Merge branch 'develop' into export-shared-only

Author: MehdiChinoune

.github/workflows/arm64_graviton.yml

@@ -88,13 +88,14 @@ jobs:
         run: |
           case "${{ matrix.build }}" in
             "make")
-              make -j$(nproc) DYNAMIC_ARCH=1 USE_OPENMP=0 FC="ccache ${{ matrix.fortran }}"
+              make -j$(nproc) DYNAMIC_ARCH=1 BUILD_BFLOAT16=1 USE_OPENMP=0 FC="ccache ${{ matrix.fortran }}"
               ;;
             "cmake")
               mkdir build && cd build
               cmake -DDYNAMIC_ARCH=1 \
                     -DNOFORTRAN=0 \
                     -DBUILD_WITHOUT_LAPACK=0 \
+                    -DBUILD_BFLOAT16=1 \
                     -DCMAKE_VERBOSE_MAKEFILE=ON \
                     -DCMAKE_BUILD_TYPE=Release \
                     -DCMAKE_Fortran_COMPILER=${{ matrix.fortran }} \

.github/workflows/windows_arm64.yml

@@ -0,0 +1,82 @@
+name: Windows ARM64 CI
+
+on:
+  push:
+    branches:
+      - develop
+      - release-**
+  pull_request:
+    branches:
+      - develop
+      - release-**
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read # to fetch code (actions/checkout)
+
+jobs:
+  build:
+    if: "github.repository == 'OpenMathLib/OpenBLAS'"
+    runs-on: windows-11-arm
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+
+      - name: Install LLVM for Win-ARM64
+        shell: pwsh
+        run: |
+          Invoke-WebRequest https://github.com/llvm/llvm-project/releases/download/llvmorg-20.1.6/LLVM-20.1.6-woa64.exe -UseBasicParsing -OutFile LLVM-woa64.exe
+          Start-Process -FilePath ".\LLVM-woa64.exe" -ArgumentList "/S" -Wait
+          echo "C:\Program Files\LLVM\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
+
+      - name: Install CMake and Ninja for Win-ARM64
+        shell: pwsh
+        run: |
+          Invoke-WebRequest https://github.com/Kitware/CMake/releases/download/v3.29.4/cmake-3.29.4-windows-arm64.msi -OutFile cmake-arm64.msi
+          Start-Process msiexec.exe -ArgumentList "/i cmake-arm64.msi /quiet /norestart" -Wait
+          echo "C:\Program Files\CMake\bin" >> $env:GITHUB_PATH
+          
+          Invoke-WebRequest https://github.com/ninja-build/ninja/releases/download/v1.13.1/ninja-winarm64.zip -OutFile ninja-winarm64.zip
+          Expand-Archive ninja-winarm64.zip -DestinationPath ninja
+          Copy-Item ninja\ninja.exe -Destination "C:\Windows\System32"
+
+      - name: Configure OpenBLAS
+        shell: cmd
+        run: |
+          CALL "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsarm64.bat"
+          mkdir build
+          cd build
+          cmake .. -G Ninja ^
+            -DCMAKE_BUILD_TYPE=Release ^
+            -DTARGET=ARMV8 ^
+            -DBINARY=64 ^
+            -DCMAKE_C_COMPILER=clang-cl ^
+            -DCMAKE_Fortran_COMPILER=flang-new ^
+            -DBUILD_SHARED_LIBS=ON ^
+            -DCMAKE_SYSTEM_PROCESSOR=arm64 ^
+            -DCMAKE_SYSTEM_NAME=Windows ^
+            -DCMAKE_INSTALL_PREFIX=C:/opt
+
+      - name: Build OpenBLAS
+        shell: cmd
+        run: |
+          cd build
+          ninja -j16
+
+      - name: Install OpenBLAS
+        shell: cmd
+        run: |
+          cd build
+          cmake --install .
+
+      - name: Run ctests 
+        shell: pwsh
+        run: |
+          $env:PATH = "C:\opt\bin;$env:PATH"
+          cd build
+          ctest
+
+

.gitignore

@@ -13,8 +13,8 @@ lapack-3.4.1.tgz
 lapack-3.4.2
 lapack-3.4.2.tgz
 lapack-netlib/make.inc
-lapack-netlib/lapacke/include/lapacke_mangling.h
 lapack-netlib/SRC/la_constants.mod
+lapack-netlib/SRC/la_xisnan.mod
 lapack-netlib/TESTING/testing_results.txt
 lapack-netlib/INSTALL/test*
 lapack-netlib/TESTING/xeigtstc
@@ -81,7 +81,10 @@ test/ZBLAT2.SUMM
 test/ZBLAT3.SUMM
 test/ZBLAT3_3M.SUMM
 test/SHBLAT3.SUMM
+test/SBBLAT2.SUMM
 test/SBBLAT3.SUMM
+test/BBLAT2.SUMM
+test/BBLAT3.SUMM
 test/cblat1
 test/cblat2
 test/cblat3
@@ -96,6 +99,9 @@ test/sblat3
 test/sblat3_3m
 test/test_shgemm
 test/test_sbgemm
+test/test_sbgemv
+test/test_bgemm
+test/test_bgemv
 test/zblat1
 test/zblat2
 test/zblat3

CMakeLists.txt

@@ -152,6 +152,9 @@ endif ()
 if (NOT DEFINED BUILD_BFLOAT16)
  set (BUILD_BFLOAT16 false)
 endif ()
+if (NOT DEFINED BUILD_HFLOAT16)
+ set (BUILD_HFLOAT16 false)
+endif ()
 # set which float types we want to build for
 if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_COMPLEX AND NOT DEFINED BUILD_COMPLEX16)
   # if none are defined, build for all
@@ -302,8 +305,8 @@ if (USE_OPENMP)
   endif()
 endif()
 
-# Fix "Argument list too long" for macOS with Intel CPUs and DYNAMIC_ARCH turned on
-if(APPLE AND DYNAMIC_ARCH AND (NOT CMAKE_HOST_SYSTEM_PROCESSOR STREQUAL "arm64"))
+# Fix "Argument list too long" for macOS with POWERPC or Intel CPUs 
+if(APPLE AND (NOT CMAKE_HOST_SYSTEM_PROCESSOR STREQUAL "arm64"))
   # Use response files
   set(CMAKE_C_USE_RESPONSE_FILE_FOR_OBJECTS 1)
   # Always build static library first
@@ -537,18 +540,18 @@ if (BUILD_SHARED_LIBS OR DELETE_STATIC_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFFI
   else ()
     if (NOT USE_PERL)
       add_custom_command(TARGET ${OpenBLAS_LIBNAME}_shared POST_BUILD
-        COMMAND sh ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BLD}" "${BBF16}" "${BS}" "${BD}" "${BC}" "${BZ}" > ${PROJECT_BINARY_DIR}/objcopy.def
-        COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/${OpenBLAS_LIBNAME}.so
+	      COMMAND sh ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BLD}" "${BBF16}" "${BS}" "${BD}" "${BC}" "${BZ}" > ${PROJECT_BINARY_DIR}/objcopy.def
+        COMMAND objcopy --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/${OpenBLAS_LIBNAME}.so
         COMMENT "renaming symbols"
-        )
+      )
     else()
       add_custom_command(TARGET ${OpenBLAS_LIBNAME}_shared POST_BUILD
         COMMAND perl ${PROJECT_SOURCE_DIR}/exports/gensymbol.pl "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BLD}" "${BBF16}" "${BS}" "${BD}" "${BC}" "${BZ}" > ${PROJECT_BINARY_DIR}/objcopy.def
-        COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
+        COMMAND objcopy --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
         COMMENT "renaming symbols"
-        )
-      endif()
+      )
     endif()
+  endif()
 endif()
 
 if (BUILD_BENCHMARKS)

CONTRIBUTORS.md

@@ -251,6 +251,18 @@ In chronological order:
 * Ye Tao <[email protected]>
   * [2025-02-03] Optimize SBGEMM kernel on NEOVERSEV1
   * [2025-02-27] Add sbgemv_n_neon kernel
+  * [2025-05-17] Impl prototype of BGEMM inferface
 
 * Abhishek Kumar <https://github.com/abhishek-iitmadras>
-  * [2025-04-22] Optimise dot kernel for NEOVERSE V1
+  * [2025-04-22] Optimise dot kernel for NEOVERSE V1
+
+* Sharif Inamdar <[email protected]>
+  * [2025-06-05] Optimize gemv_n_sve_v1x3 kernel
+
+* Guoyuan Li <https://github.com/guoyuanplct>
+  * [2025-04-11] Optimise gemv kernel for RISCV64_ZVL256B
+  * [2025-05-01] Optimise zgemv kernel for RISCV64_ZVL256B
+  * [2025-05-17] Optimise omatcopy/zomatcopy kernel for RISCV64_ZVL256B
+  * [2025-05-29] Optimise axpby kernel for RISCV64_ZVL256B
+  * [2025-06-05] Optimise hbmv kernel for RISCV64_ZVL256B
+

Makefile.arm64

@@ -191,6 +191,16 @@ endif
 endif
 endif
 
+# Detect Ampere AmpereOne(ampere1,ampere1a) processors.
+ifeq ($(CORE), AMPERE1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ12) $(ISCLANG)))
+CCOMMON_OPT += -march=armv8.6-a+crypto+crc+fp16+sha3+rng
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.6-a+crypto+crc+fp16+sha3+rng
+endif
+endif
+endif
+
 # Use a53 tunings because a55 is only available in GCC>=8.1
 ifeq ($(CORE), CORTEXA55)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))

Makefile.power

@@ -13,16 +13,16 @@ ifeq ($(CORE), POWER10)
 ifneq ($(C_COMPILER), PGI)
 ifeq ($(C_COMPILER), GCC)
 ifeq ($(GCCVERSIONGTEQ10), 1)
-CCOMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
+CCOMMON_OPT += -O3 -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
 else ifneq ($(GCCVERSIONGT4), 1)
 $(warning your compiler is too old to fully support POWER9, getting a newer version of gcc is recommended)
-CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math
+CCOMMON_OPT += -O3 -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math
 else
 $(warning your compiler is too old to fully support POWER10, getting a newer version of gcc is recommended)
-CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math
+CCOMMON_OPT += -O3 -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math
 endif
 else
-CCOMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
+CCOMMON_OPT += -O3 -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
 endif
 ifeq ($(F_COMPILER), IBM)
 FCOMMON_OPT += -O2 -qrecur -qnosave -qarch=pwr10 -qtune=pwr10 -qfloat=nomaf -qzerosize
@@ -34,7 +34,7 @@ endif
 
 ifeq ($(CORE), POWER9)
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mvsx -fno-fast-math
+CCOMMON_OPT += -O3 -mvsx -fno-fast-math
 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)
@@ -70,7 +70,7 @@ endif
 
 ifeq ($(CORE), POWER8)
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx  -fno-fast-math
+CCOMMON_OPT += -O3 -mcpu=power8 -mtune=power8 -mvsx  -fno-fast-math
 else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif

Makefile.prebuild

@@ -64,11 +64,11 @@ TARGET_FLAGS = -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d
 endif
 
 ifeq ($(TARGET), RISCV64_ZVL256B)
-TARGET_FLAGS = -march=rv64imafdcv -mabi=lp64d
+TARGET_FLAGS = -march=rv64imafdcv_zvfh_zfh -mabi=lp64d
 endif
 
 ifeq ($(TARGET), RISCV64_ZVL128B)
-TARGET_FLAGS = -march=rv64imafdcv -mabi=lp64d
+TARGET_FLAGS = -march=rv64imafdcv_zvfh_zfh -mabi=lp64d
 endif
 
 ifeq ($(TARGET), RISCV64_GENERIC)

Makefile.riscv64

@@ -7,12 +7,12 @@ CCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh_zvl512b -mabi=lp64d
 FCOMMON_OPT += -march=rv64imafdcv_zba_zbb_zfh -mabi=lp64d -static
 endif
 ifeq ($(CORE), RISCV64_ZVL256B)
-CCOMMON_OPT += -march=rv64imafdcv_zvl256b -mabi=lp64d
-FCOMMON_OPT += -march=rv64imafdcv -mabi=lp64d
+CCOMMON_OPT += -march=rv64imafdcv_zvl256b_zvfh_zfh -mabi=lp64d
+FCOMMON_OPT += -march=rv64imafdcv_zvfh_zfh -mabi=lp64d
 endif
 ifeq ($(CORE), RISCV64_ZVL128B)
-CCOMMON_OPT += -march=rv64imafdcv -mabi=lp64d 
-FCOMMON_OPT += -march=rv64imafdcv -mabi=lp64d
+CCOMMON_OPT += -march=rv64imafdcv_zvfh_zfh -mabi=lp64d 
+FCOMMON_OPT += -march=rv64imafdcv_zvfh_zfh -mabi=lp64d
 endif
 ifeq ($(CORE), RISCV64_GENERIC)
 CCOMMON_OPT += -march=rv64imafdc -mabi=lp64d

Makefile.rule

@@ -308,6 +308,8 @@ COMMON_PROF = -pg
 # If you want to enable the experimental BFLOAT16 support
 # BUILD_BFLOAT16 = 1
 
+# If you want to enable the experimental HFLOAT16 support
+# BUILD_HFLOAT16 = 1
 
 # Set the thread number threshold beyond which the job array for the threaded level3 BLAS
 # will be allocated on the heap rather than the stack. (This array alone requires