Merge branch 'develop' into sdk26

Author: martin-frbg

.github/workflows/apple_m.yml

@@ -44,7 +44,7 @@ jobs:
           elif [ "$RUNNER_OS" == "macOS" ]; then
             # It looks like "gfortran" isn't working correctly unless "gcc" is re-installed.
             brew reinstall gcc
-            brew install coreutils cmake ccache
+            brew install coreutils ccache
             brew install llvm
           else
             echo "::error::$RUNNER_OS not supported"
@@ -87,10 +87,16 @@ jobs:
           echo "max_size = 300M" > ~/.ccache/ccache.conf
           echo "compression = true" >> ~/.ccache/ccache.conf
           ccache -s
+          
+      - name: Add gfortran runtime to link path
+        if: matrix.build == 'make' && runner.os == 'macOS'
+        run: |
+          GFORTRAN_LIBDIR=$(gfortran -print-file-name=libgfortran.dylib | xargs dirname)
+          echo "Using gfortran runtime in $GFORTRAN_LIBDIR"
+          echo "LDFLAGS=-L/opt/homebrew/opt/llvm/lib -L$GFORTRAN_LIBDIR" >> $GITHUB_ENV
 
       - name: Build OpenBLAS
         run: |
-          export LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
           export CPPFLAGS="-I/opt/homebrew/opt/llvm/include"
           export CC="/opt/homebrew/opt/llvm/bin/clang"
           case "${{ matrix.build }}" in

.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/dynamic_arch.yml

@@ -49,7 +49,7 @@ jobs:
           elif [ "$RUNNER_OS" == "macOS" ]; then
             # It looks like "gfortran" isn't working correctly unless "gcc" is re-installed.
             brew reinstall gcc
-            brew install coreutils cmake ccache
+            brew install coreutils ccache
           else
             echo "::error::$RUNNER_OS not supported"
             exit 1
@@ -89,6 +89,14 @@ jobs:
           echo "max_size = 300M" > ~/.ccache/ccache.conf
           echo "compression = true" >> ~/.ccache/ccache.conf
           ccache -s
+          
+      - name: Add gfortran runtime to link path
+        if: matrix.build == 'make' && runner.os == 'macOS'
+        run: |
+          GFORTRAN_LIBDIR=$(gfortran -print-file-name=libgfortran.dylib | xargs dirname)
+          echo "Using gfortran runtime in $GFORTRAN_LIBDIR"
+          # Preserve whatever LDFLAGS may already contain
+          echo "LDFLAGS=${LDFLAGS:+$LDFLAGS }-L$GFORTRAN_LIBDIR" >> "$GITHUB_ENV"
 
       - name: Build OpenBLAS
         run: |

.github/workflows/loongarch64_clang.yml

@@ -35,7 +35,7 @@ jobs:
 
       - name: Install libffi6
         run: |
-          wget http://ftp.ca.debian.org/debian/pool/main/libf/libffi/libffi6_3.2.1-9_amd64.deb
+          wget https://download.nvidia.com/cumulus/apt.cumulusnetworks.com/pool/upstream/libf/libffi/libffi6_3.2.1-9_amd64.deb
           sudo dpkg -i libffi6_3.2.1-9_amd64.deb
 
       - name: Install APT deps

.github/workflows/riscv64_vector.yml

@@ -17,7 +17,7 @@ jobs:
       triple: riscv64-unknown-linux-gnu
       riscv_gnu_toolchain: https://github.com/riscv-collab/riscv-gnu-toolchain
       riscv_gnu_toolchain_version: 13.2.0
-      riscv_gnu_toolchain_nightly_download_path: /releases/download/2024.02.02/riscv64-glibc-ubuntu-22.04-llvm-nightly-2024.02.02-nightly.tar.gz
+      riscv_gnu_toolchain_nightly_download_path: /releases/download/2025.08.29/riscv64-glibc-ubuntu-22.04-llvm-nightly-2025.08.29-nightly.tar.xz
     strategy:
       fail-fast: false
       matrix:

.github/workflows/windows_arm64.yml

@@ -0,0 +1,84 @@
+name: Windows ARM64 CI
+
+on:
+  push:
+    branches:
+      - develop
+  pull_request:
+    branches:
+      - develop
+
+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-20.1.8/LLVM-20.1.8-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
+          #dir "C:\Program Files\LLVM\include\flang"
+          #rmdir /Q /S "C:/Program Files/Microsoft Visual Studio/2022/Enterprise/VC/Tools/Llvm/ARM64"
+          
+      - 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"
+          set PATH=C:\Program Files\LLVM\bin;%PATH%
+          
+          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

@@ -9,7 +9,7 @@ project(OpenBLAS C ASM)
 
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 30)
+set(OpenBLAS_PATCH_VERSION 30.dev)
 
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
@@ -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
@@ -305,8 +308,8 @@ 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
@@ -541,13 +544,13 @@ message(STATUS "adding postbuild instruction to rename syms")
   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 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()

CONTRIBUTORS.md

@@ -251,6 +251,19 @@ In chronological order:
 * Ye Tao <ye.tao@arm.com>
   * [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
+  * [2025-07-23] ARM64-Enable bfloat16 kernels by default 
+
+* Sharif Inamdar <sharif.inamdar@arm.com>
+  * [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
+

Jenkinsfile

@@ -10,5 +10,10 @@ pipeline {
                 sh 'make clean && make'
             }
         }
+        stage('CMakeBuild') {
+            steps {
+                sh 'sudo apt update && sudo apt install cmake -y && make clean && rm -rf build && mkdir build && cd build && cmake -DDYNAMIC_ARCH=1 .. && make'
+            }
+        }
     }
 }