter $(ARCH),arm arm64))
+EXTRALIB	+= -lm
+endif
+endif
+
 ifeq ($(OSNAME), WINNT)
 NEED_PIC = 0
 NO_EXPRECISION = 1
@@ -617,12 +655,24 @@ DYNAMIC_CORE += CORTEXA57
 DYNAMIC_CORE += CORTEXA72
 DYNAMIC_CORE += CORTEXA73
 DYNAMIC_CORE += NEOVERSEN1
+DYNAMIC_CORE += NEOVERSEV1
+DYNAMIC_CORE += NEOVERSEN2
+DYNAMIC_CORE += CORTEXA55
 DYNAMIC_CORE += FALKOR
 DYNAMIC_CORE += THUNDERX
 DYNAMIC_CORE += THUNDERX2T99
 DYNAMIC_CORE += TSV110
 DYNAMIC_CORE += EMAG8180
 DYNAMIC_CORE += THUNDERX3T110
+ifdef DYNAMIC_LIST
+override DYNAMIC_CORE = ARMV8 $(DYNAMIC_LIST)
+XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_ARMV8
+XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
+endif
+endif
+
+ifeq ($(ARCH), mips64)
+DYNAMIC_CORE = LOONGSON3R3 LOONGSON3R4
 endif
 
 ifeq ($(ARCH), zarch)
@@ -659,6 +709,7 @@ endif
 endif # ARCH zarch
 
 ifeq ($(ARCH), power)
+ifneq ($(C_COMPILER), PGI)
 DYNAMIC_CORE = POWER6
 DYNAMIC_CORE += POWER8
 ifneq ($(C_COMPILER), GCC)
@@ -672,7 +723,7 @@ DYNAMIC_CORE += POWER9
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
 endif
-LDVERSIONGTEQ35 := $(shell expr `ld --version | head -1 | cut -f2 -d "." | cut -f1 -d "-"` >= 35)
+LDVERSIONGTEQ35 := $(shell expr `$(CC) -Wl,--version 2> /dev/null | head -1 | cut -f2 -d "." | cut -f1 -d "-"` \>= 35)
 ifeq ($(GCCVERSIONGTEQ11)$(LDVERSIONGTEQ35), 11)
 DYNAMIC_CORE += POWER10
 CCOMMON_OPT += -DHAVE_P10_SUPPORT
@@ -685,6 +736,10 @@ else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)
 endif
 endif
+else
+DYNAMIC_CORE = POWER8
+DYNAMIC_CORE += POWER9
+endif
 endif
 
 # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
@@ -756,6 +811,11 @@ NO_BINARY_MODE  = 1
 BINARY_DEFINED  = 1
 endif
 
+ifeq ($(ARCH), loongarch64)
+NO_BINARY_MODE  = 1
+BINARY_DEFINED  = 1
+endif
+
 
 #
 #  C Compiler dependent settings
@@ -787,14 +847,9 @@ CCOMMON_OPT += -mabi=32
 BINARY_DEFINED = 1
 endif
 
-ifeq ($(CORE), LOONGSON3A)
-CCOMMON_OPT += -march=mips64
-FCOMMON_OPT += -march=mips64
-endif
-
-ifeq ($(CORE), LOONGSON3B)
-CCOMMON_OPT += -march=mips64
-FCOMMON_OPT += -march=mips64
+ifeq ($(CORE), $(filter $(CORE),LOONGSON3R3 LOONGSON3R4))
+CCOMMON_OPT += -march=loongson3a
+FCOMMON_OPT += -march=loongson3a
 endif
 
 ifeq ($(CORE), MIPS24K)
@@ -831,6 +886,13 @@ ifeq ($(OSNAME), AIX)
 BINARY_DEFINED = 1
 endif
 
+ifeq ($(ARCH), loongarch64)
+ifeq ($(CORE), LOONGSON3R5)
+CCOMMON_OPT += -march=loongarch64 -mabi=lp64
+FCOMMON_OPT += -march=loongarch64 -mabi=lp64
+endif
+endif
+
 endif
 
 ifndef BINARY_DEFINED
@@ -848,9 +910,29 @@ endif
 endif
 
 ifeq ($(C_COMPILER), PGI)
+PGCVERSIONGT20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 20)
+PGCVERSIONEQ20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` == 20)
+PGCMINORVERSIONGE11 := $(shell expr `$(CC) --version|sed -n "2p" |cut -d "-" -f 1 |sed -e "s/[^0-9.]//g" |cut -c 4-5` \>= 11)
+PGCVERSIONCHECK := $(PGCVERSIONGT20)$(PGCVERSIONEQ20)$(PGCMINORVERSIONGE11)
+ifeq ($(PGCVERSIONCHECK), $(filter $(PGCVERSIONCHECK), 100 101 011))
+NEWPGI := 1
+PGCVERSIONGT21 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 21)
+PGCVERSIONEQ21 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` == 21)
+PGCVERSIONCHECK2 := $(PGCVERSIONGT21)$(PGCVERSIONEQ21)$(PGCMINORVERSIONGE11)
+ifeq ($(PGCVERSIONCHECK2), $(filter $(PGCVERSIONCHECK2), 100 101 011))
+NEWPGI2 := 1
+endif
+endif
 ifdef BINARY64
 ifeq ($(ARCH), x86_64)
-CCOMMON_OPT += -tp p7-64 -D__MMX__ -Mnollvm
+ifneq ($(NEWPGI2),1)
+CCOMMON_OPT += -tp p7-64
+else
+CCOMMON_OPT += -tp px
+endif
+ifneq ($(NEWPGI),1)
+CCOMMON_OPT +=  -D__MMX__ -Mnollvm
+endif
 else
 ifeq ($(ARCH), power)
 ifeq ($(CORE), POWER8)
@@ -862,7 +944,11 @@ endif
 endif
 endif
 else
+ifneq ($(NEWPGI2),1)
 CCOMMON_OPT += -tp p7
+else
+CCOMMON_OPT += -tp px
+endif
 endif
 endif
 
@@ -878,13 +964,25 @@ endif
 #  Fortran Compiler dependent settings
 #
 
+ifeq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -dcfuns -recursive -ieee=full -w=obs -thread_safe
+ifdef INTERFACE64
+ifneq ($(INTERFACE64), 0)
+FCOMMON_OPT += -i8
+endif
+endif
+ifeq ($(USE_OPENMP), 1)
+FCOMMON_OPT += -openmp
+endif
+endif
+
 ifeq ($(F_COMPILER), FLANG)
 CCOMMON_OPT += -DF_INTERFACE_FLANG
 FCOMMON_OPT += -Mrecursive -Kieee
 ifeq ($(OSNAME), Linux)
 ifeq ($(ARCH), x86_64)
-FLANG_VENDOR := $(shell `$(FC) --version|cut -f 1 -d "."|head -1`)
-ifeq ($(FLANG_VENDOR),AOCC)
+FLANG_VENDOR := $(shell $(FC) --version|head -1 |cut -f 1 -d " ")
+ifeq ($(FLANG_VENDOR), AMD)
 FCOMMON_OPT += -fno-unroll-loops
 endif
 endif
@@ -1027,21 +1125,31 @@ FCOMMON_OPT += -i8
 endif
 endif
 ifeq ($(ARCH), x86_64)
+ifneq ($(NEWPGI2),1)
 FCOMMON_OPT += -tp p7-64
 else
+FCOMMON_OPT += -tp px
+endif
+else
 ifeq ($(ARCH), power)
+ifeq ($(CORE), POWER6)
+$(warning NVIDIA HPC compilers do not support POWER6.)
+endif
 ifeq ($(CORE), POWER8)
 FCOMMON_OPT += -tp pwr8
 endif
 ifeq ($(CORE), POWER9)
 FCOMMON_OPT += -tp pwr9
 endif
+ifeq ($(CORE), POWER10)
+$(warning NVIDIA HPC compilers do not support POWER10.)
+endif
 endif
 endif
 else
 FCOMMON_OPT += -tp p7
 endif
-FCOMMON_OPT += -Mrecursive
+FCOMMON_OPT += -Mrecursive -Kieee
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -mp
 endif
@@ -1078,11 +1186,11 @@ FCOMMON_OPT += -n32
 else
 FCOMMON_OPT += -n64
 endif
-ifeq ($(CORE), LOONGSON3A)
+ifeq ($(CORE), LOONGSON3R3)
 FCOMMON_OPT += -loongson3 -static
 endif
 
-ifeq ($(CORE), LOONGSON3B)
+ifeq ($(CORE), LOONGSON3R4)
 FCOMMON_OPT += -loongson3 -static
 endif
 
@@ -1108,11 +1216,11 @@ CCOMMON_OPT += -n32
 else
 CCOMMON_OPT += -n64
 endif
-ifeq ($(CORE), LOONGSON3A)
+ifeq ($(CORE), LOONGSON3R3)
 CCOMMON_OPT += -loongson3 -static
 endif
 
-ifeq ($(CORE), LOONGSON3B)
+ifeq ($(CORE), LOONGSON3R4)
 CCOMMON_OPT += -loongson3 -static
 endif
 
@@ -1180,6 +1288,8 @@ CCOMMON_OPT += -fPIC
 endif
 ifeq ($(F_COMPILER), SUN)
 FCOMMON_OPT  += -pic
+else ifeq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -PIC
 else
 FCOMMON_OPT += -fPIC
 endif
@@ -1223,10 +1333,8 @@ ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
 ifeq ($(ARCH), mips64)
-ifneq ($(CORE), LOONGSON3B)
 USE_SIMPLE_THREADED_LEVEL3 = 1
 endif
-endif
 
 ifeq ($(USE_OPENMP), 1)
 # USE_SIMPLE_THREADED_LEVEL3 = 1
@@ -1259,6 +1367,10 @@ CCOMMON_OPT	 += -DUSE_PAPI
 EXTRALIB	 += -lpapi -lperfctr
 endif
 
+ifdef BUFFERSIZE
+CCOMMON_OPT	 += -DBUFFERSIZE=$(BUFFERSIZE)
+endif
+
 ifdef DYNAMIC_THREADS
 CCOMMON_OPT	 += -DDYNAMIC_THREADS
 endif
@@ -1342,11 +1454,9 @@ endif
 
 ifneq ($(ARCH), x86_64)
 ifneq ($(ARCH), x86)
-ifneq ($(CORE), LOONGSON3B)
 NO_AFFINITY = 1
 endif
 endif
-endif
 
 ifdef NO_AFFINITY
 ifeq ($(NO_AFFINITY), 0)
@@ -1438,6 +1548,10 @@ LAPACK_FFLAGS := $(FFLAGS)
 LAPACK_FPFLAGS := $(FPFLAGS)
 endif
 
+ifeq ($(F_COMPILER),NAG)
+LAPACK_FFLAGS := $(filter-out -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mskylake-avx512 ,$(FFLAGS))
+endif
+
 LAPACK_CFLAGS = $(CFLAGS)
 LAPACK_CFLAGS += -DHAVE_LAPACK_CONFIG_H
 ifdef INTERFACE64
@@ -1566,8 +1680,10 @@ export HAVE_VFP
 export HAVE_VFPV3
 export HAVE_VFPV4
 export HAVE_NEON
-export HAVE_MSA
-export MSA_FLAGS
+ifndef NO_MSA
+  export HAVE_MSA
+  export MSA_FLAGS
+endif
 export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE

Makefile.x86

@@ -1,10 +1,21 @@
 # COMPILER_PREFIX = mingw32-
 
+ifneq ($(DYNAMIC_ARCH),1)
+ADD_CPUFLAGS = 1
+else
+ifdef TARGET_CORE
+ADD_CPUFLAGS = 1
+endif
+endif
+
+ifdef ADD_CPUFLAGS
 ifdef HAVE_SSE
 CCOMMON_OPT += -msse
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -msse
 endif
-
+endif
+endif
 
 ifeq ($(OSNAME), Interix)
 ARFLAGS		= -m x86

Makefile.x86_64

@@ -8,42 +8,57 @@ endif
 endif
 endif
 
+
+ifneq ($(DYNAMIC_ARCH),1)
+ADD_CPUFLAGS = 1
+else
+ifdef TARGET_CORE
+ADD_CPUFLAGS = 1
+endif
+endif
+
+ifdef ADD_CPUFLAGS
 ifdef HAVE_SSE3
 CCOMMON_OPT += -msse3
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -msse3
 endif
+endif
 ifdef HAVE_SSSE3
 CCOMMON_OPT += -mssse3
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -mssse3
 endif
+endif
 ifdef HAVE_SSE4_1
 CCOMMON_OPT += -msse4.1
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -msse4.1
 endif
+endif
 ifndef OLDGCC
 ifdef HAVE_AVX
 CCOMMON_OPT += -mavx
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -mavx
 endif
 endif
+endif
 ifndef NO_AVX2
 ifdef HAVE_AVX2
 CCOMMON_OPT += -mavx2
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -mavx2
 endif
 endif
-ifndef OLDGCC
-ifdef HAVE_FMA3
-CCOMMON_OPT += -mfma
-FCOMMON_OPT += -mfma
-endif
 endif
 
 ifeq ($(CORE), SKYLAKEX)
-ifndef DYNAMIC_ARCH
 ifndef NO_AVX512
 CCOMMON_OPT += -march=skylake-avx512
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=skylake-avx512
+endif
 ifeq ($(OSNAME), CYGWIN_NT)
 CCOMMON_OPT += -fno-asynchronous-unwind-tables
 FCOMMON_OPT += -fno-asynchronous-unwind-tables
@@ -56,17 +71,22 @@ endif
 endif
 endif
 endif
-endif
 
 ifeq ($(CORE), COOPERLAKE)
-ifndef DYNAMIC_ARCH
 ifndef NO_AVX512
 ifeq ($(C_COMPILER), GCC)
 # cooperlake support was added in 10.1
 ifeq ($(GCCVERSIONGTEQ10)$(GCCMINORVERSIONGTEQ1), 11)
 CCOMMON_OPT += -march=cooperlake
+ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=cooperlake
 endif
+else  # gcc not support, fallback to avx512
+CCOMMON_OPT += -march=skylake-avx512
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=skylake-avx512
+endif
+endif
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 CCOMMON_OPT += -fno-asynchronous-unwind-tables
@@ -80,6 +100,34 @@ endif
 endif
 endif
 endif
+
+ifeq ($(CORE), SAPPHIRERAPIDS)
+ifndef NO_AVX512
+ifeq ($(C_COMPILER), GCC)
+# sapphire rapids support was added in 11
+ifeq ($(GCCVERSIONGTEQ11), 1)
+CCOMMON_OPT += -march=sapphirerapids
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=sapphirerapids
+endif
+else  # gcc not support, fallback to avx512
+CCOMMON_OPT += -march=skylake-avx512
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=skylake-avx512
+endif
+endif
+endif
+ifeq ($(OSNAME), CYGWIN_NT)
+CCOMMON_OPT += -fno-asynchronous-unwind-tables
+FCOMMON_OPT += -fno-asynchronous-unwind-tables
+endif
+ifeq ($(OSNAME), WINNT)
+ifeq ($(C_COMPILER), GCC)
+CCOMMON_OPT += -fno-asynchronous-unwind-tables
+FCOMMON_OPT += -fno-asynchronous-unwind-tables
+endif
+endif
+endif
 endif
 
 ifdef HAVE_AVX2
@@ -112,6 +160,7 @@ endif
 endif
 endif
 
+endif
 
 
 ifeq ($(OSNAME), Interix)

README.md

@@ -2,7 +2,7 @@
 
 [![Join the chat at https://gitter.im/xianyi/OpenBLAS](https://badges.gitter.im/Join%20Chat.svg)](https://gitter.im/xianyi/OpenBLAS?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
 
-Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.svg?branch=develop)](https://travis-ci.org/xianyi/OpenBLAS)
+Travis CI: [![Build Status](https://travis-ci.com/xianyi/OpenBLAS.svg?branch=develop)](https://travis-ci.com/xianyi/OpenBLAS)
 
 AppVeyor: [![Build status](https://ci.appveyor.com/api/projects/status/09sohd35n8nkkx64/branch/develop?svg=true)](https://ci.appveyor.com/project/xianyi/openblas/branch/develop)
 
@@ -13,17 +13,21 @@ Drone CI: [![Build Status](https://cloud.drone.io/api/badges/xianyi/OpenBLAS/sta
 
 ## Introduction
 
-OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
+OpenBLAS is an optimized BLAS (Basic Linear Algebra Subprograms) library based on GotoBLAS2 1.13 BSD version.
 
 Please read the documentation on the OpenBLAS wiki pages: <https://github.com/xianyi/OpenBLAS/wiki>.
 
+For a general introduction to the BLAS routines, please refer to the extensive documentation of their reference implementation hosted at netlib:
+<https://www.netlib.org/blas>. On that site you will likewise find documentation for the reference implementation of the higher-level library LAPACK - the **L**inear **A**lgebra **Pack**age that comes included with OpenBLAS. If you are looking for a general primer or refresher on Linear Algebra, the set of six
+20-minute lecture videos by Prof. Gilbert Strang on either MIT OpenCourseWare <https://ocw.mit.edu/resources/res-18-010-a-2020-vision-of-linear-algebra-spring-2020/> or Youtube <https://www.youtube.com/playlist?list=PLUl4u3cNGP61iQEFiWLE21EJCxwmWvvek> may be helpful.
+
 ## Binary Packages
 
 We provide official binary packages for the following platform:
 
   * Windows x86/x86_64
 
-You can download them from [file hosting on sourceforge.net](https://sourceforge.net/projects/openblas/files/).
+You can download them from [file hosting on sourceforge.net](https://sourceforge.net/projects/openblas/files/) or from the Releases section of the github project page, [https://github.com/xianyi/OpenBLAS/releases](https://github.com/xianyi/OpenBLAS/releases).
 
 ## Installation from Source
 
@@ -124,6 +128,7 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 - **Intel Sandy Bridge**: Optimized Level-3 and Level-2 BLAS with AVX on x86-64.
 - **Intel Haswell**: Optimized Level-3 and Level-2 BLAS with AVX2 and FMA on x86-64.
 - **Intel Skylake-X**: Optimized Level-3 and Level-2 BLAS with AVX512 and FMA on x86-64.
+- **Intel Cooper Lake**: as Skylake-X with improved BFLOAT16 support.
 - **AMD Bobcat**: Used GotoBLAS2 Barcelona codes.
 - **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thanks to Werner Saar)
 - **AMD PILEDRIVER**: Uses Bulldozer codes with some optimizations.
@@ -149,6 +154,7 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 
 - **ARMv8**: Basic ARMV8 with small caches, optimized Level-3 and Level-2 BLAS
 - **Cortex-A53**: same as ARMV8 (different cpu specifications)
+- **Cortex-A55**: same as ARMV8 (different cpu specifications)
 - **Cortex A57**: Optimized Level-3 and Level-2 functions
 - **Cortex A72**: same as A57 ( different cpu specifications)
 - **Cortex A73**: same as A57 (different cpu specifications)
@@ -174,10 +180,11 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 
 #### RISC-V
 
-- **C910V**: Optimized Leve-3 BLAS (real) and Level-1,2 by RISC-V Vector extension 0.7.1.
+- **C910V**: Optimized Level-3 BLAS (real) and Level-1,2 by RISC-V Vector extension 0.7.1.
   ```sh
   make HOSTCC=gcc TARGET=C910V CC=riscv64-unknown-linux-gnu-gcc FC=riscv64-unknown-linux-gnu-gfortran
   ```
+  (also known to work on C906)
 
 ### Support for multiple targets in a single library
 
@@ -208,7 +215,8 @@ Please note that it is not possible to combine support for different architectur
 - **Android**: Supported by the community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-build-OpenBLAS-for-Android>.
 - **AIX**: Supported on PPC up to POWER8
 - **Haiku**: Supported by the community. We don't actively test the library on this OS.
-- **SunOS**: Supported by the community. We don't actively test the library on this OS:
+- **SunOS**: Supported by the community. We don't actively test the library on this OS.
+- **Cortex-M**: Supported by the community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-use-OpenBLAS-on-Cortex-M>.
 
 ## Usage
 

TargetList.txt

@@ -23,6 +23,7 @@ HASWELL
 SKYLAKEX
 ATOM
 COOPERLAKE
+SAPPHIRERAPIDS
 
 b)AMD CPU:
 ATHLON
@@ -92,6 +93,9 @@ CORTEXA57
 CORTEXA72
 CORTEXA73
 NEOVERSEN1
+NEOVERSEV1
+NEOVERSEN2
+CORTEXA55
 EMAG8180
 FALKOR
 THUNDERX
@@ -109,3 +113,9 @@ Z14
 RISCV64_GENERIC
 C910V
 
+11.LOONGARCH64:
+LOONGSON3R5
+
+12. Elbrus E2000:
+E2K
+

appveyor.yml

@@ -29,15 +29,15 @@ environment:
   global:
     CONDA_INSTALL_LOCN: C:\\Miniconda36-x64
   matrix:
-    - COMPILER: clang-cl
-      WITH_FORTRAN: yes
-    - COMPILER: clang-cl
-      DYNAMIC_ARCH: ON
-      WITH_FORTRAN: no
-    - COMPILER: cl
-    - COMPILER: MinGW64-gcc-7.2.0-mingw
-      DYNAMIC_ARCH: OFF
-      WITH_FORTRAN: ignore
+#    - COMPILER: clang-cl
+#      WITH_FORTRAN: ON
+#    - COMPILER: clang-cl
+#      DYNAMIC_ARCH: ON
+#      WITH_FORTRAN: OFF
+#    - COMPILER: cl
+#    - COMPILER: MinGW64-gcc-7.2.0-mingw
+#      DYNAMIC_ARCH: OFF
+#      WITH_FORTRAN: ignore
     - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
       COMPILER: MinGW-gcc-6.3.0-32   
     - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
@@ -46,13 +46,10 @@ environment:
     
 install:
   - if [%COMPILER%]==[clang-cl] call %CONDA_INSTALL_LOCN%\Scripts\activate.bat
+  - if [%COMPILER%]==[clang-cl] conda update --yes -n base conda
   - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
-  - if [%COMPILER%]==[clang-cl] conda install --yes --quiet clangdev cmake
-
-  - if [%WITH_FORTRAN%]==[no] conda install --yes --quiet ninja
-  - if [%WITH_FORTRAN%]==[yes] conda install --yes --quiet -c isuruf kitware-ninja
-  - if [%WITH_FORTRAN%]==[yes] conda install --yes --quiet flang
-
+  - if [%COMPILER%]==[clang-cl] conda config --set auto_update_conda false
+  - if [%COMPILER%]==[clang-cl] conda install --yes --quiet clangdev cmake ninja flang=11.0.1
   - if [%COMPILER%]==[clang-cl] call "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvarsall.bat" x64
   - if [%COMPILER%]==[clang-cl] set "LIB=%CONDA_INSTALL_LOCN%\Library\lib;%LIB%"
   - if [%COMPILER%]==[clang-cl] set "CPATH=%CONDA_INSTALL_LOCN%\Library\include;%CPATH%"
@@ -68,15 +65,14 @@ before_build:
   - if [%COMPILER%]==[MinGW64-gcc-7.2.0-mingw] cmake -G "MinGW Makefiles" -DNOFORTRAN=1 ..
   - if [%COMPILER%]==[MinGW-gcc-6.3.0-32] cmake -G "MSYS Makefiles" -DNOFORTRAN=1 ..
   - if [%COMPILER%]==[MinGW-gcc-5.3.0] cmake -G "MSYS Makefiles" -DNOFORTRAN=1 ..
-  - if [%WITH_FORTRAN%]==[no] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DMSVC_STATIC_CRT=ON ..
-  - if [%WITH_FORTRAN%]==[yes] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
+  - if [%WITH_FORTRAN%]==[OFF] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_MT=mt -DMSVC_STATIC_CRT=ON ..
+  - if [%WITH_FORTRAN%]==[ON] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DCMAKE_MT=mt -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
+  - if [%USE_OPENMP%]==[ON] cmake -DUSE_OPENMP=ON ..
   - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON -DDYNAMIC_LIST='CORE2;NEHALEM;SANDYBRIDGE;BULLDOZER;HASWELL' ..
 
 build_script:
   - cmake --build .
 
 test_script:
-  - echo Running Test
-  - cd utest
-  - openblas_utest
+  - ctest -j2
   

azure-pipelines.yml

@@ -4,14 +4,22 @@ trigger:
   branches:
     include:
       - develop
-
+resources:
+  containers:
+      - container: oneapi-hpckit
+        image: intel/oneapi-hpckit:latest
+        options: '-v /usr/bin/sudo:/usr/bin/sudo -v /usr/lib/sudo/libsudo_util.so.0:/usr/lib/sudo/libsudo_util.so.0 -v /usr/lib/sudo/sudoers.so:/usr/lib/sudo/sudoers.so'
+      - container: oneapi-basekit
+        image: intel/oneapi-basekit:latest
+        options: '-v /usr/bin/sudo:/usr/bin/sudo -v /usr/lib/sudo/libsudo_util.so.0:/usr/lib/sudo/libsudo_util.so.0 -v /usr/lib/sudo/sudoers.so:/usr/lib/sudo/sudoers.so'
+ 
 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'
+    vmImage: 'ubuntu-latest'
   steps:
   - script: |
       echo "FROM quay.io/pypa/manylinux1_x86_64
@@ -27,7 +35,7 @@ jobs:
     displayName: Run manylinux1 docker build
 - job: Intel_SDE_skx
   pool:
-    vmImage: 'ubuntu-16.04'
+    vmImage: 'ubuntu-latest'
   steps:
   - script: |
       # at the time of writing the available Azure Ubuntu vm image
@@ -67,5 +75,189 @@ jobs:
       cd utest
       dir
       openblas_utest.exe
-  
+
+- job: Windows_mingw_gmake
+  pool:
+     vmImage: 'windows-latest'
+  steps:   
+  - script: |
+      mingw32-make CC=gcc FC=gfortran DYNAMIC_ARCH=1 DYNAMIC_LIST="NEHALEM SANDYBRIDGE HASWELL"
+
+- job: Windows_clang_cmake
+  pool:
+     vmImage: 'windows-latest'
+  steps:
+  - script: |
+      set "PATH=C:\Miniconda\Scripts;C:\Miniconda\Library\bin;C:\Miniconda\Library\usr\bin;C:\Miniconda\condabin;%PATH%"
+      set "LIB=C:\Miniconda\Library\lib;%LIB%"
+      set "CPATH=C:\Miniconda\Library\include;%CPATH%
+      conda config --add channels conda-forge --force
+      conda config --set auto_update_conda false
+      conda install --yes ninja
+      call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
+      mkdir build 
+      cd build
+      cmake -G "Ninja" -DCMAKE_C_COMPILER=clang-cl -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_MT=mt -DCMAKE_BUILD_TYPE=Release -DNOFORTRAN=1 -DMSVC_STATIC_CRT=ON ..
+      cmake --build . --config Release
+      ctest
+
+- job: Windows_flang_clang
+  pool:
+     vmImage: 'windows-latest'
+  steps:
+  - script: |
+      set "PATH=C:\Miniconda\Scripts;C:\Miniconda\Library\bin;C:\Miniconda\Library\usr\bin;C:\Miniconda\condabin;%PATH%"
+      set "LIB=C:\Miniconda\Library\lib;%LIB%"
+      set "CPATH=C:\Miniconda\Library\include;%CPATH%"
+      conda config --add channels conda-forge --force
+      conda config --set auto_update_conda false
+      conda install --yes --quiet ninja flang
+      mkdir build 
+      cd build
+      call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
+      cmake -G "Ninja" -DCMAKE_C_COMPILER=clang-cl -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DCMAKE_MT=mt -DCMAKE_BUILD_TYPE=Release -DMSVC_STATIC_CRT=ON ..
+      cmake --build . --config Release
+      ctest
+
+- job: OSX_OpenMP
+  pool:
+     vmImage: 'macOS-10.15'
+  steps:   
+  - script: |
+      brew update
+      make TARGET=CORE2 DYNAMIC_ARCH=1 USE_OPENMP=1 INTERFACE64=1 CC=gcc-10 FC=gfortran-10
+      make TARGET=CORE2 DYNAMIC_ARCH=1 USE_OPENMP=1 INTERFACE64=1 CC=gcc-10 FC=gfortran-10 PREFIX=../blasinst install
+      ls -lR ../blasinst
+     
+- job: OSX_GCC_Nothreads
+  pool:
+     vmImage: 'macOS-10.15'
+  steps:   
+  - script: |
+      brew update
+      make USE_THREADS=0 CC=gcc-10 FC=gfortran-10
+     
+- job: OSX_OpenMP_Clang
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     LD_LIBRARY_PATH: /usr/local/opt/llvm/lib
+     LIBRARY_PATH: /usr/local/opt/llvm/lib
+  steps:   
+  - script: |
+      brew update
+      brew install llvm libomp
+      make TARGET=CORE2 USE_OPENMP=1 INTERFACE64=1 DYNAMIC_ARCH=1 CC=/usr/local/opt/llvm/bin/clang FC=gfortran-10
+
+- job: OSX_OpenMP_Clang_cmake
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     LD_LIBRARY_PATH: /usr/local/opt/llvm/lib
+     LIBRARY_PATH: /usr/local/opt/llvm/lib
+  steps:   
+  - script: |
+      brew update
+      brew install llvm libomp
+      mkdir build
+      cd build
+      cmake -DTARGET=CORE2 -DUSE_OPENMP=1 -DINTERFACE64=1 -DDYNAMIC_ARCH=1 -DCMAKE_C_COMPILER=/usr/local/opt/llvm/bin/clang -DNOFORTRAN=1 -DNO_AVX512=1 ..
+      make
+      ctest
+      
+- job: OSX_dynarch_cmake
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     LD_LIBRARY_PATH: /usr/local/opt/llvm/lib
+     LIBRARY_PATH: /usr/local/opt/llvm/lib
+  steps:   
+  - script: |
+      mkdir build
+      cd build
+      cmake -DTARGET=CORE2 -DDYNAMIC_ARCH=1 -DCMAKE_C_COMPILER=gcc-10 -DCMAKE_Fortran_COMPILER=gfortran-10 -DBUILD_SHARED_LIBS=ON ..
+      cmake --build .
+      ctest
       
+- job: OSX_Ifort_Clang
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     LD_LIBRARY_PATH: /usr/local/opt/llvm/lib
+     MACOS_HPCKIT_URL: https://registrationcenter-download.intel.com/akdlm/irc_nas/17643/m_HPCKit_p_2021.2.0.2903_offline.dmg
+     LIBRARY_PATH: /usr/local/opt/llvm/lib
+     MACOS_FORTRAN_COMPONENTS: intel.oneapi.mac.ifort-compiler
+  steps:   
+  - script: |
+      brew update
+      brew install llvm libomp
+      sudo mkdir -p /opt/intel
+      sudo chown $USER /opt/intel
+    displayName: prepare for cache restore
+  - task: Cache@2
+    inputs:
+      path: /opt/intel/oneapi
+      key: '"install" | "$(MACOS_HPCKIT_URL)" | "$(MACOS_FORTRAN_COMPONENTS)"'
+      cacheHitVar: CACHE_RESTORED
+  - script: | 
+      curl --output webimage.dmg --url $(MACOS_HPCKIT_URL) --retry 5 --retry-delay 5
+      hdiutil attach webimage.dmg
+      sudo /Volumes/"$(basename "$(MACOS_HPCKIT_URL)" .dmg)"/bootstrapper.app/Contents/MacOS/bootstrapper -s --action install --components="$(MACOS_FORTRAN_COMPONENTS)" --eula=accept --continue-with-optional-error=yes --log-dir=.
+      installer_exit_code=$?
+      hdiutil detach /Volumes/"$(basename "$URL" .dmg)" -quiet
+      exit $installer_exit_code
+    displayName: install
+    condition: ne(variables.CACHE_RESTORED, 'true')
+  - script: | 
+      source /opt/intel/oneapi/setvars.sh
+      make CC=/usr/local/opt/llvm/bin/clang FC=ifort
+ 
+- job: OSX_NDK_ARMV7
+  pool:
+     vmImage: 'macOS-10.15'
+  steps:   
+  - script: | 
+      brew update
+      brew install --cask android-ndk
+      export ANDROID_NDK_HOME=/usr/local/share/android-ndk
+      make TARGET=ARMV7 ONLY_CBLAS=1 CC=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/darwin-x86_64/bin/armv7a-linux-androideabi21-clang AR=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/darwin-x86_64/bin/llvm-ar HOSTCC=gcc ARM_SOFTFP_ABI=1 -j4
+
+- job: OSX_IOS_ARMV8
+  pool:
+     vmImage: 'macOS-11'
+  variables:
+     CC: /Applications/Xcode_12.4.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+     CFLAGS: -O2 -Wno-macro-redefined -isysroot /Applications/Xcode_12.4.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS14.4.sdk -arch arm64 -miphoneos-version-min=10.0
+  steps:
+  - script: |
+     make TARGET=ARMV8 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
+
+- job: OSX_IOS_ARMV7
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     CC: /Applications/Xcode_12.4.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+     CFLAGS: -O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode_12.4.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS14.4.sdk -arch armv7 -miphoneos-version-min=5.1
+  steps:
+  - script: |
+     make TARGET=ARMV7 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
+
+- job: ALPINE_MUSL
+  pool:
+     vmImage: 'ubuntu-latest'
+  steps:
+  - script: |
+        wget https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.13.2/alpine-chroot-install \
+          && echo '60c7e0b5d82e21d1a549fc9a46ba3b36688c09dc  alpine-chroot-install' | sha1sum -c \
+          || exit 1
+        alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
+        sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers sudo'
+        alpine make DYNAMIC_ARCH=1 BINARY=64
+        alpine make DYNAMIC_ARCH=1 BINARY=64 PREFIX=mytestdir install
+        alpine ls -l mytestdir/include
+        alpine echo "// tests that inclusion of openblas_config.h works with musl" >test_install.c
+        alpine echo "#include <openblas_config.h>" >>test_install.c
+        alpine echo "int main(){" >> test_install.c
+        alpine echo "cpu_set_t* cpu_set = NULL;}" >>test_install.c
+        alpine gcc -Imytestdir/include test_install.c -Lmytestdir/lib -lopenblas -lpthread -lgfortran -o test_install
+        

benchmark/bench.h

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

benchmark/gemv.c

@@ -125,7 +125,7 @@ int main(int argc, char *argv[]){
    		fprintf(stderr, " %6dx%d : ", (int)m,(int)n);
    		for(j = 0; j < m; j++){
       			for(i = 0; i < n * COMPSIZE; i++){
-				a[(long)j + (long)i * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+				a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       			}
    		}
 
@@ -162,7 +162,7 @@ int main(int argc, char *argv[]){
    		fprintf(stderr, " %6dx%d : ", (int)m,(int)n);
    		for(j = 0; j < m; j++){
       			for(i = 0; i < n * COMPSIZE; i++){
-				a[(long)j + (long)i * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+				a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       			}
    		}
 

benchmark/getri.c

@@ -72,13 +72,17 @@ int main(int argc, char *argv[]){
   FLOAT *a,*work;
   FLOAT wkopt[4];
   blasint *ipiv;
-  blasint m, i, j, info,lwork;
+  blasint m, i, j, l, info,lwork;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  double time1,timeg;
+  
+  char *p;
+  char btest = 'I';
 
   argc--;argv++;
 
@@ -86,6 +90,9 @@ int main(int argc, char *argv[]){
   if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
   if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
 
+  if ((p = getenv("OPENBLAS_TEST"))) btest=*p;
+  
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
 
   fprintf(stderr, "From : %3d  To : %3d Step = %3d\n", from, to, step);
 
@@ -124,32 +131,41 @@ int main(int argc, char *argv[]){
   fprintf(stderr, "   SIZE           FLops           Time          Lwork\n");
 
   for(m = from; m <= to; m += step){
-
+    timeg = 0.;
     fprintf(stderr, " %6d : ", (int)m);
 
-    GETRF (&m, &m, a, &m, ipiv, &info);
+    for (l = 0; l < loops; l++) {
 
+    if (btest == 'F') begin();
+    GETRF (&m, &m, a, &m, ipiv, &info);
+    if (btest == 'F') {
+      end();
+      timeg += getsec();
+    }
     if (info) {
       fprintf(stderr, "Matrix is not singular .. %d\n", info);
       exit(1);
     }
 
-    begin();
+    if (btest == 'I') begin();
 
     lwork = -1;
     GETRI(&m, a, &m, ipiv, wkopt, &lwork, &info);
 
     lwork = (blasint)wkopt[0];
     GETRI(&m, a, &m, ipiv, work, &lwork, &info);
-    end();
+    if (btest == 'I') end();
 
     if (info) {
       fprintf(stderr, "failed compute inverse matrix .. %d\n", info);
       exit(1);
     }
 
-    time1 = getsec();
-
+    if (btest == 'I') 
+      timeg += getsec();
+    
+    } // loops
+    time1 = timeg/(double)loops;
     fprintf(stderr,
 	    " %10.2f MFlops : %10.2f Sec : %d\n",
 	    COMPSIZE * COMPSIZE * (4.0/3.0 * (double)m * (double)m *(double)m - (double)m *(double)m + 5.0/3.0* (double)m) / time1 * 1.e-6,time1,lwork);

benchmark/linpack.c

@@ -72,17 +72,21 @@ int main(int argc, char *argv[]){
   FLOAT *a, *b;
   blasint *ipiv;
 
-  blasint m, i, j, info;
+  blasint m, i, j, l, info;
   blasint unit =   1;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
   FLOAT maxerr;
 
-  double time1, time2;
+  double time1, time2, timeg1,timeg2;
 
+  char *p;
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+  
   argc--;argv++;
 
   if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
@@ -110,9 +114,9 @@ int main(int argc, char *argv[]){
   fprintf(stderr, "   SIZE       Residual     Decompose            Solve           Total\n");
 
   for(m = from; m <= to; m += step){
-
+    timeg1 = timeg2 = 0.;
     fprintf(stderr, " %6d : ", (int)m);
-
+    for (l = 0; l < loops; l++) {
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
 	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
@@ -138,7 +142,7 @@ int main(int argc, char *argv[]){
       exit(1);
     }
 
-    time1 = getsec();
+    timeg1 += getsec();
 
     begin();
 
@@ -151,8 +155,10 @@ int main(int argc, char *argv[]){
       exit(1);
     }
 
-    time2 = getsec();
-
+    timeg2 += getsec();
+    } //loops
+    time1=timeg1/(double)loops;
+    time2=timeg2/(double)loops;
     maxerr = 0.;
 
     for(i = 0; i < m; i++){

benchmark/potrf.c

@@ -99,14 +99,15 @@ int main(int argc, char *argv[]){
   char *p;
   char btest = 'F';
 
-  blasint m, i, j, info, uplos=0;
-  double flops;
+  blasint m, i, j, l, info, uplos=0;
+  double flops = 0.;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  double time1, timeg;
 
   argc--;argv++;
 
@@ -119,6 +120,8 @@ int main(int argc, char *argv[]){
 
   if ((p = getenv("OPENBLAS_TEST"))) btest=*p;
 
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+
   fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c\n", from, to, step,*uplo[uplos]);
 
   if (( a    = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
@@ -129,19 +132,21 @@ int main(int argc, char *argv[]){
     fprintf(stderr,"Out of Memory!!\n");exit(1);
   }
 
-  for(m = from; m <= to; m += step){
 
+  for(m = from; m <= to; m += step){
+    timeg=0.;
+	  for (l = 0; l < loops; l++) {
 #ifndef COMPLEX
       if (uplos & 1) {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = 0.;
-	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
 	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	}
       } else {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
 	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = 0.;
 	}
       }
@@ -192,8 +197,8 @@ int main(int argc, char *argv[]){
 	exit(1);
       }
 
-      time1 = getsec();
-      flops = COMPSIZE * COMPSIZE * (1.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 1.0/6.0* (double)m) / time1 * 1.e-6;
+      if ( btest == 'F')
+	      timeg += getsec();
 
       if ( btest == 'S' )
       {
@@ -214,9 +219,7 @@ int main(int argc, char *argv[]){
 		fprintf(stderr, "Potrs info = %d\n", info);
 		exit(1);
         }
-        time1 = getsec();
-        flops = COMPSIZE * COMPSIZE * (2.0 * (double)m * (double)m *(double)m ) / time1 * 1.e-6;
-
+        timeg += getsec();
       }
 	
       if ( btest == 'I' )
@@ -232,11 +235,17 @@ int main(int argc, char *argv[]){
 		fprintf(stderr, "Potri info = %d\n", info);
 		exit(1);
         }
-
-        time1 = getsec();
-        flops = COMPSIZE * COMPSIZE * (2.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 5.0/6.0* (double)m) / time1 * 1.e-6;
+        timeg += getsec();
       }
-	
+    } // loops
+
+      time1 = timeg/(double)loops;
+	if ( btest == 'F')
+		flops = COMPSIZE * COMPSIZE * (1.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 1.0/6.0* (double)m) / time1 * 1.e-6;
+	if ( btest == 'S')
+		flops = COMPSIZE * COMPSIZE * (2.0 * (double)m * (double)m *(double)m ) / time1 * 1.e-6;
+	if ( btest == 'I')
+		flops = COMPSIZE * COMPSIZE * (2.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 5.0/6.0* (double)m) / time1 * 1.e-6;
       fprintf(stderr, "%8d : %10.2f MFlops : %10.3f Sec : Test=%c\n",m,flops ,time1,btest);
 
 

benchmark/syr2.c

@@ -46,14 +46,17 @@ int main(int argc, char *argv[]){
 
   if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p; 
 
-  blasint m, i, j;
+  blasint m, i, j, l;
   blasint inc_x= 1;
   blasint inc_y= 1;
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+
+  double time1,timeg;
 
   argc--;argv++;
 
@@ -85,8 +88,9 @@ int main(int argc, char *argv[]){
 
   for(m = from; m <= to; m += step)
   {
-	
+    timeg = 0.;	
     fprintf(stderr, " %6d : ", (int)m);
+    for (l = 0; l < loops; l++) {
 	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
 	x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
    	}
@@ -107,8 +111,10 @@ int main(int argc, char *argv[]){
 
     end();
 
-    time1 = getsec();
+    timeg += getsec();
+    } // loops
 
+    time1 = timeg/(double)loops;
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / time1 * 1.e-6);

benchmark/syrk.c

@@ -56,17 +56,20 @@ int main(int argc, char *argv[]){
 
   char uplo='U';
   char trans='N';
-
+  
   if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
   if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
 
-  blasint m, i, j;
+  blasint m, i, j, l;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
+
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
 
-  double time1;
+  double time1,timeg;
 
   argc--;argv++;
 
@@ -95,9 +98,12 @@ int main(int argc, char *argv[]){
 
   for(m = from; m <= to; m += step)
   {
+    timeg = 0.;
 
     fprintf(stderr, " %6d : ", (int)m);
 
+    for(l = 0; l < loops; l++) {
+
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
 	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
@@ -111,8 +117,10 @@ int main(int argc, char *argv[]){
 
     end();
 
-    time1 = getsec();
-
+    timeg += getsec();
+    
+    } //loops
+    time1 = timeg / (double)loops;
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m * (double)m / time1 * 1.e-6);

c_check

@@ -1,11 +1,11 @@
-#!/usr/bin/perl
+#!/usr/bin/env perl
 
 #use File::Basename;
 # use File::Temp qw(tempfile);
 
 # Checking cross compile
 $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
-$hostarch = `uname -m | sed -e s/i.86/x86/`;chop($hostarch);
+$hostarch = `uname -m | sed -e s/i.86/x86/`;
 $hostarch = `uname -p` if ($hostos eq "AIX" || $hostos eq "SunOS");
 chop($hostarch);
 $hostarch = "x86_64" if ($hostarch eq "amd64");
@@ -82,18 +82,20 @@ $os = Interix         if ($data =~ /OS_INTERIX/);
 $os = Android         if ($data =~ /OS_ANDROID/);
 $os = Haiku           if ($data =~ /OS_HAIKU/);
 
-$architecture = x86    if ($data =~ /ARCH_X86/);
-$architecture = x86_64 if ($data =~ /ARCH_X86_64/);
-$architecture = power  if ($data =~ /ARCH_POWER/);
-$architecture = mips   if ($data =~ /ARCH_MIPS/);
-$architecture = mips64 if ($data =~ /ARCH_MIPS64/);
-$architecture = alpha  if ($data =~ /ARCH_ALPHA/);
-$architecture = sparc  if ($data =~ /ARCH_SPARC/);
-$architecture = ia64   if ($data =~ /ARCH_IA64/);
-$architecture = arm    if ($data =~ /ARCH_ARM/);
-$architecture = arm64  if ($data =~ /ARCH_ARM64/);
-$architecture = zarch  if ($data =~ /ARCH_ZARCH/);
-$architecture = riscv64 if ($data =~ /ARCH_RISCV64/);
+$architecture = x86          if ($data =~ /ARCH_X86/);
+$architecture = x86_64       if ($data =~ /ARCH_X86_64/);
+$architecture = e2k          if ($data =~ /ARCH_E2K/);
+$architecture = power        if ($data =~ /ARCH_POWER/);
+$architecture = mips         if ($data =~ /ARCH_MIPS/);
+$architecture = mips64       if ($data =~ /ARCH_MIPS64/);
+$architecture = alpha        if ($data =~ /ARCH_ALPHA/);
+$architecture = sparc        if ($data =~ /ARCH_SPARC/);
+$architecture = ia64         if ($data =~ /ARCH_IA64/);
+$architecture = arm          if ($data =~ /ARCH_ARM/);
+$architecture = arm64        if ($data =~ /ARCH_ARM64/);
+$architecture = zarch        if ($data =~ /ARCH_ZARCH/);
+$architecture = riscv64      if ($data =~ /ARCH_RISCV64/);
+$architecture = loongarch64  if ($data =~ /ARCH_LOONGARCH64/);
 
 $defined = 0;
 
@@ -123,6 +125,11 @@ if ($architecture eq "zarch") {
     $binary = 64;
 }
 
+if ($architecture eq "e2k") {
+    $defined = 1;
+    $binary = 64;
+}
+
 if ($architecture eq "alpha") {
     $defined = 1;
     $binary = 64;
@@ -143,6 +150,11 @@ if ($architecture eq "riscv64") {
     $binary = 64;
 }
 
+if ($architecture eq "loongarch64") {
+    $defined = 1;
+    $binary = 64;
+}
+
 if ($compiler eq "PGI") {
     $compiler_name .= " -tp p7"    if ($binary eq "32");
     $compiler_name .= " -tp p7-64" if ($binary eq "64");
@@ -199,7 +211,7 @@ if (($architecture eq "mips") || ($architecture eq "mips64")) {
     } else {
 	$tmpf = new File::Temp( SUFFIX => '.c' , UNLINK => 1 );
 	$code = '"addvi.b $w0, $w1, 1"';
-	$msa_flags = "-mmsa -mfp64 -msched-weight -mload-store-pairs";
+	$msa_flags = "-mmsa -mfp64 -mload-store-pairs";
 	print $tmpf "#include <msa.h>\n\n";
 	print $tmpf "void main(void){ __asm__ volatile($code); }\n";
 
@@ -215,17 +227,19 @@ if (($architecture eq "mips") || ($architecture eq "mips64")) {
     }
 }
 
-$architecture = x86    if ($data =~ /ARCH_X86/);
-$architecture = x86_64 if ($data =~ /ARCH_X86_64/);
-$architecture = power  if ($data =~ /ARCH_POWER/);
-$architecture = mips   if ($data =~ /ARCH_MIPS/);
-$architecture = mips64 if ($data =~ /ARCH_MIPS64/);
-$architecture = alpha  if ($data =~ /ARCH_ALPHA/);
-$architecture = sparc  if ($data =~ /ARCH_SPARC/);
-$architecture = ia64   if ($data =~ /ARCH_IA64/);
-$architecture = arm    if ($data =~ /ARCH_ARM/);
-$architecture = arm64  if ($data =~ /ARCH_ARM64/);
-$architecture = zarch  if ($data =~ /ARCH_ZARCH/);
+$architecture = x86          if ($data =~ /ARCH_X86/);
+$architecture = x86_64       if ($data =~ /ARCH_X86_64/);
+$architecture = e2k          if ($data =~ /ARCH_E2K/);
+$architecture = power        if ($data =~ /ARCH_POWER/);
+$architecture = mips         if ($data =~ /ARCH_MIPS/);
+$architecture = mips64       if ($data =~ /ARCH_MIPS64/);
+$architecture = alpha        if ($data =~ /ARCH_ALPHA/);
+$architecture = sparc        if ($data =~ /ARCH_SPARC/);
+$architecture = ia64         if ($data =~ /ARCH_IA64/);
+$architecture = arm          if ($data =~ /ARCH_ARM/);
+$architecture = arm64        if ($data =~ /ARCH_ARM64/);
+$architecture = zarch        if ($data =~ /ARCH_ZARCH/);
+$architecture = loongarch64  if ($data =~ /ARCH_LOONGARCH64/);
 
 $binformat    = bin32;
 $binformat    = bin64  if ($data =~ /BINARY_64/);

cblas.h

@@ -125,9 +125,14 @@ void cblas_zswap(OPENBLAS_CONST blasint n, void *x, OPENBLAS_CONST blasint incx,
 
 void cblas_srot(OPENBLAS_CONST blasint N, float *X, OPENBLAS_CONST blasint incX, float *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float c, OPENBLAS_CONST float s);
 void cblas_drot(OPENBLAS_CONST blasint N, double *X, OPENBLAS_CONST blasint incX, double *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double c, OPENBLAS_CONST double  s);
+void cblas_csrot(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float c, OPENBLAS_CONST float s);
+void cblas_zdrot(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double c, OPENBLAS_CONST double s);
 
 void cblas_srotg(float *a, float *b, float *c, float *s);
 void cblas_drotg(double *a, double *b, double *c, double *s);
+void cblas_crotg(void *a, void *b, float *c, void *s);
+void cblas_zrotg(void *a, void *b, double *c, void *s);
+
 
 void cblas_srotm(OPENBLAS_CONST blasint N, float *X, OPENBLAS_CONST blasint incX, float *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float *P);
 void cblas_drotm(OPENBLAS_CONST blasint N, double *X, OPENBLAS_CONST blasint incX, double *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double *P);
@@ -395,6 +400,8 @@ void   cblas_dbf16tod(OPENBLAS_CONST blasint n, OPENBLAS_CONST bfloat16 *in, OPE
 float  cblas_sbdot(OPENBLAS_CONST blasint n, OPENBLAS_CONST bfloat16 *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST bfloat16 *y, OPENBLAS_CONST blasint incy);
 void   cblas_sbgemv(OPENBLAS_CONST enum CBLAS_ORDER order,  OPENBLAS_CONST enum CBLAS_TRANSPOSE trans,  OPENBLAS_CONST blasint m, OPENBLAS_CONST blasint n, OPENBLAS_CONST float alpha, OPENBLAS_CONST bfloat16 *a, OPENBLAS_CONST blasint lda, OPENBLAS_CONST bfloat16 *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float beta, float *y, OPENBLAS_CONST blasint incy);
 
+void   cblas_sbgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
+		    OPENBLAS_CONST float alpha, OPENBLAS_CONST bfloat16 *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST bfloat16 *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 #ifdef __cplusplus
 }
 #endif  /* __cplusplus */

cmake/arch.cmake

@@ -44,7 +44,10 @@ endif ()
 
 if (DYNAMIC_ARCH)
   if (ARM64)
-    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
+	  set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA55 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 NEOVERSEV1 NEOVERSEN2 THUNDERX3T110)
+    if (DYNAMIC_LIST)
+	    set(DYNAMIC_CORE ARMV8 ${DYNAMIC_LIST})
+    endif ()
   endif ()
   
   if (POWER)
@@ -106,7 +109,11 @@ if (${ARCH} STREQUAL "ia64")
   endif ()
 endif ()
 
-if (MIPS64)
+if (MIPS32 OR MIPS64)
+  set(NO_BINARY_MODE 1)
+endif ()
+
+if (LOONGARCH64)
   set(NO_BINARY_MODE 1)
 endif ()
 

cmake/cc.cmake

@@ -15,6 +15,11 @@ if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "LS
 
   if (NO_BINARY_MODE)
 
+    if (MIPS32)
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=32")
+      set(BINARY_DEFINED 1)
+    endif ()
+
     if (MIPS64)
       if (BINARY64)
         set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=64")
@@ -29,6 +34,15 @@ if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "LS
       set(FCOMMON_OPT "${FCOMMON_OPT} -march=mips64")
     endif ()
 
+    if (LOONGARCH64)
+      if (BINARY64)
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=lp64")
+      else ()
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=lp32")
+      endif ()
+      set(BINARY_DEFINED 1)
+    endif ()
+
     if (CMAKE_SYSTEM_NAME STREQUAL "AIX")
       set(BINARY_DEFINED 1)
     endif ()
@@ -117,6 +131,65 @@ if (${CORE} STREQUAL COOPERLAKE)
   endif ()
 endif ()
 
+if (${CORE} STREQUAL SAPPHIRERAPIDS)
+  if (NOT DYNAMIC_ARCH)
+    if (NOT NO_AVX512)
+      execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+      if (${GCC_VERSION} VERSION_GREATER 11.0 OR ${GCC_VERSION} VERSION_EQUAL 11.0)
+        set (CCOMMON_OPT  "${CCOMMON_OPT} -march=sapphirerapids")
+      else ()
+        set (CCOMMON_OPT "${CCOMMON_OPT} -march=skylake-avx512")
+      endif()  
+    endif ()
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL A64FX)
+  if (NOT DYNAMIC_ARCH)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 11.0 OR ${GCC_VERSION} VERSION_EQUAL 11.0)
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.2-a+sve -mtune=a64fx")
+    else ()
+      set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.2-a+sve")
+    endif()
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL ARMV8SVE)
+  if (NOT DYNAMIC_ARCH)
+    set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8-a+sve")
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL POWER10)
+  if (NOT DYNAMIC_ARCH)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 10.2 OR ${GCC_VERSION} VERSION_EQUAL 10.2)
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math")
+    else ()
+      message(FATAL_ERROR "Compiler GCC.${GCC_VERSION} does not support Power10." )
+    endif()
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL POWER9)
+  if (NOT DYNAMIC_ARCH)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 5.0 OR ${GCC_VERSION} VERSION_EQUAL 5.0)
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math")
+    else ()
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+      message(WARNING "Compiler GCC.${GCC_VERSION} does not fully support Power9.")
+    endif ()
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL POWER8)
+  if (NOT DYNAMIC_ARCH)
+    set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+  endif ()
+endif ()
+
 if (NOT DYNAMIC_ARCH)
 	if (HAVE_AVX2)
         set (CCOMMON_OPT  "${CCOMMON_OPT} -mavx2")
@@ -124,9 +197,9 @@ if (NOT DYNAMIC_ARCH)
 	if (HAVE_AVX)
         set (CCOMMON_OPT  "${CCOMMON_OPT} -mavx")
 	endif ()
-	if (HAVE_FMA3)
-	set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
-	endif ()
+	#	if (HAVE_FMA3)
+	#set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
+	#endif ()
 	if (HAVE_SSE)
 	set (CCOMMON_OPT  "${CCOMMON_OPT} -msse")
 	endif ()

cmake/f_check.cmake

@@ -20,19 +20,16 @@
 # NEEDBUNDERSCORE
 # NEED2UNDERSCORES
 
-if (NOT NO_LAPACK)
-  include(CheckLanguage)
-  check_language(Fortran)
-  if(CMAKE_Fortran_COMPILER)
-    enable_language(Fortran)
-  else()
-  message(STATUS "No Fortran compiler found, can build only BLAS but not LAPACK")
+include(CheckLanguage)
+check_language(Fortran)
+if(CMAKE_Fortran_COMPILER)
+  enable_language(Fortran)
+else()
+  if (NOT NO_LAPACK)
+    message(STATUS "No Fortran compiler found, can build only BLAS but not LAPACK")
+  endif()
   set (NOFORTRAN 1)
   set (NO_LAPACK 1)
-  endif()
-else()
-  include(CMakeForceCompiler)
-  CMAKE_FORCE_Fortran_COMPILER(gfortran GNU)
 endif()
 
 if (NOT ONLY_CBLAS)

cmake/fc.cmake

@@ -3,11 +3,6 @@
 ## Description: Ported from portion of OpenBLAS/Makefile.system
 ##              Sets Fortran related variables.
 
-if (INTERFACE64)
-  set(SUFFIX64 64)
-  set(SUFFIX64_UNDERSCORE _64)
-endif()
-
 if (${F_COMPILER} STREQUAL "FLANG")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_FLANG")
   if (BINARY64 AND INTERFACE64)
@@ -61,6 +56,13 @@ if (${F_COMPILER} STREQUAL "GFORTRAN")
         set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=n32")
       endif ()
     endif ()
+    if (LOONGARCH64)
+      if (BINARY64)
+        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=lp64")
+      else ()
+        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=lp32")
+      endif ()
+    endif ()
   else ()
     if (BINARY64)
       set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
@@ -97,7 +99,7 @@ endif ()
 
 if (${F_COMPILER} STREQUAL "IBM")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_IBM")
-  # FCOMMON_OPT	+= -qarch=440
+  set(FCOMMON_OPT "${FCOMMON_OPT} -qrecur")
   if (BINARY64)
     set(FCOMMON_OPT "${FCOMMON_OPT} -q64")
     if (INTERFACE64)

cmake/kernel.cmake

@@ -1,212 +1,218 @@
 # helper functions for the kernel CMakeLists.txt
 
+function(SetFallback KERNEL SOURCE_PATH)
+  if (NOT (DEFINED ${KERNEL}))
+    set(${KERNEL} ${SOURCE_PATH} PARENT_SCOPE)
+  endif ()
+endfunction()
 
-# 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)
-  set(QAMAXKERNEL amax.S)
-  set(CAMAXKERNEL zamax.S)
-  set(ZAMAXKERNEL zamax.S)
-  set(XAMAXKERNEL zamax.S)
-  set(SAMINKERNEL amin.S)
-  set(DAMINKERNEL amin.S)
-  set(QAMINKERNEL amin.S)
-  set(CAMINKERNEL zamin.S)
-  set(ZAMINKERNEL zamin.S)
-  set(XAMINKERNEL zamin.S)
-  set(SMAXKERNEL max.S)
-  set(DMAXKERNEL max.S)
-  set(QMAXKERNEL max.S)
-  set(SMINKERNEL min.S)
-  set(DMINKERNEL min.S)
-  set(QMINKERNEL min.S)
-  set(ISAMAXKERNEL iamax.S)
-  set(IDAMAXKERNEL iamax.S)
-  set(IQAMAXKERNEL iamax.S)
-  set(ICAMAXKERNEL izamax.S)
-  set(IZAMAXKERNEL izamax.S)
-  set(IXAMAXKERNEL izamax.S)
-  set(ISAMINKERNEL iamin.S)
-  set(IDAMINKERNEL iamin.S)
-  set(IQAMINKERNEL iamin.S)
-  set(ICAMINKERNEL izamin.S)
-  set(IZAMINKERNEL izamin.S)
-  set(IXAMINKERNEL izamin.S)
-  set(ISMAXKERNEL iamax.S)
-  set(IDMAXKERNEL iamax.S)
-  set(IQMAXKERNEL iamax.S)
-  set(ISMINKERNEL iamin.S)
-  set(IDMINKERNEL iamin.S)
-  set(IQMINKERNEL iamin.S)
-  set(SASUMKERNEL asum.S)
-  set(DASUMKERNEL asum.S)
-  set(CASUMKERNEL zasum.S)
-  set(ZASUMKERNEL zasum.S)
-  set(QASUMKERNEL asum.S)
-  set(XASUMKERNEL zasum.S)
-  set(SAXPYKERNEL axpy.S)
-  set(DAXPYKERNEL axpy.S)
-  set(CAXPYKERNEL zaxpy.S)
-  set(ZAXPYKERNEL zaxpy.S)
-  set(QAXPYKERNEL axpy.S)
-  set(XAXPYKERNEL zaxpy.S)
-  set(SCOPYKERNEL copy.S)
-  set(DCOPYKERNEL copy.S)
-  set(CCOPYKERNEL zcopy.S)
-  set(ZCOPYKERNEL zcopy.S)
-  set(QCOPYKERNEL copy.S)
-  set(XCOPYKERNEL zcopy.S)
-  set(SDOTKERNEL dot.S)
-  set(DDOTKERNEL dot.S)
-  set(CDOTKERNEL zdot.S)
-  set(ZDOTKERNEL zdot.S)
-  set(QDOTKERNEL dot.S)
-  set(XDOTKERNEL zdot.S)
-  set(SNRM2KERNEL nrm2.S)
-  set(DNRM2KERNEL nrm2.S)
-  set(QNRM2KERNEL nrm2.S)
-  set(CNRM2KERNEL znrm2.S)
-  set(ZNRM2KERNEL znrm2.S)
-  set(XNRM2KERNEL znrm2.S)
-  set(SROTKERNEL rot.S)
-  set(DROTKERNEL rot.S)
-  set(QROTKERNEL rot.S)
-  set(CROTKERNEL zrot.S)
-  set(ZROTKERNEL zrot.S)
-  set(XROTKERNEL zrot.S)
-  set(SSCALKERNEL scal.S)
-  set(DSCALKERNEL scal.S)
-  set(CSCALKERNEL zscal.S)
-  set(ZSCALKERNEL zscal.S)
-  set(QSCALKERNEL scal.S)
-  set(XSCALKERNEL zscal.S)
-  set(SSWAPKERNEL swap.S)
-  set(DSWAPKERNEL swap.S)
-  set(CSWAPKERNEL zswap.S)
-  set(ZSWAPKERNEL zswap.S)
-  set(QSWAPKERNEL swap.S)
-  set(XSWAPKERNEL zswap.S)
-  set(SGEMVNKERNEL gemv_n.S)
-  set(SGEMVTKERNEL gemv_t.S)
-  set(DGEMVNKERNEL gemv_n.S)
-  set(DGEMVTKERNEL gemv_t.S)
-  set(CGEMVNKERNEL zgemv_n.S)
-  set(CGEMVTKERNEL zgemv_t.S)
-  set(ZGEMVNKERNEL zgemv_n.S)
-  set(ZGEMVTKERNEL zgemv_t.S)
-  set(QGEMVNKERNEL gemv_n.S)
-  set(QGEMVTKERNEL gemv_t.S)
-  set(XGEMVNKERNEL zgemv_n.S)
-  set(XGEMVTKERNEL zgemv_t.S)
-  set(SCABS_KERNEL ../generic/cabs.c)
-  set(DCABS_KERNEL ../generic/cabs.c)
-  set(QCABS_KERNEL ../generic/cabs.c)
-  set(LSAME_KERNEL ../generic/lsame.c)
-  set(SAXPBYKERNEL ../arm/axpby.c)
-  set(DAXPBYKERNEL ../arm/axpby.c)
-  set(CAXPBYKERNEL ../arm/zaxpby.c)
-  set(ZAXPBYKERNEL ../arm/zaxpby.c)
-  set(SSUMKERNEL sum.S)
-  set(DSUMKERNEL sum.S)
-  set(CSUMKERNEL zsum.S)
-  set(ZSUMKERNEL zsum.S)
-  set(QSUMKERNEL sum.S)
-  set(XSUMKERNEL zsum.S)
+  SetFallback(SAMAXKERNEL amax.S)
+  SetFallback(DAMAXKERNEL amax.S)
+  SetFallback(QAMAXKERNEL amax.S)
+  SetFallback(CAMAXKERNEL zamax.S)
+  SetFallback(ZAMAXKERNEL zamax.S)
+  SetFallback(XAMAXKERNEL zamax.S)
+  SetFallback(SAMINKERNEL amin.S)
+  SetFallback(DAMINKERNEL amin.S)
+  SetFallback(QAMINKERNEL amin.S)
+  SetFallback(CAMINKERNEL zamin.S)
+  SetFallback(ZAMINKERNEL zamin.S)
+  SetFallback(XAMINKERNEL zamin.S)
+  SetFallback(SMAXKERNEL max.S)
+  SetFallback(DMAXKERNEL max.S)
+  SetFallback(QMAXKERNEL max.S)
+  SetFallback(SMINKERNEL min.S)
+  SetFallback(DMINKERNEL min.S)
+  SetFallback(QMINKERNEL min.S)
+  SetFallback(ISAMAXKERNEL iamax.S)
+  SetFallback(IDAMAXKERNEL iamax.S)
+  SetFallback(IQAMAXKERNEL iamax.S)
+  SetFallback(ICAMAXKERNEL izamax.S)
+  SetFallback(IZAMAXKERNEL izamax.S)
+  SetFallback(IXAMAXKERNEL izamax.S)
+  SetFallback(ISAMINKERNEL iamin.S)
+  SetFallback(IDAMINKERNEL iamin.S)
+  SetFallback(IQAMINKERNEL iamin.S)
+  SetFallback(ICAMINKERNEL izamin.S)
+  SetFallback(IZAMINKERNEL izamin.S)
+  SetFallback(IXAMINKERNEL izamin.S)
+  SetFallback(ISMAXKERNEL iamax.S)
+  SetFallback(IDMAXKERNEL iamax.S)
+  SetFallback(IQMAXKERNEL iamax.S)
+  SetFallback(ISMINKERNEL iamin.S)
+  SetFallback(IDMINKERNEL iamin.S)
+  SetFallback(IQMINKERNEL iamin.S)
+  SetFallback(SASUMKERNEL asum.S)
+  SetFallback(DASUMKERNEL asum.S)
+  SetFallback(CASUMKERNEL zasum.S)
+  SetFallback(ZASUMKERNEL zasum.S)
+  SetFallback(QASUMKERNEL asum.S)
+  SetFallback(XASUMKERNEL zasum.S)
+  SetFallback(SAXPYKERNEL axpy.S)
+  SetFallback(DAXPYKERNEL axpy.S)
+  SetFallback(CAXPYKERNEL zaxpy.S)
+  SetFallback(ZAXPYKERNEL zaxpy.S)
+  SetFallback(QAXPYKERNEL axpy.S)
+  SetFallback(XAXPYKERNEL zaxpy.S)
+  SetFallback(SCOPYKERNEL copy.S)
+  SetFallback(DCOPYKERNEL copy.S)
+  SetFallback(CCOPYKERNEL zcopy.S)
+  SetFallback(ZCOPYKERNEL zcopy.S)
+  SetFallback(QCOPYKERNEL copy.S)
+  SetFallback(XCOPYKERNEL zcopy.S)
+  SetFallback(SDOTKERNEL dot.S)
+  SetFallback(DDOTKERNEL dot.S)
+  SetFallback(CDOTKERNEL zdot.S)
+  SetFallback(ZDOTKERNEL zdot.S)
+  SetFallback(QDOTKERNEL dot.S)
+  SetFallback(XDOTKERNEL zdot.S)
+  SetFallback(SNRM2KERNEL nrm2.S)
+  SetFallback(DNRM2KERNEL nrm2.S)
+  SetFallback(QNRM2KERNEL nrm2.S)
+  SetFallback(CNRM2KERNEL znrm2.S)
+  SetFallback(ZNRM2KERNEL znrm2.S)
+  SetFallback(XNRM2KERNEL znrm2.S)
+  SetFallback(SROTKERNEL rot.S)
+  SetFallback(DROTKERNEL rot.S)
+  SetFallback(QROTKERNEL rot.S)
+  SetFallback(CROTKERNEL zrot.S)
+  SetFallback(ZROTKERNEL zrot.S)
+  SetFallback(XROTKERNEL zrot.S)
+  SetFallback(SSCALKERNEL scal.S)
+  SetFallback(DSCALKERNEL scal.S)
+  SetFallback(CSCALKERNEL zscal.S)
+  SetFallback(ZSCALKERNEL zscal.S)
+  SetFallback(QSCALKERNEL scal.S)
+  SetFallback(XSCALKERNEL zscal.S)
+  SetFallback(SSWAPKERNEL swap.S)
+  SetFallback(DSWAPKERNEL swap.S)
+  SetFallback(CSWAPKERNEL zswap.S)
+  SetFallback(ZSWAPKERNEL zswap.S)
+  SetFallback(QSWAPKERNEL swap.S)
+  SetFallback(XSWAPKERNEL zswap.S)
+  SetFallback(SGEMVNKERNEL gemv_n.S)
+  SetFallback(SGEMVTKERNEL gemv_t.S)
+  SetFallback(DGEMVNKERNEL gemv_n.S)
+  SetFallback(DGEMVTKERNEL gemv_t.S)
+  SetFallback(CGEMVNKERNEL zgemv_n.S)
+  SetFallback(CGEMVTKERNEL zgemv_t.S)
+  SetFallback(ZGEMVNKERNEL zgemv_n.S)
+  SetFallback(ZGEMVTKERNEL zgemv_t.S)
+  SetFallback(QGEMVNKERNEL gemv_n.S)
+  SetFallback(QGEMVTKERNEL gemv_t.S)
+  SetFallback(XGEMVNKERNEL zgemv_n.S)
+  SetFallback(XGEMVTKERNEL zgemv_t.S)
+  SetFallback(SCABS_KERNEL ../generic/cabs.c)
+  SetFallback(DCABS_KERNEL ../generic/cabs.c)
+  SetFallback(QCABS_KERNEL ../generic/cabs.c)
+  SetFallback(LSAME_KERNEL ../generic/lsame.c)
+  SetFallback(SAXPBYKERNEL ../arm/axpby.c)
+  SetFallback(DAXPBYKERNEL ../arm/axpby.c)
+  SetFallback(CAXPBYKERNEL ../arm/zaxpby.c)
+  SetFallback(ZAXPBYKERNEL ../arm/zaxpby.c)
+  SetFallback(SSUMKERNEL sum.S)
+  SetFallback(DSUMKERNEL sum.S)
+  SetFallback(CSUMKERNEL zsum.S)
+  SetFallback(ZSUMKERNEL zsum.S)
+  SetFallback(QSUMKERNEL sum.S)
+  SetFallback(XSUMKERNEL zsum.S)
 if (BUILD_BFLOAT16)
-  set(SHAMINKERNEL ../arm/amin.c)
-  set(SHAMAXKERNEL ../arm/amax.c)
-  set(SHMAXKERNEL ../arm/max.c)
-  set(SHMINKERNEL ../arm/min.c)
-  set(ISHAMAXKERNEL ../arm/iamax.c)
-  set(ISHAMINKERNEL ../arm/iamin.c)
-  set(ISHMAXKERNEL ../arm/imax.c)
-  set(ISHMINKERNEL ../arm/imin.c)
-  set(SHASUMKERNEL ../arm/asum.c)
-  set(SHAXPYKERNEL ../arm/axpy.c)
-  set(SHAXPBYKERNEL ../arm/axpby.c)
-  set(SHCOPYKERNEL ../arm/copy.c)
-  set(SBDOTKERNEL ../x86_64/sbdot.c)
-  set(SHROTKERNEL ../arm/rot.c)
-  set(SHSCALKERNEL ../arm/scal.c)
-  set(SHNRM2KERNEL ../arm/nrm2.c)
-  set(SHSUMKERNEL ../arm/sum.c)
-  set(SHSWAPKERNEL ../arm/swap.c)
-  set(TOBF16KERNEL ../x86_64/tobf16.c)
-  set(BF16TOKERNEL ../x86_64/bf16to.c)
+  SetFallback(SHAMINKERNEL ../arm/amin.c)
+  SetFallback(SHAMAXKERNEL ../arm/amax.c)
+  SetFallback(SHMAXKERNEL ../arm/max.c)
+  SetFallback(SHMINKERNEL ../arm/min.c)
+  SetFallback(ISHAMAXKERNEL ../arm/iamax.c)
+  SetFallback(ISHAMINKERNEL ../arm/iamin.c)
+  SetFallback(ISHMAXKERNEL ../arm/imax.c)
+  SetFallback(ISHMINKERNEL ../arm/imin.c)
+  SetFallback(SHASUMKERNEL ../arm/asum.c)
+  SetFallback(SHAXPYKERNEL ../arm/axpy.c)
+  SetFallback(SHAXPBYKERNEL ../arm/axpby.c)
+  SetFallback(SHCOPYKERNEL ../arm/copy.c)
+  SetFallback(SBDOTKERNEL ../x86_64/sbdot.c)
+  SetFallback(SHROTKERNEL ../arm/rot.c)
+  SetFallback(SHSCALKERNEL ../arm/scal.c)
+  SetFallback(SHNRM2KERNEL ../arm/nrm2.c)
+  SetFallback(SHSUMKERNEL ../arm/sum.c)
+  SetFallback(SHSWAPKERNEL ../arm/swap.c)
+  SetFallback(TOBF16KERNEL ../x86_64/tobf16.c)
+  SetFallback(BF16TOKERNEL ../x86_64/bf16to.c)
+  SetFallback(SBGEMVNKERNEL ../x86_64/sbgemv_n.c)
+  SetFallback(SBGEMVTKERNEL ../x86_64/sbgemv_t.c)
 endif ()
 endmacro ()
 
 macro(SetDefaultL2)
-  set(SGEMVNKERNEL ../arm/gemv_n.c)
-  set(SGEMVTKERNEL ../arm/gemv_t.c)
-  set(DGEMVNKERNEL gemv_n.S)
-  set(DGEMVTKERNEL gemv_t.S)
-  set(CGEMVNKERNEL zgemv_n.S)
-  set(CGEMVTKERNEL zgemv_t.S)
-  set(ZGEMVNKERNEL zgemv_n.S)
-  set(ZGEMVTKERNEL zgemv_t.S)
-  set(QGEMVNKERNEL gemv_n.S)
-  set(QGEMVTKERNEL gemv_t.S)
-  set(XGEMVNKERNEL zgemv_n.S)
-  set(XGEMVTKERNEL zgemv_t.S)
-  set(SGERKERNEL ../generic/ger.c)
-  set(DGERKERNEL ../generic/ger.c)
-  set(QGERKERNEL ../generic/ger.c)
-  set(CGERUKERNEL ../generic/zger.c)
-  set(CGERCKERNEL ../generic/zger.c)
-  set(ZGERUKERNEL ../generic/zger.c)
-  set(ZGERCKERNEL ../generic/zger.c)
-  set(XGERUKERNEL ../generic/zger.c)
-  set(XGERCKERNEL ../generic/zger.c)
-  set(SSYMV_U_KERNEL ../generic/symv_k.c)
-  set(SSYMV_L_KERNEL ../generic/symv_k.c)
-  set(DSYMV_U_KERNEL ../generic/symv_k.c)
-  set(DSYMV_L_KERNEL ../generic/symv_k.c)
-  set(QSYMV_U_KERNEL ../generic/symv_k.c)
-  set(QSYMV_L_KERNEL ../generic/symv_k.c)
-  set(CSYMV_U_KERNEL ../generic/zsymv_k.c)
-  set(CSYMV_L_KERNEL ../generic/zsymv_k.c)
-  set(ZSYMV_U_KERNEL ../generic/zsymv_k.c)
-  set(ZSYMV_L_KERNEL ../generic/zsymv_k.c)
-  set(XSYMV_U_KERNEL ../generic/zsymv_k.c)
-  set(XSYMV_L_KERNEL ../generic/zsymv_k.c)
-  set(CHEMV_U_KERNEL ../generic/zhemv_k.c)
-  set(CHEMV_L_KERNEL ../generic/zhemv_k.c)
-  set(CHEMV_V_KERNEL ../generic/zhemv_k.c)
-  set(CHEMV_M_KERNEL ../generic/zhemv_k.c)
-  set(ZHEMV_U_KERNEL ../generic/zhemv_k.c)
-  set(ZHEMV_L_KERNEL ../generic/zhemv_k.c)
-  set(ZHEMV_V_KERNEL ../generic/zhemv_k.c)
-  set(ZHEMV_M_KERNEL ../generic/zhemv_k.c)
-  set(XHEMV_U_KERNEL ../generic/zhemv_k.c)
-  set(XHEMV_L_KERNEL ../generic/zhemv_k.c)
-  set(XHEMV_V_KERNEL ../generic/zhemv_k.c)
-  set(XHEMV_M_KERNEL ../generic/zhemv_k.c)
+  SetFallback(SGEMVNKERNEL ../arm/gemv_n.c)
+  SetFallback(SGEMVTKERNEL ../arm/gemv_t.c)
+  SetFallback(DGEMVNKERNEL gemv_n.S)
+  SetFallback(DGEMVTKERNEL gemv_t.S)
+  SetFallback(CGEMVNKERNEL zgemv_n.S)
+  SetFallback(CGEMVTKERNEL zgemv_t.S)
+  SetFallback(ZGEMVNKERNEL zgemv_n.S)
+  SetFallback(ZGEMVTKERNEL zgemv_t.S)
+  SetFallback(QGEMVNKERNEL gemv_n.S)
+  SetFallback(QGEMVTKERNEL gemv_t.S)
+  SetFallback(XGEMVNKERNEL zgemv_n.S)
+  SetFallback(XGEMVTKERNEL zgemv_t.S)
+  SetFallback(SGERKERNEL ../generic/ger.c)
+  SetFallback(DGERKERNEL ../generic/ger.c)
+  SetFallback(QGERKERNEL ../generic/ger.c)
+  SetFallback(CGERUKERNEL ../generic/zger.c)
+  SetFallback(CGERCKERNEL ../generic/zger.c)
+  SetFallback(ZGERUKERNEL ../generic/zger.c)
+  SetFallback(ZGERCKERNEL ../generic/zger.c)
+  SetFallback(XGERUKERNEL ../generic/zger.c)
+  SetFallback(XGERCKERNEL ../generic/zger.c)
+  SetFallback(SSYMV_U_KERNEL ../generic/symv_k.c)
+  SetFallback(SSYMV_L_KERNEL ../generic/symv_k.c)
+  SetFallback(DSYMV_U_KERNEL ../generic/symv_k.c)
+  SetFallback(DSYMV_L_KERNEL ../generic/symv_k.c)
+  SetFallback(QSYMV_U_KERNEL ../generic/symv_k.c)
+  SetFallback(QSYMV_L_KERNEL ../generic/symv_k.c)
+  SetFallback(CSYMV_U_KERNEL ../generic/zsymv_k.c)
+  SetFallback(CSYMV_L_KERNEL ../generic/zsymv_k.c)
+  SetFallback(ZSYMV_U_KERNEL ../generic/zsymv_k.c)
+  SetFallback(ZSYMV_L_KERNEL ../generic/zsymv_k.c)
+  SetFallback(XSYMV_U_KERNEL ../generic/zsymv_k.c)
+  SetFallback(XSYMV_L_KERNEL ../generic/zsymv_k.c)
+  SetFallback(CHEMV_U_KERNEL ../generic/zhemv_k.c)
+  SetFallback(CHEMV_L_KERNEL ../generic/zhemv_k.c)
+  SetFallback(CHEMV_V_KERNEL ../generic/zhemv_k.c)
+  SetFallback(CHEMV_M_KERNEL ../generic/zhemv_k.c)
+  SetFallback(ZHEMV_U_KERNEL ../generic/zhemv_k.c)
+  SetFallback(ZHEMV_L_KERNEL ../generic/zhemv_k.c)
+  SetFallback(ZHEMV_V_KERNEL ../generic/zhemv_k.c)
+  SetFallback(ZHEMV_M_KERNEL ../generic/zhemv_k.c)
+  SetFallback(XHEMV_U_KERNEL ../generic/zhemv_k.c)
+  SetFallback(XHEMV_L_KERNEL ../generic/zhemv_k.c)
+  SetFallback(XHEMV_V_KERNEL ../generic/zhemv_k.c)
+  SetFallback(XHEMV_M_KERNEL ../generic/zhemv_k.c)
 if (BUILD_BFLOAT16)
-  set(SBGEMVNKERNEL ../x86_64/sbgemv_n.c)
-  set(SBGEMVTKERNEL ../x86_64/sbgemv_t.c)
-  set(SHGERKERNEL ../generic/ger.c)
+  SetFallback(SBGEMVNKERNEL ../x86_64/sbgemv_n.c)
+  SetFallback(SBGEMVTKERNEL ../x86_64/sbgemv_t.c)
+  SetFallback(SHGERKERNEL ../generic/ger.c)
 endif ()
 endmacro ()
 
 macro(SetDefaultL3)
-  set(SGEADD_KERNEL ../generic/geadd.c)
-  set(DGEADD_KERNEL ../generic/geadd.c)
-  set(CGEADD_KERNEL ../generic/zgeadd.c)
-  set(ZGEADD_KERNEL ../generic/zgeadd.c)
+  SetFallback(SGEADD_KERNEL ../generic/geadd.c)
+  SetFallback(DGEADD_KERNEL ../generic/geadd.c)
+  SetFallback(CGEADD_KERNEL ../generic/zgeadd.c)
+  SetFallback(ZGEADD_KERNEL ../generic/zgeadd.c)
 if (BUILD_BFLOAT16)
-  set(SHGEADD_KERNEL ../generic/geadd.c)
-  set(SBGEMMKERNEL ../generic/gemmkernel_2x2.c)
-  set(SBGEMM_BETA  ../generic/gemm_beta.c)
-  set(SBGEMMINCOPY ../generic/gemm_ncopy_2.c)
-  set(SBGEMMITCOPY ../generic/gemm_tcopy_2.c)
-  set(SBGEMMONCOPY ../generic/gemm_ncopy_2.c)
-  set(SBGEMMOTCOPY ../generic/gemm_tcopy_2.c)
-  set(SBGEMMINCOPYOBJ sbgemm_incopy.o)
-  set(SBGEMMITCOPYOBJ sbgemm_itcopy.o)
-  set(SBGEMMONCOPYOBJ sbgemm_oncopy.o)
-  set(SBGEMMOTCOPYOBJ sbgemm_otcopy.o)
+  SetFallback(SHGEADD_KERNEL ../generic/geadd.c)
+  SetFallback(SBGEMMKERNEL ../generic/gemmkernel_2x2.c)
+  SetFallback(SBGEMM_BETA  ../generic/gemm_beta.c)
+  SetFallback(SBGEMMINCOPY ../generic/gemm_ncopy_2.c)
+  SetFallback(SBGEMMITCOPY ../generic/gemm_tcopy_2.c)
+  SetFallback(SBGEMMONCOPY ../generic/gemm_ncopy_2.c)
+  SetFallback(SBGEMMOTCOPY ../generic/gemm_tcopy_2.c)
+  SetFallback(SBGEMMINCOPYOBJ sbgemm_incopy.o)
+  SetFallback(SBGEMMITCOPYOBJ sbgemm_itcopy.o)
+  SetFallback(SBGEMMONCOPYOBJ sbgemm_oncopy.o)
+  SetFallback(SBGEMMOTCOPYOBJ sbgemm_otcopy.o)
 endif ()
 
 endmacro ()

cmake/lapack.cmake

@@ -66,7 +66,7 @@ set(SLASRC
    slaqgb.f slaqge.f slaqp2.f slaqps.f slaqsb.f slaqsp.f slaqsy.f
    slaqr0.f slaqr1.f slaqr2.f slaqr3.f slaqr4.f slaqr5.f
    slaqtr.f slar1v.f slar2v.f ilaslr.f ilaslc.f
-   slarf.f  slarfb.f slarfg.f slarfgp.f slarft.f slarfx.f slarfy.f slargv.f
+   slarf.f  slarfb.f slarfb_gett.f slarfg.f slarfgp.f slarft.f slarfx.f slarfy.f slargv.f
    slarrv.f slartv.f
    slarz.f  slarzb.f slarzt.f slasy2.f
    slasyf.f slasyf_rook.f slasyf_rk.f slasyf_aa.f
@@ -112,14 +112,14 @@ set(SLASRC
    sgeqrt.f sgeqrt2.f sgeqrt3.f sgemqrt.f
    stpqrt.f stpqrt2.f stpmqrt.f stprfb.f
    sgelqt.f sgelqt3.f sgemlqt.f
-   sgetsls.f sgeqr.f slatsqr.f slamtsqr.f sgemqr.f
+   sgetsls.f sgetsqrhrt.f sgeqr.f slatsqr.f slamtsqr.f sgemqr.f
    sgelq.f slaswlq.f slamswlq.f sgemlq.f
    stplqt.f stplqt2.f stpmlqt.f
    ssytrd_2stage.f ssytrd_sy2sb.f ssytrd_sb2st.F ssb2st_kernels.f
    ssyevd_2stage.f ssyev_2stage.f ssyevx_2stage.f ssyevr_2stage.f
    ssbev_2stage.f ssbevx_2stage.f ssbevd_2stage.f ssygv_2stage.f
    sgesvdq.f slaorhr_col_getrfnp.f
-   slaorhr_col_getrfnp2.f sorgtsqr.f sorhr_col.f )
+   slaorhr_col_getrfnp2.f sorgtsqr.f sorgtsqr_row.f sorhr_col.f )
 
 set(SXLASRC sgesvxx.f sgerfsx.f sla_gerfsx_extended.f sla_geamv.f
    sla_gercond.f sla_gerpvgrw.f ssysvxx.f ssyrfsx.f
@@ -171,7 +171,7 @@ set(CLASRC
    claqhb.f claqhe.f claqhp.f claqp2.f claqps.f claqsb.f
    claqr0.f claqr1.f claqr2.f claqr3.f claqr4.f claqr5.f
    claqsp.f claqsy.f clar1v.f clar2v.f ilaclr.f ilaclc.f
-   clarf.f  clarfb.f clarfg.f clarfgp.f clarft.f
+   clarf.f  clarfb.f clarfb_gett.f clarfg.f clarfgp.f clarft.f
    clarfx.f clarfy.f clargv.f clarnv.f clarrv.f clartg.f clartv.f
    clarz.f  clarzb.f clarzt.f clascl.f claset.f clasr.f  classq.f
    clasyf.f clasyf_rook.f clasyf_rk.f clasyf_aa.f
@@ -209,14 +209,14 @@ set(CLASRC
    cgeqrt.f cgeqrt2.f cgeqrt3.f cgemqrt.f
    ctpqrt.f ctpqrt2.f ctpmqrt.f ctprfb.f
    cgelqt.f cgelqt3.f cgemlqt.f
-   cgetsls.f cgeqr.f clatsqr.f clamtsqr.f cgemqr.f
+   cgetsls.f cgetsqrhrt.f cgeqr.f clatsqr.f clamtsqr.f cgemqr.f
    cgelq.f claswlq.f clamswlq.f cgemlq.f
    ctplqt.f ctplqt2.f ctpmlqt.f
    chetrd_2stage.f chetrd_he2hb.f chetrd_hb2st.F chb2st_kernels.f
    cheevd_2stage.f cheev_2stage.f cheevx_2stage.f cheevr_2stage.f
    chbev_2stage.f chbevx_2stage.f chbevd_2stage.f chegv_2stage.f
    cgesvdq.f claunhr_col_getrfnp.f claunhr_col_getrfnp2.f 
-   cungtsqr.f cunhr_col.f )
+   cungtsqr.f cungtsqr_row.f cunhr_col.f )
 
 set(CXLASRC cgesvxx.f cgerfsx.f cla_gerfsx_extended.f cla_geamv.f
    cla_gercond_c.f cla_gercond_x.f cla_gerpvgrw.f
@@ -253,7 +253,7 @@ set(DLASRC
    dlaqgb.f dlaqge.f dlaqp2.f dlaqps.f dlaqsb.f dlaqsp.f dlaqsy.f
    dlaqr0.f dlaqr1.f dlaqr2.f dlaqr3.f dlaqr4.f dlaqr5.f
    dlaqtr.f dlar1v.f dlar2v.f iladlr.f iladlc.f
-   dlarf.f  dlarfb.f dlarfg.f dlarfgp.f dlarft.f dlarfx.f dlarfy.f
+   dlarf.f  dlarfb.f dlarfb_gett.f dlarfg.f dlarfgp.f dlarft.f dlarfx.f dlarfy.f
    dlargv.f dlarrv.f dlartv.f
    dlarz.f  dlarzb.f dlarzt.f dlasy2.f
    dlasyf.f dlasyf_rook.f dlasyf_rk.f dlasyf_aa.f
@@ -300,14 +300,14 @@ set(DLASRC
    dgeqrt.f dgeqrt2.f dgeqrt3.f dgemqrt.f
    dtpqrt.f dtpqrt2.f dtpmqrt.f dtprfb.f
    dgelqt.f dgelqt3.f dgemlqt.f
-   dgetsls.f dgeqr.f dlatsqr.f dlamtsqr.f dgemqr.f
+   dgetsls.f dgetsqrhrt.f dgeqr.f dlatsqr.f dlamtsqr.f dgemqr.f
    dgelq.f dlaswlq.f dlamswlq.f dgemlq.f
    dtplqt.f dtplqt2.f dtpmlqt.f
    dsytrd_2stage.f dsytrd_sy2sb.f dsytrd_sb2st.F dsb2st_kernels.f
    dsyevd_2stage.f dsyev_2stage.f dsyevx_2stage.f dsyevr_2stage.f
    dsbev_2stage.f dsbevx_2stage.f dsbevd_2stage.f dsygv_2stage.f
    dcombssq.f dgesvdq.f dlaorhr_col_getrfnp.f
-   dlaorhr_col_getrfnp2.f dorgtsqr.f dorhr_col.f )
+   dlaorhr_col_getrfnp2.f dorgtsqr.f dorgtsqr_row.f dorhr_col.f )
 
 set(DXLASRC dgesvxx.f dgerfsx.f dla_gerfsx_extended.f dla_geamv.f
    dla_gercond.f dla_gerpvgrw.f dsysvxx.f dsyrfsx.f
@@ -360,7 +360,7 @@ set(ZLASRC
    zlaqhb.f zlaqhe.f zlaqhp.f zlaqp2.f zlaqps.f zlaqsb.f
    zlaqr0.f zlaqr1.f zlaqr2.f zlaqr3.f zlaqr4.f zlaqr5.f
    zlaqsp.f zlaqsy.f zlar1v.f zlar2v.f ilazlr.f ilazlc.f
-   zlarcm.f zlarf.f  zlarfb.f
+   zlarcm.f zlarf.f  zlarfb.f zlarfb_gett.f
    zlarfg.f zlarfgp.f zlarft.f
    zlarfx.f zlarfy.f zlargv.f zlarnv.f zlarrv.f zlartg.f zlartv.f
    zlarz.f  zlarzb.f zlarzt.f zlascl.f zlaset.f zlasr.f
@@ -402,13 +402,13 @@ set(ZLASRC
    ztpqrt.f ztpqrt2.f ztpmqrt.f ztprfb.f
    ztplqt.f ztplqt2.f ztpmlqt.f
    zgelqt.f zgelqt3.f zgemlqt.f
-   zgetsls.f zgeqr.f zlatsqr.f zlamtsqr.f zgemqr.f
+   zgetsls.f zgetsqrhrt.f zgeqr.f zlatsqr.f zlamtsqr.f zgemqr.f
    zgelq.f zlaswlq.f zlamswlq.f zgemlq.f
    zhetrd_2stage.f zhetrd_he2hb.f zhetrd_hb2st.F zhb2st_kernels.f
    zheevd_2stage.f zheev_2stage.f zheevx_2stage.f zheevr_2stage.f
    zhbev_2stage.f zhbevx_2stage.f zhbevd_2stage.f zhegv_2stage.f
    zgesvdq.f zlaunhr_col_getrfnp.f zlaunhr_col_getrfnp2.f
-   zungtsqr.f zunhr_col.f)
+   zungtsqr.f zungtsqr_row.f zunhr_col.f)
 
 set(ZXLASRC zgesvxx.f zgerfsx.f zla_gerfsx_extended.f zla_geamv.f
    zla_gercond_c.f zla_gercond_x.f zla_gerpvgrw.f zsysvxx.f zsyrfsx.f

cmake/lapacke.cmake

@@ -114,6 +114,8 @@ set(CSRC
   lapacke_cgetrs_work.c
   lapacke_cgetsls.c
   lapacke_cgetsls_work.c
+  lapacke_cgetsqrhrt.c
+  lapacke_cgetsqrhrt_work.c 
   lapacke_cggbak.c
   lapacke_cggbak_work.c
   lapacke_cggbal.c
@@ -590,6 +592,8 @@ set(CSRC
   lapacke_cungrq_work.c
   lapacke_cungtr.c
   lapacke_cungtr_work.c
+  lapacke_cungtsqr_row.c
+  lapacke_cungtsqr_row_work.c
   lapacke_cunmbr.c
   lapacke_cunmbr_work.c
   lapacke_cunmhr.c
@@ -735,6 +739,8 @@ set(DSRC
   lapacke_dgetrs_work.c
   lapacke_dgetsls.c
   lapacke_dgetsls_work.c
+  lapacke_dgetsqrhrt.c
+  lapacke_dgetsqrhrt_work.c
   lapacke_dggbak.c
   lapacke_dggbak_work.c
   lapacke_dggbal.c
@@ -862,6 +868,8 @@ set(DSRC
   lapacke_dorgrq_work.c
   lapacke_dorgtr.c
   lapacke_dorgtr_work.c
+  lapacke_dorgtsqr_row.c
+  lapacke_dorgtsqr_row_work.c
   lapacke_dormbr.c
   lapacke_dormbr_work.c
   lapacke_dormhr.c
@@ -1309,6 +1317,8 @@ set(SSRC
   lapacke_sgetrs_work.c
   lapacke_sgetsls.c
   lapacke_sgetsls_work.c
+  lapacke_sgetsqrhrt.c
+  lapacke_sgetsqrhrt_work.c
   lapacke_sggbak.c
   lapacke_sggbak_work.c
   lapacke_sggbal.c
@@ -1435,6 +1445,8 @@ set(SSRC
   lapacke_sorgrq_work.c
   lapacke_sorgtr.c
   lapacke_sorgtr_work.c
+  lapacke_sorgtsqr_row.c
+  lapacke_sorgtsqr_row_work.c
   lapacke_sormbr.c
   lapacke_sormbr_work.c
   lapacke_sormhr.c
@@ -1877,6 +1889,8 @@ set(ZSRC
   lapacke_zgetrs_work.c
   lapacke_zgetsls.c
   lapacke_zgetsls_work.c
+  lapacke_zgetsqrhrt.c
+  lapacke_zgetsqrhrt_work.c
   lapacke_zggbak.c
   lapacke_zggbak_work.c
   lapacke_zggbal.c
@@ -2351,6 +2365,8 @@ set(ZSRC
   lapacke_zungrq_work.c
   lapacke_zungtr.c
   lapacke_zungtr_work.c
+  lapacke_zungtsqr_row.c
+  lapacke_zungtsqr_row_work.c
   lapacke_zunmbr.c
   lapacke_zunmbr_work.c
   lapacke_zunmhr.c
@@ -2499,6 +2515,5 @@ foreach (Utils_FILE ${Utils_SRC})
 endforeach ()
 
 set(lapacke_include_dir "${NETLIB_LAPACK_DIR}/LAPACKE/include")
-configure_file("${lapacke_include_dir}/lapacke_mangling_with_flags.h.in" "${lapacke_include_dir}/lapacke_mangling.h" COPYONLY)
 include_directories(${lapacke_include_dir})
 set_source_files_properties(${LAPACKE_SOURCES} PROPERTIES COMPILE_FLAGS "${LAPACK_CFLAGS}")

cmake/prebuild.cmake

@@ -127,6 +127,10 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
       "#define DLOCAL_BUFFER_SIZE\t16384\n"
       "#define CLOCAL_BUFFER_SIZE\t16384\n"
       "#define ZLOCAL_BUFFER_SIZE\t16384\n")
+      set(HAVE_SSE 1)
+      set(HAVE_SSE2 1)
+      set(HAVE_SSE3 1)
+      set(HAVE_SSSE3 1)
       set(SGEMM_UNROLL_M 8)
       set(SGEMM_UNROLL_N 4)
       set(DGEMM_UNROLL_M 4)
@@ -177,7 +181,7 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
     set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
-  elseif ("${TCORE}" STREQUAL "CORTEXA57" OR "${TCORE}" STREQUAL "CORTEXA53")
+  elseif ("${TCORE}" STREQUAL "CORTEXA57" OR "${TCORE}" STREQUAL "CORTEXA53" OR "${TCORE}" STREQUAL "CORTEXA55")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_CODE_SIZE\t32768\n"
       "#define L1_CODE_LINESIZE\t64\n"
@@ -237,6 +241,61 @@ endif ()
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
   elseif ("${TCORE}" STREQUAL "NEOVERSEN1")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t65536\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t4\n"
+      "#define L1_DATA_SIZE\t65536\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t4\n"
+      "#define L2_SIZE\t1048576\n\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t8\n"
+      "#define DTB_DEFAULT_ENTRIES\t48\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define HAVE_VFPV4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n"
+      "#define HAVE_NEON\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "NEOVERSEV1")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t65536\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t4\n"
+      "#define L1_DATA_SIZE\t65536\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t4\n"
+      "#define L2_SIZE\t1048576\n\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t8\n"
+      "#define DTB_DEFAULT_ENTRIES\t48\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define HAVE_VFPV4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n"
+      "#define HAVE_NEON\n"
+      "#define HAVE_SVE\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "NEOVERSEN2")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_CODE_SIZE\t65536\n"
       "#define L1_CODE_LINESIZE\t64\n"
@@ -246,13 +305,14 @@ endif ()
       "#define L1_DATA_ASSOCIATIVE\t2\n"
       "#define L2_SIZE\t1048576\n\n"
       "#define L2_LINESIZE\t64\n"
-      "#define L2_ASSOCIATIVE\t16\n"
-      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define L2_ASSOCIATIVE\t8\n"
+      "#define DTB_DEFAULT_ENTRIES\t48\n"
       "#define DTB_SIZE\t4096\n"
       "#define HAVE_VFPV4\n"
       "#define HAVE_VFPV3\n"
       "#define HAVE_VFP\n"
       "#define HAVE_NEON\n"
+      "#define HAVE_SVE\n"
       "#define ARMV8\n")
     set(SGEMM_UNROLL_M 16)
     set(SGEMM_UNROLL_N 4)
@@ -416,7 +476,7 @@ endif ()
     set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
-elseif ("${TCORE}" STREQUAL "VORTEX")
+  elseif ("${TCORE}" STREQUAL "VORTEX")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define ARMV8\n"
       "#define L1_CODE_SIZE\t32768\n"
@@ -439,6 +499,34 @@ elseif ("${TCORE}" STREQUAL "VORTEX")
     set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "P5600")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L2_SIZE 1048576\n"
+      "#define DTB_SIZE 4096\n"
+      "#define DTB_DEFAULT_ENTRIES 64\n")
+    set(SGEMM_UNROLL_M 2)
+    set(SGEMM_UNROLL_N 2)
+    set(DGEMM_UNROLL_M 2)
+    set(DGEMM_UNROLL_N 2)
+    set(CGEMM_UNROLL_M 2)
+    set(CGEMM_UNROLL_N 2)
+    set(ZGEMM_UNROLL_M 2)
+    set(ZGEMM_UNROLL_N 2)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" MATCHES "MIPS")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L2_SIZE 262144\n"
+      "#define DTB_SIZE 4096\n"
+      "#define DTB_DEFAULT_ENTRIES 64\n")
+    set(SGEMM_UNROLL_M 2)
+    set(SGEMM_UNROLL_N 2)
+    set(DGEMM_UNROLL_M 2)
+    set(DGEMM_UNROLL_N 2)
+    set(CGEMM_UNROLL_M 2)
+    set(CGEMM_UNROLL_N 2)
+    set(ZGEMM_UNROLL_M 2)
+    set(ZGEMM_UNROLL_N 2)
+    set(SYMV_P 16)
   elseif ("${TCORE}" STREQUAL "POWER6")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE 32768\n"

cmake/system.cmake

@@ -33,15 +33,18 @@ endif ()
 if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
   message(STATUS "Compiling a ${BINARY}-bit binary.")
   set(NO_AVX 1)
-  if (${TARGET} STREQUAL "HASWELL" OR ${TARGET} STREQUAL "SANDYBRIDGE" OR ${TARGET} STREQUAL "SKYLAKEX" OR ${TARGET} STREQUAL "COOPERLAKE")
+  if (${TARGET} STREQUAL "HASWELL" OR ${TARGET} STREQUAL "SANDYBRIDGE" OR ${TARGET} STREQUAL "SKYLAKEX" OR ${TARGET} STREQUAL "COOPERLAKE" OR ${TARGET} STREQUAL "SAPPHIRERAPIDS")
     set(TARGET "NEHALEM")
   endif ()
   if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "ZEN")
     set(TARGET "BARCELONA")
   endif ()
-  if (${TARGET} STREQUAL "ARMV8" OR ${TARGET} STREQUAL "CORTEXA57" OR ${TARGET} STREQUAL "CORTEXA53")
+  if (${TARGET} STREQUAL "ARMV8" OR ${TARGET} STREQUAL "CORTEXA57" OR ${TARGET} STREQUAL "CORTEXA53" OR ${TARGET} STREQUAL "CORTEXA55")
     set(TARGET "ARMV7")
   endif ()
+  if (${TARGET} STREQUAL "POWER8" OR ${TARGET} STREQUAL "POWER9" OR ${TARGET} STREQUAL "POWER10")
+    set(TARGET "POWER6")
+  endif ()
 endif ()
 
 
@@ -102,6 +105,18 @@ if (CMAKE_C_COMPILER STREQUAL loongcc)
   set(GETARCH_FLAGS	"${GETARCH_FLAGS} -static")
 endif ()
 
+if (POWER)
+  set(NO_WARMUP 1)
+  set(HAVE_GAS 1)
+  if (CMAKE_ASM_COMPILER_ID STREQUAL "GNU")
+    set(HAVE_GAS 0)
+  elseif (CMAKE_ASM_COMPILER_ID STREQUAL "Clang")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -fno-integrated-as")
+    set(HAVE_GAS 0)
+  endif ()
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DHAVE_GAS=${HAVE_GAS}")
+endif ()
+
 #if don't use Fortran, it will only compile CBLAS.
 if (ONLY_CBLAS)
   set(NO_LAPACK 1)
@@ -148,16 +163,36 @@ endif ()
 include("${PROJECT_SOURCE_DIR}/cmake/prebuild.cmake")
 if (DEFINED TARGET)
   if (${TARGET} STREQUAL COOPERLAKE AND NOT NO_AVX512)
-#    if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
       execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
-        if (${GCC_VERSION} VERSION_GREATER 10.1 OR ${GCC_VERSION} VERSION_EQUAL 10.1)
+        if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 10.09)
           set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=cooperlake")
         else()
           set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
         endif()
-#    elseif (${CMAKE_C_COMPILER_ID} STREQUAL "CLANG")
-#      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
-#    endif()    
+    elseif (${CMAKE_C_COMPILER_ID} STREQUAL "Clang" OR ${CMAKE_C_COMPILER_ID} STREQUAL "AppleClang")
+         if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 8.99)
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=cooperlake")
+        else()
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
+        endif()
+    endif()    
+  endif()
+  if (${TARGET} STREQUAL SAPPHIRERAPIDS AND NOT NO_AVX512)
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
+      execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+        if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 11.0)
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=sapphirerapids")
+        else()
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
+        endif()
+    elseif (${CMAKE_C_COMPILER_ID} STREQUAL "Clang" OR ${CMAKE_C_COMPILER_ID} STREQUAL "AppleClang")
+         if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 12.0)
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=sapphirerapids")
+        else()
+          set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
+        endif()
+    endif()    
   endif()
   if (${TARGET} STREQUAL SKYLAKEX AND NOT NO_AVX512)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
@@ -182,11 +217,11 @@ if (DEFINED TARGET)
       	  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
 	endif()
   endif()
-  if (DEFINED HAVE_FMA3)
-	if (NOT NO_AVX2)
-    set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
-	endif()
-  endif()
+  #  if (DEFINED HAVE_FMA3)
+  #	if (NOT NO_AVX2)
+  #  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
+  #	endif()
+  #  endif()
     if (DEFINED HAVE_SSE)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
   endif()
@@ -202,6 +237,27 @@ if (DEFINED TARGET)
     if (DEFINED HAVE_SSE4_1)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse4.1")
   endif()
+
+  if (${TARGET} STREQUAL POWER10)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 10.2 OR ${GCC_VERSION} VERSION_EQUAL 10.2)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math")
+    else ()
+      message(FATAL_ERROR "Compiler GCC.${GCC_VERSION} does not support Power10.")
+    endif()
+  endif()
+  if (${TARGET} STREQUAL POWER9)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 5.0 OR ${GCC_VERSION} VERSION_EQUAL 5.0)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math")
+    else ()
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+      message(WARNING "Compiler GCC.${GCC_VERSION} does not support fully Power9.")
+    endif()
+  endif()
+  if (${TARGET} STREQUAL POWER8)
+    set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+  endif()
 endif()
 if (DEFINED BINARY)
   message(STATUS "Compiling a ${BINARY}-bit binary.")
@@ -219,6 +275,11 @@ include("${PROJECT_SOURCE_DIR}/cmake/arch.cmake")
 # C Compiler dependent settings
 include("${PROJECT_SOURCE_DIR}/cmake/cc.cmake")
 
+if (INTERFACE64)
+  set(SUFFIX64 64)
+  set(SUFFIX64_UNDERSCORE _64)
+endif()
+
 if (NOT NOFORTRAN)
   # Fortran Compiler dependent settings
   include("${PROJECT_SOURCE_DIR}/cmake/fc.cmake")
@@ -233,6 +294,11 @@ if (BINARY64)
   endif ()
 endif ()
 
+if(EMBEDDED)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DOS_EMBEDDED")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -mthumb -mcpu=cortex-m4 -mfloat-abi=hard -mfpu=fpv4-sp-d16")
+endif()
+
 if (NEED_PIC)
   if (${CMAKE_C_COMPILER} STREQUAL "IBM")
     set(CCOMMON_OPT "${CCOMMON_OPT} -qpic=large")
@@ -249,8 +315,15 @@ if (NEED_PIC)
   endif()
 endif ()
 
+if (X86_64 OR ${CORE} STREQUAL POWER10)
+  set(SMALL_MATRIX_OPT TRUE)
+endif ()
+if (SMALL_MATRIX_OPT)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DSMALL_MATRIX_OPT")
+endif ()
+
 if (DYNAMIC_ARCH)
-  if (X86 OR X86_64 OR ARM64 OR PPC)
+  if (X86 OR X86_64 OR ARM64 OR POWER)
     set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
     if (DYNAMIC_OLDER)
       set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
@@ -290,6 +363,10 @@ if (NO_AVX2)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX2")
 endif ()
 
+if (NO_AVX512)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX512")
+endif ()
+
 if (USE_THREAD)
   # USE_SIMPLE_THREADED_LEVEL3 = 1
   # NO_AFFINITY = 1
@@ -449,6 +526,9 @@ endif()
 if (BUILD_COMPLEX16)
 	set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_COMPLEX16")
 endif()
+if (BUILD_BFLOAT16)
+       set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_BFLOAT16")
+endif()
 if(NOT MSVC)
 set(CMAKE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${CCOMMON_OPT}")
 endif()

cmake/system_check.cmake

@@ -20,11 +20,11 @@ endif()
 
 
 
-if(CMAKE_COMPILER_IS_GNUCC AND WIN32)
+if(MINGW)
     execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpmachine
-              OUTPUT_VARIABLE OPENBLAS_GCC_TARGET_MACHINE
+              OUTPUT_VARIABLE OPENBLAS_MINGW_TARGET_MACHINE
               OUTPUT_STRIP_TRAILING_WHITESPACE)
-    if(OPENBLAS_GCC_TARGET_MACHINE MATCHES "amd64|x86_64|AMD64")
+    if(OPENBLAS_MINGW_TARGET_MACHINE MATCHES "amd64|x86_64|AMD64")
       set(MINGW64 1)
     endif()
 endif()
@@ -35,9 +35,11 @@ if(CMAKE_CL_64 OR MINGW64)
 elseif(MINGW OR (MSVC AND NOT CMAKE_CROSSCOMPILING))
   set(X86 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc.*|power.*|Power.*")
-  set(PPC 1)
+  set(POWER 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
   set(MIPS64 1)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64.*")
+  set(LOONGARCH64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
   if (NOT BINARY)
     if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
@@ -71,6 +73,8 @@ elseif (${CMAKE_CROSSCOMPILING})
     else ()
        set(X86 1)
     endif()
+   elseif (${TARGET} STREQUAL "P5600" OR ${TARGET} MATCHES "MIPS.*")
+       set(MIPS32 1)
    elseif (${TARGET} STREQUAL "ARMV7")
        set(ARM 1)
    else()
@@ -84,8 +88,12 @@ if (X86_64)
   set(ARCH "x86_64")
 elseif(X86)
   set(ARCH "x86")
-elseif(PPC)
+elseif(POWER)
   set(ARCH "power")
+elseif(MIPS32)
+  set(ARCH "mips")
+elseif(MIPS64)
+  set(ARCH "mips64")
 elseif(ARM)
   set(ARCH "arm")
 elseif(ARM64)
@@ -95,7 +103,7 @@ else()
 endif ()
 
 if (NOT BINARY)
-  if (X86_64 OR ARM64 OR PPC OR MIPS64)
+  if (X86_64 OR ARM64 OR POWER OR MIPS64 OR LOONGARCH64)
     set(BINARY 64)
   else ()
     set(BINARY 32)

cmake/utils.cmake

@@ -15,35 +15,83 @@ endfunction ()
 # Reads a Makefile into CMake vars.
 macro(ParseMakefileVars MAKEFILE_IN)
   message(STATUS "Reading vars from ${MAKEFILE_IN}...")
-        set (IfElse 0)
-        set (ElseSeen 0)
+  set (C_COMPILER ${CMAKE_C_COMPILER_ID})
+  set (IfElse 0)
+  set (ElseSeen 0)
+  set (SkipIfs 0)
+  set (SkipElse 0)
   file(STRINGS ${MAKEFILE_IN} makefile_contents)
   foreach (makefile_line ${makefile_contents})
-#message(STATUS "parsing ${makefile_line}")
+    #message(STATUS "parsing ${makefile_line}")
+    # Skip the entire scope of the else statement given that the if statement that precedes it has the valid condition.
+    # The variable SkipIfs is used to identify which endif statement closes the scope of the else statement.
+    if (${SkipElse} EQUAL 1)
+      #message(STATUS "skipping ${makefile_line}")
+      string(REGEX MATCH "(ifeq|ifneq|ifdef|ifndef) .*$" line_match "${makefile_line}")
+      if (NOT "${line_match}" STREQUAL "")
+        MATH(EXPR SkipIfs "${SkipIfs}+1")
+      endif ()
+      string(REGEX MATCH "endif[ \t]*" line_match "${makefile_line}")
+      if (NOT "${line_match}" STREQUAL "")
+        if (${SkipIfs} EQUAL 0)
+          set (SkipElse 0)
+        else ()
+          MATH(EXPR SkipIfs "${SkipIfs}-1")
+        endif ()
+      endif ()
+      continue ()
+    endif ()
+    # The variable IfElse is greater than 0 if and only if the previously parsed line is an if statement.
     if (${IfElse} GREATER 0)
+      # If the current scope is the one that has to be skipped, the if/endif/else statements
+      # along with it till the endif that closes the current scope have to be ignored as well.
+      string(REGEX MATCH "(ifeq|ifneq|ifdef|ifndef) .*$" line_match "${makefile_line}")
+      if (NOT "${line_match}" STREQUAL "")
+        if ((${IfElse} EQUAL 2 AND ${ElseSeen} EQUAL 0) OR (${IfElse} EQUAL 1 AND ${ElseSeen} EQUAL 1))
+          #message(STATUS "skipping ${makefile_line}")
+          MATH(EXPR SkipIfs "${SkipIfs}+1")
+          continue ()
+        endif ()
+      endif ()
       string(REGEX MATCH "endif[ \t]*" line_match "${makefile_line}")
       if (NOT "${line_match}" STREQUAL "")
-#           message(STATUS "ENDIF ${makefile_line}")
-        set (IfElse 0)
-        set (ElseSeen 0)
+        if (${SkipIfs} EQUAL 0)
+          #message(STATUS "ENDIF ${makefile_line}")
+          set (IfElse 0)
+          set (ElseSeen 0)
+        else ()
+          #message(STATUS "skipping ${makefile_line}")
+          MATH(EXPR SkipIfs "${SkipIfs}-1")
+        endif ()
         continue ()
       endif ()
       string(REGEX MATCH "else[ \t]*" line_match "${makefile_line}")
-        if (NOT "${line_match}" STREQUAL "") 
-#           message(STATUS "ELSE ${makefile_line}")
-           set (ElseSeen 1)        
-           continue ()  
-        endif()
-      if ( (${IfElse} EQUAL 2 AND ${ElseSeen} EQUAL 0) OR ( ${IfElse} EQUAL 1 AND ${ElseSeen} EQUAL 1))
-#           message(STATUS "skipping ${makefile_line}")
-         continue ()
+      if (NOT "${line_match}" STREQUAL "") 
+        if (${SkipIfs} EQUAL 0)
+          #message(STATUS "ELSE ${makefile_line}")
+          set (ElseSeen 1)        
+        else ()
+          #message(STATUS "skipping ${makefile_line}")
+        endif ()
+        continue ()  
+      endif()
+      # Skip the lines that are not part of the path that has to be taken.
+      if ((${IfElse} EQUAL 2 AND ${ElseSeen} EQUAL 0) OR (${IfElse} EQUAL 1 AND ${ElseSeen} EQUAL 1) OR (${SkipIfs} GREATER 0))
+        #message(STATUS "skipping ${makefile_line}")
+        continue ()
       endif ()    
-    endif ()    
+    endif ()
+    # Skip commented lines (the ones that start with '#')
+    string(REGEX MATCH "[ \t]*\\#.*$" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      #message(STATUS "skipping ${makefile_line}")
+      continue ()
+    endif ()
     string(REGEX MATCH "([0-9_a-zA-Z]+)[ \t]*=[ \t]*(.+)$" line_match "${makefile_line}")
     if (NOT "${line_match}" STREQUAL "")
-#message(STATUS "match on ${line_match}")
+      #message(STATUS "match on ${line_match}")
       set(var_name ${CMAKE_MATCH_1})
-#      set(var_value ${CMAKE_MATCH_2})
+      #set(var_value ${CMAKE_MATCH_2})
       string(STRIP ${CMAKE_MATCH_2} var_value)
       # check for Makefile variables in the string, e.g. $(TSUFFIX)
       string(REGEX MATCHALL "\\$\\(([0-9_a-zA-Z]+)\\)" make_var_matches ${var_value})
@@ -54,36 +102,93 @@ macro(ParseMakefileVars MAKEFILE_IN)
         string(REPLACE "$(${make_var})" "${${make_var}}" var_value ${var_value})
       endforeach ()
       set(${var_name} ${var_value})
-    else ()
-      string(REGEX MATCH "include \\$\\(KERNELDIR\\)/(.+)$" line_match "${makefile_line}")
-      if (NOT "${line_match}" STREQUAL "")
-#message(STATUS "match on include ${line_match}")
-        ParseMakefileVars(${KERNELDIR}/${CMAKE_MATCH_1})
+      continue ()
+    endif ()
+    # Include a new file to be parsed
+    string(REGEX MATCH "include \\$\\(KERNELDIR\\)/(.+)$" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      #message(STATUS "match on include ${line_match}")
+      ParseMakefileVars(${KERNELDIR}/${CMAKE_MATCH_1})
+      continue ()
+    endif ()
+    # The if statement that precedes this else has the path taken
+    # Thus, this else statement has to be skipped.
+    string(REGEX MATCH "else[ \t]*" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "") 
+      #message(STATUS "skipping ${makefile_line}")
+      set (SkipElse 1)
+      continue()
+    endif()
+    # Example 1: ifdef HAVE_MSA
+    # Example 2: ifndef ZNRM2KERNEL
+    string(REGEX MATCH "(ifdef|ifndef) ([0-9_A-Z]+)" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      #message(STATUS "${CMAKE_MATCH_1} first: ${CMAKE_MATCH_2}")
+      set (ElseSeen 0)
+      if (${CMAKE_MATCH_2})
+        if (${CMAKE_MATCH_1} STREQUAL "ifdef")
+          #message (STATUS "condition is true")
+          set (IfElse 1)
+        else ()
+          set (IfElse 2)
+        endif ()
       else ()
-#        message(STATUS "unmatched line ${line_match}")
-        string(REGEX MATCH "ifeq \\(\\$\\(([_A-Z]+)\\),[ \t]*([0-9_A-Z]+)\\)" line_match "${makefile_line}")
-        if (NOT "${line_match}" STREQUAL "")
-#          message(STATUS "IFEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_2}")
-          if (DEFINED ${${CMAKE_MATCH_1}} AND ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2})
-#            message (STATUS "condition is true")
-            set (IfElse 1)
-          else ()
-            set (IfElse 2)
-          endif ()
+        if (${CMAKE_MATCH_1} STREQUAL "ifdef")
+          set (IfElse 2)
         else ()
-          string(REGEX MATCH "ifneq \\(\\$\\(([_A-Z]+)\\),[ \t]*([0-9_A-Z]+)\\)" line_match "${makefile_line}")
-          if (NOT "${line_match}" STREQUAL "")
-#            message(STATUS "IFNEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_2}")
-            if (NOT ( ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2}))
-#              message (STATUS "condition is true")
-              set (IfElse 1)
-            else ()
-              set (IfElse 2)
-            endif ()
-          endif ()
+          #message (STATUS "condition is true")
+          set (IfElse 1)
+        endif ()
+      endif ()
+      continue ()
+    endif ()
+    # Example 1: ifeq ($(SGEMM_UNROLL_M), 16)
+    # Example 2: ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 8x8)
+    # Example 3: ifeq ($(__BYTE_ORDER__)$(ELF_VERSION),__ORDER_BIG_ENDIAN__2)
+    # Ignore the second group since (?:...) does not work on cmake
+    string(REGEX MATCH "ifeq \\(\\$\\(([0-9_A-Z]+)\\)(([0-9_A-Za-z]*)\\$\\(([0-9_A-Z]+)\\))?,[ \t]*([0-9_A-Za-z]+)\\)" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      #message(STATUS "IFEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_3} third: ${CMAKE_MATCH_4} fourth: ${CMAKE_MATCH_5}")
+      if (DEFINED ${CMAKE_MATCH_1})
+        if (DEFINED ${CMAKE_MATCH_4})
+          set (STR ${${CMAKE_MATCH_1}}${CMAKE_MATCH_3}${${CMAKE_MATCH_4}})
+        else ()
+          set (STR ${${CMAKE_MATCH_1}})
+        endif ()
+        if (${STR} STREQUAL ${CMAKE_MATCH_5})
+          #message (STATUS "condition is true")
+          set (IfElse 1)
+          continue ()
         endif ()
       endif ()
+      set (IfElse 2)
+      continue ()
     endif ()
+    # Example 1 (Group 3): ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+    # Example 2 (Group 4): ifneq ($(C_COMPILER), PGI)
+    string(REGEX MATCH "ifneq \\(\\$\\(([0-9_A-Z]+)\\),[ \t]*(\\$\\(([0-9_A-Z]+)\\)|([0-9_A-Z]+))\\)" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      #message(STATUS "IFNEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_3} third: ${CMAKE_MATCH_4}")
+      set (ElseSeen 0)
+      set (HasValidGroup 0)
+      if (DEFINED ${CMAKE_MATCH_3})
+        set (HasValidGroup 1)
+        set (STR ${${CMAKE_MATCH_3}})
+      elseif (NOT ${CMAKE_MATCH_4} STREQUAL "")
+        set (HasValidGroup 1)
+        set (STR ${CMAKE_MATCH_4})
+      endif ()
+      if (DEFINED ${CMAKE_MATCH_1} AND ${HasValidGroup} EQUAL 1)
+        if (NOT (${${CMAKE_MATCH_1}} STREQUAL ${STR}))
+          #message (STATUS "condition is true")
+          set (IfElse 1)
+          continue ()
+        endif ()
+      endif ()
+      set (IfElse 2)
+      continue ()
+    endif ()
+    #message(STATUS "unmatched line ${line_match}")
   endforeach ()
 endmacro ()
 
@@ -154,31 +259,31 @@ endfunction ()
 #                               STRING - compiles only the given type (e.g. DOUBLE)
 function(GenerateNamedObjects sources_in)
 
-  if (DEFINED ARGV1)
+  if (${ARGC} GREATER 1)
     set(defines_in ${ARGV1})
   endif ()
 
-  if (DEFINED ARGV2 AND NOT "${ARGV2}" STREQUAL "")
+  if (${ARGC} GREATER 2 AND NOT "${ARGV2}" STREQUAL "")
     set(name_in ${ARGV2})
     # strip off extension for kernel files that pass in the object name.
     get_filename_component(name_in ${name_in} NAME_WE)
   endif ()
 
-  if (DEFINED ARGV3)
+  if (${ARGC} GREATER 3)
     set(use_cblas ${ARGV3})
   else ()
     set(use_cblas false)
   endif ()
 
-  if (DEFINED ARGV4)
+  if (${ARGC} GREATER 4)
     set(replace_last_with ${ARGV4})
   endif ()
 
-  if (DEFINED ARGV5)
+  if (${ARGC} GREATER 5)
     set(append_with ${ARGV5})
   endif ()
 
-  if (DEFINED ARGV6)
+  if (${ARGC} GREATER 6)
     set(no_float_type ${ARGV6})
   else ()
     set(no_float_type false)
@@ -193,7 +298,7 @@ function(GenerateNamedObjects sources_in)
   set(real_only false)
   set(complex_only false)
   set(mangle_complex_sources false)
-  if (DEFINED ARGV7 AND NOT "${ARGV7}" STREQUAL "")
+  if (${ARGC} GREATER 7 AND NOT "${ARGV7}" STREQUAL "")
     if (${ARGV7} EQUAL 1)
       set(real_only true)
     elseif (${ARGV7} EQUAL 2)
@@ -251,6 +356,19 @@ function(GenerateNamedObjects sources_in)
       # now add the object and set the defines
       set(obj_defines ${defines_in})
 
+      list(FIND obj_defines "RC" def_idx)
+      if (${def_idx} GREATER -1) 
+	      #	      list(REMOVE_AT ${obj_defines} ${def_idx})
+	      list (REMOVE_ITEM obj_defines "RC")
+	      list(APPEND obj_defines  "RC=RC")
+      endif ()
+      list(FIND obj_defines "CR" def_idx)
+      if (${def_idx} GREATER -1) 
+	      #	      list(REMOVE_AT ${obj_defines} ${def_idx})
+	      list (REMOVE_ITEM obj_defines "CR")
+	      list(APPEND obj_defines  "CR=CR")
+      endif ()
+
       if (use_cblas)
         set(obj_name "cblas_${obj_name}")
         list(APPEND obj_defines "CBLAS")
@@ -295,7 +413,15 @@ function(GenerateNamedObjects sources_in)
       configure_file(${new_source_file}.tmp ${new_source_file} COPYONLY)
       file(REMOVE ${new_source_file}.tmp)
       list(APPEND SRC_LIST_OUT ${new_source_file})
-
+      message (STATUS ${new_source_file})
+      if (DEFINED HAVE_FMA3)
+        if ( ${new_source_file} MATCHES "(s|d?)rot_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+        if ( ${new_source_file} MATCHES "dgemv_t_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+      endif ()
     endforeach ()
   endforeach ()
 
@@ -318,17 +444,17 @@ endfunction ()
 function(GenerateCombinationObjects sources_in defines_in absent_codes_in all_defines_in replace_scheme)
 
   set(alternate_name_in "")
-  if (DEFINED ARGV5)
+  if (${ARGC} GREATER 5)
     set(alternate_name_in ${ARGV5})
   endif ()
 
   set(no_float_type false)
-  if (DEFINED ARGV6)
+  if (${ARGC} GREATER 6)
     set(no_float_type ${ARGV6})
   endif ()
 
   set(complex_filename_scheme "")
-  if (DEFINED ARGV7)
+  if (${ARGC} GREATER 7)
     set(complex_filename_scheme ${ARGV7})
   endif ()
 

common.h

@@ -122,7 +122,7 @@ extern "C" {
 #define ATOM GOTO_ATOM
 #undef  GOTO_ATOM
 #endif
-#else
+#elif !defined(OS_EMBEDDED)
 #include <sys/mman.h>
 #ifndef NO_SYSV_IPC
 #include <sys/shm.h>
@@ -134,6 +134,9 @@ extern "C" {
 #if defined(SMP) || defined(USE_LOCKING)
 #include <pthread.h>
 #endif
+#else
+#include <time.h>
+#include <math.h>
 #endif
 
 #if defined(OS_SUNOS)
@@ -413,6 +416,15 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_alpha.h"
 #endif
 
+#if (defined(ARCH_X86) || defined(ARCH_X86_64)) && defined(__CET__) && defined(__has_include)
+#if __has_include(<cet.h>)
+#include <cet.h>
+#endif
+#endif
+#ifndef _CET_ENDBR
+#define _CET_ENDBR
+#endif
+
 #ifdef ARCH_X86
 #include "common_x86.h"
 #endif
@@ -437,7 +449,7 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_mips.h"
 #endif
 
-    
+
 #ifdef ARCH_RISCV64
 #include "common_riscv64.h"
 #endif
@@ -458,6 +470,14 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_zarch.h"
 #endif
 
+#ifdef ARCH_LOONGARCH64
+#include "common_loongarch64.h"
+#endif
+
+#ifdef ARCH_E2K
+#include "common_e2k.h"
+#endif
+
 #ifndef ASSEMBLER
 #ifdef OS_WINDOWSSTORE
 typedef char env_var_t[MAX_PATH];
@@ -488,10 +508,12 @@ static inline unsigned long long rpcc(void){
   struct timespec ts;
   clock_gettime(CLOCK_MONOTONIC, &ts);
   return (unsigned long long)ts.tv_sec * 1000000000ull + ts.tv_nsec;
-#else
+#elif !defined(OS_EMBEDDED)
   struct timeval tv;
   gettimeofday(&tv,NULL);
   return (unsigned long long)tv.tv_sec * 1000000000ull + tv.tv_usec * 1000;
+#else
+  return 0;
 #endif
 }
 #define RPCC_DEFINED
@@ -521,6 +543,10 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #include "common_linux.h"
 #endif
 
+#ifdef OS_EMBEDDED
+#define DTB_DEFAULT_ENTRIES 64
+#endif
+
 #define MMAP_ACCESS (PROT_READ | PROT_WRITE)
 
 #ifdef __NetBSD__

common_arm64.h

@@ -39,7 +39,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define INLINE inline
 
-#ifdef F_INTERFACE_FLANG
+#if defined( F_INTERFACE_FLANG) || defined(F_INTERFACE_PGI)
 #define RETURN_BY_STACK
 #else
 #define RETURN_BY_COMPLEX
@@ -120,7 +120,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 	.text ;
 	.p2align 2 ;
 	.global	REALNAME ;
-#ifndef __APPLE__
+#if !defined(__APPLE__) && !defined(_WIN32)
 	.type	REALNAME, %function ;
 #endif
 REALNAME:

common_c.h

@@ -232,6 +232,8 @@
 
 #define CGEADD_K                cgeadd_k 
 
+#define CGEMM_SMALL_MATRIX_PERMIT	cgemm_small_matrix_permit
+
 #else
 
 #define	CAMAX_K			gotoblas -> camax_k
@@ -426,8 +428,51 @@
 
 #define CGEADD_K                gotoblas -> cgeadd_k 
 
+#define CGEMM_SMALL_MATRIX_PERMIT	gotoblas -> cgemm_small_matrix_permit
+
 #endif
 
+#define CGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(cgemm_small_kernel_nn)
+#define CGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(cgemm_small_kernel_nt)
+#define CGEMM_SMALL_KERNEL_NR		FUNC_OFFSET(cgemm_small_kernel_nr)
+#define CGEMM_SMALL_KERNEL_NC		FUNC_OFFSET(cgemm_small_kernel_nc)
+
+#define CGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(cgemm_small_kernel_tn)
+#define CGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(cgemm_small_kernel_tt)
+#define CGEMM_SMALL_KERNEL_TR		FUNC_OFFSET(cgemm_small_kernel_tr)
+#define CGEMM_SMALL_KERNEL_TC		FUNC_OFFSET(cgemm_small_kernel_tc)
+
+#define CGEMM_SMALL_KERNEL_RN		FUNC_OFFSET(cgemm_small_kernel_rn)
+#define CGEMM_SMALL_KERNEL_RT		FUNC_OFFSET(cgemm_small_kernel_rt)
+#define CGEMM_SMALL_KERNEL_RR		FUNC_OFFSET(cgemm_small_kernel_rr)
+#define CGEMM_SMALL_KERNEL_RC		FUNC_OFFSET(cgemm_small_kernel_rc)
+
+#define CGEMM_SMALL_KERNEL_CN		FUNC_OFFSET(cgemm_small_kernel_cn)
+#define CGEMM_SMALL_KERNEL_CT		FUNC_OFFSET(cgemm_small_kernel_ct)
+#define CGEMM_SMALL_KERNEL_CR		FUNC_OFFSET(cgemm_small_kernel_cr)
+#define CGEMM_SMALL_KERNEL_CC		FUNC_OFFSET(cgemm_small_kernel_cc)
+
+#define CGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(cgemm_small_kernel_b0_nn)
+#define CGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(cgemm_small_kernel_b0_nt)
+#define CGEMM_SMALL_KERNEL_B0_NR	FUNC_OFFSET(cgemm_small_kernel_b0_nr)
+#define CGEMM_SMALL_KERNEL_B0_NC	FUNC_OFFSET(cgemm_small_kernel_b0_nc)
+
+#define CGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(cgemm_small_kernel_b0_tn)
+#define CGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(cgemm_small_kernel_b0_tt)
+#define CGEMM_SMALL_KERNEL_B0_TR	FUNC_OFFSET(cgemm_small_kernel_b0_tr)
+#define CGEMM_SMALL_KERNEL_B0_TC	FUNC_OFFSET(cgemm_small_kernel_b0_tc)
+
+#define CGEMM_SMALL_KERNEL_B0_RN	FUNC_OFFSET(cgemm_small_kernel_b0_rn)
+#define CGEMM_SMALL_KERNEL_B0_RT	FUNC_OFFSET(cgemm_small_kernel_b0_rt)
+#define CGEMM_SMALL_KERNEL_B0_RR	FUNC_OFFSET(cgemm_small_kernel_b0_rr)
+#define CGEMM_SMALL_KERNEL_B0_RC	FUNC_OFFSET(cgemm_small_kernel_b0_rc)
+
+#define CGEMM_SMALL_KERNEL_B0_CN	FUNC_OFFSET(cgemm_small_kernel_b0_cn)
+#define CGEMM_SMALL_KERNEL_B0_CT	FUNC_OFFSET(cgemm_small_kernel_b0_ct)
+#define CGEMM_SMALL_KERNEL_B0_CR	FUNC_OFFSET(cgemm_small_kernel_b0_cr)
+#define CGEMM_SMALL_KERNEL_B0_CC	FUNC_OFFSET(cgemm_small_kernel_b0_cc)
+
+
 #define	CGEMM_NN		cgemm_nn
 #define	CGEMM_CN		cgemm_cn
 #define	CGEMM_TN		cgemm_tn

common_d.h

@@ -157,6 +157,8 @@
 #define DIMATCOPY_K_RT      dimatcopy_k_rt
 #define DGEADD_K                dgeadd_k 
 
+#define DGEMM_SMALL_MATRIX_PERMIT	dgemm_small_matrix_permit
+
 #else
 
 #define	DAMAX_K			gotoblas -> damax_k
@@ -281,8 +283,21 @@
 
 #define DGEADD_K                gotoblas -> dgeadd_k 
 
+#define DGEMM_SMALL_MATRIX_PERMIT	gotoblas -> dgemm_small_matrix_permit
+
 #endif
 
+#define DGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(dgemm_small_kernel_nn)
+#define DGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(dgemm_small_kernel_nt)
+#define DGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(dgemm_small_kernel_tn)
+#define DGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(dgemm_small_kernel_tt)
+
+#define DGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(dgemm_small_kernel_b0_nn)
+#define DGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(dgemm_small_kernel_b0_nt)
+#define DGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(dgemm_small_kernel_b0_tn)
+#define DGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(dgemm_small_kernel_b0_tt)
+
+
 #define	DGEMM_NN		dgemm_nn
 #define	DGEMM_CN		dgemm_tn
 #define	DGEMM_TN		dgemm_tn

common_e2k.h

@@ -0,0 +1,64 @@
+/*****************************************************************************
+Copyright (c) 2011-2016, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of 
+      its contributors may be used to endorse or promote products 
+      derived from this software without specific prior written 
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+
+#ifndef COMMON_E2K
+#define COMMON_E2K
+
+#ifdef ASSEMBLER
+#error
+#endif
+
+#define MB do { __asm__ __volatile__("": : :"memory"); } while (0)
+#define WMB do { __asm__ __volatile__("": : :"memory"); } while (0)
+#define RMB
+
+#define INLINE __attribute__((__always_inline__)) inline
+
+static inline int blas_quickdivide(blasint x, blasint y) {
+  return x / y;
+}
+
+#ifndef PAGESIZE
+#define PAGESIZE	( 4 << 10)
+#endif
+#define HUGE_PAGESIZE	( 2 << 20)
+
+#ifndef BUFFERSIZE
+#define BUFFER_SIZE	(32 << 20)
+#else
+#define BUFFER_SIZE	(32 << BUFFERSIZE)
+#endif
+
+#define SEEK_ADDRESS
+
+#endif
+

common_interface.h

@@ -709,6 +709,13 @@ int BLASFUNC(cpotrf)(char *, blasint *, float  *, blasint *, blasint *);
 int BLASFUNC(zpotrf)(char *, blasint *, double *, blasint *, blasint *);
 int BLASFUNC(xpotrf)(char *, blasint *, xdouble *, blasint *, blasint *);
 
+int BLASFUNC(spotri)(char *, blasint *, float  *, blasint *, blasint *);
+int BLASFUNC(dpotri)(char *, blasint *, double *, blasint *, blasint *);
+int BLASFUNC(qpotri)(char *, blasint *, xdouble *, blasint *, blasint *);
+int BLASFUNC(cpotri)(char *, blasint *, float  *, blasint *, blasint *);
+int BLASFUNC(zpotri)(char *, blasint *, double *, blasint *, blasint *);
+int BLASFUNC(xpotri)(char *, blasint *, xdouble *, blasint *, blasint *);
+
 int BLASFUNC(spotrs)(char *, blasint *, blasint *, float   *, blasint *, float   *, blasint *, blasint *);
 int BLASFUNC(dpotrs)(char *, blasint *, blasint *, double  *, blasint *, double  *, blasint *, blasint *);
 int BLASFUNC(qpotrs)(char *, blasint *, blasint *, xdouble *, blasint *, xdouble *, blasint *, blasint *);

common_level3.h

@@ -515,6 +515,129 @@ int qgemm_kernel(BLASLONG, BLASLONG, BLASLONG, xidouble *, xidouble *, xidouble
 int qgemm_kernel(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
 #endif
 
+#ifdef SMALL_MATRIX_OPT
+int sbgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+int sbgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+int sgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+int sgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+int dgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha, double beta);
+
+int dgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+int dgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+int dgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+int dgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+
+int sbgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+
+int sgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+
+int dgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+int dgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+int dgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+int dgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+
+int cgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha0, float alpha1, float beta0, float beta1);
+
+int cgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_nr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_nc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+	
+int cgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_tr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_tc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_rn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_rt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_rr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_rc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_cn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_ct(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_cr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_cc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+int zgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha0, double alpha1, double beta0, double beta1);
+
+int zgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_nr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_nc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+	
+int zgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_tr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_tc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_rn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_rt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_rr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_rc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_cn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_ct(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_cr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_cc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+int cgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);	
+int cgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_nr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_nc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+	
+int cgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_tr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_tc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_b0_rn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_rt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_rr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_rc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_b0_cn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_ct(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_cr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_cc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+int zgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);	
+int zgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_nr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_nc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+	
+int zgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_tr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_tc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_b0_rn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_rt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_rr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_rc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_b0_cn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_ct(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_cr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_cc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+#endif
+
 int cgemm_kernel_n(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float  *, float  *, BLASLONG);
 int cgemm_kernel_l(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float  *, float  *, BLASLONG);
 int cgemm_kernel_r(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float  *, float  *, BLASLONG);

common_linux.h

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

common_loongarch64.h

@@ -0,0 +1,199 @@
+/*****************************************************************************
+Copyright (c) 2011-2020, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of
+      its contributors may be used to endorse or promote products
+      derived from this software without specific prior written
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#ifndef COMMON_LOONGARCH64
+#define COMMON_LOONGARCH64
+
+#define MB  __sync_synchronize()
+#define WMB __sync_synchronize()
+#define RMB __sync_synchronize()
+
+#define INLINE inline
+
+#ifndef ASSEMBLER
+
+static inline int blas_quickdivide(blasint x, blasint y){
+  return x / y;
+}
+
+#ifdef DOUBLE
+#define GET_IMAGE(res)  __asm__ __volatile__("fmov.d %0, $f2" : "=f"(res)  : : "memory")
+#else
+#define GET_IMAGE(res)  __asm__ __volatile__("fmov.s %0, $f2" : "=f"(res)  : : "memory")
+#endif
+
+#define GET_IMAGE_CANCEL
+
+#else
+
+#ifdef DOUBLE
+#define LD      fld.d
+#define ST      fst.d
+#define MADD    fmadd.d
+#define NMADD   fnmadd.d
+#define MSUB    fmsub.d
+#define NMSUB   fnmsub.d
+#define ADD     fadd.d
+#define SUB     fsub.d
+#define MUL     fmul.d
+#define MOV     fmov.d
+#define CMOVT   fsel
+#define MTC     movgr2fr.d
+#define FABS    fabs.d
+#define CMPEQ   fcmp.ceq.d
+#define CMPLE   fcmp.cle.d
+#define CMPLT   fcmp.clt.d
+#define NEG     fneg.d
+#else
+#define LD      fld.s
+#define ST      fst.s
+#define MADD    fmadd.s
+#define NMADD   fnmadd.s
+#define MSUB    fmsub.s
+#define NMSUB   fnmsub.s
+#define ADD     fadd.s
+#define SUB     fsub.s
+#define MUL     fmul.s
+#define MOV     fmov.s
+#define CMOVT   fsel
+#define MTC     movgr2fr.w
+#define FABS    fabs.s
+#define CMPEQ   fcmp.ceq.s
+#define CMPLE   fcmp.cle.s
+#define CMPLT   fcmp.clt.s
+#define NEG     fneg.s
+#endif /* defined(DOUBLE) */
+
+#if defined(__64BIT__) && defined(USE64BITINT)
+#define LDINT   ld.d
+#define LDARG   ld.d
+#define SDARG   st.d
+#elif defined(__64BIT__) && !defined(USE64BITINT)
+#define LDINT   ld.w
+#define LDARG   ld.d
+#define SDARG   st.d
+#else
+#define LDINT   ld.w
+#define LDARG   ld.w
+#define SDARG   st.w
+#endif
+
+
+#ifndef F_INTERFACE
+#define REALNAME ASMNAME
+#else
+#define REALNAME ASMFNAME
+#endif /* defined(F_INTERFACE) */
+
+#if defined(ASSEMBLER) && !defined(NEEDPARAM)
+
+#define PROLOGUE \
+    .text ;\
+    .align 5 ;\
+    .globl  REALNAME ;\
+    .type   REALNAME, @function ;\
+REALNAME: ;\
+
+#if defined(__linux__) && defined(__ELF__)
+#define GNUSTACK .section .note.GNU-stack,"",@progbits
+#else
+#define GNUSTACK
+#endif /* defined(__linux__) && defined(__ELF__) */
+
+#define EPILOGUE      \
+    .end    REALNAME ;\
+    GNUSTACK
+
+#define PROFCODE
+
+#define MOVT(dst, src, cc)  \
+    bceqz  cc,   1f;        \
+    add.d dst,  src,  $r0;  \
+    1:
+
+#endif /* defined(ASSEMBLER) && !defined(NEEDPARAM) */
+
+#endif /* defined(ASSEMBLER) */
+
+#define SEEK_ADDRESS
+
+#define BUFFER_SIZE     ( 32 << 20)
+
+#define PAGESIZE        (16UL << 10)
+#define FIXED_PAGESIZE  (16UL << 10)
+#define HUGE_PAGESIZE   ( 2 << 20)
+
+#define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
+
+#ifndef MAP_ANONYMOUS
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+
+#endif

common_macro.h

@@ -644,6 +644,17 @@
 
 #define GEADD_K                 DGEADD_K
 
+#define GEMM_SMALL_MATRIX_PERMIT	DGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    DGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    DGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_TN    DGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    DGEMM_SMALL_KERNEL_TT
+#define GEMM_SMALL_KERNEL_B0_NN    DGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    DGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_TN    DGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    DGEMM_SMALL_KERNEL_B0_TT
+
 #elif defined(BFLOAT16)
 
 #define D_TO_BF16_K     SBDTOBF16_K
@@ -931,6 +942,18 @@
 
 #define GEADD_K 		SGEADD_K
 
+#define GEMM_SMALL_MATRIX_PERMIT	SBGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    SBGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    SBGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_TN    SBGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    SBGEMM_SMALL_KERNEL_TT
+
+#define GEMM_SMALL_KERNEL_B0_NN    SBGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    SBGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_TN    SBGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    SBGEMM_SMALL_KERNEL_B0_TT
+
 #endif
 
 #else
@@ -1236,6 +1259,19 @@
 #define IMATCOPY_K_RT		SIMATCOPY_K_RT
 
 #define GEADD_K 		SGEADD_K
+
+#define GEMM_SMALL_MATRIX_PERMIT	SGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    SGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    SGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_TN    SGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    SGEMM_SMALL_KERNEL_TT
+
+#define GEMM_SMALL_KERNEL_B0_NN    SGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    SGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_TN    SGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    SGEMM_SMALL_KERNEL_B0_TT
+
 #endif
 #else
 #ifdef XDOUBLE
@@ -2063,6 +2099,48 @@
 
 #define GEADD_K                 ZGEADD_K
 
+#define GEMM_SMALL_MATRIX_PERMIT	ZGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    ZGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    ZGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_NR    ZGEMM_SMALL_KERNEL_NR
+#define GEMM_SMALL_KERNEL_NC    ZGEMM_SMALL_KERNEL_NC
+
+#define GEMM_SMALL_KERNEL_TN    ZGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    ZGEMM_SMALL_KERNEL_TT
+#define GEMM_SMALL_KERNEL_TR    ZGEMM_SMALL_KERNEL_TR
+#define GEMM_SMALL_KERNEL_TC    ZGEMM_SMALL_KERNEL_TC
+
+#define GEMM_SMALL_KERNEL_RN    ZGEMM_SMALL_KERNEL_RN
+#define GEMM_SMALL_KERNEL_RT    ZGEMM_SMALL_KERNEL_RT
+#define GEMM_SMALL_KERNEL_RR    ZGEMM_SMALL_KERNEL_RR
+#define GEMM_SMALL_KERNEL_RC    ZGEMM_SMALL_KERNEL_RC
+
+#define GEMM_SMALL_KERNEL_CN    ZGEMM_SMALL_KERNEL_CN
+#define GEMM_SMALL_KERNEL_CT    ZGEMM_SMALL_KERNEL_CT
+#define GEMM_SMALL_KERNEL_CR    ZGEMM_SMALL_KERNEL_CR
+#define GEMM_SMALL_KERNEL_CC    ZGEMM_SMALL_KERNEL_CC
+
+#define GEMM_SMALL_KERNEL_B0_NN    ZGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    ZGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_NR    ZGEMM_SMALL_KERNEL_B0_NR
+#define GEMM_SMALL_KERNEL_B0_NC    ZGEMM_SMALL_KERNEL_B0_NC
+
+#define GEMM_SMALL_KERNEL_B0_TN    ZGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    ZGEMM_SMALL_KERNEL_B0_TT
+#define GEMM_SMALL_KERNEL_B0_TR    ZGEMM_SMALL_KERNEL_B0_TR
+#define GEMM_SMALL_KERNEL_B0_TC    ZGEMM_SMALL_KERNEL_B0_TC
+
+#define GEMM_SMALL_KERNEL_B0_RN    ZGEMM_SMALL_KERNEL_B0_RN
+#define GEMM_SMALL_KERNEL_B0_RT    ZGEMM_SMALL_KERNEL_B0_RT
+#define GEMM_SMALL_KERNEL_B0_RR    ZGEMM_SMALL_KERNEL_B0_RR
+#define GEMM_SMALL_KERNEL_B0_RC    ZGEMM_SMALL_KERNEL_B0_RC
+
+#define GEMM_SMALL_KERNEL_B0_CN    ZGEMM_SMALL_KERNEL_B0_CN
+#define GEMM_SMALL_KERNEL_B0_CT    ZGEMM_SMALL_KERNEL_B0_CT
+#define GEMM_SMALL_KERNEL_B0_CR    ZGEMM_SMALL_KERNEL_B0_CR
+#define GEMM_SMALL_KERNEL_B0_CC    ZGEMM_SMALL_KERNEL_B0_CC
+
 #else
 
 #define	AMAX_K			CAMAX_K
@@ -2486,11 +2564,54 @@
 
 #define GEADD_K                 CGEADD_K
 
+#define GEMM_SMALL_MATRIX_PERMIT	CGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    CGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    CGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_NR    CGEMM_SMALL_KERNEL_NR
+#define GEMM_SMALL_KERNEL_NC    CGEMM_SMALL_KERNEL_NC
+
+#define GEMM_SMALL_KERNEL_TN    CGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    CGEMM_SMALL_KERNEL_TT
+#define GEMM_SMALL_KERNEL_TR    CGEMM_SMALL_KERNEL_TR
+#define GEMM_SMALL_KERNEL_TC    CGEMM_SMALL_KERNEL_TC
+
+#define GEMM_SMALL_KERNEL_RN    CGEMM_SMALL_KERNEL_RN
+#define GEMM_SMALL_KERNEL_RT    CGEMM_SMALL_KERNEL_RT
+#define GEMM_SMALL_KERNEL_RR    CGEMM_SMALL_KERNEL_RR
+#define GEMM_SMALL_KERNEL_RC    CGEMM_SMALL_KERNEL_RC
+
+#define GEMM_SMALL_KERNEL_CN    CGEMM_SMALL_KERNEL_CN
+#define GEMM_SMALL_KERNEL_CT    CGEMM_SMALL_KERNEL_CT
+#define GEMM_SMALL_KERNEL_CR    CGEMM_SMALL_KERNEL_CR
+#define GEMM_SMALL_KERNEL_CC    CGEMM_SMALL_KERNEL_CC
+
+#define GEMM_SMALL_KERNEL_B0_NN    CGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    CGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_NR    CGEMM_SMALL_KERNEL_B0_NR
+#define GEMM_SMALL_KERNEL_B0_NC    CGEMM_SMALL_KERNEL_B0_NC
+
+#define GEMM_SMALL_KERNEL_B0_TN    CGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    CGEMM_SMALL_KERNEL_B0_TT
+#define GEMM_SMALL_KERNEL_B0_TR    CGEMM_SMALL_KERNEL_B0_TR
+#define GEMM_SMALL_KERNEL_B0_TC    CGEMM_SMALL_KERNEL_B0_TC
+
+#define GEMM_SMALL_KERNEL_B0_RN    CGEMM_SMALL_KERNEL_B0_RN
+#define GEMM_SMALL_KERNEL_B0_RT    CGEMM_SMALL_KERNEL_B0_RT
+#define GEMM_SMALL_KERNEL_B0_RR    CGEMM_SMALL_KERNEL_B0_RR
+#define GEMM_SMALL_KERNEL_B0_RC    CGEMM_SMALL_KERNEL_B0_RC
+
+#define GEMM_SMALL_KERNEL_B0_CN    CGEMM_SMALL_KERNEL_B0_CN
+#define GEMM_SMALL_KERNEL_B0_CT    CGEMM_SMALL_KERNEL_B0_CT
+#define GEMM_SMALL_KERNEL_B0_CR    CGEMM_SMALL_KERNEL_B0_CR
+#define GEMM_SMALL_KERNEL_B0_CC    CGEMM_SMALL_KERNEL_B0_CC
+
 #endif
 #endif
 
 #ifndef ASSEMBLER
-#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)
+#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)\
+|| defined(ARCH_LOONGARCH64) || defined(ARCH_E2K)
 extern BLASLONG gemm_offset_a;
 extern BLASLONG gemm_offset_b;
 extern BLASLONG sbgemm_p;

common_mips64.h

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

common_param.h

@@ -145,6 +145,19 @@ BLASLONG (*isbmin_k) (BLASLONG, float *, BLASLONG);
   int	 (*sbneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sblaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
 
+#ifdef SMALL_MATRIX_OPT
+  int    (*sbgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+  int    (*sbgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+  int    (*sbgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+#endif
 #endif
 
 #if defined(BUILD_SINGLE) || defined(BUILD_COMPLEX)
@@ -207,6 +220,20 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int    (*sgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
 #endif
 #ifdef BUILD_SINGLE  
+#ifdef SMALL_MATRIX_OPT
+  int    (*sgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+  int    (*sgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+  int    (*sgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+#endif
+
   int    (*strsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*strsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*strsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
@@ -314,6 +341,19 @@ BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
   int    (*dgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
 #endif
 #ifdef BUILD_DOUBLE
+#ifdef SMALL_MATRIX_OPT
+  int    (*dgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha, double beta);
+
+  int    (*dgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+
+  int    (*dgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+#endif
   int    (*dtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
   int    (*dtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
   int    (*dtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
@@ -513,6 +553,50 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   int    (*cgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*cgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
 
+#ifdef SMALL_MATRIX_OPT
+  int    (*cgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha0, float alpha1, float beta0, float beta1);
+
+  int    (*cgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_nr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_nc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_tr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_tc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_rn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_rt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_rr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_rc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_cn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_ct    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_cr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_cc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_nr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_nc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_tr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_tc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_rn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_rt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_rr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_rc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_cn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_ct )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_cr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_cc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+#endif
+
   int    (*ctrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*ctrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*ctrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
@@ -679,6 +763,50 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
   int    (*zgemm_oncopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
   int    (*zgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
 
+#ifdef SMALL_MATRIX_OPT
+  int    (*zgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha0, double alpha1, double beta0, double beta1);
+
+  int    (*zgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_nr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_nc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_tr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_tc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_rn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_rt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_rr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_rc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_cn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_ct    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_cr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_cc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_nr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_nc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_tr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_tc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_rn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_rt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_rr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_rc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_cn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_ct )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_cr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_cc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+#endif
+
   int    (*ztrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
   int    (*ztrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
   int    (*ztrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
@@ -1069,6 +1197,8 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
 
 extern gotoblas_t *gotoblas;
 
+#define FUNC_OFFSET(func)	(size_t)(&((gotoblas_t *)NULL)->func)
+
 #define DTB_ENTRIES  gotoblas -> dtb_entries
 #define GEMM_OFFSET_A	gotoblas -> offsetA
 #define GEMM_OFFSET_B	gotoblas -> offsetB
@@ -1174,6 +1304,8 @@ extern gotoblas_t *gotoblas;
 
 #else
 
+#define FUNC_OFFSET(func)	(size_t)(func)
+
 #define DTB_ENTRIES  DTB_DEFAULT_ENTRIES
 
 #define GEMM_OFFSET_A	GEMM_DEFAULT_OFFSET_A

common_s.h

@@ -164,6 +164,8 @@
 
 #define SGEADD_K                sgeadd_k 
 
+#define SGEMM_SMALL_MATRIX_PERMIT	sgemm_small_matrix_permit
+
 #else
 
 #define	SAMAX_K			gotoblas -> samax_k
@@ -299,8 +301,21 @@
 
 #define SGEADD_K                gotoblas -> sgeadd_k 
 
+#define SGEMM_SMALL_MATRIX_PERMIT	gotoblas -> sgemm_small_matrix_permit
+
 #endif
 
+#define SGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(sgemm_small_kernel_nn)
+#define SGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(sgemm_small_kernel_nt)
+#define SGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(sgemm_small_kernel_tn)
+#define SGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(sgemm_small_kernel_tt)
+
+#define SGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(sgemm_small_kernel_b0_nn)
+#define SGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(sgemm_small_kernel_b0_nt)
+#define SGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(sgemm_small_kernel_b0_tn)
+#define SGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(sgemm_small_kernel_b0_tt)
+
+
 #define	SGEMM_NN		sgemm_nn
 #define	SGEMM_CN		sgemm_tn
 #define	SGEMM_TN		sgemm_tn

common_sb.h

@@ -24,6 +24,7 @@
 #define	SBGEMM_BETA		sbgemm_beta
 #define SBGEMM_KERNEL            sbgemm_kernel
 
+#define SBGEMM_SMALL_MATRIX_PERMIT	sbgemm_small_matrix_permit
 #else
 
 #define SBDOT_K             gotoblas -> sbdot_k
@@ -41,8 +42,19 @@
 #define	SBGEMM_BETA		gotoblas -> sbgemm_beta
 #define	SBGEMM_KERNEL		gotoblas -> sbgemm_kernel
 
+#define SBGEMM_SMALL_MATRIX_PERMIT	gotoblas -> sbgemm_small_matrix_permit
 #endif
 
+#define SBGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(sbgemm_small_kernel_nn)
+#define SBGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(sbgemm_small_kernel_nt)
+#define SBGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(sbgemm_small_kernel_tn)
+#define SBGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(sbgemm_small_kernel_tt)
+
+#define SBGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(sbgemm_small_kernel_b0_nn)
+#define SBGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(sbgemm_small_kernel_b0_nt)
+#define SBGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(sbgemm_small_kernel_b0_tn)
+#define SBGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(sbgemm_small_kernel_b0_tt)
+
 #define	SBGEMM_NN		sbgemm_nn
 #define	SBGEMM_CN		sbgemm_tn
 #define	SBGEMM_TN		sbgemm_tn

common_x86.h

@@ -340,7 +340,8 @@ REALNAME:
 	.align 16; \
 	.globl REALNAME ;\
        .type REALNAME, @function; \
-REALNAME:
+REALNAME: \
+	_CET_ENDBR
 
 #ifdef PROFILE
 #define PROFCODE call mcount

common_x86_64.h

@@ -451,7 +451,8 @@ REALNAME:
 	.align 512; \
 	.globl REALNAME ;\
        .type REALNAME, @function; \
-REALNAME:
+REALNAME: \
+	_CET_ENDBR
 
 #ifdef PROFILE
 #define PROFCODE call *mcount@GOTPCREL(%rip)

common_z.h

@@ -232,6 +232,8 @@
 
 #define ZGEADD_K                zgeadd_k 
 
+#define ZGEMM_SMALL_MATRIX_PERMIT	zgemm_small_matrix_permit
+
 #else
 
 #define	ZAMAX_K			gotoblas -> zamax_k
@@ -426,8 +428,51 @@
 
 #define ZGEADD_K                gotoblas -> zgeadd_k
 
+#define ZGEMM_SMALL_MATRIX_PERMIT	gotoblas -> zgemm_small_matrix_permit
+
 #endif
 
+#define ZGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(zgemm_small_kernel_nn)
+#define ZGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(zgemm_small_kernel_nt)
+#define ZGEMM_SMALL_KERNEL_NR		FUNC_OFFSET(zgemm_small_kernel_nr)
+#define ZGEMM_SMALL_KERNEL_NC		FUNC_OFFSET(zgemm_small_kernel_nc)
+
+#define ZGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(zgemm_small_kernel_tn)
+#define ZGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(zgemm_small_kernel_tt)
+#define ZGEMM_SMALL_KERNEL_TR		FUNC_OFFSET(zgemm_small_kernel_tr)
+#define ZGEMM_SMALL_KERNEL_TC		FUNC_OFFSET(zgemm_small_kernel_tc)
+
+#define ZGEMM_SMALL_KERNEL_RN		FUNC_OFFSET(zgemm_small_kernel_rn)
+#define ZGEMM_SMALL_KERNEL_RT		FUNC_OFFSET(zgemm_small_kernel_rt)
+#define ZGEMM_SMALL_KERNEL_RR		FUNC_OFFSET(zgemm_small_kernel_rr)
+#define ZGEMM_SMALL_KERNEL_RC		FUNC_OFFSET(zgemm_small_kernel_rc)
+
+#define ZGEMM_SMALL_KERNEL_CN		FUNC_OFFSET(zgemm_small_kernel_cn)
+#define ZGEMM_SMALL_KERNEL_CT		FUNC_OFFSET(zgemm_small_kernel_ct)
+#define ZGEMM_SMALL_KERNEL_CR		FUNC_OFFSET(zgemm_small_kernel_cr)
+#define ZGEMM_SMALL_KERNEL_CC		FUNC_OFFSET(zgemm_small_kernel_cc)
+
+#define ZGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(zgemm_small_kernel_b0_nn)
+#define ZGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(zgemm_small_kernel_b0_nt)
+#define ZGEMM_SMALL_KERNEL_B0_NR	FUNC_OFFSET(zgemm_small_kernel_b0_nr)
+#define ZGEMM_SMALL_KERNEL_B0_NC	FUNC_OFFSET(zgemm_small_kernel_b0_nc)
+
+#define ZGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(zgemm_small_kernel_b0_tn)
+#define ZGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(zgemm_small_kernel_b0_tt)
+#define ZGEMM_SMALL_KERNEL_B0_TR	FUNC_OFFSET(zgemm_small_kernel_b0_tr)
+#define ZGEMM_SMALL_KERNEL_B0_TC	FUNC_OFFSET(zgemm_small_kernel_b0_tc)
+
+#define ZGEMM_SMALL_KERNEL_B0_RN	FUNC_OFFSET(zgemm_small_kernel_b0_rn)
+#define ZGEMM_SMALL_KERNEL_B0_RT	FUNC_OFFSET(zgemm_small_kernel_b0_rt)
+#define ZGEMM_SMALL_KERNEL_B0_RR	FUNC_OFFSET(zgemm_small_kernel_b0_rr)
+#define ZGEMM_SMALL_KERNEL_B0_RC	FUNC_OFFSET(zgemm_small_kernel_b0_rc)
+
+#define ZGEMM_SMALL_KERNEL_B0_CN	FUNC_OFFSET(zgemm_small_kernel_b0_cn)
+#define ZGEMM_SMALL_KERNEL_B0_CT	FUNC_OFFSET(zgemm_small_kernel_b0_ct)
+#define ZGEMM_SMALL_KERNEL_B0_CR	FUNC_OFFSET(zgemm_small_kernel_b0_cr)
+#define ZGEMM_SMALL_KERNEL_B0_CC	FUNC_OFFSET(zgemm_small_kernel_b0_cc)
+
+
 #define	ZGEMM_NN		zgemm_nn
 #define	ZGEMM_CN		zgemm_cn
 #define	ZGEMM_TN		zgemm_tn

cpp_thread_test/Makefile

@@ -1,13 +1,14 @@
-include ../Makefile.rule
+TOPDIR = ..
+include $(TOPDIR)/Makefile.system
 
 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
+	$(CXX) $(COMMON_OPT) -Wall -Wextra -Wshadow -fopenmp -std=c++11 dgemv_thread_safety.cpp ../$(LIBNAME) $(EXTRALIB) $(FEXTRALIB) -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
+	$(CXX) $(COMMON_OPT) -Wall -Wextra -Wshadow -fopenmp -std=c++11 dgemm_thread_safety.cpp ../$(LIBNAME) $(EXTRALIB) $(FEXTRALIB) -o dgemm_tester
 	./dgemm_tester
 
 clean ::

cpuid.h

@@ -54,6 +54,7 @@
 #define VENDOR_TRANSMETA  9
 #define VENDOR_NSC	 10
 #define VENDOR_HYGON	 11
+#define VENDOR_ZHAOXIN   12
 #define VENDOR_UNKNOWN   99
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
@@ -119,6 +120,7 @@
 #define CORE_SKYLAKEX    28
 #define CORE_DHYANA	 29
 #define CORE_COOPERLAKE  30
+#define CORE_SAPPHIRERAPIDS  31
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -144,6 +146,7 @@
 #define HAVE_AVX512VL (1 << 21)
 #define HAVE_AVX2     (1 << 22)
 #define HAVE_AVX512BF16  (1 << 23)
+#define HAVE_AMXBF16  (1 << 24)
 
 #define CACHE_INFO_L1_I     1
 #define CACHE_INFO_L1_D     2
@@ -221,6 +224,7 @@ typedef struct {
 #define CPUTYPE_SKYLAKEX		52
 #define CPUTYPE_DHYANA			53
 #define CPUTYPE_COOPERLAKE		54
+#define CPUTYPE_SAPPHIRERAPIDS		55
 
 #define CPUTYPE_HYGON_UNKNOWN		99
 

cpuid_arm64.c

@@ -26,20 +26,25 @@
   *****************************************************************************/
 
 #include <string.h>
-#ifdef OS_DARWIN
+#ifdef __APPLE__
 #include <sys/sysctl.h>
 int32_t value;
 size_t length=sizeof(value);
+int64_t value64;
+size_t length64=sizeof(value64);
 #endif
 
 #define CPU_UNKNOWN     	0
 #define CPU_ARMV8       	1
 // Arm
 #define CPU_CORTEXA53     2
+#define CPU_CORTEXA55     14
 #define CPU_CORTEXA57     3
 #define CPU_CORTEXA72     4
 #define CPU_CORTEXA73     5
 #define CPU_NEOVERSEN1    11
+#define CPU_NEOVERSEV1    16
+#define CPU_NEOVERSEN2    17
 // Qualcomm
 #define CPU_FALKOR        6
 // Cavium
@@ -52,6 +57,8 @@ size_t length=sizeof(value);
 #define CPU_EMAG8180	 10
 // Apple
 #define CPU_VORTEX       13
+// Fujitsu
+#define CPU_A64FX	 15
 
 static char *cpuname[] = {
   "UNKNOWN",
@@ -66,8 +73,12 @@ static char *cpuname[] = {
   "TSV110",
   "EMAG8180",
   "NEOVERSEN1",
+  "NEOVERSEV1"
+  "NEOVERSEN2"
   "THUNDERX3T110",
-  "VORTEX"	
+  "VORTEX",
+  "CORTEXA55",
+  "A64FX"
 };
 
 static char *cpuname_lower[] = {
@@ -83,8 +94,12 @@ static char *cpuname_lower[] = {
   "tsv110",
   "emag8180",
   "neoversen1",
+  "neoversev1",
+  "neoversen2",
   "thunderx3t110",
-  "vortex"	
+  "vortex",
+  "cortexa55",
+  "a64fx"
 };
 
 int get_feature(char *search)
@@ -161,6 +176,12 @@ int detect(void)
         return CPU_CORTEXA73;
       else if (strstr(cpu_part, "0xd0c"))
         return CPU_NEOVERSEN1;
+      else if (strstr(cpu_part, "0xd40"))
+        return CPU_NEOVERSEV1;
+      else if (strstr(cpu_part, "0xd49"))
+        return CPU_NEOVERSEN2;
+      else if (strstr(cpu_part, "0xd05"))
+	return CPU_CORTEXA55;
     }
     // Qualcomm
     else if (strstr(cpu_implementer, "0x51") && strstr(cpu_part, "0xc00"))
@@ -178,6 +199,9 @@ int detect(void)
     // Ampere
     else if (strstr(cpu_implementer, "0x50") && strstr(cpu_part, "0x000"))
                         return CPU_EMAG8180;
+    // Fujitsu
+    else if (strstr(cpu_implementer, "0x46") && strstr(cpu_part, "0x001"))
+                        return CPU_A64FX;
 	}
 
 	p = (char *) NULL ;
@@ -207,9 +231,9 @@ int detect(void)
 
 	}
 #else
-#ifdef DARWIN
+#ifdef __APPLE__
 	sysctlbyname("hw.cpufamily",&value,&length,NULL,0);
-	if (value ==131287967) return CPU_VORTEX;
+	if (value ==131287967|| value == 458787763 ) return CPU_VORTEX;
 #endif
 	return CPU_ARMV8;	
 #endif
@@ -260,7 +284,7 @@ int n=0;
 
 	printf("#define NUM_CORES %d\n",n);
 #endif
-#ifdef DARWIN
+#ifdef __APPLE__
 	sysctlbyname("hw.physicalcpu_max",&value,&length,NULL,0);
 	printf("#define NUM_CORES %d\n",value);
 #endif	
@@ -280,153 +304,196 @@ void get_cpuconfig(void)
 	switch (d)
 	{
 
-    case CPU_CORTEXA53:
-      printf("#define %s\n", cpuname[d]);
-      // Fall-through
-		case CPU_ARMV8:
-      // Minimum parameters for ARMv8 (based on A53)
-    	printf("#define L1_DATA_SIZE 32768\n");
-    	printf("#define L1_DATA_LINESIZE 64\n");
-    	printf("#define L2_SIZE 262144\n");
-    	printf("#define L2_LINESIZE 64\n");
-    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
-    	printf("#define DTB_SIZE 4096\n");
-    	printf("#define L2_ASSOCIATIVE 4\n");
+	    case CPU_CORTEXA53:
+	    case CPU_CORTEXA55:
+	        printf("#define %s\n", cpuname[d]);
+	      // Fall-through
+	    case CPU_ARMV8:
+	      // Minimum parameters for ARMv8 (based on A53)
+	    	printf("#define L1_DATA_SIZE 32768\n");
+	    	printf("#define L1_DATA_LINESIZE 64\n");
+ 	   	printf("#define L2_SIZE 262144\n");
+	    	printf("#define L2_LINESIZE 64\n");
+	    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
+ 	   	printf("#define DTB_SIZE 4096\n");
+ 	   	printf("#define L2_ASSOCIATIVE 4\n");
 			break;
 
-		case CPU_CORTEXA57:
-		case CPU_CORTEXA72:
-		case CPU_CORTEXA73:
+	    case CPU_CORTEXA57:
+	    case CPU_CORTEXA72:
+	    case CPU_CORTEXA73:
       // Common minimum settings for these Arm cores
       // Can change a lot, but we need to be conservative
       // TODO: detect info from /sys if possible
-      printf("#define %s\n", cpuname[d]);
-			printf("#define L1_CODE_SIZE 49152\n");
-			printf("#define L1_CODE_LINESIZE 64\n");
-			printf("#define L1_CODE_ASSOCIATIVE 3\n");
-			printf("#define L1_DATA_SIZE 32768\n");
-			printf("#define L1_DATA_LINESIZE 64\n");
-			printf("#define L1_DATA_ASSOCIATIVE 2\n");
-      printf("#define L2_SIZE 524288\n");
-			printf("#define L2_LINESIZE 64\n");
-			printf("#define L2_ASSOCIATIVE 16\n");
-			printf("#define DTB_DEFAULT_ENTRIES 64\n");
-			printf("#define DTB_SIZE 4096\n");
-			break;
-		case CPU_NEOVERSEN1:
-			printf("#define %s\n", cpuname[d]);
-			printf("#define L1_CODE_SIZE 65536\n");
-			printf("#define L1_CODE_LINESIZE 64\n");
-			printf("#define L1_CODE_ASSOCIATIVE 4\n");
-			printf("#define L1_DATA_SIZE 65536\n");
-			printf("#define L1_DATA_LINESIZE 64\n");
-			printf("#define L1_DATA_ASSOCIATIVE 4\n");
-			printf("#define L2_SIZE 1048576\n");
-			printf("#define L2_LINESIZE 64\n");
-			printf("#define L2_ASSOCIATIVE 16\n");
-			printf("#define DTB_DEFAULT_ENTRIES 64\n");
-			printf("#define DTB_SIZE 4096\n");
-			break;
-
-    case CPU_FALKOR:
-      printf("#define FALKOR\n");
-      printf("#define L1_CODE_SIZE 65536\n");
-      printf("#define L1_CODE_LINESIZE 64\n");
-      printf("#define L1_DATA_SIZE 32768\n");
-      printf("#define L1_DATA_LINESIZE 128\n");
-      printf("#define L2_SIZE 524288\n");
-      printf("#define L2_LINESIZE 64\n");
-      printf("#define DTB_DEFAULT_ENTRIES 64\n");
-      printf("#define DTB_SIZE 4096\n");
-      printf("#define L2_ASSOCIATIVE 16\n");
-      break;
-
-		case CPU_THUNDERX:
-			printf("#define THUNDERX\n");
-			printf("#define L1_DATA_SIZE 32768\n");
-			printf("#define L1_DATA_LINESIZE 128\n");
-			printf("#define L2_SIZE 16777216\n");
-			printf("#define L2_LINESIZE 128\n");
-			printf("#define DTB_DEFAULT_ENTRIES 64\n");
-			printf("#define DTB_SIZE 4096\n");
-			printf("#define L2_ASSOCIATIVE 16\n");
-			break;
-
-		case CPU_THUNDERX2T99:
-			printf("#define THUNDERX2T99                  \n");
-			printf("#define L1_CODE_SIZE         32768    \n");
-			printf("#define L1_CODE_LINESIZE     64       \n");
-			printf("#define L1_CODE_ASSOCIATIVE  8        \n");
-			printf("#define L1_DATA_SIZE         32768    \n");
-			printf("#define L1_DATA_LINESIZE     64       \n");
-			printf("#define L1_DATA_ASSOCIATIVE  8        \n");
-			printf("#define L2_SIZE              262144   \n");
-			printf("#define L2_LINESIZE          64       \n");
-			printf("#define L2_ASSOCIATIVE       8        \n");
-			printf("#define L3_SIZE              33554432 \n");
-			printf("#define L3_LINESIZE          64       \n");
-			printf("#define L3_ASSOCIATIVE       32       \n");
-			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
-			printf("#define DTB_SIZE             4096     \n");
-			break;
+      		printf("#define %s\n", cpuname[d]);
+		printf("#define L1_CODE_SIZE 49152\n");
+		printf("#define L1_CODE_LINESIZE 64\n");
+		printf("#define L1_CODE_ASSOCIATIVE 3\n");
+		printf("#define L1_DATA_SIZE 32768\n");
+		printf("#define L1_DATA_LINESIZE 64\n");
+		printf("#define L1_DATA_ASSOCIATIVE 2\n");
+		printf("#define L2_SIZE 524288\n");
+		printf("#define L2_LINESIZE 64\n");
+		printf("#define L2_ASSOCIATIVE 16\n");
+		printf("#define DTB_DEFAULT_ENTRIES 64\n");
+		printf("#define DTB_SIZE 4096\n");
+		break;
+	    case CPU_NEOVERSEN1:
+		printf("#define %s\n", cpuname[d]);
+		printf("#define L1_CODE_SIZE 65536\n");
+		printf("#define L1_CODE_LINESIZE 64\n");
+		printf("#define L1_CODE_ASSOCIATIVE 4\n");
+		printf("#define L1_DATA_SIZE 65536\n");
+		printf("#define L1_DATA_LINESIZE 64\n");
+		printf("#define L1_DATA_ASSOCIATIVE 4\n");
+		printf("#define L2_SIZE 1048576\n");
+		printf("#define L2_LINESIZE 64\n");
+		printf("#define L2_ASSOCIATIVE 8\n");
+		printf("#define DTB_DEFAULT_ENTRIES 48\n");
+		printf("#define DTB_SIZE 4096\n");
+		break;
+
+	    case CPU_NEOVERSEV1:
+                printf("#define %s\n", cpuname[d]);
+                printf("#define L1_CODE_SIZE 65536\n");
+                printf("#define L1_CODE_LINESIZE 64\n");
+                printf("#define L1_CODE_ASSOCIATIVE 4\n");
+                printf("#define L1_DATA_SIZE 65536\n");
+                printf("#define L1_DATA_LINESIZE 64\n");
+                printf("#define L1_DATA_ASSOCIATIVE 4\n");
+                printf("#define L2_SIZE 1048576\n");
+                printf("#define L2_LINESIZE 64\n");
+                printf("#define L2_ASSOCIATIVE 8\n");
+                printf("#define DTB_DEFAULT_ENTRIES 48\n");
+                printf("#define DTB_SIZE 4096\n");
+                break;
+
+	    case CPU_NEOVERSEN2:
+                printf("#define %s\n", cpuname[d]);
+                printf("#define L1_CODE_SIZE 65536\n");
+                printf("#define L1_CODE_LINESIZE 64\n");
+                printf("#define L1_CODE_ASSOCIATIVE 4\n");
+                printf("#define L1_DATA_SIZE 65536\n");
+                printf("#define L1_DATA_LINESIZE 64\n");
+                printf("#define L1_DATA_ASSOCIATIVE 4\n");
+                printf("#define L2_SIZE 1048576\n");
+                printf("#define L2_LINESIZE 64\n");
+                printf("#define L2_ASSOCIATIVE 8\n");
+                printf("#define DTB_DEFAULT_ENTRIES 48\n");
+                printf("#define DTB_SIZE 4096\n");
+                break;
+
+	    case CPU_FALKOR:
+	        printf("#define FALKOR\n");
+	        printf("#define L1_CODE_SIZE 65536\n");
+	        printf("#define L1_CODE_LINESIZE 64\n");
+	        printf("#define L1_DATA_SIZE 32768\n");
+	        printf("#define L1_DATA_LINESIZE 128\n");
+	        printf("#define L2_SIZE 524288\n");
+	        printf("#define L2_LINESIZE 64\n");
+	        printf("#define DTB_DEFAULT_ENTRIES 64\n");
+	        printf("#define DTB_SIZE 4096\n");
+	        printf("#define L2_ASSOCIATIVE 16\n");
+	        break;
+
+	    case CPU_THUNDERX:
+		printf("#define THUNDERX\n");
+		printf("#define L1_DATA_SIZE 32768\n");
+		printf("#define L1_DATA_LINESIZE 128\n");
+		printf("#define L2_SIZE 16777216\n");
+		printf("#define L2_LINESIZE 128\n");
+		printf("#define DTB_DEFAULT_ENTRIES 64\n");
+		printf("#define DTB_SIZE 4096\n");
+		printf("#define L2_ASSOCIATIVE 16\n");
+		break;
+
+	    case CPU_THUNDERX2T99:
+		printf("#define THUNDERX2T99                  \n");
+		printf("#define L1_CODE_SIZE         32768    \n");
+		printf("#define L1_CODE_LINESIZE     64       \n");
+		printf("#define L1_CODE_ASSOCIATIVE  8        \n");
+		printf("#define L1_DATA_SIZE         32768    \n");
+		printf("#define L1_DATA_LINESIZE     64       \n");
+		printf("#define L1_DATA_ASSOCIATIVE  8        \n");
+		printf("#define L2_SIZE              262144   \n");
+		printf("#define L2_LINESIZE          64       \n");
+		printf("#define L2_ASSOCIATIVE       8        \n");
+		printf("#define L3_SIZE              33554432 \n");
+		printf("#define L3_LINESIZE          64       \n");
+		printf("#define L3_ASSOCIATIVE       32       \n");
+		printf("#define DTB_DEFAULT_ENTRIES  64       \n");
+		printf("#define DTB_SIZE             4096     \n");
+		break;
 			
-		case CPU_TSV110:
-			printf("#define TSV110                        \n");
-			printf("#define L1_CODE_SIZE         65536    \n");
-			printf("#define L1_CODE_LINESIZE     64       \n");
-			printf("#define L1_CODE_ASSOCIATIVE  4        \n");
-			printf("#define L1_DATA_SIZE         65536    \n");
-			printf("#define L1_DATA_LINESIZE     64       \n");
-			printf("#define L1_DATA_ASSOCIATIVE  4        \n");
-			printf("#define L2_SIZE              524228   \n");
-			printf("#define L2_LINESIZE          64       \n");
-			printf("#define L2_ASSOCIATIVE       8        \n");
-			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
-			printf("#define DTB_SIZE             4096     \n");
-			break;	
-
-		case CPU_EMAG8180:
-      // Minimum parameters for ARMv8 (based on A53)
-	printf("#define EMAG8180\n");
-    	printf("#define L1_CODE_SIZE 32768\n");
-    	printf("#define L1_DATA_SIZE 32768\n");
-    	printf("#define L1_DATA_LINESIZE 64\n");
-    	printf("#define L2_SIZE 262144\n");
-    	printf("#define L2_LINESIZE 64\n");
-    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
-    	printf("#define DTB_SIZE 4096\n");
-			break;
-
-		case CPU_THUNDERX3T110:
-			printf("#define THUNDERX3T110                 \n");
-			printf("#define L1_CODE_SIZE         65536    \n");
-			printf("#define L1_CODE_LINESIZE     64       \n");
-			printf("#define L1_CODE_ASSOCIATIVE  8        \n");
-			printf("#define L1_DATA_SIZE         32768    \n");
-			printf("#define L1_DATA_LINESIZE     64       \n");
-			printf("#define L1_DATA_ASSOCIATIVE  8        \n");
-			printf("#define L2_SIZE              524288   \n");
-			printf("#define L2_LINESIZE          64       \n");
-			printf("#define L2_ASSOCIATIVE       8        \n");
-			printf("#define L3_SIZE              94371840 \n");
-			printf("#define L3_LINESIZE          64       \n");
-			printf("#define L3_ASSOCIATIVE       32       \n");
-			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
-			printf("#define DTB_SIZE             4096     \n");
-			break;
-#ifdef DARWIN
-		case CPU_VORTEX:
-			printf("#define VORTEX			      \n");
-			sysctlbyname("hw.l1icachesize",&value,&length,NULL,0);
-			printf("#define L1_CODE_SIZE	     %d       \n",value);
-			sysctlbyname("hw.cachelinesize",&value,&length,NULL,0);
-			printf("#define L1_CODE_LINESIZE     %d       \n",value);
-			sysctlbyname("hw.l1dcachesize",&value,&length,NULL,0);
-			printf("#define L1_DATA_SIZE	     %d       \n",value);
-			sysctlbyname("hw.l2dcachesize",&value,&length,NULL,0);
-			printf("#define L2_SIZE	     %d       \n",value);
-			break;
+	    case CPU_TSV110:
+		printf("#define TSV110                        \n");
+		printf("#define L1_CODE_SIZE         65536    \n");
+		printf("#define L1_CODE_LINESIZE     64       \n");
+		printf("#define L1_CODE_ASSOCIATIVE  4        \n");
+		printf("#define L1_DATA_SIZE         65536    \n");
+		printf("#define L1_DATA_LINESIZE     64       \n");
+		printf("#define L1_DATA_ASSOCIATIVE  4        \n");
+		printf("#define L2_SIZE              524228   \n");
+		printf("#define L2_LINESIZE          64       \n");
+		printf("#define L2_ASSOCIATIVE       8        \n");
+		printf("#define DTB_DEFAULT_ENTRIES  64       \n");
+		printf("#define DTB_SIZE             4096     \n");
+		break;	
+
+	    case CPU_EMAG8180:
+     		 // Minimum parameters for ARMv8 (based on A53)
+		printf("#define EMAG8180\n");
+    		printf("#define L1_CODE_SIZE 32768\n");
+    		printf("#define L1_DATA_SIZE 32768\n");
+    		printf("#define L1_DATA_LINESIZE 64\n");
+    		printf("#define L2_SIZE 262144\n");
+    		printf("#define L2_LINESIZE 64\n");
+	    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
+	    	printf("#define DTB_SIZE 4096\n");
+		break;
+
+	    case CPU_THUNDERX3T110:
+		printf("#define THUNDERX3T110                 \n");
+		printf("#define L1_CODE_SIZE         65536    \n");
+		printf("#define L1_CODE_LINESIZE     64       \n");
+		printf("#define L1_CODE_ASSOCIATIVE  8        \n");
+		printf("#define L1_DATA_SIZE         32768    \n");
+		printf("#define L1_DATA_LINESIZE     64       \n");
+		printf("#define L1_DATA_ASSOCIATIVE  8        \n");
+		printf("#define L2_SIZE              524288   \n");
+		printf("#define L2_LINESIZE          64       \n");
+		printf("#define L2_ASSOCIATIVE       8        \n");
+		printf("#define L3_SIZE              94371840 \n");
+		printf("#define L3_LINESIZE          64       \n");
+		printf("#define L3_ASSOCIATIVE       32       \n");
+		printf("#define DTB_DEFAULT_ENTRIES  64       \n");
+		printf("#define DTB_SIZE             4096     \n");
+		break;
+#ifdef __APPLE__
+	    case CPU_VORTEX:
+		printf("#define VORTEX			      \n");
+		sysctlbyname("hw.l1icachesize",&value64,&length64,NULL,0);
+		printf("#define L1_CODE_SIZE	     %lld       \n",value64);
+		sysctlbyname("hw.cachelinesize",&value64,&length64,NULL,0);
+		printf("#define L1_CODE_LINESIZE     %lld       \n",value64);
+		sysctlbyname("hw.l1dcachesize",&value64,&length64,NULL,0);
+		printf("#define L1_DATA_SIZE	     %lld       \n",value64);
+		sysctlbyname("hw.l2cachesize",&value64,&length64,NULL,0);
+		printf("#define L2_SIZE	     %lld       \n",value64);
+		printf("#define DTB_DEFAULT_ENTRIES  64       \n");
+		printf("#define DTB_SIZE             4096     \n");
+		break;
 #endif			
+	    case CPU_A64FX:
+		printf("#define A64FX\n");
+    		printf("#define L1_CODE_SIZE 65535\n");
+    		printf("#define L1_DATA_SIZE 65535\n");
+    		printf("#define L1_DATA_LINESIZE 256\n");
+    		printf("#define L2_SIZE 8388608\n");
+    		printf("#define L2_LINESIZE 256\n");
+	    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
+	    	printf("#define DTB_SIZE 4096\n");
+		break;
 	}
 	get_cpucount();
 }

cpuid_loongarch64.c

@@ -0,0 +1,110 @@
+/*****************************************************************************
+Copyright (c) 2011-2020, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of
+      its contributors may be used to endorse or promote products
+      derived from this software without specific prior written
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+**********************************************************************************/
+
+#include <stdint.h>
+
+#define CPU_UNKNOWN     0
+#define CPU_LOONGSON3R5 1
+
+#define LOONGARCH_CFG2  0x02
+#define LOONGARCH_LASX  1<<7
+
+static char *cpuname[] = {
+  "UNKNOWN",
+  "LOONGSON3R5"
+};
+
+int detect(void) {
+    uint32_t reg = 0;
+
+    __asm__ volatile (
+        "cpucfg %0, %1 \n\t"
+        : "+&r"(reg)
+        : "r"(LOONGARCH_CFG2)
+    );
+
+    if (reg & LOONGARCH_LASX)
+        return CPU_LOONGSON3R5;
+    else
+        return CPU_UNKNOWN;
+}
+
+char *get_corename(void) {
+  return cpuname[detect()];
+}
+
+void get_architecture(void) {
+  printf("LOONGARCH64");
+}
+
+void get_subarchitecture(void) {
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("LOONGSON3R5");
+  } else {
+    printf("UNKNOWN");
+  }
+}
+
+void get_subdirname(void) {
+  printf("loongarch64");
+}
+
+void get_cpuconfig(void) {
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("#define LOONGSON3R5\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 64\n");
+    printf("#define L2_SIZE 1048576\n");
+    printf("#define L2_LINESIZE 64\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 16\n");
+  } else {
+    printf("#define LOONGSON3R5\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 64\n");
+    printf("#define L2_SIZE 1048576\n");
+    printf("#define L2_LINESIZE 64\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 16\n");
+  }
+}
+
+void get_libname(void){
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("loongson3r5\n");
+  } else {
+    printf("loongarch64\n");
+  }
+}

cpuid_mips.c

@@ -165,6 +165,7 @@ void get_cpuconfig(void){
   }else{
     printf("#define UNKNOWN\n");
   }
+  if (!get_feature("msa")) printf("#define NO_MSA\n");
 }
 
 void get_libname(void){
@@ -178,3 +179,38 @@ void get_libname(void){
     printf("mips\n");
   }
 }
+
+int get_feature(char *search)
+{
+
+#ifdef __linux
+        FILE *infile;
+        char buffer[2048], *p,*t;
+        p = (char *) NULL ;
+
+        infile = fopen("/proc/cpuinfo", "r");
+
+        while (fgets(buffer, sizeof(buffer), infile))
+        {
+
+                if (!strncmp("Features", buffer, 8) || !strncmp("ASEs implemented", buffer, 16))
+                {
+                        p = strchr(buffer, ':') + 2;
+                        break;
+                }
+        }
+
+        fclose(infile);
+
+        if( p == NULL ) return 0;
+
+        t = strtok(p," ");
+        while( t = strtok(NULL," "))
+        {
+                if (strstr(t, search))   { return(1); }
+        }
+
+#endif
+        return(0);
+}
+

cpuid_mips64.c

@@ -70,19 +70,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* or implied, of The University of Texas at Austin.                 */
 /*********************************************************************/
 
-#define CPU_UNKNOWN     0
-#define CPU_SICORTEX    1
-#define CPU_LOONGSON3A  2
-#define CPU_LOONGSON3B  3
-#define CPU_I6400       4
-#define CPU_P6600       5
-#define CPU_I6500       6
+#define CPU_UNKNOWN      0
+#define CPU_SICORTEX     1
+#define CPU_LOONGSON3R3  2
+#define CPU_LOONGSON3R4  3
+#define CPU_I6400        4
+#define CPU_P6600        5
+#define CPU_I6500        6
 
 static char *cpuname[] = {
   "UNKNOWN",
   "SICORTEX",
-  "LOONGSON3A",
-  "LOONGSON3B",
+  "LOONGSON3R3",
+  "LOONGSON3R4",
   "I6400",
   "P6600",
   "I6500"
@@ -90,48 +90,13 @@ static char *cpuname[] = {
 
 int detect(void){
 
-#ifdef __linux
+#ifdef linux
   FILE *infile;
   char buffer[512], *p;
 
   p = (char *)NULL;
-  infile = fopen("/proc/cpuinfo", "r");
-  while (fgets(buffer, sizeof(buffer), infile)){
-    if (!strncmp("cpu", buffer, 3)){
-	p = strchr(buffer, ':') + 2;
-#if 0
-	fprintf(stderr, "%s\n", p);
-#endif
-	break;
-      }
-  }
-
-  fclose(infile);
-
-  if(p != NULL){
-  if (strstr(p, "Loongson-3A")){
-    return CPU_LOONGSON3A;
-  }else if(strstr(p, "Loongson-3B")){
-    return CPU_LOONGSON3B;
-  }else if (strstr(p, "Loongson-3")){
-    infile = fopen("/proc/cpuinfo", "r");
-    p = (char *)NULL;
-    while (fgets(buffer, sizeof(buffer), infile)){
-      if (!strncmp("system type", buffer, 11)){
-	p = strchr(buffer, ':') + 2;
-	break;
-      }
-    }
-    fclose(infile);
-    if (strstr(p, "loongson3a"))
-      return CPU_LOONGSON3A;
-  }else{
-    return CPU_SICORTEX;
-  }
-  }
   //Check model name for Loongson3
   infile = fopen("/proc/cpuinfo", "r");
-  p = (char *)NULL;
   while (fgets(buffer, sizeof(buffer), infile)){
     if (!strncmp("model name", buffer, 10)){
       p = strchr(buffer, ':') + 2;
@@ -139,12 +104,14 @@ int detect(void){
     }
   }
   fclose(infile);
-  if(p != NULL){
-  if (strstr(p, "Loongson-3A")){
-    return CPU_LOONGSON3A;
-  }else if(strstr(p, "Loongson-3B")){
-    return CPU_LOONGSON3B;
-  }
+  if (p != NULL){
+    if (strstr(p, "Loongson-3A3000") || strstr(p, "Loongson-3B3000")){
+      return CPU_LOONGSON3R3;
+    } else if (strstr(p, "Loongson-3A4000") || strstr(p, "Loongson-3B4000")){
+      return CPU_LOONGSON3R4;
+    } else{
+      return CPU_SICORTEX;
+    }
   }
 #endif
     return CPU_UNKNOWN;
@@ -159,10 +126,10 @@ void get_architecture(void){
 }
 
 void get_subarchitecture(void){
-  if(detect()==CPU_LOONGSON3A) {
-    printf("LOONGSON3A");
-  }else if(detect()==CPU_LOONGSON3B){
-    printf("LOONGSON3B");
+  if(detect()==CPU_LOONGSON3R3) {
+    printf("LOONGSON3R3");
+  }else if(detect()==CPU_LOONGSON3R4){
+    printf("LOONGSON3R4");
   }else if(detect()==CPU_I6400){
     printf("I6400");
   }else if(detect()==CPU_P6600){
@@ -179,8 +146,8 @@ void get_subdirname(void){
 }
 
 void get_cpuconfig(void){
-  if(detect()==CPU_LOONGSON3A) {
-    printf("#define LOONGSON3A\n");
+  if(detect()==CPU_LOONGSON3R3) {
+    printf("#define LOONGSON3R3\n");
     printf("#define L1_DATA_SIZE 65536\n");
     printf("#define L1_DATA_LINESIZE 32\n");
     printf("#define L2_SIZE 512488\n");
@@ -188,8 +155,8 @@ void get_cpuconfig(void){
     printf("#define DTB_DEFAULT_ENTRIES 64\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 4\n");
-  }else if(detect()==CPU_LOONGSON3B){
-    printf("#define LOONGSON3B\n");
+  }else if(detect()==CPU_LOONGSON3R4){
+    printf("#define LOONGSON3R4\n");
     printf("#define L1_DATA_SIZE 65536\n");
     printf("#define L1_DATA_LINESIZE 32\n");
     printf("#define L2_SIZE 512488\n");
@@ -234,13 +201,14 @@ void get_cpuconfig(void){
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 8\n");
   }
+  if (!get_feature("msa")) printf("#define NO_MSA\n");
 }
 
 void get_libname(void){
-  if(detect()==CPU_LOONGSON3A) {
-    printf("loongson3a\n");
-  }else if(detect()==CPU_LOONGSON3B) {
-    printf("loongson3b\n");
+  if(detect()==CPU_LOONGSON3R3) {
+    printf("loongson3r3\n");
+  }else if(detect()==CPU_LOONGSON3R4) {
+    printf("loongson3r4\n");
   }else if(detect()==CPU_I6400) {
     printf("i6400\n");
   }else if(detect()==CPU_P6600) {
@@ -251,3 +219,38 @@ void get_libname(void){
     printf("mips64\n");
   }
 }
+
+int get_feature(char *search)
+{
+
+#ifdef __linux
+        FILE *infile;
+        char buffer[2048], *p,*t;
+        p = (char *) NULL ;
+
+        infile = fopen("/proc/cpuinfo", "r");
+
+        while (fgets(buffer, sizeof(buffer), infile))
+        {
+
+                if (!strncmp("Features", buffer, 8) || !strncmp("ASEs implemented", buffer, 16))
+                {
+                        p = strchr(buffer, ':') + 2;
+                        break;
+                }
+        }
+
+        fclose(infile);
+
+        if( p == NULL ) return 0;
+
+        t = strtok(p," ");
+        while( t = strtok(NULL," "))
+        {
+                if (strstr(t, search))   { return(1); }
+        }
+
+#endif
+        return(0);
+}
+

cpuid_x86.c

@@ -1,3 +1,4 @@
+//{
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
 /* All rights reserved.                                              */
@@ -266,6 +267,31 @@ int support_avx512_bf16(){
 #endif
 }
 
+#define BIT_AMX_TILE	0x01000000
+#define BIT_AMX_BF16	0x00400000
+#define BIT_AMX_ENBD	0x00060000
+
+int support_amx_bf16() {
+#if !defined(NO_AVX) && !defined(NO_AVX512)
+  int eax, ebx, ecx, edx;
+  int ret=0;
+
+  if (!support_avx512())
+    return 0;
+  // CPUID.7.0:EDX indicates AMX support
+  cpuid_count(7, 0, &eax, &ebx, &ecx, &edx);
+  if ((edx & BIT_AMX_TILE) && (edx & BIT_AMX_BF16)) {
+    // CPUID.D.0:EAX[17:18] indicates AMX enabled
+    cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
+    if ((eax & BIT_AMX_ENBD) == BIT_AMX_ENBD)
+      ret = 1;
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
 int get_vendor(void){
   int eax, ebx, ecx, edx;
   char vendor[13];
@@ -283,6 +309,7 @@ int get_vendor(void){
   if (!strcmp(vendor, "CyrixInstead")) return VENDOR_CYRIX;
   if (!strcmp(vendor, "NexGenDriven")) return VENDOR_NEXGEN;
   if (!strcmp(vendor, "CentaurHauls")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor, "  Shanghai  ")) return VENDOR_ZHAOXIN;
   if (!strcmp(vendor, "RiseRiseRise")) return VENDOR_RISE;
   if (!strcmp(vendor, " SiS SiS SiS")) return VENDOR_SIS;
   if (!strcmp(vendor, "GenuineTMx86")) return VENDOR_TRANSMETA;
@@ -296,9 +323,11 @@ int get_vendor(void){
 
 int get_cputype(int gettype){
   int eax, ebx, ecx, edx;
+/*
   int extend_family, family;
   int extend_model, model;
   int type, stepping;
+*/
   int feature = 0;
 
   cpuid(1, &eax, &ebx, &ecx, &edx);
@@ -352,6 +381,7 @@ int get_cputype(int gettype){
     if (support_avx2()) feature |= HAVE_AVX2;
     if (support_avx512()) feature |= HAVE_AVX512VL;
     if (support_avx512_bf16()) feature |= HAVE_AVX512BF16;
+    if (support_amx_bf16()) feature |= HAVE_AMXBF16;
     if ((ecx & (1 << 12)) != 0) feature |= HAVE_FMA3;
 #endif
 
@@ -400,7 +430,8 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
   cpuid(0, &cpuid_level, &ebx, &ecx, &edx);
 
   if (cpuid_level > 1) {
-    int numcalls =0 ;
+    int numcalls;
+    
     cpuid(2, &eax, &ebx, &ecx, &edx);
     numcalls = BITMASK(eax, 0, 0xff); //FIXME some systems may require repeated calls to read all entries
     info[ 0] = BITMASK(eax,  8, 0xff);
@@ -1066,7 +1097,8 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
 
   if ((get_vendor() == VENDOR_AMD) ||
       (get_vendor() == VENDOR_HYGON) ||
-      (get_vendor() == VENDOR_CENTAUR)) {
+      (get_vendor() == VENDOR_CENTAUR) ||
+      (get_vendor() == VENDOR_ZHAOXIN)) {
     cpuid(0x80000005, &eax, &ebx, &ecx, &edx);
 
     LDTB.size        = 4096;
@@ -1189,7 +1221,7 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
 
 int get_cpuname(void){
 
-  int family, exfamily, model, vendor, exmodel;
+  int family, exfamily, model, vendor, exmodel, stepping;
 
   if (!have_cpuid()) return CPUTYPE_80386;
 
@@ -1197,6 +1229,7 @@ int get_cpuname(void){
   exfamily = get_cputype(GET_EXFAMILY);
   model    = get_cputype(GET_MODEL);
   exmodel  = get_cputype(GET_EXMODEL);
+  stepping = get_cputype(GET_STEPPING);
 
   vendor = get_vendor();
 
@@ -1398,6 +1431,17 @@ int get_cpuname(void){
 	    return CPUTYPE_SANDYBRIDGE;
 	  else
 	  return CPUTYPE_NEHALEM;
+	case 10: // Ice Lake SP
+	  if(support_avx512_bf16())
+            return CPUTYPE_COOPERLAKE;	
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	  return CPUTYPE_NEHALEM;	
         }
       break;
       case 7: // family 6 exmodel 7
@@ -1415,9 +1459,18 @@ int get_cpuname(void){
 	  return CPUTYPE_NEHALEM;
         }
       break;
-      case 9:
       case 8:      
         switch (model) {
+        case 12: // Tiger Lake
+        case 13: // Tiger Lake (11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz)
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+            return CPUTYPE_SANDYBRIDGE;
+          else
+          return CPUTYPE_NEHALEM;
 	case 14: // Kaby Lake and refreshes
           if(support_avx2())
             return CPUTYPE_HASWELL;
@@ -1425,21 +1478,74 @@ int get_cpuname(void){
 	    return CPUTYPE_SANDYBRIDGE;
           else
 	    return CPUTYPE_NEHALEM;
-    }
-      case 10: //family 6 exmodel 10
+	case 15: // Sapphire Rapids
+	  if(support_avx512_bf16())
+            return CPUTYPE_COOPERLAKE;	
+          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:
         switch (model) {
-    case 5: // Comet Lake H and S
-    case 6: // Comet Lake U
+        case 7: // Alder Lake desktop
+        case 10: // Alder Lake mobile
+	  if(support_avx512_bf16())
+            return CPUTYPE_COOPERLAKE;	
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
           if(support_avx2())
             return CPUTYPE_HASWELL;
           if(support_avx())
-        return CPUTYPE_SANDYBRIDGE;
+            return CPUTYPE_SANDYBRIDGE;
           else
-        return CPUTYPE_NEHALEM;
-	}
-	break;    
-      }
+          return CPUTYPE_NEHALEM;
+        case 13: // Ice Lake NNPI
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+            return CPUTYPE_SANDYBRIDGE;
+          else
+          return CPUTYPE_NEHALEM;
+	case 14: // Kaby Lake and refreshes
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+          else
+	    return CPUTYPE_NEHALEM;
+        }
       break;
+      case 10: //family 6 exmodel 10
+        switch (model) {
+          case 5: // Comet Lake H and S
+          case 6: // Comet Lake U
+            if(support_avx2())
+              return CPUTYPE_HASWELL;
+            if(support_avx())
+              return CPUTYPE_SANDYBRIDGE;
+            else
+              return CPUTYPE_NEHALEM;
+          case 7: // Rocket Lake           
+	    if(support_avx512())
+              return CPUTYPE_SKYLAKEX;
+            if(support_avx2())
+              return CPUTYPE_HASWELL;
+            if(support_avx())
+	      return CPUTYPE_SANDYBRIDGE;
+	    else
+	    return CPUTYPE_NEHALEM;
+        }
+        break;
+      }
+      break;    
     case 0x7:
       return CPUTYPE_ITANIUM;
     case 0xf:
@@ -1538,7 +1644,6 @@ int get_cpuname(void){
 	  else
 	    return CPUTYPE_BARCELONA;
         }
-	break;	      
       case 10: // Zen3		      
 	if(support_avx())
 #ifndef NO_AVX2
@@ -1598,13 +1703,20 @@ int get_cpuname(void){
     switch (family) {
     case 0x5:
       return CPUTYPE_CENTAURC6;
-      break;
     case 0x6:
-      return CPUTYPE_NANO;
-      break;
-
+      if (model == 0xf && stepping < 0xe)
+        return CPUTYPE_NANO;
+      return CPUTYPE_NEHALEM;
+    default:
+      if (family >= 0x7)
+        return CPUTYPE_NEHALEM;
+      else
+        return CPUTYPE_VIAC3;
     }
-    return CPUTYPE_VIAC3;
+  }
+
+  if (vendor == VENDOR_ZHAOXIN){
+    return CPUTYPE_NEHALEM;
   }
 
   if (vendor == VENDOR_RISE){
@@ -1837,7 +1949,7 @@ char *get_lower_cpunamechar(void){
 
 int get_coretype(void){
 
-  int family, exfamily, model, exmodel, vendor;
+  int family, exfamily, model, exmodel, vendor, stepping;
 
   if (!have_cpuid()) return CORE_80486;
 
@@ -1845,6 +1957,7 @@ int get_coretype(void){
   exfamily = get_cputype(GET_EXFAMILY);
   model    = get_cputype(GET_MODEL);
   exmodel  = get_cputype(GET_EXMODEL);
+  stepping = get_cputype(GET_STEPPING);
 
   vendor = get_vendor();
 
@@ -2002,19 +2115,7 @@ int get_coretype(void){
 	    return CORE_NEHALEM;
         }
         break;
-      case 10:
-        switch (model) {
-	  case 5: // Comet Lake H and S
-    	  case 6: // Comet Lake U
-            if(support_avx())
-  #ifndef NO_AVX2
-              return CORE_HASWELL;
-  #else
-              return CORE_SANDYBRIDGE;
-  #endif
-            else
-              return CORE_NEHALEM;
-        }
+
       case 5:
         switch (model) {
 	case 6:
@@ -2068,6 +2169,7 @@ int get_coretype(void){
 	    return CORE_NEHALEM;
         }
 	break;
+
       case 6:
         if (model == 6)
 #ifndef NO_AVX512
@@ -2081,12 +2183,27 @@ int get_coretype(void){
 #endif
 	  else
 	    return CORE_NEHALEM;
-#endif			
-        break;    	
+#endif
+	if (model == 10 || model == 12)
+#ifndef NO_AVX512
+	  if(support_avx512_bf16())
+            return CORE_COOPERLAKE;
+	  return CORE_SKYLAKEX;
+#else
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_HASWELL;
+#else
+	    return CORE_SANDYBRIDGE;
+#endif
+	  else
+	    return CORE_NEHALEM;
+#endif	
+
       case 7:
         if (model == 10) 
             return CORE_NEHALEM;
-        if (model == 14)
+        if (model == 13 || model == 14) // Ice Lake
 #ifndef NO_AVX512
 	    return CORE_SKYLAKEX;
 #else
@@ -2100,9 +2217,19 @@ int get_coretype(void){
 	    return CORE_NEHALEM;
 #endif			
         break;    	
-      case 9:
+
       case 8:
-        if (model == 14) { // Kaby Lake 
+        if (model == 12 || model == 13) { // Tiger Lake
+          if(support_avx512())
+            return CORE_SKYLAKEX;
+          if(support_avx2())
+            return CORE_HASWELL;
+          if(support_avx())
+            return CORE_SANDYBRIDGE;
+          else
+          return CORE_NEHALEM;
+        }
+        if (model == 14) { // Kaby Lake mobile
 	  if(support_avx())
 #ifndef NO_AVX2
 	    return CORE_HASWELL;
@@ -2112,12 +2239,82 @@ int get_coretype(void){
 	  else
             return CORE_NEHALEM;
 	}
-      }
+        if (model == 15) { // Sapphire Rapids
+	  if(support_avx512_bf16())
+            return CPUTYPE_COOPERLAKE;	
+          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:
+        if (model == 7 || model == 10) { // Alder Lake
+          if(support_avx2())
+            return CORE_HASWELL;
+          if(support_avx())
+            return CORE_SANDYBRIDGE;
+          else
+          return CORE_NEHALEM;
+        }
+        if (model == 13) { // Ice Lake NNPI
+          if(support_avx512())
+            return CORE_SKYLAKEX;
+          if(support_avx2())
+            return CORE_HASWELL;
+          if(support_avx())
+            return CORE_SANDYBRIDGE;
+          else
+          return CORE_NEHALEM;
+        }
+        if (model == 14) { // Kaby Lake desktop
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_HASWELL;
+#else
+	    return CORE_SANDYBRIDGE;
+#endif
+	  else
+            return CORE_NEHALEM;
+	}
+      break;
+
+      case 10:
+        switch (model) {
+	  case 5: // Comet Lake H and S
+    	  case 6: // Comet Lake U
+            if(support_avx())
+  #ifndef NO_AVX2
+              return CORE_HASWELL;
+  #else
+              return CORE_SANDYBRIDGE;
+  #endif
+            else
+              return CORE_NEHALEM;
+	  case 7:// Rocket Lake
+#ifndef NO_AVX512
+	  if(support_avx512())
+            return CORE_SKYLAKEX;
+#endif
+#ifndef NO_AVX2
+	  if(support_avx2())
+            return CORE_HASWELL;
+#endif
+	  if(support_avx())
+	    return CORE_SANDYBRIDGE;
+	  else
+	  return CORE_NEHALEM;
+        }
+
       case 15:
 	if (model <= 0x2) return CORE_NORTHWOOD;
 	else return CORE_PRESCOTT;
+      }
     }
   }
 
@@ -2216,10 +2413,19 @@ int get_coretype(void){
   if (vendor == VENDOR_CENTAUR) {
     switch (family) {
     case 0x6:
-      return CORE_NANO;
-      break;
+      if (model == 0xf && stepping < 0xe)
+        return CORE_NANO;
+      return CORE_NEHALEM;
+    default:
+      if (family >= 0x7)
+        return CORE_NEHALEM;
+      else
+        return CORE_VIAC3;
     }
-    return CORE_VIAC3;
+  }
+
+  if (vendor == VENDOR_ZHAOXIN) {
+     return CORE_NEHALEM;
   }
 
   return CORE_UNKNOWN;
@@ -2302,6 +2508,7 @@ void get_cpuconfig(void){
     if (features & HAVE_AVX2 )    printf("#define HAVE_AVX2\n");
     if (features & HAVE_AVX512VL )    printf("#define HAVE_AVX512VL\n");
     if (features & HAVE_AVX512BF16 )    printf("#define HAVE_AVX512BF16\n");
+    if (features & HAVE_AMXBF16 )    printf("#define HAVE_AMXBF16\n");
     if (features & HAVE_3DNOWEX) printf("#define HAVE_3DNOWEX\n");
     if (features & HAVE_3DNOW)   printf("#define HAVE_3DNOW\n");
     if (features & HAVE_FMA4 )    printf("#define HAVE_FMA4\n");
@@ -2373,9 +2580,11 @@ void get_sse(void){
   if (features & HAVE_AVX2 )    printf("HAVE_AVX2=1\n");
   if (features & HAVE_AVX512VL )    printf("HAVE_AVX512VL=1\n");
   if (features & HAVE_AVX512BF16 )    printf("HAVE_AVX512BF16=1\n");
+  if (features & HAVE_AMXBF16 )    printf("HAVE_AMXBF16=1\n");
   if (features & HAVE_3DNOWEX) printf("HAVE_3DNOWEX=1\n");
   if (features & HAVE_3DNOW)   printf("HAVE_3DNOW=1\n");
   if (features & HAVE_FMA4 )    printf("HAVE_FMA4=1\n");
   if (features & HAVE_FMA3 )    printf("HAVE_FMA3=1\n");
 
 }
+//}

cpuid_zarch.c

@@ -27,57 +27,11 @@
 
 #include <string.h>
 
-#define CPU_GENERIC     0
-#define CPU_Z13         1
-#define CPU_Z14         2
-#define CPU_Z15         3
+#include "cpuid_zarch.h"
 
-static char *cpuname[] = {
-  "ZARCH_GENERIC",
-  "Z13",
-  "Z14",
-  "Z15"
-};
-
-static char *cpuname_lower[] = {
-  "zarch_generic",
-  "z13",
-  "z14",
-  "z15"
-};
-
-int detect(void)
-{
-  FILE *infile;
-  char buffer[512], *p;
-
-  p = (char *)NULL;
-  infile = fopen("/proc/sysinfo", "r");
-  while (fgets(buffer, sizeof(buffer), infile)){
-    if (!strncmp("Type", buffer, 4)){
-        p = strchr(buffer, ':') + 2;
-#if 0
-        fprintf(stderr, "%s\n", p);
-#endif
-        break;
-      }
-  }
-
-  fclose(infile);
-
-  if (strstr(p, "2964")) return CPU_Z13;
-  if (strstr(p, "2965")) return CPU_Z13;
-  if (strstr(p, "3906")) return CPU_Z14;
-  if (strstr(p, "3907")) return CPU_Z14;
-  if (strstr(p, "8561")) return CPU_Z14;        // fallback z15 to z14
-  if (strstr(p, "8562")) return CPU_Z14;        // fallback z15 to z14
-
-  return CPU_GENERIC;
-}
 
 void get_libname(void)
 {
-
 	int d = detect();
 	printf("%s", cpuname_lower[d]);
 }

cpuid_zarch.h

@@ -0,0 +1,101 @@
+#include <stdlib.h>
+
+#define CPU_GENERIC     0
+#define CPU_Z13         1
+#define CPU_Z14         2
+#define CPU_Z15         3
+
+static char *cpuname[] = {
+  "ZARCH_GENERIC",
+  "Z13",
+  "Z14",
+  "Z15"
+};
+
+static char *cpuname_lower[] = {
+  "zarch_generic",
+  "z13",
+  "z14",
+  "z15"
+};
+
+// Guard the use of getauxval() on glibc version >= 2.16
+#ifdef __GLIBC__
+#include <features.h>
+#if __GLIBC_PREREQ(2, 16)
+#include <sys/auxv.h>
+#define HAVE_GETAUXVAL 1
+
+static unsigned long get_hwcap(void)
+{
+	unsigned long hwcap = getauxval(AT_HWCAP);
+	char *maskenv;
+
+	// honor requests for not using specific CPU features in LD_HWCAP_MASK
+	maskenv = getenv("LD_HWCAP_MASK");
+	if (maskenv)
+		hwcap &= strtoul(maskenv, NULL, 0);
+
+	return hwcap;
+	// note that a missing auxval is interpreted as no capabilities
+	// available, which is safe.
+}
+
+#else // __GLIBC_PREREQ(2, 16)
+#warn "Cannot detect SIMD support in Z13 or newer architectures since glibc is older than 2.16"
+
+static unsigned long get_hwcap(void) {
+	// treat missing support for getauxval() as no capabilities available,
+	// which is safe.
+	return 0;
+}
+#endif // __GLIBC_PREREQ(2, 16)
+#endif // __GLIBC
+
+static int detect(void)
+{
+	unsigned long hwcap = get_hwcap();
+
+	// Choose the architecture level for optimized kernels based on hardware
+	// capability bits (just like glibc chooses optimized implementations).
+	//
+	// The hardware capability bits that are used here indicate both
+	// hardware support for a particular ISA extension and the presence of
+	// software support to enable its use. For example, when HWCAP_S390_VX
+	// is set then both the CPU can execute SIMD instructions and the Linux
+	// kernel can manage applications using the vector registers and SIMD
+	// instructions.
+	//
+	// See glibc's sysdeps/s390/dl-procinfo.h for an overview (also in
+	// sysdeps/unix/sysv/linux/s390/bits/hwcap.h) of the defined hardware
+	// capability bits. They are derived from the information that the
+	// "store facility list (extended)" instructions provide.
+	// (https://sourceware.org/git/?p=glibc.git;a=blob_plain;f=sysdeps/s390/dl-procinfo.h;hb=HEAD)
+	//
+	// currently used:
+	// HWCAP_S390_VX - vector facility for z/Architecture (introduced with
+	//                 IBM z13), enables level CPU_Z13 (SIMD)
+	// HWCAP_S390_VXE - vector enhancements facility 1 (introduced with IBM
+	//                  z14), together with VX enables level CPU_Z14
+	//                  (single-precision SIMD instructions)
+	//
+	// When you add optimized kernels that make use of other ISA extensions
+	// (e.g., for exploiting the vector-enhancements facility 2 that was introduced
+	// with IBM z15), then add a new architecture level (e.g., CPU_Z15) and gate
+	// it on the hwcap that represents it here (e.g., HWCAP_S390_VXRS_EXT2
+	// for the z15 vector enhancements).
+	//
+	// To learn the value of hwcaps on a given system, set the environment
+	// variable LD_SHOW_AUXV and let ld.so dump it (e.g., by running
+	// LD_SHOW_AUXV=1 /bin/true).
+	// Also, the init function for dynamic arch support will print hwcaps
+	// when OPENBLAS_VERBOSE is set to 2 or higher.
+	if ((hwcap & HWCAP_S390_VX) && (hwcap & HWCAP_S390_VXE))
+		return CPU_Z14;
+
+	if (hwcap & HWCAP_S390_VX)
+		return CPU_Z13;
+
+	return CPU_GENERIC;
+}
+

ctest.c

@@ -84,7 +84,7 @@ OS_AIX
 OS_OSF
 #endif
 
-#if defined(__WIN32) || defined(__WIN64) || defined(__WINNT)
+#if defined(__WIN32) || defined(__WIN64) || defined(_WIN32) || defined(_WIN64) || defined(__WINNT)
 OS_WINNT
 #endif
 
@@ -141,7 +141,7 @@ ARCH_SPARC
 ARCH_IA64
 #endif
 
-#if defined(__LP64) || defined(__LP64__) || defined(__ptr64) || defined(__x86_64__) || defined(__amd64__) || defined(__64BIT__)
+#if defined(__LP64) || defined(__LP64__) || defined(__ptr64) || defined(__x86_64__) || defined(__amd64__) || defined(__64BIT__) || defined(__aarch64__)
 BINARY_64
 #endif
 
@@ -157,7 +157,15 @@ ARCH_ARM64
 ARCH_RISCV64
 #endif
 
+#ifdef __loongarch64
+ARCH_LOONGARCH64
+#endif
+
 #if (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L)
 HAVE_C11
 #endif
 
+#if defined(__e2k__)
+ARCH_E2K
+#endif
+

ctest/CMakeLists.txt

@@ -4,10 +4,22 @@ include_directories(${PROJECT_BINARY_DIR})
 enable_language(Fortran)
 
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DADD${BU} -DCBLAS")
+if (CMAKE_Fortran_COMPILER_ID STREQUAL GNU)
+	set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fno-tree-vectorize")
+endif()
 
+if(WIN32)
+FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.ps1
+"$ErrorActionPreference = \"Stop\"\n"
+"Get-Content $args[1] | & $args[0]\n"
+)
+set(test_helper powershell -ExecutionPolicy Bypass "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.ps1")
+else()
 FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh
 "$1 < $2\n"
 )
+set(test_helper sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh")
+endif()
 
 foreach(float_type ${FLOAT_TYPES})
   string(SUBSTRING ${float_type} 0 1 float_char_upper)
@@ -21,7 +33,7 @@ foreach(float_type ${FLOAT_TYPES})
     c_${float_char}blas1.c)
   target_link_libraries(x${float_char}cblat1 ${OpenBLAS_LIBNAME})
   add_test(NAME "x${float_char}cblat1"
-    COMMAND "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat1")
+    COMMAND $<TARGET_FILE:x${float_char}cblat1>)
 
   #level2
   add_executable(x${float_char}cblat2
@@ -33,7 +45,7 @@ foreach(float_type ${FLOAT_TYPES})
     constant.c)
   target_link_libraries(x${float_char}cblat2 ${OpenBLAS_LIBNAME})
   add_test(NAME "x${float_char}cblat2"
-    COMMAND sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh" "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat2" "${PROJECT_SOURCE_DIR}/ctest/${float_char}in2")
+    COMMAND ${test_helper} $<TARGET_FILE:x${float_char}cblat2> "${PROJECT_SOURCE_DIR}/ctest/${float_char}in2")
 
   #level3
   add_executable(x${float_char}cblat3
@@ -45,6 +57,6 @@ foreach(float_type ${FLOAT_TYPES})
     constant.c)
   target_link_libraries(x${float_char}cblat3 ${OpenBLAS_LIBNAME})
   add_test(NAME "x${float_char}cblat3"
-    COMMAND sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh" "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat3" "${PROJECT_SOURCE_DIR}/ctest/${float_char}in3")
+    COMMAND ${test_helper} $<TARGET_FILE:x${float_char}cblat3> "${PROJECT_SOURCE_DIR}/ctest/${float_char}in3")
 
 endforeach()

ctest/Makefile

@@ -6,6 +6,9 @@ TOPDIR = ..
 include $(TOPDIR)/Makefile.system
 
 override CFLAGS += -DADD$(BU) -DCBLAS
+ifeq ($(F_COMPILER),GFORTRAN)
+	override FFLAGS += -fno-tree-vectorize
+endif
 override TARGET_ARCH=
 override TARGET_MACH=
 
@@ -212,6 +215,9 @@ ifeq ($(C_COMPILER), CLANG)
 CEXTRALIB = -lomp
 endif
 endif
+ifeq ($(F_COMPILER), NAG)
+CEXTRALIB = -lgomp
+endif
 endif
 
 ifeq ($(BUILD_SINGLE),1)

ctest/c_cblas2.c

@@ -20,7 +20,7 @@ void F77_cgemv(int *order, char *transp, int *m, int *n,
   get_transpose_type(transp, &trans);
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A  = (CBLAS_TEST_COMPLEX *)malloc( (*m)*LDA*sizeof( CBLAS_TEST_COMPLEX) );
+     A  = (CBLAS_TEST_COMPLEX *)malloc( (*m)*(size_t)LDA*sizeof( CBLAS_TEST_COMPLEX) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -50,7 +50,7 @@ void F77_cgbmv(int *order, char *transp, int *m, int *n, int *kl, int *ku,
   get_transpose_type(transp, &trans);
   if (*order == TEST_ROW_MJR) {
      LDA = *ku+*kl+2;
-     A=( CBLAS_TEST_COMPLEX* )malloc((*n+*kl)*LDA*sizeof(CBLAS_TEST_COMPLEX));
+     A=( CBLAS_TEST_COMPLEX* )malloc((*n+*kl)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX));
      for( i=0; i<*ku; i++ ){
         irow=*ku+*kl-i;
         jcol=(*ku)-i;
@@ -94,7 +94,7 @@ void F77_cgeru(int *order, int *m, int *n, CBLAS_TEST_COMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A=(CBLAS_TEST_COMPLEX*)malloc((*m)*LDA*sizeof(CBLAS_TEST_COMPLEX));
+     A=(CBLAS_TEST_COMPLEX*)malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX));
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -122,7 +122,7 @@ void F77_cgerc(int *order, int *m, int *n, CBLAS_TEST_COMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A=(CBLAS_TEST_COMPLEX* )malloc((*m)*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
+     A=(CBLAS_TEST_COMPLEX* )malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -154,7 +154,7 @@ void F77_chemv(int *order, char *uplow, int *n, CBLAS_TEST_COMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A = (CBLAS_TEST_COMPLEX *)malloc((*n)*LDA*sizeof(CBLAS_TEST_COMPLEX));
+     A = (CBLAS_TEST_COMPLEX *)malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX));
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -190,7 +190,7 @@ int i,irow,j,jcol,LDA;
 		 *incx, beta, y, *incy );
      else {
         LDA = *k+2;
-        A =(CBLAS_TEST_COMPLEX*)malloc((*n+*k)*LDA*sizeof(CBLAS_TEST_COMPLEX));
+        A =(CBLAS_TEST_COMPLEX*)malloc((*n+*k)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX));
         if (uplo == CblasUpper) {
            for( i=0; i<*k; i++ ){
               irow=*k-i;
@@ -251,8 +251,8 @@ void F77_chpmv(int *order, char *uplow, int *n, CBLAS_TEST_COMPLEX *alpha,
 	         beta, y, *incy);
      else {
         LDA = *n;
-        A = (CBLAS_TEST_COMPLEX* )malloc(LDA*LDA*sizeof(CBLAS_TEST_COMPLEX ));
-        AP = (CBLAS_TEST_COMPLEX* )malloc( (((LDA+1)*LDA)/2)*
+        A = (CBLAS_TEST_COMPLEX* )malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_COMPLEX ));
+        AP = (CBLAS_TEST_COMPLEX* )malloc( ((((size_t)LDA+1)*LDA)/2)*
 	        sizeof( CBLAS_TEST_COMPLEX ));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -311,7 +311,7 @@ void F77_ctbmv(int *order, char *uplow, char *transp, char *diagn,
 	x, *incx);
      else {
         LDA = *k+2;
-        A=(CBLAS_TEST_COMPLEX *)malloc((*n+*k)*LDA*sizeof(CBLAS_TEST_COMPLEX));
+        A=(CBLAS_TEST_COMPLEX *)malloc((*n+*k)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX));
         if (uplo == CblasUpper) {
            for( i=0; i<*k; i++ ){
               irow=*k-i;
@@ -375,7 +375,7 @@ void F77_ctbsv(int *order, char *uplow, char *transp, char *diagn,
 	         *incx);
      else {
         LDA = *k+2;
-        A=(CBLAS_TEST_COMPLEX*)malloc((*n+*k)*LDA*sizeof(CBLAS_TEST_COMPLEX ));
+        A=(CBLAS_TEST_COMPLEX*)malloc((*n+*k)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX ));
         if (uplo == CblasUpper) {
            for( i=0; i<*k; i++ ){
               irow=*k-i;
@@ -436,8 +436,8 @@ void F77_ctpmv(int *order, char *uplow, char *transp, char *diagn,
         cblas_ctpmv( CblasRowMajor, UNDEFINED, trans, diag, *n, ap, x, *incx );
      else {
         LDA = *n;
-        A=(CBLAS_TEST_COMPLEX*)malloc(LDA*LDA*sizeof(CBLAS_TEST_COMPLEX));
-        AP=(CBLAS_TEST_COMPLEX*)malloc((((LDA+1)*LDA)/2)*
+        A=(CBLAS_TEST_COMPLEX*)malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_COMPLEX));
+        AP=(CBLAS_TEST_COMPLEX*)malloc(((((size_t)LDA+1)*LDA)/2)*
 	 	sizeof(CBLAS_TEST_COMPLEX));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -491,8 +491,8 @@ void F77_ctpsv(int *order, char *uplow, char *transp, char *diagn,
         cblas_ctpsv( CblasRowMajor, UNDEFINED, trans, diag, *n, ap, x, *incx );
      else {
         LDA = *n;
-        A=(CBLAS_TEST_COMPLEX*)malloc(LDA*LDA*sizeof(CBLAS_TEST_COMPLEX));
-        AP=(CBLAS_TEST_COMPLEX*)malloc((((LDA+1)*LDA)/2)*
+        A=(CBLAS_TEST_COMPLEX*)malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_COMPLEX));
+        AP=(CBLAS_TEST_COMPLEX*)malloc(((((size_t)LDA+1)*LDA)/2)*
 		sizeof(CBLAS_TEST_COMPLEX));
      	if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -544,7 +544,7 @@ void F77_ctrmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA=*n+1;
-     A=(CBLAS_TEST_COMPLEX*)malloc((*n)*LDA*sizeof(CBLAS_TEST_COMPLEX));
+     A=(CBLAS_TEST_COMPLEX*)malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX));
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ ) {
 	  A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -573,7 +573,7 @@ void F77_ctrsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A =(CBLAS_TEST_COMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
+     A =(CBLAS_TEST_COMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ) {
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -601,8 +601,8 @@ void F77_chpr(int *order, char *uplow, int *n, float *alpha,
         cblas_chpr(CblasRowMajor, UNDEFINED, *n, *alpha, x, *incx, ap );
      else {
         LDA = *n;
-        A = (CBLAS_TEST_COMPLEX* )malloc(LDA*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
-        AP = ( CBLAS_TEST_COMPLEX* )malloc( (((LDA+1)*LDA)/2)*
+        A = (CBLAS_TEST_COMPLEX* )malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
+        AP = ( CBLAS_TEST_COMPLEX* )malloc( ((((size_t)LDA+1)*LDA)/2)*
 		sizeof( CBLAS_TEST_COMPLEX ));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -678,8 +678,8 @@ void F77_chpr2(int *order, char *uplow, int *n, CBLAS_TEST_COMPLEX *alpha,
 		     *incy, ap );
      else {
         LDA = *n;
-        A=(CBLAS_TEST_COMPLEX*)malloc( LDA*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
-        AP=(CBLAS_TEST_COMPLEX*)malloc( (((LDA+1)*LDA)/2)*
+        A=(CBLAS_TEST_COMPLEX*)malloc( (size_t)LDA*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
+        AP=(CBLAS_TEST_COMPLEX*)malloc( ((((size_t)LDA+1)*LDA)/2)*
 	sizeof( CBLAS_TEST_COMPLEX ));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -750,7 +750,7 @@ void F77_cher(int *order, char *uplow, int *n, float *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A=(CBLAS_TEST_COMPLEX*)malloc((*n)*LDA*sizeof( CBLAS_TEST_COMPLEX ));
+     A=(CBLAS_TEST_COMPLEX*)malloc((*n)*(size_t)LDA*sizeof( CBLAS_TEST_COMPLEX ));
 
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ ) {
@@ -784,7 +784,7 @@ void F77_cher2(int *order, char *uplow, int *n, CBLAS_TEST_COMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A= ( CBLAS_TEST_COMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_COMPLEX ) );
+     A= ( CBLAS_TEST_COMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_COMPLEX ) );
 
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ ) {

ctest/c_dblas2.c

@@ -19,7 +19,7 @@ void F77_dgemv(int *order, char *transp, int *m, int *n, double *alpha,
   get_transpose_type(transp, &trans);
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*m)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*m)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -43,7 +43,7 @@ void F77_dger(int *order, int *m, int *n, double *alpha, double *x, int *incx,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*m)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*m)*(size_t)LDA*sizeof( double ) );
 
      for( i=0; i<*m; i++ ) {
        for( j=0; j<*n; j++ )
@@ -74,7 +74,7 @@ void F77_dtrmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ )
          A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -102,7 +102,7 @@ void F77_dtrsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -123,7 +123,7 @@ void F77_dsymv(int *order, char *uplow, int *n, double *alpha, double *a,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -146,7 +146,7 @@ void F77_dsyr(int *order, char *uplow, int *n, double *alpha, double *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -170,7 +170,7 @@ void F77_dsyr2(int *order, char *uplow, int *n, double *alpha, double *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -196,7 +196,7 @@ void F77_dgbmv(int *order, char *transp, int *m, int *n, int *kl, int *ku,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *ku+*kl+2;
-     A   = ( double* )malloc( (*n+*kl)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n+*kl)*(size_t)LDA*sizeof( double ) );
      for( i=0; i<*ku; i++ ){
         irow=*ku+*kl-i;
         jcol=(*ku)-i;
@@ -236,7 +236,7 @@ void F77_dtbmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *k+1;
-     A = ( double* )malloc( (*n+*k)*LDA*sizeof( double ) );
+     A = ( double* )malloc( (*n+*k)*(size_t)LDA*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( i=0; i<*k; i++ ){
            irow=*k-i;
@@ -282,7 +282,7 @@ void F77_dtbsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *k+1;
-     A = ( double* )malloc( (*n+*k)*LDA*sizeof( double ) );
+     A = ( double* )malloc( (*n+*k)*(size_t)LDA*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( i=0; i<*k; i++ ){
         irow=*k-i;
@@ -325,7 +325,7 @@ void F77_dsbmv(int *order, char *uplow, int *n, int *k, double *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *k+1;
-     A   = ( double* )malloc( (*n+*k)*LDA*sizeof( double ) );
+     A   = ( double* )malloc( (*n+*k)*(size_t)LDA*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( i=0; i<*k; i++ ){
            irow=*k-i;
@@ -369,8 +369,8 @@ void F77_dspmv(int *order, char *uplow, int *n, double *alpha, double *ap,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( double* )malloc( LDA*LDA*sizeof( double ) );
-     AP  = ( double* )malloc( (((LDA+1)*LDA)/2)*sizeof( double ) );
+     A   = ( double* )malloc( (size_t)LDA*LDA*sizeof( double ) );
+     AP  = ( double* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -411,8 +411,8 @@ void F77_dtpmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( double* )malloc( LDA*LDA*sizeof( double ) );
-     AP  = ( double* )malloc( (((LDA+1)*LDA)/2)*sizeof( double ) );
+     A   = ( double* )malloc( (size_t)LDA*LDA*sizeof( double ) );
+     AP  = ( double* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -451,8 +451,8 @@ void F77_dtpsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( double* )malloc( LDA*LDA*sizeof( double ) );
-     AP  = ( double* )malloc( (((LDA+1)*LDA)/2)*sizeof( double ) );
+     A   = ( double* )malloc( (size_t)LDA*LDA*sizeof( double ) );
+     AP  = ( double* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -488,8 +488,8 @@ void F77_dspr(int *order, char *uplow, int *n, double *alpha, double *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( double* )malloc( LDA*LDA*sizeof( double ) );
-     AP  = ( double* )malloc( (((LDA+1)*LDA)/2)*sizeof( double ) );
+     A   = ( double* )malloc( (size_t)LDA*LDA*sizeof( double ) );
+     AP  = ( double* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -540,8 +540,8 @@ void F77_dspr2(int *order, char *uplow, int *n, double *alpha, double *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( double* )malloc( LDA*LDA*sizeof( double ) );
-     AP  = ( double* )malloc( (((LDA+1)*LDA)/2)*sizeof( double ) );
+     A   = ( double* )malloc( (size_t)LDA*LDA*sizeof( double ) );
+     AP  = ( double* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( double ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )

ctest/c_dblas3.c

@@ -26,34 +26,34 @@ void F77_dgemm(int *order, char *transpa, char *transpb, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (transa == CblasNoTrans) {
         LDA = *k+1;
-        A = (double *)malloc( (*m)*LDA*sizeof( double ) );
+        A = (double *)malloc( (*m)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*k; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else {
         LDA = *m+1;
-        A   = ( double* )malloc( LDA*(*k)*sizeof( double ) );
+        A   = ( double* )malloc( (size_t)LDA*(*k)*sizeof( double ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      if (transb == CblasNoTrans) {
         LDB = *n+1;
-        B   = ( double* )malloc( (*k)*LDB*sizeof( double ) );
+        B   = ( double* )malloc( (*k)*(size_t)LDB*sizeof( double ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ )
               B[i*LDB+j]=b[j*(*ldb)+i];
      }
      else {
         LDB = *k+1;
-        B   = ( double* )malloc( LDB*(*n)*sizeof( double ) );
+        B   = ( double* )malloc( (size_t)LDB*(*n)*sizeof( double ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ )
               B[i*LDB+j]=b[j*(*ldb)+i];
      }
      LDC = *n+1;
-     C   = ( double* )malloc( (*m)*LDC*sizeof( double ) );
+     C   = ( double* )malloc( (*m)*(size_t)LDC*sizeof( double ) );
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -89,25 +89,25 @@ void F77_dsymm(int *order, char *rtlf, char *uplow, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A   = ( double* )malloc( (*m)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*m)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B   = ( double* )malloc( (*m)*LDB*sizeof( double ) );
+     B   = ( double* )malloc( (*m)*(size_t)LDB*sizeof( double ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];
      LDC = *n+1;
-     C   = ( double* )malloc( (*m)*LDC*sizeof( double ) );
+     C   = ( double* )malloc( (*m)*(size_t)LDC*sizeof( double ) );
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -143,20 +143,20 @@ void F77_dsyrk(int *order, char *uplow, char *transp, int *n, int *k,
   if (*order == TEST_ROW_MJR) {
      if (trans == CblasNoTrans) {
         LDA = *k+1;
-        A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( double* )malloc( (*k)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*k)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDC = *n+1;
-     C   = ( double* )malloc( (*n)*LDC*sizeof( double ) );
+     C   = ( double* )malloc( (*n)*(size_t)LDC*sizeof( double ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -191,8 +191,8 @@ void F77_dsyr2k(int *order, char *uplow, char *transp, int *n, int *k,
      if (trans == CblasNoTrans) {
         LDA = *k+1;
         LDB = *k+1;
-        A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
-        B   = ( double* )malloc( (*n)*LDB*sizeof( double ) );
+        A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
+        B   = ( double* )malloc( (*n)*(size_t)LDB*sizeof( double ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j]=a[j*(*lda)+i];
@@ -202,8 +202,8 @@ void F77_dsyr2k(int *order, char *uplow, char *transp, int *n, int *k,
      else {
         LDA = *n+1;
         LDB = *n+1;
-        A   = ( double* )malloc( LDA*(*k)*sizeof( double ) );
-        B   = ( double* )malloc( LDB*(*k)*sizeof( double ) );
+        A   = ( double* )malloc( (size_t)LDA*(*k)*sizeof( double ) );
+        B   = ( double* )malloc( (size_t)LDB*(*k)*sizeof( double ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ){
               A[i*LDA+j]=a[j*(*lda)+i];
@@ -211,7 +211,7 @@ void F77_dsyr2k(int *order, char *uplow, char *transp, int *n, int *k,
            }
      }
      LDC = *n+1;
-     C   = ( double* )malloc( (*n)*LDC*sizeof( double ) );
+     C   = ( double* )malloc( (*n)*(size_t)LDC*sizeof( double ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -249,20 +249,20 @@ void F77_dtrmm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A   = ( double* )malloc( (*m)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*m)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B   = ( double* )malloc( (*m)*LDB*sizeof( double ) );
+     B   = ( double* )malloc( (*m)*(size_t)LDB*sizeof( double ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];
@@ -300,20 +300,20 @@ void F77_dtrsm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A   = ( double* )malloc( (*m)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*m)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( double* )malloc( (*n)*LDA*sizeof( double ) );
+        A   = ( double* )malloc( (*n)*(size_t)LDA*sizeof( double ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B   = ( double* )malloc( (*m)*LDB*sizeof( double ) );
+     B   = ( double* )malloc( (*m)*(size_t)LDB*sizeof( double ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];

ctest/c_sblas2.c

@@ -19,7 +19,7 @@ void F77_sgemv(int *order, char *transp, int *m, int *n, float *alpha,
   get_transpose_type(transp, &trans);
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*m)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*m)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -43,7 +43,7 @@ void F77_sger(int *order, int *m, int *n, float *alpha, float *x, int *incx,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*m)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*m)*(size_t)LDA*sizeof( float ) );
 
      for( i=0; i<*m; i++ ) {
        for( j=0; j<*n; j++ )
@@ -74,7 +74,7 @@ void F77_strmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ )
          A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -102,7 +102,7 @@ void F77_strsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -123,7 +123,7 @@ void F77_ssymv(int *order, char *uplow, int *n, float *alpha, float *a,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -146,7 +146,7 @@ void F77_ssyr(int *order, char *uplow, int *n, float *alpha, float *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -170,7 +170,7 @@ void F77_ssyr2(int *order, char *uplow, int *n, float *alpha, float *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            A[ LDA*i+j ]=a[ (*lda)*j+i ];
@@ -196,7 +196,7 @@ void F77_sgbmv(int *order, char *transp, int *m, int *n, int *kl, int *ku,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *ku+*kl+2;
-     A   = ( float* )malloc( (*n+*kl)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n+*kl)*(size_t)LDA*sizeof( float ) );
      for( i=0; i<*ku; i++ ){
         irow=*ku+*kl-i;
         jcol=(*ku)-i;
@@ -236,7 +236,7 @@ void F77_stbmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *k+1;
-     A = ( float* )malloc( (*n+*k)*LDA*sizeof( float ) );
+     A = ( float* )malloc( (*n+*k)*(size_t)LDA*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( i=0; i<*k; i++ ){
            irow=*k-i;
@@ -282,7 +282,7 @@ void F77_stbsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *k+1;
-     A = ( float* )malloc( (*n+*k)*LDA*sizeof( float ) );
+     A = ( float* )malloc( (*n+*k)*(size_t)LDA*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( i=0; i<*k; i++ ){
         irow=*k-i;
@@ -325,7 +325,7 @@ void F77_ssbmv(int *order, char *uplow, int *n, int *k, float *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *k+1;
-     A   = ( float* )malloc( (*n+*k)*LDA*sizeof( float ) );
+     A   = ( float* )malloc( (*n+*k)*(size_t)LDA*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( i=0; i<*k; i++ ){
            irow=*k-i;
@@ -369,8 +369,8 @@ void F77_sspmv(int *order, char *uplow, int *n, float *alpha, float *ap,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( float* )malloc( LDA*LDA*sizeof( float ) );
-     AP  = ( float* )malloc( (((LDA+1)*LDA)/2)*sizeof( float ) );
+     A   = ( float* )malloc( (size_t)LDA*LDA*sizeof( float ) );
+     AP  = ( float* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -410,8 +410,8 @@ void F77_stpmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( float* )malloc( LDA*LDA*sizeof( float ) );
-     AP  = ( float* )malloc( (((LDA+1)*LDA)/2)*sizeof( float ) );
+     A   = ( float* )malloc( (size_t)LDA*LDA*sizeof( float ) );
+     AP  = ( float* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -449,8 +449,8 @@ void F77_stpsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( float* )malloc( LDA*LDA*sizeof( float ) );
-     AP  = ( float* )malloc( (((LDA+1)*LDA)/2)*sizeof( float ) );
+     A   = ( float* )malloc( (size_t)LDA*LDA*sizeof( float ) );
+     AP  = ( float* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -485,8 +485,8 @@ void F77_sspr(int *order, char *uplow, int *n, float *alpha, float *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( float* )malloc( LDA*LDA*sizeof( float ) );
-     AP  = ( float* )malloc( (((LDA+1)*LDA)/2)*sizeof( float ) );
+     A   = ( float* )malloc( (size_t)LDA*LDA*sizeof( float ) );
+     AP  = ( float* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )
@@ -536,8 +536,8 @@ void F77_sspr2(int *order, char *uplow, int *n, float *alpha, float *x,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n;
-     A   = ( float* )malloc( LDA*LDA*sizeof( float ) );
-     AP  = ( float* )malloc( (((LDA+1)*LDA)/2)*sizeof( float ) );
+     A   = ( float* )malloc( (size_t)LDA*LDA*sizeof( float ) );
+     AP  = ( float* )malloc( ((((size_t)LDA+1)*LDA)/2)*sizeof( float ) );
      if (uplo == CblasUpper) {
         for( j=0, k=0; j<*n; j++ )
            for( i=0; i<j+1; i++, k++ )

ctest/c_sblas3.c

@@ -23,34 +23,34 @@ void F77_sgemm(int *order, char *transpa, char *transpb, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (transa == CblasNoTrans) {
         LDA = *k+1;
-        A = (float *)malloc( (*m)*LDA*sizeof( float ) );
+        A = (float *)malloc( (*m)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*k; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else {
         LDA = *m+1;
-        A   = ( float* )malloc( LDA*(*k)*sizeof( float ) );
+        A   = ( float* )malloc( (size_t)LDA*(*k)*sizeof( float ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      if (transb == CblasNoTrans) {
         LDB = *n+1;
-        B   = ( float* )malloc( (*k)*LDB*sizeof( float ) );
+        B   = ( float* )malloc( (*k)*(size_t)LDB*sizeof( float ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ )
               B[i*LDB+j]=b[j*(*ldb)+i];
      }
      else {
         LDB = *k+1;
-        B   = ( float* )malloc( LDB*(*n)*sizeof( float ) );
+        B   = ( float* )malloc( (size_t)LDB*(*n)*sizeof( float ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ )
               B[i*LDB+j]=b[j*(*ldb)+i];
      }
      LDC = *n+1;
-     C   = ( float* )malloc( (*m)*LDC*sizeof( float ) );
+     C   = ( float* )malloc( (*m)*(size_t)LDC*sizeof( float ) );
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -85,25 +85,25 @@ void F77_ssymm(int *order, char *rtlf, char *uplow, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A   = ( float* )malloc( (*m)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*m)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B   = ( float* )malloc( (*m)*LDB*sizeof( float ) );
+     B   = ( float* )malloc( (*m)*(size_t)LDB*sizeof( float ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];
      LDC = *n+1;
-     C   = ( float* )malloc( (*m)*LDC*sizeof( float ) );
+     C   = ( float* )malloc( (*m)*(size_t)LDC*sizeof( float ) );
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -139,20 +139,20 @@ void F77_ssyrk(int *order, char *uplow, char *transp, int *n, int *k,
   if (*order == TEST_ROW_MJR) {
      if (trans == CblasNoTrans) {
         LDA = *k+1;
-        A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( float* )malloc( (*k)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*k)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDC = *n+1;
-     C   = ( float* )malloc( (*n)*LDC*sizeof( float ) );
+     C   = ( float* )malloc( (*n)*(size_t)LDC*sizeof( float ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -187,8 +187,8 @@ void F77_ssyr2k(int *order, char *uplow, char *transp, int *n, int *k,
      if (trans == CblasNoTrans) {
         LDA = *k+1;
         LDB = *k+1;
-        A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
-        B   = ( float* )malloc( (*n)*LDB*sizeof( float ) );
+        A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
+        B   = ( float* )malloc( (*n)*(size_t)LDB*sizeof( float ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j]=a[j*(*lda)+i];
@@ -198,8 +198,8 @@ void F77_ssyr2k(int *order, char *uplow, char *transp, int *n, int *k,
      else {
         LDA = *n+1;
         LDB = *n+1;
-        A   = ( float* )malloc( LDA*(*k)*sizeof( float ) );
-        B   = ( float* )malloc( LDB*(*k)*sizeof( float ) );
+        A   = ( float* )malloc( (size_t)LDA*(*k)*sizeof( float ) );
+        B   = ( float* )malloc( (size_t)LDB*(*k)*sizeof( float ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ){
               A[i*LDA+j]=a[j*(*lda)+i];
@@ -207,7 +207,7 @@ void F77_ssyr2k(int *order, char *uplow, char *transp, int *n, int *k,
            }
      }
      LDC = *n+1;
-     C   = ( float* )malloc( (*n)*LDC*sizeof( float ) );
+     C   = ( float* )malloc( (*n)*(size_t)LDC*sizeof( float ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -245,20 +245,20 @@ void F77_strmm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A   = ( float* )malloc( (*m)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*m)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B   = ( float* )malloc( (*m)*LDB*sizeof( float ) );
+     B   = ( float* )malloc( (*m)*(size_t)LDB*sizeof( float ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];
@@ -296,20 +296,20 @@ void F77_strsm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A   = ( float* )malloc( (*m)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*m)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A   = ( float* )malloc( (*n)*LDA*sizeof( float ) );
+        A   = ( float* )malloc( (*n)*(size_t)LDA*sizeof( float ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B   = ( float* )malloc( (*m)*LDB*sizeof( float ) );
+     B   = ( float* )malloc( (*m)*(size_t)LDB*sizeof( float ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];

ctest/c_zblas2.c

@@ -20,7 +20,7 @@ void F77_zgemv(int *order, char *transp, int *m, int *n,
   get_transpose_type(transp, &trans);
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A  = (CBLAS_TEST_ZOMPLEX *)malloc( (*m)*LDA*sizeof( CBLAS_TEST_ZOMPLEX) );
+     A  = (CBLAS_TEST_ZOMPLEX *)malloc( (*m)*(size_t)LDA*sizeof( CBLAS_TEST_ZOMPLEX) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -50,7 +50,7 @@ void F77_zgbmv(int *order, char *transp, int *m, int *n, int *kl, int *ku,
   get_transpose_type(transp, &trans);
   if (*order == TEST_ROW_MJR) {
      LDA = *ku+*kl+2;
-     A=( CBLAS_TEST_ZOMPLEX* )malloc((*n+*kl)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+     A=( CBLAS_TEST_ZOMPLEX* )malloc((*n+*kl)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*ku; i++ ){
         irow=*ku+*kl-i;
         jcol=(*ku)-i;
@@ -94,7 +94,7 @@ void F77_zgeru(int *order, int *m, int *n, CBLAS_TEST_ZOMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A=(CBLAS_TEST_ZOMPLEX*)malloc((*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+     A=(CBLAS_TEST_ZOMPLEX*)malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -122,7 +122,7 @@ void F77_zgerc(int *order, int *m, int *n, CBLAS_TEST_ZOMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     A=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -154,7 +154,7 @@ void F77_zhemv(int *order, char *uplow, int *n, CBLAS_TEST_ZOMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A = (CBLAS_TEST_ZOMPLEX *)malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+     A = (CBLAS_TEST_ZOMPLEX *)malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ){
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -190,7 +190,7 @@ int i,irow,j,jcol,LDA;
 		 *incx, beta, y, *incy );
      else {
         LDA = *k+2;
-        A =(CBLAS_TEST_ZOMPLEX*)malloc((*n+*k)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A =(CBLAS_TEST_ZOMPLEX*)malloc((*n+*k)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         if (uplo == CblasUpper) {
            for( i=0; i<*k; i++ ){
               irow=*k-i;
@@ -251,8 +251,8 @@ void F77_zhpmv(int *order, char *uplow, int *n, CBLAS_TEST_ZOMPLEX *alpha,
 	         beta, y, *incy);
      else {
         LDA = *n;
-        A = (CBLAS_TEST_ZOMPLEX* )malloc(LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX ));
-        AP = (CBLAS_TEST_ZOMPLEX* )malloc( (((LDA+1)*LDA)/2)*
+        A = (CBLAS_TEST_ZOMPLEX* )malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX ));
+        AP = (CBLAS_TEST_ZOMPLEX* )malloc( ((((size_t)LDA+1)*LDA)/2)*
 	        sizeof( CBLAS_TEST_ZOMPLEX ));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -311,7 +311,7 @@ void F77_ztbmv(int *order, char *uplow, char *transp, char *diagn,
 	x, *incx);
      else {
         LDA = *k+2;
-        A=(CBLAS_TEST_ZOMPLEX *)malloc((*n+*k)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX *)malloc((*n+*k)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         if (uplo == CblasUpper) {
            for( i=0; i<*k; i++ ){
               irow=*k-i;
@@ -375,7 +375,7 @@ void F77_ztbsv(int *order, char *uplow, char *transp, char *diagn,
 	         *incx);
      else {
         LDA = *k+2;
-        A=(CBLAS_TEST_ZOMPLEX*)malloc((*n+*k)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ));
+        A=(CBLAS_TEST_ZOMPLEX*)malloc((*n+*k)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ));
         if (uplo == CblasUpper) {
            for( i=0; i<*k; i++ ){
               irow=*k-i;
@@ -436,8 +436,8 @@ void F77_ztpmv(int *order, char *uplow, char *transp, char *diagn,
         cblas_ztpmv( CblasRowMajor, UNDEFINED, trans, diag, *n, ap, x, *incx );
      else {
         LDA = *n;
-        A=(CBLAS_TEST_ZOMPLEX*)malloc(LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
-        AP=(CBLAS_TEST_ZOMPLEX*)malloc((((LDA+1)*LDA)/2)*
+        A=(CBLAS_TEST_ZOMPLEX*)malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        AP=(CBLAS_TEST_ZOMPLEX*)malloc(((((size_t)LDA+1)*LDA)/2)*
 	 	sizeof(CBLAS_TEST_ZOMPLEX));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -491,8 +491,8 @@ void F77_ztpsv(int *order, char *uplow, char *transp, char *diagn,
         cblas_ztpsv( CblasRowMajor, UNDEFINED, trans, diag, *n, ap, x, *incx );
      else {
         LDA = *n;
-        A=(CBLAS_TEST_ZOMPLEX*)malloc(LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
-        AP=(CBLAS_TEST_ZOMPLEX*)malloc((((LDA+1)*LDA)/2)*
+        A=(CBLAS_TEST_ZOMPLEX*)malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        AP=(CBLAS_TEST_ZOMPLEX*)malloc(((((size_t)LDA+1)*LDA)/2)*
 		sizeof(CBLAS_TEST_ZOMPLEX));
      	if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -544,7 +544,7 @@ void F77_ztrmv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA=*n+1;
-     A=(CBLAS_TEST_ZOMPLEX*)malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+     A=(CBLAS_TEST_ZOMPLEX*)malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ ) {
 	  A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -573,7 +573,7 @@ void F77_ztrsv(int *order, char *uplow, char *transp, char *diagn,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A =(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     A =(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ) {
            A[ LDA*i+j ].real=a[ (*lda)*j+i ].real;
@@ -601,8 +601,8 @@ void F77_zhpr(int *order, char *uplow, int *n, double *alpha,
         cblas_zhpr(CblasRowMajor, UNDEFINED, *n, *alpha, x, *incx, ap );
      else {
         LDA = *n;
-        A = (CBLAS_TEST_ZOMPLEX* )malloc(LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
-        AP = ( CBLAS_TEST_ZOMPLEX* )malloc( (((LDA+1)*LDA)/2)*
+        A = (CBLAS_TEST_ZOMPLEX* )malloc((size_t)LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        AP = ( CBLAS_TEST_ZOMPLEX* )malloc( ((((size_t)LDA+1)*LDA)/2)*
 		sizeof( CBLAS_TEST_ZOMPLEX ));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -678,8 +678,8 @@ void F77_zhpr2(int *order, char *uplow, int *n, CBLAS_TEST_ZOMPLEX *alpha,
 		     *incy, ap );
      else {
         LDA = *n;
-        A=(CBLAS_TEST_ZOMPLEX*)malloc( LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
-        AP=(CBLAS_TEST_ZOMPLEX*)malloc( (((LDA+1)*LDA)/2)*
+        A=(CBLAS_TEST_ZOMPLEX*)malloc( (size_t)LDA*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        AP=(CBLAS_TEST_ZOMPLEX*)malloc( ((((size_t)LDA+1)*LDA)/2)*
 	sizeof( CBLAS_TEST_ZOMPLEX ));
         if (uplo == CblasUpper) {
            for( j=0, k=0; j<*n; j++ )
@@ -750,7 +750,7 @@ void F77_zher(int *order, char *uplow, int *n, double *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A=(CBLAS_TEST_ZOMPLEX*)malloc((*n)*LDA*sizeof( CBLAS_TEST_ZOMPLEX ));
+     A=(CBLAS_TEST_ZOMPLEX*)malloc((*n)*(size_t)LDA*sizeof( CBLAS_TEST_ZOMPLEX ));
 
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ ) {
@@ -784,7 +784,7 @@ void F77_zher2(int *order, char *uplow, int *n, CBLAS_TEST_ZOMPLEX *alpha,
 
   if (*order == TEST_ROW_MJR) {
      LDA = *n+1;
-     A= ( CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     A= ( CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
 
      for( i=0; i<*n; i++ )
        for( j=0; j<*n; j++ ) {

ctest/c_zblas3.c

@@ -26,7 +26,7 @@ void F77_zgemm(int *order, char *transpa, char *transpb, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (transa == CblasNoTrans) {
         LDA = *k+1;
-        A=(CBLAS_TEST_ZOMPLEX*)malloc((*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX*)malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*m; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -35,7 +35,7 @@ void F77_zgemm(int *order, char *transpa, char *transpb, int *m, int *n,
      }
      else {
         LDA = *m+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc(LDA*(*k)*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((size_t)LDA*(*k)*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*k; i++ )
            for( j=0; j<*m; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -45,7 +45,7 @@ void F77_zgemm(int *order, char *transpa, char *transpb, int *m, int *n,
 
      if (transb == CblasNoTrans) {
         LDB = *n+1;
-        B=(CBLAS_TEST_ZOMPLEX* )malloc((*k)*LDB*sizeof(CBLAS_TEST_ZOMPLEX) );
+        B=(CBLAS_TEST_ZOMPLEX* )malloc((*k)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ) {
               B[i*LDB+j].real=b[j*(*ldb)+i].real;
@@ -54,7 +54,7 @@ void F77_zgemm(int *order, char *transpa, char *transpb, int *m, int *n,
      }
      else {
         LDB = *k+1;
-        B=(CBLAS_TEST_ZOMPLEX* )malloc(LDB*(*n)*sizeof(CBLAS_TEST_ZOMPLEX));
+        B=(CBLAS_TEST_ZOMPLEX* )malloc((size_t)LDB*(*n)*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               B[i*LDB+j].real=b[j*(*ldb)+i].real;
@@ -63,7 +63,7 @@ void F77_zgemm(int *order, char *transpa, char *transpb, int *m, int *n,
      }
 
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDC*sizeof(CBLAS_TEST_ZOMPLEX));
+     C=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX));
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ ) {
            C[i*LDC+j].real=c[j*(*ldc)+i].real;
@@ -103,7 +103,7 @@ void F77_zhemm(int *order, char *rtlf, char *uplow, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A= (CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A= (CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -112,7 +112,7 @@ void F77_zhemm(int *order, char *rtlf, char *uplow, int *m, int *n,
      }
      else{
         LDA = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -120,14 +120,14 @@ void F77_zhemm(int *order, char *rtlf, char *uplow, int *m, int *n,
            }
      }
      LDB = *n+1;
-     B=(CBLAS_TEST_ZOMPLEX* )malloc( (*m)*LDB*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     B=(CBLAS_TEST_ZOMPLEX* )malloc( (*m)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ) {
            B[i*LDB+j].real=b[j*(*ldb)+i].real;
            B[i*LDB+j].imag=b[j*(*ldb)+i].imag;
         }
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     C=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ ) {
            C[i*LDC+j].real=c[j*(*ldc)+i].real;
@@ -167,25 +167,25 @@ void F77_zsymm(int *order, char *rtlf, char *uplow, int *m, int *n,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      else{
         LDA = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ )
               A[i*LDA+j]=a[j*(*lda)+i];
      }
      LDB = *n+1;
-     B=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDB*sizeof(CBLAS_TEST_ZOMPLEX ));
+     B=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX ));
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ )
            B[i*LDB+j]=b[j*(*ldb)+i];
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDC*sizeof(CBLAS_TEST_ZOMPLEX));
+     C=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX));
      for( j=0; j<*n; j++ )
         for( i=0; i<*m; i++ )
            C[i*LDC+j]=c[j*(*ldc)+i];
@@ -221,7 +221,7 @@ void F77_zherk(int *order, char *uplow, char *transp, int *n, int *k,
   if (*order == TEST_ROW_MJR) {
      if (trans == CblasNoTrans) {
         LDA = *k+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -230,7 +230,7 @@ void F77_zherk(int *order, char *uplow, char *transp, int *n, int *k,
      }
      else{
         LDA = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*k)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*k)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -238,7 +238,7 @@ void F77_zherk(int *order, char *uplow, char *transp, int *n, int *k,
            }
      }
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     C=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ) {
            C[i*LDC+j].real=c[j*(*ldc)+i].real;
@@ -277,7 +277,7 @@ void F77_zsyrk(int *order, char *uplow, char *transp, int *n, int *k,
   if (*order == TEST_ROW_MJR) {
      if (trans == CblasNoTrans) {
         LDA = *k+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -286,7 +286,7 @@ void F77_zsyrk(int *order, char *uplow, char *transp, int *n, int *k,
      }
      else{
         LDA = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*k)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*k)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -294,7 +294,7 @@ void F77_zsyrk(int *order, char *uplow, char *transp, int *n, int *k,
            }
      }
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     C=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ) {
            C[i*LDC+j].real=c[j*(*ldc)+i].real;
@@ -333,8 +333,8 @@ void F77_zher2k(int *order, char *uplow, char *transp, int *n, int *k,
      if (trans == CblasNoTrans) {
         LDA = *k+1;
         LDB = *k+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ));
-        B=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDB*sizeof(CBLAS_TEST_ZOMPLEX ));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ));
+        B=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX ));
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -346,8 +346,8 @@ void F77_zher2k(int *order, char *uplow, char *transp, int *n, int *k,
      else {
         LDA = *n+1;
         LDB = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc( LDA*(*k)*sizeof(CBLAS_TEST_ZOMPLEX ) );
-        B=(CBLAS_TEST_ZOMPLEX* )malloc( LDB*(*k)*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc( (size_t)LDA*(*k)*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        B=(CBLAS_TEST_ZOMPLEX* )malloc( (size_t)LDB*(*k)*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ){
 	      A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -357,7 +357,7 @@ void F77_zher2k(int *order, char *uplow, char *transp, int *n, int *k,
            }
      }
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc( (*n)*LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
+     C=(CBLAS_TEST_ZOMPLEX* )malloc( (*n)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX ) );
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ) {
            C[i*LDC+j].real=c[j*(*ldc)+i].real;
@@ -397,8 +397,8 @@ void F77_zsyr2k(int *order, char *uplow, char *transp, int *n, int *k,
      if (trans == CblasNoTrans) {
         LDA = *k+1;
         LDB = *k+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
-        B=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDB*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        B=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*n; i++ )
            for( j=0; j<*k; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -410,8 +410,8 @@ void F77_zsyr2k(int *order, char *uplow, char *transp, int *n, int *k,
      else {
         LDA = *n+1;
         LDB = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc(LDA*(*k)*sizeof(CBLAS_TEST_ZOMPLEX));
-        B=(CBLAS_TEST_ZOMPLEX* )malloc(LDB*(*k)*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((size_t)LDA*(*k)*sizeof(CBLAS_TEST_ZOMPLEX));
+        B=(CBLAS_TEST_ZOMPLEX* )malloc((size_t)LDB*(*k)*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*k; i++ )
            for( j=0; j<*n; j++ ){
 	      A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -421,7 +421,7 @@ void F77_zsyr2k(int *order, char *uplow, char *transp, int *n, int *k,
            }
      }
      LDC = *n+1;
-     C=(CBLAS_TEST_ZOMPLEX* )malloc( (*n)*LDC*sizeof(CBLAS_TEST_ZOMPLEX));
+     C=(CBLAS_TEST_ZOMPLEX* )malloc( (*n)*(size_t)LDC*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*n; i++ )
         for( j=0; j<*n; j++ ) {
            C[i*LDC+j].real=c[j*(*ldc)+i].real;
@@ -463,7 +463,7 @@ void F77_ztrmm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -472,7 +472,7 @@ void F77_ztrmm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
      }
      else{
         LDA = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -480,7 +480,7 @@ void F77_ztrmm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
            }
      }
      LDB = *n+1;
-     B=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDB*sizeof(CBLAS_TEST_ZOMPLEX));
+     B=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ) {
            B[i*LDB+j].real=b[j*(*ldb)+i].real;
@@ -522,7 +522,7 @@ void F77_ztrsm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
   if (*order == TEST_ROW_MJR) {
      if (side == CblasLeft) {
         LDA = *m+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc( (*m)*LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
+        A=(CBLAS_TEST_ZOMPLEX* )malloc( (*m)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX ) );
         for( i=0; i<*m; i++ )
            for( j=0; j<*m; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -531,7 +531,7 @@ void F77_ztrsm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
      }
      else{
         LDA = *n+1;
-        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*LDA*sizeof(CBLAS_TEST_ZOMPLEX));
+        A=(CBLAS_TEST_ZOMPLEX* )malloc((*n)*(size_t)LDA*sizeof(CBLAS_TEST_ZOMPLEX));
         for( i=0; i<*n; i++ )
            for( j=0; j<*n; j++ ) {
               A[i*LDA+j].real=a[j*(*lda)+i].real;
@@ -539,7 +539,7 @@ void F77_ztrsm(int *order, char *rtlf, char *uplow, char *transp, char *diagn,
            }
      }
      LDB = *n+1;
-     B=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*LDB*sizeof(CBLAS_TEST_ZOMPLEX));
+     B=(CBLAS_TEST_ZOMPLEX* )malloc((*m)*(size_t)LDB*sizeof(CBLAS_TEST_ZOMPLEX));
      for( i=0; i<*m; i++ )
         for( j=0; j<*n; j++ ) {
            B[i*LDB+j].real=b[j*(*ldb)+i].real;

ctest/constant.c

@@ -1,3 +1,4 @@
+#include "cblas_test.h"
 int CBLAS_CallFromC;
 int RowMajorStrg;
 

driver/level2/CMakeLists.txt

@@ -81,6 +81,7 @@ foreach (float_type ${FLOAT_TYPES})
     GenerateNamedObjects("gbmv_thread.c" "TRANSA" "gbmv_thread_t" false "" "" false ${float_type})
   endif ()
 
+# special defines for complex
   if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
 
     foreach (u_source ${U_SOURCES})
@@ -197,6 +198,13 @@ foreach (float_type ${FLOAT_TYPES})
   endif ()
 endforeach ()
 
+if (BUILD_BFLOAT16)
+  if (USE_THREAD)
+    GenerateNamedObjects("sbgemv_thread.c" "" "gemv_thread_n" false "" "" false "BFLOAT16")
+    GenerateNamedObjects("sbgemv_thread.c" "TRANSA" "gemv_thread_t" false "" "" false "BFLOAT16")
+  endif ()
+endif ()
+
 if ( BUILD_COMPLEX AND NOT  BUILD_SINGLE)
   if (USE_THREAD)
 	  GenerateNamedObjects("gemv_thread.c" "" "gemv_thread_n" false "" "" false "SINGLE")

driver/level2/Makefile

@@ -64,9 +64,9 @@ CBLASOBJS += \
 	chpmv_U.$(SUFFIX) chpmv_L.$(SUFFIX) chpmv_V.$(SUFFIX) chpmv_M.$(SUFFIX) \
 	chpr_U.$(SUFFIX)  chpr_L.$(SUFFIX)  chpr_V.$(SUFFIX)  chpr_M.$(SUFFIX) \
 	chpr2_U.$(SUFFIX) chpr2_L.$(SUFFIX) chpr2_V.$(SUFFIX) chpr2_M.$(SUFFIX) \
-	csbmv_U.$(SUFFIX) csbmv_L.$(SUFFIX) cspmv_U.$(SUFFIX) cspmv_L.$(SUFFIX) \
-	cspr_U.$(SUFFIX)  cspr_L.$(SUFFIX)  cspr2_U.$(SUFFIX) cspr2_L.$(SUFFIX) \
-	csyr_U.$(SUFFIX)  csyr_L.$(SUFFIX)  csyr2_U.$(SUFFIX) csyr2_L.$(SUFFIX) \
+	csbmv_U.$(SUFFIX) csbmv_L.$(SUFFIX) \
+	cspr2_U.$(SUFFIX) cspr2_L.$(SUFFIX) \
+	csyr2_U.$(SUFFIX) csyr2_L.$(SUFFIX) \
 	ctbmv_NUU.$(SUFFIX) ctbmv_NUN.$(SUFFIX) ctbmv_NLU.$(SUFFIX) ctbmv_NLN.$(SUFFIX) \
 	ctbmv_TUU.$(SUFFIX) ctbmv_TUN.$(SUFFIX) ctbmv_TLU.$(SUFFIX) ctbmv_TLN.$(SUFFIX) \
 	ctbmv_RUU.$(SUFFIX) ctbmv_RUN.$(SUFFIX) ctbmv_RLU.$(SUFFIX) ctbmv_RLN.$(SUFFIX) \
@@ -92,6 +92,13 @@ CBLASOBJS += \
 	ctrsv_RUU.$(SUFFIX) ctrsv_RUN.$(SUFFIX) ctrsv_RLU.$(SUFFIX) ctrsv_RLN.$(SUFFIX) \
 	ctrsv_CUU.$(SUFFIX) ctrsv_CUN.$(SUFFIX) ctrsv_CLU.$(SUFFIX) ctrsv_CLN.$(SUFFIX)
 
+ifndef NO_LAPACK
+CBLASOBJS += \
+	cspmv_U.$(SUFFIX) cspmv_L.$(SUFFIX) \
+	cspr_U.$(SUFFIX)  cspr_L.$(SUFFIX)  \
+	csyr_U.$(SUFFIX)  csyr_L.$(SUFFIX)  
+endif
+
 ZBLASOBJS += \
 	zgbmv_n.$(SUFFIX) zgbmv_t.$(SUFFIX) zgbmv_r.$(SUFFIX) zgbmv_c.$(SUFFIX) \
 	zgbmv_o.$(SUFFIX) zgbmv_u.$(SUFFIX) zgbmv_s.$(SUFFIX) zgbmv_d.$(SUFFIX) \

driver/level3/CMakeLists.txt

@@ -12,6 +12,12 @@ foreach (GEMM_DEFINE ${GEMM_DEFINES})
   if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
     GenerateNamedObjects("gemm.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm_thread_${GEMM_DEFINE_LC}" 0)
   endif ()
+  if (BUILD_BFLOAT16)
+    GenerateNamedObjects("gemm.c" "${GEMM_DEFINE}" "gemm_${GEMM_DEFINE_LC}" 0 "" "" false "BFLOAT16")
+    if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
+      GenerateNamedObjects("gemm.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm_thread_${GEMM_DEFINE_LC}" 0 "" "" false "BFLOAT16")
+    endif ()
+  endif ()
 endforeach ()
 
 if ( BUILD_COMPLEX16 AND NOT  BUILD_DOUBLE)

driver/level3/Makefile

@@ -425,7 +425,7 @@ cgemm_rr.$(SUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_rc.$(SUFFIX) : gemm.c level3.c  ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_cn.$(SUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -473,7 +473,7 @@ zgemm_rr.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_rc.$(SUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_cn.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -521,7 +521,7 @@ xgemm_rr.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_rc.$(SUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_cn.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -632,7 +632,7 @@ cgemm_thread_rr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_thread_rc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_thread_cn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -680,7 +680,7 @@ zgemm_thread_rr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_thread_rc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_thread_cn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -728,7 +728,7 @@ xgemm_thread_rr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_thread_rc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_thread_cn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -1895,7 +1895,7 @@ cgemm3m_rr.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_rc.$(SUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_cn.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -1943,7 +1943,7 @@ zgemm3m_rr.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_rc.$(SUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_cn.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -1991,7 +1991,7 @@ xgemm3m_rr.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_rc.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_cn.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2048,7 +2048,7 @@ cgemm3m_thread_rr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_thread_rc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_thread_cn.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2096,7 +2096,7 @@ zgemm3m_thread_rr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_thread_rc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_thread_cn.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2144,7 +2144,7 @@ xgemm3m_thread_rr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_thread_rc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_thread_cn.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2817,7 +2817,7 @@ cgemm_rr.$(PSUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_rc.$(PSUFFIX) : gemm.c level3.c  ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_cn.$(PSUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2865,7 +2865,7 @@ zgemm_rr.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_rc.$(PSUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_cn.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2913,7 +2913,7 @@ xgemm_rr.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_rc.$(PSUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_cn.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -3025,7 +3025,7 @@ cgemm_thread_rr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_thread_rc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_thread_cn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -3073,7 +3073,7 @@ zgemm_thread_rr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_thread_rc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_thread_cn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -3121,7 +3121,7 @@ xgemm_thread_rr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_thread_rc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_thread_cn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4288,7 +4288,7 @@ cgemm3m_rr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_rc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_cn.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4336,7 +4336,7 @@ zgemm3m_rr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_rc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_cn.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4384,7 +4384,7 @@ xgemm3m_rr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_rc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_cn.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4441,7 +4441,7 @@ cgemm3m_thread_rr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_thread_rc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_thread_cn.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4489,7 +4489,7 @@ zgemm3m_thread_rr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_thread_rc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_thread_cn.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4537,7 +4537,7 @@ xgemm3m_thread_rr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_thread_rc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_thread_cn.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)

driver/level3/level3.c

@@ -333,14 +333,16 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 #else
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
         if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
         else
-        	if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
+/*
+		if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
         	else
+*/
           		if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 #endif
 

driver/level3/level3_thread.c

@@ -367,14 +367,16 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       /* Split local region of B into parts */
       for(jjs = js; jjs < MIN(n_to, js + div_n); jjs += min_jj){
 	min_jj = MIN(n_to, js + div_n) - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
 	if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
 	else
+/*
           if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
           else
+*/
             if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 #endif
         /* Copy part of local region of B into workspace */

driver/level3/trmm_L.c

@@ -138,7 +138,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
     for(jjs = js; jjs < js + min_j; jjs += min_jj){
       min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
       /* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
       if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -215,7 +215,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -320,7 +320,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
     for(jjs = js; jjs < js + min_j; jjs += min_jj){
       min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
       /* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
       if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -399,7 +399,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else

driver/level3/trmm_R.c

@@ -122,7 +122,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < ls - js; jjs += min_jj){
 	min_jj = ls - js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -146,7 +146,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < min_l; jjs += min_jj){
 	min_jj = min_l - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -203,7 +203,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -258,7 +258,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < min_l; jjs += min_jj){
 	min_jj = min_l - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -283,7 +283,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < js - ls - min_l; jjs += min_jj){
 	min_jj = js - ls - min_l - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else
@@ -344,7 +344,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-#if defined(SKYLAKEX) || defined(COOPERLAKE)
+#if defined(SKYLAKEX) || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 	/* the current AVX512 s/d/c/z GEMM kernel requires n>=6*GEMM_UNROLL_N to achieve the best performance */
 	if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
 #else

driver/others/CMakeLists.txt

@@ -49,6 +49,8 @@ GenerateNamedObjects("openblas_get_config.c;openblas_get_parallel.c" "" "" 0 ""
 if (DYNAMIC_ARCH)
   if (ARM64)
     list(APPEND COMMON_SOURCES dynamic_arm64.c)
+  elseif (POWER)
+    list(APPEND COMMON_SOURCES dynamic_power.c)
   else ()  
     list(APPEND COMMON_SOURCES dynamic.c)
   endif ()  

driver/others/Makefile

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

driver/others/blas_server.c

@@ -209,7 +209,8 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
 	    /* REAL / Double */
 	    void (*afunc)(BLASLONG, BLASLONG, BLASLONG, double,
 			  double *, BLASLONG, double *, BLASLONG,
-			  double *, BLASLONG, void *) = func;
+			  double *, BLASLONG, void *) =  (void (*)(BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, 
+			  double *, BLASLONG, double *, BLASLONG, void *)) func;
 
 	    afunc(args -> m, args -> n, args -> k,
 		  ((double *)args -> alpha)[0],
@@ -220,7 +221,10 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
             /* REAL / Single */
             void (*afunc)(BLASLONG, BLASLONG, BLASLONG, float,
                           float *, BLASLONG, float *, BLASLONG,
-                          float *, BLASLONG, void *) = func;
+                          float *, BLASLONG, void *) = (void (*)
+                          (BLASLONG, BLASLONG, BLASLONG, float,
+                          float *, BLASLONG, float *, BLASLONG,
+                          float *, BLASLONG, void *)) func;
 
             afunc(args -> m, args -> n, args -> k,
                   ((float *)args -> alpha)[0],
@@ -232,7 +236,9 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
             /* REAL / BFLOAT16 */
             void (*afunc)(BLASLONG, BLASLONG, BLASLONG, bfloat16,
                           bfloat16 *, BLASLONG, bfloat16 *, BLASLONG,
-                          bfloat16 *, BLASLONG, void *) = func;
+                          bfloat16 *, BLASLONG, void *) = (void (*)(BLASLONG, BLASLONG, BLASLONG, bfloat16,
+                          bfloat16 *, BLASLONG, bfloat16 *, BLASLONG,
+                          bfloat16 *, BLASLONG, void *)) func;
 
             afunc(args -> m, args -> n, args -> k,
                   ((bfloat16 *)args -> alpha)[0],
@@ -243,7 +249,9 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
             /* REAL / BLAS_STOBF16 */
             void (*afunc)(BLASLONG, BLASLONG, BLASLONG, float,
                           float *, BLASLONG, bfloat16 *, BLASLONG,
-                          float *, BLASLONG, void *) = func;
+                          float *, BLASLONG, void *) = (void (*)(BLASLONG, BLASLONG, BLASLONG, float,
+                          float *, BLASLONG, bfloat16 *, BLASLONG,
+                          float *, BLASLONG, void *)) func;
 
             afunc(args -> m, args -> n, args -> k,
                   ((float *)args -> alpha)[0],
@@ -254,7 +262,9 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
             /* REAL / BLAS_DTOBF16 */
             void (*afunc)(BLASLONG, BLASLONG, BLASLONG, double,
                           double *, BLASLONG, bfloat16 *, BLASLONG,
-                          double *, BLASLONG, void *) = func;
+                          double *, BLASLONG, void *) = (void (*)(BLASLONG, BLASLONG, BLASLONG, double,
+                          double *, BLASLONG, bfloat16 *, BLASLONG,
+                          double *, BLASLONG, void *)) func;
 
             afunc(args -> m, args -> n, args -> k,
                   ((double *)args -> alpha)[0],
@@ -271,7 +281,9 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
 	  /* COMPLEX / Extended Double */
 	  void (*afunc)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,
 			xdouble *, BLASLONG, xdouble *, BLASLONG,
-			xdouble *, BLASLONG, void *) = func;
+			xdouble *, BLASLONG, void *) = (void (*)(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble,
+                        xdouble *, BLASLONG, xdouble *, BLASLONG,
+                        xdouble *, BLASLONG, void *)) func;
 
 	  afunc(args -> m, args -> n, args -> k,
 		((xdouble *)args -> alpha)[0],
@@ -285,7 +297,9 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
 	    /* COMPLEX / Double */
 	  void (*afunc)(BLASLONG, BLASLONG, BLASLONG, double, double,
 			double *, BLASLONG, double *, BLASLONG,
-			double *, BLASLONG, void *) = func;
+			double *, BLASLONG, void *) = (void (*)(BLASLONG, BLASLONG, BLASLONG, double, double,
+                        double *, BLASLONG, double *, BLASLONG,
+                        double *, BLASLONG, void *)) func;
 
 	  afunc(args -> m, args -> n, args -> k,
 		((double *)args -> alpha)[0],
@@ -297,7 +311,9 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
 	    /* COMPLEX / Single */
 	  void (*afunc)(BLASLONG, BLASLONG, BLASLONG, float, float,
 			float *, BLASLONG, float *, BLASLONG,
-			float *, BLASLONG, void *) = func;
+			float *, BLASLONG, void *) = (void (*)(BLASLONG, BLASLONG, BLASLONG, float, float,
+                        float *, BLASLONG, float *, BLASLONG,
+                        float *, BLASLONG, void *)) func;
 
 	  afunc(args -> m, args -> n, args -> k,
 		((float *)args -> alpha)[0],
@@ -425,7 +441,7 @@ blas_queue_t *tscq;
 #endif
 
     if (queue) {
-      int (*routine)(blas_arg_t *, void *, void *, void *, void *, BLASLONG) = queue -> routine;
+      int (*routine)(blas_arg_t *, void *, void *, void *, void *, BLASLONG) = (int (*)(blas_arg_t *, void *, void *, void *, void *, BLASLONG))queue -> routine;
 
       atomic_store_queue(&thread_status[cpu].queue, (blas_queue_t *)1);
 
@@ -503,7 +519,7 @@ blas_queue_t *tscq;
 	legacy_exec(routine, queue -> mode, queue -> args, sb);
       } else
 	if (queue -> mode & BLAS_PTHREAD) {
-	  void (*pthreadcompat)(void *) = queue -> routine;
+	  void (*pthreadcompat)(void *) = (void(*)(void*))queue -> routine;
 	  (pthreadcompat)(queue -> args);
 	} else
 	  (routine)(queue -> args, queue -> range_m, queue -> range_n, sa, sb, queue -> position);
@@ -871,13 +887,13 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
   fprintf(STDERR, "\n");
 #endif
 
-  routine = queue -> routine;
+  routine = (int (*)(blas_arg_t *, void *, void *, double *, double *, BLASLONG))queue -> routine;
 
   if (queue -> mode & BLAS_LEGACY) {
     legacy_exec(routine, queue -> mode, queue -> args, queue -> sb);
   } else
     if (queue -> mode & BLAS_PTHREAD) {
-      void (*pthreadcompat)(void *) = queue -> routine;
+      void (*pthreadcompat)(void *) = (void (*)(void*))queue -> routine;
       (pthreadcompat)(queue -> args);
     } else
       (routine)(queue -> args, queue -> range_m, queue -> range_n,
@@ -967,9 +983,11 @@ void goto_set_num_threads(int num_threads) {
   blas_cpu_number  = num_threads;
 
 #if defined(ARCH_MIPS64)
+#ifndef DYNAMIC_ARCH
   //set parameters for different number of threads.
   blas_set_parameter();
 #endif
+#endif
 
 }
 
@@ -1022,38 +1040,39 @@ int BLASFUNC(blas_thread_shutdown)(void){
 
   int i;
 
-  if (!blas_server_avail) return 0;
-
   LOCK_COMMAND(&server_lock);
 
-  for (i = 0; i < blas_num_threads - 1; i++) {
+  if (blas_server_avail) {
 
+    for (i = 0; i < blas_num_threads - 1; i++) {
 
-    pthread_mutex_lock (&thread_status[i].lock);
 
-    atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)-1);
-    thread_status[i].status = THREAD_STATUS_WAKEUP;
-    pthread_cond_signal (&thread_status[i].wakeup);
+      pthread_mutex_lock (&thread_status[i].lock);
 
-    pthread_mutex_unlock(&thread_status[i].lock);
+      atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)-1);
+      thread_status[i].status = THREAD_STATUS_WAKEUP;
+      pthread_cond_signal (&thread_status[i].wakeup);
 
-  }
+      pthread_mutex_unlock(&thread_status[i].lock);
 
-  for(i = 0; i < blas_num_threads - 1; i++){
-    pthread_join(blas_threads[i], NULL);
-  }
+    }
 
-  for(i = 0; i < blas_num_threads - 1; i++){
-    pthread_mutex_destroy(&thread_status[i].lock);
-    pthread_cond_destroy (&thread_status[i].wakeup);
-  }
+    for(i = 0; i < blas_num_threads - 1; i++){
+      pthread_join(blas_threads[i], NULL);
+    }
+
+    for(i = 0; i < blas_num_threads - 1; i++){
+      pthread_mutex_destroy(&thread_status[i].lock);
+      pthread_cond_destroy (&thread_status[i].wakeup);
+    }
 
 #ifdef NEED_STACKATTR
-  pthread_attr_destory(&attr);
+    pthread_attr_destroy(&attr);
 #endif
 
-  blas_server_avail = 0;
+    blas_server_avail = 0;
 
+  }
   UNLOCK_COMMAND(&server_lock);
 
   return 0;

driver/others/blas_server_win32.c

@@ -40,7 +40,7 @@
 #include <stdlib.h>
 #include "common.h"
 
-#if defined(OS_CYGWIN_NT) && !defined(unlikely)
+#if !defined(unlikely)
 #ifdef __GNUC__
 #define unlikely(x) __builtin_expect(!!(x), 0)
 #else
@@ -391,8 +391,9 @@ int blas_thread_init(void){
 
 int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
 
-#if defined(SMP_SERVER) && defined(OS_CYGWIN_NT)
+#if defined(SMP_SERVER)
   // Handle lazy re-init of the thread-pool after a POSIX fork
+  // on Cygwin or as delayed init when a static library	is used
   if (unlikely(blas_server_avail == 0)) blas_thread_init();
 #endif
 

driver/others/dynamic.c

@@ -292,6 +292,7 @@ extern gotoblas_t  gotoblas_COOPERLAKE;
 #define VENDOR_AMD        2
 #define VENDOR_CENTAUR    3
 #define VENDOR_HYGON	  4
+#define VENDOR_ZHAOXIN    5
 #define VENDOR_UNKNOWN   99
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
@@ -404,6 +405,7 @@ static int get_vendor(void){
   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, "  Shanghai  ")) return VENDOR_ZHAOXIN;
   if (!strcmp(vendor.vchar, "HygonGenuine")) return VENDOR_HYGON;
 
   if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
@@ -414,7 +416,7 @@ static int get_vendor(void){
 static gotoblas_t *get_coretype(void){
 
   int eax, ebx, ecx, edx;
-  int family, exfamily, model, vendor, exmodel;
+  int family, exfamily, model, vendor, exmodel, stepping;
 
   cpuid(1, &eax, &ebx, &ecx, &edx);
 
@@ -422,6 +424,7 @@ static gotoblas_t *get_coretype(void){
   exfamily = BITMASK(eax, 20, 0xff);
   model    = BITMASK(eax,  4, 0x0f);
   exmodel  = BITMASK(eax, 16, 0x0f);
+  stepping = BITMASK(eax,  0, 0x0f);
 
   vendor = get_vendor();
 
@@ -621,11 +624,27 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_NEHALEM;
 	  }
         }
+	if (model == 10 || model == 12){
+          // Ice Lake SP
+	   if(support_avx512_bf16())
+             return &gotoblas_COOPERLAKE;
+          if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM;
+	  }
+        }
         return NULL;  
       case 7:
 	if (model == 10) // Goldmont Plus
 	   return &gotoblas_NEHALEM;
-        if (model == 14) {
+        if (model == 13 || model == 14) {
 	// Ice Lake
           if (support_avx512()) 
 	    return &gotoblas_SKYLAKEX;
@@ -642,8 +661,68 @@ static gotoblas_t *get_coretype(void){
           }
         }
         return NULL;  
-      case 9:
       case 8:
+        if (model == 12 || model == 13) { // Tiger Lake
+          if (support_avx512()) 
+            return &gotoblas_SKYLAKEX;
+          if(support_avx2()){
+            openblas_warning(FALLBACK_VERBOSE, HASWELL_FALLBACK);
+            return &gotoblas_HASWELL;
+          }
+          if(support_avx()) {
+            openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+            return &gotoblas_SANDYBRIDGE;
+          } else {
+          openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+          return &gotoblas_NEHALEM;
+          }
+        }
+	if (model == 14 ) { // Kaby Lake, Coffee Lake
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  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 == 15){          // Sapphire Rapids
+	   if(support_avx512_bf16())
+             return &gotoblas_COOPERLAKE;
+          if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM;
+	  }
+        }
+	return NULL;
+	
+	
+      case 9:
+        if (model == 7 || model == 10) { // Alder Lake
+	   if(support_avx512_bf16())
+             return &gotoblas_COOPERLAKE;
+          if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+          if(support_avx2()){
+            return &gotoblas_HASWELL;
+          }
+          if(support_avx()) {
+            openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+            return &gotoblas_SANDYBRIDGE;
+          } else {
+          openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+          return &gotoblas_NEHALEM;
+          }
+        }
 	if (model == 14 ) { // Kaby Lake, Coffee Lake
 	  if(support_avx2())
 	    return &gotoblas_HASWELL;
@@ -655,8 +734,9 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
+	return NULL;
       case 10:
-    if (model == 5 || model == 6) {
+        if (model == 5 || model == 6) {
 	  if(support_avx2())
 	    return &gotoblas_HASWELL;
 	  if(support_avx()) {
@@ -666,7 +746,20 @@ static gotoblas_t *get_coretype(void){
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
-    }
+        }
+        if (model == 7) {
+	  if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+        }      
 	return NULL;
       }
       case 0xf:
@@ -779,10 +872,19 @@ static gotoblas_t *get_coretype(void){
   if (vendor == VENDOR_CENTAUR) {
     switch (family) {
     case 0x6:
-      return &gotoblas_NANO;
+      if (model == 0xf && stepping < 0xe)
+        return &gotoblas_NANO;
+      return &gotoblas_NEHALEM;
+    default:
+      if (family >= 0x7)
+        return &gotoblas_NEHALEM;
     }
   }
 
+  if (vendor == VENDOR_ZHAOXIN) {
+      return &gotoblas_NEHALEM;
+  }
+
   return NULL;
 }
 
@@ -962,7 +1064,13 @@ void gotoblas_dynamic_init(void) {
 #ifdef ARCH_X86
   if (gotoblas == NULL) gotoblas = &gotoblas_KATMAI;
 #else
-  if (gotoblas == NULL) gotoblas = &gotoblas_PRESCOTT;
+  if (gotoblas == NULL) {
+   if (support_avx512_bf16()) gotoblas = &gotoblas_COOPERLAKE;
+   else if (support_avx512()) gotoblas = &gotoblas_SKYLAKEX;
+   else if   (support_avx2()) gotoblas = &gotoblas_HASWELL;
+   else if    (support_avx()) gotoblas = &gotoblas_SANDYBRIDGE;
+   else                       gotoblas = &gotoblas_PRESCOTT;
+  }
   /* sanity check, if 64bit pointer we can't have a 32 bit cpu */
   if (sizeof(void*) == 8) {
       if (gotoblas == &gotoblas_KATMAI ||

driver/others/dynamic_arm64.c

@@ -43,6 +43,68 @@
 #endif
 
 extern gotoblas_t  gotoblas_ARMV8;
+#ifdef DYNAMIC_LIST
+#ifdef DYN_CORTEXA53
+extern gotoblas_t  gotoblas_CORTEXA53;
+#else
+#define gotoblas_CORTEXA53 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEXA57
+extern gotoblas_t  gotoblas_CORTEXA57;
+#else
+#define gotoblas_CORTEXA57 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEXA72
+extern gotoblas_t  gotoblas_CORTEXA72;
+#else
+#define gotoblas_CORTEXA72 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEXA73
+extern gotoblas_t  gotoblas_CORTEXA73;
+#else
+#define gotoblas_CORTEXA73 gotoblas_ARMV8
+#endif
+#ifdef DYN_FALKOR
+extern gotoblas_t  gotoblas_FALKOR;
+#else
+#define gotoblas_FALKOR gotoblas_ARMV8
+#endif
+#ifdef DYN_TSV110
+extern gotoblas_t  gotoblas_TSV110;
+#else
+#define gotoblas_TSV110 gotoblas_ARMV8
+#endif
+#ifdef DYN_THUNDERX
+extern gotoblas_t  gotoblas_THUNDERX;
+#else
+#define gotoblas_THUNDERX gotoblas_ARMV8
+#endif
+#ifdef DYN_THUNDERX2T99
+extern gotoblas_t  gotoblas_THUNDERX2T99;
+#else
+#define gotoblas_THUNDERX2T99 gotoblas_ARMV8
+#endif
+#ifdef DYN_THUNDERX3T110
+extern gotoblas_t  gotoblas_THUNDERX3T110;
+#else
+#define gotoblas_THUNDERX3T110 gotoblas_ARMV8
+#endif
+#ifdef DYN_EMAG8180
+extern gotoblas_t  gotoblas_EMAG8180;
+#else
+#define gotoblas_EMAG8180 gotoblas_ARMV8
+#endif
+#ifdef DYN_NEOVERSEN1
+extern gotoblas_t  gotoblas_NEOVERSEN1;
+#else
+#define gotoblas_NEOVERSEN1 gotoblas_ARMV8
+#endif
+#ifdef DYN_CORTEX_A55
+extern gotoblas_t  gotoblas_CORTEXA55;
+#else
+#define gotoblas_CORTEXA55 gotoblas_ARMV8
+#endif
+#else
 extern gotoblas_t  gotoblas_CORTEXA53;
 extern gotoblas_t  gotoblas_CORTEXA57;
 extern gotoblas_t  gotoblas_CORTEXA72;
@@ -54,10 +116,12 @@ extern gotoblas_t  gotoblas_TSV110;
 extern gotoblas_t  gotoblas_EMAG8180;
 extern gotoblas_t  gotoblas_NEOVERSEN1;
 extern gotoblas_t  gotoblas_THUNDERX3T110;
+extern gotoblas_t  gotoblas_CORTEXA55;
+#endif
 
 extern void openblas_warning(int verbose, const char * msg);
 
-#define NUM_CORETYPES   12
+#define NUM_CORETYPES   13
 
 /*
  * In case asm/hwcap.h is outdated on the build system, make sure
@@ -68,7 +132,7 @@ extern void openblas_warning(int verbose, const char * msg);
 #endif
 
 #define get_cpu_ftr(id, var) ({					\
-		__asm__("mrs %0, "#id : "=r" (var));		\
+		__asm__ __volatile__ ("mrs %0, "#id : "=r" (var));		\
 	})
 
 static char *corename[] = {
@@ -83,7 +147,10 @@ static char *corename[] = {
   "tsv110",
   "emag8180",
   "neoversen1",
+  "neoversev1",
+  "neoversen2",
   "thunderx3t110",
+  "cortexa55",
   "unknown"
 };
 
@@ -100,6 +167,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_EMAG8180)     return corename[ 9];
   if (gotoblas == &gotoblas_NEOVERSEN1)   return corename[10];
   if (gotoblas == &gotoblas_THUNDERX3T110) return corename[11];
+  if (gotoblas == &gotoblas_CORTEXA55)    return corename[12];
   return corename[NUM_CORETYPES];
 }
 
@@ -131,6 +199,7 @@ static gotoblas_t *force_coretype(char *coretype) {
     case  9: return (&gotoblas_EMAG8180);
     case 10: return (&gotoblas_NEOVERSEN1);
     case 11: return (&gotoblas_THUNDERX3T110);
+    case 12: return (&gotoblas_CORTEXA55);
   }
   snprintf(message, 128, "Core not found: %s\n", coretype);
   openblas_warning(1, message);
@@ -189,6 +258,8 @@ static gotoblas_t *get_coretype(void) {
           return &gotoblas_CORTEXA73;
         case 0xd0c: // Neoverse N1
           return &gotoblas_NEOVERSEN1;
+	case 0xd05: // Cortex A55
+	  return &gotoblas_CORTEXA55;
       }
       break;
     case 0x42: // Broadcom

driver/others/dynamic_mips64.c

@@ -0,0 +1,230 @@
+/*****************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of
+      its contributors may be used to endorse or promote products
+      derived from this software without specific prior written
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+
+#include <sys/wait.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/resource.h>
+#include "common.h"
+
+extern gotoblas_t  gotoblas_LOONGSON3R3;
+extern gotoblas_t  gotoblas_LOONGSON3R4;
+
+extern void openblas_warning(int verbose, const char * msg);
+
+#define NUM_CORETYPES    2
+
+static char *corename[] = {
+  "loongson3r3",
+  "loongson3r4",
+  "UNKNOWN"
+};
+
+char *gotoblas_corename(void) {
+  if (gotoblas == &gotoblas_LOONGSON3R3)    return corename[0];
+  if (gotoblas == &gotoblas_LOONGSON3R4)    return corename[1];
+  return corename[NUM_CORETYPES];
+}
+
+static gotoblas_t *force_coretype(char *coretype) {
+  int i;
+  int found = -1;
+  char message[128];
+
+  for ( i=0 ; i < NUM_CORETYPES; i++)
+  {
+    if (!strncasecmp(coretype, corename[i], 20))
+    {
+        found = i;
+        break;
+    }
+  }
+
+  switch (found)
+  {
+    case  0: return (&gotoblas_LOONGSON3R3);
+    case  1: return (&gotoblas_LOONGSON3R4);
+  }
+  snprintf(message, 128, "Core not found: %s\n", coretype);
+  openblas_warning(1, message);
+  return NULL;
+}
+
+#define MMI_MASK    0x00000010
+#define MSA_MASK    0x00000020
+
+int fd[2];
+int support_cpucfg;
+
+static void handler(int signum)
+{
+    close(fd[1]);
+    exit(1);
+}
+
+/* Brief :  Function to check if cpucfg supported on loongson
+ * Return:  1   supported
+ *          0   not supported
+ */
+static int cpucfg_test(void) {
+    pid_t pid;
+    int status = 0;
+
+    support_cpucfg = 0;
+    pipe(fd);
+    pid = fork();
+    if (pid == 0) { /* Subprocess */
+        struct sigaction act;
+        close(fd[0]);
+        /* Set signal action for SIGILL. */
+        act.sa_handler = handler;
+        sigaction(SIGILL,&act,NULL);
+
+        /* Execute cpucfg in subprocess. */
+        __asm__ volatile(
+            ".insn              \n\t"
+            ".word (0xc8080118) \n\t"
+            :::
+        );
+        support_cpucfg = 1;
+        write(fd[1],&support_cpucfg,sizeof(support_cpucfg));
+        close(fd[1]);
+        exit(0);
+    } else if (pid > 0){ /* Parent process*/
+        close(fd[1]);
+        if ((waitpid(pid,&status,0) <= 0) ||
+            (read(fd[0],&support_cpucfg,sizeof(support_cpucfg)) <= 0))
+            support_cpucfg = 0;
+        close(fd[0]);
+    } else {
+        support_cpucfg = 0;
+    }
+
+    return support_cpucfg;
+}
+
+static gotoblas_t *get_coretype_from_cpucfg(void) {
+    int flag = 0;
+    __asm__ volatile(
+        ".insn                     \n\t"
+        "dli    $8,    0x01        \n\t"
+        ".word (0xc9084918)        \n\t"
+        "usw    $9,    0x00(%0)    \n\t"
+        :
+        : "r"(&flag)
+        : "memory"
+    );
+    if (flag & MSA_MASK)
+        return (&gotoblas_LOONGSON3R4);
+    if (flag & MMI_MASK)
+        return (&gotoblas_LOONGSON3R3);
+    return NULL;
+}
+
+static gotoblas_t *get_coretype_from_cpuinfo(void) {
+#ifdef linux
+  FILE *infile;
+  char buffer[512], *p;
+
+  p = (char *)NULL;
+  //Check model name for Loongson3
+  infile = fopen("/proc/cpuinfo", "r");
+  while (fgets(buffer, sizeof(buffer), infile)){
+    if (!strncmp("model name", buffer, 10)){
+      p = strchr(buffer, ':') + 2;
+      break;
+    }
+  }
+  fclose(infile);
+  if(p != NULL){
+   if (strstr(p, "Loongson-3A3000") || strstr(p, "Loongson-3B3000"))
+     return (&gotoblas_LOONGSON3R3);
+   else if(strstr(p, "Loongson-3A4000") || strstr(p, "Loongson-3B4000"))
+     return (&gotoblas_LOONGSON3R4);
+   else
+     return NULL;
+  }
+#endif
+    return NULL;
+}
+
+static gotoblas_t *get_coretype(void) {
+    int ret = 0;
+
+    ret = cpucfg_test();
+    if (ret == 1)
+        return get_coretype_from_cpucfg();
+    else
+        return get_coretype_from_cpuinfo();
+}
+
+void gotoblas_dynamic_init(void) {
+  char coremsg[128];
+  char coren[22];
+  char *p;
+
+  if (gotoblas) return;
+
+  p = getenv("OPENBLAS_CORETYPE");
+  if ( p )
+  {
+    gotoblas = force_coretype(p);
+  }
+  else
+  {
+    gotoblas = get_coretype();
+  }
+
+  if (gotoblas == NULL)
+  {
+    snprintf(coremsg, 128, "Falling back to loongson3r3 core\n");
+    openblas_warning(1, coremsg);
+    gotoblas = &gotoblas_LOONGSON3R3;
+  }
+
+  if (gotoblas && gotoblas->init) {
+    strncpy(coren, gotoblas_corename(), 20);
+    sprintf(coremsg, "Core: %s\n", coren);
+    openblas_warning(2, coremsg);
+    gotoblas -> init();
+  } else {
+    openblas_warning(0, "OpenBLAS : Architecture Initialization failed. No initialization function found.\n");
+    exit(1);
+  }
+
+}
+
+void gotoblas_dynamic_quit(void) {
+  gotoblas = NULL;
+}

driver/others/dynamic_power.c

@@ -6,10 +6,6 @@ extern gotoblas_t gotoblas_POWER8;
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 extern gotoblas_t gotoblas_POWER9;
 #endif
-//#if (!defined __GNUC__) || ( __GNUC__ >= 11) \
-//     || (__GNUC__ == 10 && __GNUC_MINOR__ >= 2)
-//#define HAVE_P10_SUPPORT 1
-//#endif
 #ifdef HAVE_P10_SUPPORT
 extern gotoblas_t gotoblas_POWER10;
 #endif
@@ -27,7 +23,9 @@ static char *corename[] = {
 #define NUM_CORETYPES 4
 
 char *gotoblas_corename(void) {
+#ifndef C_PGI
 	if (gotoblas == &gotoblas_POWER6)	return corename[1];
+#endif
 	if (gotoblas == &gotoblas_POWER8)	return corename[2];
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	if (gotoblas == &gotoblas_POWER9)	return corename[3];
@@ -38,10 +36,164 @@ char *gotoblas_corename(void) {
 	return corename[0];
 }
 
+#if defined(__clang__)
+static int __builtin_cpu_supports(char* arg) 
+{
+	return 0;
+}
+#endif
+
+#if defined(C_PGI) || defined(__clang__)
+/*
+ * NV HPC compilers do not yet implement __builtin_cpu_is().
+ * Fake a version here for use in the CPU detection code below.
+ *
+ * Strategy here is to first check the CPU to see what it actually is,
+ * and then test the input to see if what the CPU actually is matches
+ * what was requested.
+ */
+
+#include <string.h>
+
+/*
+ *  Define POWER processor version table.
+ *
+ *  NOTE NV HPC SDK compilers only support POWER8 and POWER9 at this time
+ */
+
+#define CPU_UNKNOWN 0
+#define CPU_POWER5 5
+#define CPU_POWER6 6
+#define CPU_POWER8 8
+#define CPU_POWER9 9
+#define CPU_POWER10 10
+
+static  struct {
+    uint32_t    pvr_mask;
+    uint32_t    pvr_value;
+    const char* cpu_name;
+    uint32_t    cpu_type;
+} pvrPOWER [] = {
+
+    {   /* POWER6 in P5+ mode; 2.04-compliant processor */
+        .pvr_mask       = 0xffffffff,
+        .pvr_value      = 0x0f000001,
+        .cpu_name       = "POWER5+",
+        .cpu_type       = CPU_POWER5,
+    },
+
+    {   /* Power6 aka POWER6X*/
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x003e0000,
+        .cpu_name       = "POWER6 (raw)",
+        .cpu_type       = CPU_POWER6,
+    },
+
+    {   /* Power7 */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x003f0000,
+        .cpu_name       = "POWER7 (raw)",
+        .cpu_type       = CPU_POWER6,
+    },
+
+    {   /* Power7+ */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004A0000,
+        .cpu_name       = "POWER7+ (raw)",
+        .cpu_type       = CPU_POWER6,
+    },
+
+    {   /* Power8E */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004b0000,
+        .cpu_name       = "POWER8E (raw)",
+        .cpu_type       = CPU_POWER8,
+    },
+
+    {   /* Power8NVL */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004c0000,
+        .cpu_name       = "POWER8NVL (raw)",
+        .cpu_type       = CPU_POWER8,
+    },
+
+    {   /* Power8 */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004d0000,
+        .cpu_name       = "POWER8 (raw)",
+        .cpu_type       = CPU_POWER8,
+    },
+
+    {   /* Power9 DD2.0 */
+        .pvr_mask       = 0xffffefff,
+        .pvr_value      = 0x004e0200,
+        .cpu_name       = "POWER9 (raw)",
+        .cpu_type       = CPU_POWER9,
+    },
+
+    {   /* Power9 DD 2.1 */
+        .pvr_mask       = 0xffffefff,
+        .pvr_value      = 0x004e0201,
+        .cpu_name       = "POWER9 (raw)",
+        .cpu_type       = CPU_POWER9,
+    },
+
+    {   /* Power9 DD2.2 or later */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x004e0000,
+        .cpu_name       = "POWER9 (raw)",
+        .cpu_type       = CPU_POWER9,
+    },
+
+    {   /* Power10 */
+        .pvr_mask       = 0xffff0000,
+        .pvr_value      = 0x00800000,
+        .cpu_name       = "POWER10 (raw)",
+        .cpu_type       = CPU_POWER10,
+    },
+
+    {   /* End of table, pvr_mask and pvr_value must be zero */
+        .pvr_mask       = 0x0,
+        .pvr_value      = 0x0,
+        .cpu_name       = "Unknown",
+        .cpu_type       = CPU_UNKNOWN,
+    },
+};
+
+static int __builtin_cpu_is(const char *cpu) {
+	int i;
+	uint32_t pvr;
+	uint32_t cpu_type;
+
+	asm("mfpvr    %0" : "=r"(pvr));
+
+	for (i = 0 ; i < sizeof pvrPOWER / sizeof *pvrPOWER ; ++i) {
+		if ((pvr & pvrPOWER[i].pvr_mask) == pvrPOWER[i].pvr_value) {
+			break;
+		}
+	}
+
+#if defined(DEBUG)
+	printf("%s: returning CPU=%s, cpu_type=%p\n", __func__,
+		pvrPOWER[i].cpu_name, pvrPOWER[i].cpu_type);
+#endif
+	cpu_type = pvrPOWER[i].cpu_type;
+
+	if (!strcmp(cpu, "power8"))
+		return cpu_type == CPU_POWER8;
+	if (!strcmp(cpu, "power9"))
+		return cpu_type == CPU_POWER9;
+	return 0;
+}
+
+#endif  /* C_PGI */
+
 static gotoblas_t *get_coretype(void) {
 
+#ifndef C_PGI
 	if (__builtin_cpu_is("power6") || __builtin_cpu_is("power6x"))
 		return &gotoblas_POWER6;
+#endif
 	if (__builtin_cpu_is("power8"))
 		return &gotoblas_POWER8;
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
@@ -52,6 +204,11 @@ static gotoblas_t *get_coretype(void) {
 	if (__builtin_cpu_supports ("arch_3_1") && __builtin_cpu_supports ("mma"))
 		return &gotoblas_POWER10;
 #endif
+	/* Fall back to the POWER9 implementation if the toolchain is too old or the MMA feature is not set */
+#if (!defined __GNUC__) || ( __GNUC__ >= 11) || (__GNUC__ == 10 && __GNUC_MINOR__ >= 2)
+	if (__builtin_cpu_is("power10"))
+		return &gotoblas_POWER9;
+#endif	
 	return NULL;
 }
 
@@ -72,7 +229,9 @@ static gotoblas_t *force_coretype(char * coretype) {
 
 	switch (found)
 	{
+#ifndef C_PGI
 	case  1: return (&gotoblas_POWER6);
+#endif
 	case  2: return (&gotoblas_POWER8);
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	case  3: return (&gotoblas_POWER9);

driver/others/dynamic_zarch.c

@@ -1,38 +1,7 @@
 #include "common.h"
+#include "cpuid_zarch.h"
 #include <stdbool.h>
 
-// Guard the use of getauxval() on glibc version >= 2.16
-#ifdef __GLIBC__
-#include <features.h>
-#if __GLIBC_PREREQ(2, 16)
-#include <sys/auxv.h>
-#define HAVE_GETAUXVAL 1
-
-static unsigned long get_hwcap(void)
-{
-	unsigned long hwcap = getauxval(AT_HWCAP);
-	char *maskenv;
-
-	// honor requests for not using specific CPU features in LD_HWCAP_MASK
-	maskenv = getenv("LD_HWCAP_MASK");
-	if (maskenv)
-		hwcap &= strtoul(maskenv, NULL, 0);
-
-	return hwcap;
-	// note that a missing auxval is interpreted as no capabilities
-	// available, which is safe.
-}
-
-#else // __GLIBC_PREREQ(2, 16)
-#warn "Cannot detect SIMD support in Z13 or newer architectures since glibc is older than 2.16"
-
-static unsigned long get_hwcap(void) {
-	// treat missing support for getauxval() as no capabilities available,
-	// which is safe.
-	return 0;
-}
-#endif // __GLIBC_PREREQ(2, 16)
-#endif // __GLIBC
 
 extern gotoblas_t gotoblas_ZARCH_GENERIC;
 #ifdef DYN_Z13
@@ -44,25 +13,19 @@ extern gotoblas_t gotoblas_Z14;
 
 #define NUM_CORETYPES 4
 
+extern int openblas_verbose();
 extern void openblas_warning(int verbose, const char* msg);
 
-static char* corename[] = {
-	"unknown",
-	"Z13",
-	"Z14",
-	"ZARCH_GENERIC",
-};
-
 char* gotoblas_corename(void) {
 #ifdef DYN_Z13
-	if (gotoblas == &gotoblas_Z13)	return corename[1];
+	if (gotoblas == &gotoblas_Z13)	return cpuname[CPU_Z13];
 #endif
 #ifdef DYN_Z14
-	if (gotoblas == &gotoblas_Z14)	return corename[2];
+	if (gotoblas == &gotoblas_Z14)	return cpuname[CPU_Z14];
 #endif
-	if (gotoblas == &gotoblas_ZARCH_GENERIC) return corename[3];
+	if (gotoblas == &gotoblas_ZARCH_GENERIC) return cpuname[CPU_GENERIC];
 
-	return corename[0];
+	return "unknown";
 }
 
 #ifndef HWCAP_S390_VXE
@@ -79,25 +42,28 @@ char* gotoblas_corename(void) {
  */
 static gotoblas_t* get_coretype(void) {
 
-	unsigned long hwcap __attribute__((unused)) = get_hwcap();
+	int cpu = detect();
 
-#ifdef DYN_Z14
+	switch(cpu) {
 	// z14 and z15 systems: exploit Vector Facility (SIMD) and
 	// Vector-Enhancements Facility 1 (float SIMD instructions), if present.
-	if ((hwcap & HWCAP_S390_VX) && (hwcap & HWCAP_S390_VXE))
+	case CPU_Z14:
+#ifdef DYN_Z14
 		return &gotoblas_Z14;
 #endif
 
-#ifdef DYN_Z13
 	// z13: Vector Facility (SIMD for double)
-	if (hwcap & HWCAP_S390_VX)
+	case CPU_Z13:
+#ifdef DYN_Z13
 		return &gotoblas_Z13;
 #endif
 
+	default:
 	// fallback in case of missing compiler support, systems before z13, or
 	// when the OS does not advertise support for the Vector Facility (e.g.,
 	// missing support in the OS kernel)
-	return &gotoblas_ZARCH_GENERIC;
+		return &gotoblas_ZARCH_GENERIC;
+	}
 }
 
 static gotoblas_t* force_coretype(char* coretype) {
@@ -108,28 +74,28 @@ static gotoblas_t* force_coretype(char* coretype) {
 
 	for (i = 0; i < NUM_CORETYPES; i++)
 	{
-		if (!strncasecmp(coretype, corename[i], 20))
+		if (!strncasecmp(coretype, cpuname[i], 20))
 		{
 			found = i;
 			break;
 		}
 	}
 
-	if (found == 1) {
+	if (found == CPU_Z13) {
 #ifdef DYN_Z13
 		return &gotoblas_Z13;
 #else
 		openblas_warning(1, "Z13 support not compiled in");
 		return NULL;
 #endif
-	} else if (found == 2) {
+	} else if (found == CPU_Z14) {
 #ifdef DYN_Z14
 		return &gotoblas_Z14;
 #else
 		openblas_warning(1, "Z14 support not compiled in");
 		return NULL;
 #endif
-	} else if (found == 3) {
+	} else if (found == CPU_GENERIC) {
 		return &gotoblas_ZARCH_GENERIC;
 	}
 
@@ -155,6 +121,11 @@ void gotoblas_dynamic_init(void) {
 	else
 	{
 		gotoblas = get_coretype();
+		if (openblas_verbose() >= 2) {
+			snprintf(coremsg, sizeof(coremsg), "Choosing kernels based on getauxval(AT_HWCAP)=0x%lx\n",
+				 getauxval(AT_HWCAP));
+			openblas_warning(2, coremsg);
+		}
 	}
 
 	if (gotoblas == NULL)
@@ -165,9 +136,11 @@ void gotoblas_dynamic_init(void) {
 	}
 
 	if (gotoblas && gotoblas->init) {
-		strncpy(coren, gotoblas_corename(), 20);
-		sprintf(coremsg, "Core: %s\n", coren);
-		openblas_warning(2, coremsg);
+		if (openblas_verbose() >= 2) {
+			strncpy(coren, gotoblas_corename(), 20);
+			sprintf(coremsg, "Core: %s\n", coren);
+			openblas_warning(2, coremsg);
+		}
 		gotoblas->init();
 	}
 	else {

driver/others/memory.c

@@ -73,6 +73,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
+#ifndef likely
+#ifdef __GNUC__
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+#else
+#define likely(x) (x)
+#define unlikely(x) (x)
+#endif
+#endif
+
 #if defined(USE_TLS) && defined(SMP)
 #define COMPILE_TLS
 
@@ -222,11 +232,11 @@ int get_num_procs(void);
 #else
 int get_num_procs(void) {
   static int nums = 0;
+  int ret;
+#if defined(__GLIBC_PREREQ)
   cpu_set_t cpuset,*cpusetp;
   size_t size;
-  int ret;
 
-#if defined(__GLIBC_PREREQ)
 #if !__GLIBC_PREREQ(2, 7)
   int i;
 #if !__GLIBC_PREREQ(2, 6)
@@ -236,6 +246,15 @@ int get_num_procs(void) {
 #endif
 
   if (!nums) nums = sysconf(_SC_NPROCESSORS_CONF);
+
+#if defined(USE_OPENMP)
+#if _OPENMP >= 201511
+    ret = omp_get_num_places();
+    if (ret >0 ) nums = ret;
+#endif
+    return nums;
+#endif
+
 #if !defined(OS_LINUX)
   return nums;
 #endif
@@ -428,7 +447,7 @@ extern int openblas_goto_num_threads_env();
 extern int openblas_omp_num_threads_env();
 
 int blas_get_cpu_number(void){
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   int max_num;
 #endif
   int blas_goto_num   = 0;
@@ -436,7 +455,7 @@ int blas_get_cpu_number(void){
 
   if (blas_num_threads) return blas_num_threads;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   max_num = get_num_procs();
 #endif
 
@@ -460,7 +479,7 @@ int blas_get_cpu_number(void){
   else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
   else blas_num_threads = MAX_CPU_NUMBER;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif
 
@@ -1241,7 +1260,7 @@ UNLOCK_COMMAND(&alloc_lock);
 
       func = &memoryalloc[0];
 
-      while ((func != NULL) && (map_address == (void *) -1)) {
+      while ((*func != NULL) && (map_address == (void *) -1)) {
 
         map_address = (*func)((void *)base_address);
 
@@ -1291,7 +1310,12 @@ UNLOCK_COMMAND(&alloc_lock);
   return (void *)(((char *)alloc_info) + sizeof(struct alloc_t));
 
  error:
-  printf("OpenBLAS : Program will terminate because you tried to allocate too many memory regions.\n");
+  printf("OpenBLAS : Program will terminate because you tried to allocate too many TLS memory regions.\n");
+  printf("This library was built to support a maximum of %d threads - either rebuild OpenBLAS\n", NUM_BUFFERS);
+  printf("with a larger NUM_THREADS value or set the environment variable OPENBLAS_NUM_THREADS to\n");
+  printf("a sufficiently small number. This error typically occurs when the software that relies on\n");
+  printf("OpenBLAS calls BLAS functions from many threads in parallel, or when your computer has more\n");
+  printf("cpu cores than what OpenBLAS was configured to handle.\n"); 
 
   return NULL;
 }
@@ -1619,10 +1643,12 @@ static int on_process_term(void)
 #else
 #pragma data_seg(".CRT$XLB")
 #endif
-static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
+
 #ifdef _WIN64
+static const PIMAGE_TLS_CALLBACK dll_callback(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
 #pragma const_seg()
 #else
+static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
 #pragma data_seg()
 #endif
 
@@ -1631,10 +1657,12 @@ static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOI
 #else
 #pragma data_seg(".CRT$XTU")
 #endif
-static int(*p_process_term)(void) = on_process_term;
+
 #ifdef _WIN64
+static const int(*p_process_term)(void) = on_process_term;
 #pragma const_seg()
 #else
+static int(*p_process_term)(void) = on_process_term;
 #pragma data_seg()
 #endif
 #endif
@@ -1668,16 +1696,23 @@ void gotoblas_dummy_for_PGI(void) {
 #ifndef MEM_LARGE_PAGES
 #define MEM_LARGE_PAGES  0x20000000
 #endif
-#else
+#elif !defined(OS_EMBEDDED)
 #define ALLOC_MMAP
 #define ALLOC_MALLOC
+#else
+#define ALLOC_MALLOC
+
+inline int puts(const char *str) { return 0; }
+inline int printf(const char *format, ...) { return 0; }
+inline char *getenv(const char *name) { return ""; }
+inline int atoi(const char *str) { return 0; }
 #endif
 
 #include <stdlib.h>
 #include <stdio.h>
 #include <fcntl.h>
 
-#if !defined(OS_WINDOWS) || defined(OS_CYGWIN_NT)
+#if (!defined(OS_WINDOWS) || defined(OS_CYGWIN_NT)) && !defined(OS_EMBEDDED)
 #include <sys/mman.h>
 #ifndef NO_SYSV_IPC
 #include <sys/shm.h>
@@ -1691,7 +1726,6 @@ void gotoblas_dummy_for_PGI(void) {
 #include <sys/sysinfo.h>
 #include <sched.h>
 #include <errno.h>
-#include <linux/unistd.h>
 #include <sys/syscall.h>
 #include <sys/time.h>
 #include <sys/resource.h>
@@ -1767,11 +1801,12 @@ int get_num_procs(void);
 int get_num_procs(void) {
 
   static int nums = 0;
-
+  int ret;
+	
 #if defined(__GLIBC_PREREQ)
   cpu_set_t cpuset,*cpusetp;
   size_t size;
-  int ret;
+
 #if !__GLIBC_PREREQ(2, 7)
   int i;
 #if !__GLIBC_PREREQ(2, 6)
@@ -1781,10 +1816,20 @@ int get_num_procs(void) {
 #endif
 
   if (!nums) nums = sysconf(_SC_NPROCESSORS_CONF);
+
+#if defined(USE_OPENMP)
+/*  if (omp_get_proc_bind() != omp_proc_bind_false) */
+#if _OPENMP >= 201511
+    ret = omp_get_num_places();
+    if (ret >0 ) nums = ret;
+#endif
+    return nums;
+#endif
+
 #if !defined(OS_LINUX)
   return nums;
 #endif
-
+	
 #if !defined(__GLIBC_PREREQ)
   return nums;
 #else
@@ -1969,7 +2014,7 @@ extern int openblas_goto_num_threads_env();
 extern int openblas_omp_num_threads_env();
 
 int blas_get_cpu_number(void){
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   int max_num;
 #endif
   int blas_goto_num   = 0;
@@ -1977,7 +2022,7 @@ int blas_get_cpu_number(void){
 
   if (blas_num_threads) return blas_num_threads;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   max_num = get_num_procs();
 #endif
 
@@ -2001,7 +2046,7 @@ int blas_get_cpu_number(void){
   else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
   else blas_num_threads = MAX_CPU_NUMBER;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
   if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif
 
@@ -2045,6 +2090,7 @@ struct release_t {
 int hugetlb_allocated = 0;
 
 static struct release_t release_info[NUM_BUFFERS];
+static struct release_t *new_release_info;
 static int release_pos = 0;
 
 #if defined(OS_LINUX) && !defined(NO_WARMUP)
@@ -2095,8 +2141,13 @@ static void *alloc_mmap(void *address){
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     LOCK_COMMAND(&alloc_lock);
 #endif
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_mmap_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_mmap_free;
+    }
     release_pos ++;
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     UNLOCK_COMMAND(&alloc_lock);
@@ -2259,8 +2310,13 @@ static void *alloc_mmap(void *address){
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     LOCK_COMMAND(&alloc_lock);
 #endif
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_mmap_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_mmap_free;
+    }
     release_pos ++;
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     UNLOCK_COMMAND(&alloc_lock);
@@ -2292,8 +2348,13 @@ static void *alloc_malloc(void *address){
   if (map_address == (void *)NULL) map_address = (void *)-1;
 
   if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_malloc_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_malloc_free;
+    }
     release_pos ++;
   }
 
@@ -2326,8 +2387,13 @@ static void *alloc_qalloc(void *address){
   if (map_address == (void *)NULL) map_address = (void *)-1;
 
   if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_qalloc_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_qalloc_free;
+    }
     release_pos ++;
   }
 
@@ -2355,8 +2421,13 @@ static void *alloc_windows(void *address){
   if (map_address == (void *)NULL) map_address = (void *)-1;
 
   if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_windows_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_windows_free;
+    }
     release_pos ++;
   }
 
@@ -2399,9 +2470,15 @@ static void *alloc_devicedirver(void *address){
                      fd, 0);
 
   if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].attr    = fd;
     release_info[release_pos].func    = alloc_devicedirver_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].attr    = fd;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_devicedirver_free;
+    }
     release_pos ++;
   }
 
@@ -2435,9 +2512,15 @@ static void *alloc_shm(void *address){
 
     shmctl(shmid, IPC_RMID, 0);
 
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].attr    = shmid;
     release_info[release_pos].func    = alloc_shm_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].attr    = shmid;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_shm_free;
+    }
     release_pos ++;
   }
 
@@ -2541,8 +2624,13 @@ static void *alloc_hugetlb(void *address){
 #endif
 
   if (map_address != (void *)-1){
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].func    = alloc_hugetlb_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_hugetlb_free;
+    }
     release_pos ++;
   }
 
@@ -2589,9 +2677,15 @@ static void *alloc_hugetlbfile(void *address){
                      fd, 0);
 
   if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
     release_info[release_pos].address = map_address;
     release_info[release_pos].attr    = fd;
     release_info[release_pos].func    = alloc_hugetlbfile_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].attr    = fd;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_hugetlbfile_free;
+    }
     release_pos ++;
   }
 
@@ -2621,8 +2715,25 @@ static volatile struct {
 
 } memory[NUM_BUFFERS];
 
-static int memory_initialized = 0;
+struct newmemstruct 
+{
+  BLASULONG lock;
+  void *addr;
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+  int   pos;
+#endif
+  int used;
+#ifndef __64BIT__
+  char dummy[48];
+#else
+  char dummy[40];
+#endif
 
+};
+static volatile struct newmemstruct *newmemory;
+
+static int memory_initialized = 0;
+static int memory_overflowed = 0;
 /*       Memory allocation routine           */
 /* procpos ... indicates where it comes from */
 /*                0 : Level 3 functions      */
@@ -2631,6 +2742,8 @@ static int memory_initialized = 0;
 
 void *blas_memory_alloc(int procpos){
 
+  int i;
+  
   int position;
 #if defined(WHEREAMI) && !defined(USE_OPENMP)
   int mypos = 0;
@@ -2761,6 +2874,25 @@ void *blas_memory_alloc(int procpos){
     position ++;
 
   } while (position < NUM_BUFFERS);
+
+  if (memory_overflowed) {
+
+    do {
+      RMB;
+#if defined(USE_OPENMP)
+      if (!newmemory[position-NUM_BUFFERS].used) {
+        blas_lock(&newmemory[position-NUM_BUFFERS].lock);
+#endif
+        if (!newmemory[position-NUM_BUFFERS].used) goto allocation2;
+
+#if defined(USE_OPENMP)
+        blas_unlock(&newmemory[position-NUM_BUFFERS].lock);
+      }
+#endif
+      position ++;
+
+    } while (position < 512+NUM_BUFFERS);
+  }
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   UNLOCK_COMMAND(&alloc_lock);
 #endif
@@ -2788,7 +2920,7 @@ void *blas_memory_alloc(int procpos){
 
       func = &memoryalloc[0];
 
-      while ((func != NULL) && (map_address == (void *) -1)) {
+      while ((*func != NULL) && (map_address == (void *) -1)) {
 
         map_address = (*func)((void *)base_address);
 
@@ -2868,8 +3000,102 @@ void *blas_memory_alloc(int procpos){
   return (void *)memory[position].addr;
 
  error:
-  printf("BLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  LOCK_COMMAND(&alloc_lock);
+#endif
+ if (memory_overflowed) goto terminate;
+  fprintf(stderr,"OpenBLAS warning: precompiled NUM_THREADS exceeded, adding auxiliary array for thread metadata.\n");
+  memory_overflowed=1;
+  new_release_info = (struct release_t*) malloc(512*sizeof(struct release_t));
+  newmemory = (struct newmemstruct*) malloc(512*sizeof(struct newmemstruct));
+  for (i = 0; i < 512; i++) {
+  newmemory[i].addr   = (void *)0;
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+  newmemory[i].pos    = -1;
+#endif
+  newmemory[i].used   = 0;
+  newmemory[i].lock   = 0;
+}
+  
+allocation2:
+  newmemory[position-NUM_BUFFERS].used = 1;
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  UNLOCK_COMMAND(&alloc_lock);
+#else
+  blas_unlock(&newmemory[position-NUM_BUFFERS].lock);
+#endif
+    do {
+#ifdef DEBUG
+      printf("Allocation Start : %lx\n", base_address);
+#endif
+
+      map_address = (void *)-1;
+
+      func = &memoryalloc[0];
+
+      while ((*func != NULL) && (map_address == (void *) -1)) {
+
+        map_address = (*func)((void *)base_address);
+
+#ifdef ALLOC_DEVICEDRIVER
+        if ((*func ==  alloc_devicedirver) && (map_address == (void *)-1)) {
+            fprintf(stderr, "OpenBLAS Warning ... Physically contiguous allocation was failed.\n");
+        }
+#endif
+
+#ifdef ALLOC_HUGETLBFILE
+        if ((*func == alloc_hugetlbfile) && (map_address == (void *)-1)) {
+#ifndef OS_WINDOWS
+            fprintf(stderr, "OpenBLAS Warning ... HugeTLB(File) allocation was failed.\n");
+#endif
+        }
+#endif
+
+#if (defined ALLOC_SHM) && (defined OS_LINUX  || defined OS_AIX  || defined __sun__  || defined OS_WINDOWS)
+        if ((*func == alloc_hugetlb) && (map_address != (void *)-1)) hugetlb_allocated = 1;
+#endif
+
+        func ++;
+      }
 
+#ifdef DEBUG
+      printf("  Success -> %08lx\n", map_address);
+#endif
+      if (((BLASLONG) map_address) == -1) base_address = 0UL;
+
+      if (base_address) base_address += BUFFER_SIZE + FIXED_PAGESIZE;
+
+    } while ((BLASLONG)map_address == -1);
+
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+    LOCK_COMMAND(&alloc_lock);
+#endif
+    newmemory[position-NUM_BUFFERS].addr = map_address;
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+    UNLOCK_COMMAND(&alloc_lock);
+#endif
+
+#ifdef DEBUG
+    printf("  Mapping Succeeded. %p(%d)\n", (void *)newmemory[position-NUM_BUFFERS].addr, position);
+#endif
+
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+
+  if (newmemory[position-NUM_BUFFERS].pos == -1) newmemory[position-NUM_BUFFERS].pos = mypos;
+
+#endif
+  return (void *)newmemory[position-NUM_BUFFERS].addr;
+
+terminate:
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+    UNLOCK_COMMAND(&alloc_lock);
+#endif
+  printf("OpenBLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
+  printf("This library was built to support a maximum of %d threads - either rebuild OpenBLAS\n", NUM_BUFFERS);
+  printf("with a larger NUM_THREADS value or set the environment variable OPENBLAS_NUM_THREADS to\n");
+  printf("a sufficiently small number. This error typically occurs when the software that relies on\n");
+  printf("OpenBLAS calls BLAS functions from many threads in parallel, or when your computer has more\n");
+  printf("cpu cores than what OpenBLAS was configured to handle.\n"); 
   return NULL;
 }
 
@@ -2888,13 +3114,28 @@ void blas_memory_free(void *free_area){
   while ((position < NUM_BUFFERS) && (memory[position].addr != free_area))
     position++;
 
-  if (position >= NUM_BUFFERS) goto error;
+  if (position >= NUM_BUFFERS && !memory_overflowed) goto error;
 
 #ifdef DEBUG
   if (memory[position].addr != free_area) goto error;
   printf("  Position : %d\n", position);
 #endif
+  if (unlikely(memory_overflowed && position >= NUM_BUFFERS)) {
+    while ((position < NUM_BUFFERS+512) && (newmemory[position-NUM_BUFFERS].addr != free_area))
+      position++;
+  // arm: ensure all writes are finished before other thread takes this memory
+  WMB;
+
+  newmemory[position].used = 0;
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  UNLOCK_COMMAND(&alloc_lock);
+#endif
 
+#ifdef DEBUG
+  printf("Unmap from overflow area succeeded.\n\n");
+#endif
+  return;
+} else {
   // arm: ensure all writes are finished before other thread takes this memory
   WMB;
 
@@ -2908,7 +3149,7 @@ void blas_memory_free(void *free_area){
 #endif
 
   return;
-
+}
  error:
   printf("BLAS : Bad memory unallocation! : %4d  %p\n", position,  free_area);
 
@@ -2943,7 +3184,10 @@ void blas_shutdown(void){
   LOCK_COMMAND(&alloc_lock);
 
   for (pos = 0; pos < release_pos; pos ++) {
+    if (likely(pos < NUM_BUFFERS))
     release_info[pos].func(&release_info[pos]);
+    else
+    new_release_info[pos-NUM_BUFFERS].func(&new_release_info[pos-NUM_BUFFERS]);
   }
 
 #ifdef SEEK_ADDRESS
@@ -2960,6 +3204,15 @@ void blas_shutdown(void){
 #endif
     memory[pos].lock   = 0;
   }
+  if (memory_overflowed)
+    for (pos = 0; pos < 512; pos ++){
+      newmemory[pos].addr   = (void *)0;
+      newmemory[pos].used   = 0;
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+      newmemory[pos].pos    = -1;
+#endif
+      newmemory[pos].lock   = 0;
+  }
 
   UNLOCK_COMMAND(&alloc_lock);
 

driver/others/parameter.c

@@ -183,7 +183,7 @@ int get_L2_size(void){
     defined(CORE_PRESCOTT) || defined(CORE_CORE2)       || defined(PENRYN) || defined(DUNNINGTON) || \
     defined(CORE_NEHALEM)  || defined(CORE_SANDYBRIDGE) || defined(ATOM)   || defined(GENERIC)    || \
     defined(PILEDRIVER)    || defined(HASWELL)          || defined(STEAMROLLER) || defined(EXCAVATOR) || \
-    defined(ZEN)           || defined(SKYLAKEX)         || defined(COOPERLAKE)
+    defined(ZEN)           || defined(SKYLAKEX)         || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
 
   cpuid(0x80000006, &eax, &ebx, &ecx, &edx);
 
@@ -269,7 +269,7 @@ void blas_set_parameter(void){
   int factor;
 #if defined(BULLDOZER) || defined(PILEDRIVER)  || defined(SANDYBRIDGE) || defined(NEHALEM) || \
     defined(HASWELL)   || defined(STEAMROLLER) || defined(EXCAVATOR)   || defined(ZEN)     || \
-    defined(SKYLAKEX)  || defined(COOPERLAKE)
+    defined(SKYLAKEX)  || defined(COOPERLAKE) || defined(SAPPHIRERAPIDS)
   int size = 16;
 #else
   int size = get_L2_size();
@@ -524,6 +524,9 @@ void blas_set_parameter(void){
   xgemm_p = ((xgemm_p + XGEMM_UNROLL_M - 1)/XGEMM_UNROLL_M) * XGEMM_UNROLL_M;
 #endif
 
+#ifdef BUILD_BFLOAT16
+  sbgemm_r = (((BUFFER_SIZE - ((SBGEMM_P * SBGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SBGEMM_Q *  4)) - 15) & ~15;
+#endif
   sgemm_r = (((BUFFER_SIZE - ((SGEMM_P * SGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SGEMM_Q *  4)) - 15) & ~15;
   dgemm_r = (((BUFFER_SIZE - ((DGEMM_P * DGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (DGEMM_Q *  8)) - 15) & ~15;
   cgemm_r = (((BUFFER_SIZE - ((CGEMM_P * CGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (CGEMM_Q *  8)) - 15) & ~15;
@@ -629,7 +632,9 @@ void blas_set_parameter(void){
   xgemm_p =  16 * (size + 1);
 #endif
 
+#ifdef BUILD_BFLOAT16
   sbgemm_r = (((BUFFER_SIZE - ((SBGEMM_P * SBGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SBGEMM_Q *  4)) - 15) & ~15;
+#endif
   sgemm_r = (((BUFFER_SIZE - ((SGEMM_P * SGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SGEMM_Q *  4)) - 15) & ~15;
   dgemm_r = (((BUFFER_SIZE - ((DGEMM_P * DGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (DGEMM_Q *  8)) - 15) & ~15;
   cgemm_r = (((BUFFER_SIZE - ((CGEMM_P * CGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (CGEMM_Q *  8)) - 15) & ~15;
@@ -717,7 +722,7 @@ void blas_set_parameter(void){
 
 #if defined(ARCH_MIPS64)
 void blas_set_parameter(void){
-#if defined(LOONGSON3A)
+#if defined(LOONGSON3R3) || defined(LOONGSON3R4)
 #ifdef SMP
   if(blas_num_threads == 1){
 #endif
@@ -731,20 +736,6 @@ void blas_set_parameter(void){
 #endif
 #endif
 
-#if defined(LOONGSON3B)
-#ifdef SMP
-  if(blas_num_threads == 1 || blas_num_threads == 2){
-#endif
-    //single thread
-    dgemm_r = 640;
-#ifdef SMP
-  }else{
-    //multi thread
-    dgemm_r = 160;
-  }
-#endif
-#endif
-
 }
 #endif
 

exports/Makefile

@@ -139,9 +139,17 @@ endif
 ifneq (,$(filter 1 2,$(NOFORTRAN)))
 #only build without Fortran
 	$(CC) $(CFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+else
+ifeq ($(F_COMPILER), INTEL)
+	$(FC) $(FFLAGS) $(LDFLAGS) -all-load -headerpad_max_install_names -install_name "$(CURDIR)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def
+else
+ifeq ($(F_COMPILER), FLANG)
+	$(FC) $(FFLAGS) $(LDFLAGS) -fno-fortran-main -Mnomain -all_load -headerpad_max_install_names -install_name "$(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)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 endif
+endif
+endif
 
 dllinit.$(SUFFIX) : dllinit.c
 	$(CC) $(CFLAGS) -c -o $(@F) -s $<

exports/gensymbol

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