Dev222

.github/workflows/build.yaml

@@ -32,62 +32,42 @@ jobs:
       fail-fast: false
 
       matrix:
-        compiler: [gcc, clang]
-        cuda: [with, without]
+        compiler: [gcc]
+        cuda: [with]
         openmp: [with]
         link: [both]
         include:
           - compiler: gcc
-            compiler-pkgs: "g++ gcc"
-            cc: "gcc"
-            cxx: "g++"
-          - compiler: clang
-            compiler-pkgs: "clang libomp-dev"
-            cc: "clang"
-            cxx: "clang++"
-          # Clang seems to generally require less cache size (smaller object files?).
+            cuda: with
+            openmp: with
+            link: both
           - compiler: gcc
-            ccache-max: 600M
+            cuda: without
+            openmp: with
+            link: both
           - compiler: clang
-            ccache-max: 500M
-          - cuda: with
-            cuda-pkgs: "nvidia-cuda-dev nvidia-cuda-toolkit"
-            cuda-cmake-flags:
-              -DSUITESPARSE_USE_CUDA=ON
-              -DSUITESPARSE_USE_STRICT=ON
-              -DCUDAToolkit_INCLUDE_DIRS="/usr/include"
-              -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"
+            cuda: with
+            openmp: with
+            link: both
+          - compiler: clang
+            cuda: without
+            openmp: with
+            link: both
           - compiler: gcc
-            compiler-pkgs: "g++ gcc"
-            cc: "gcc"
-            cxx: "g++"
-            ccache-max: 600M
             cuda: without
             openmp: without
-            openmp-cmake-flags: "-DSUITESPARSE_USE_OPENMP=OFF"
+            link: both
           - compiler: gcc
-            compiler-pkgs: "g++ gcc"
-            cc: "gcc"
-            cxx: "g++"
-            ccache-max: 600M
             cuda: with
-            cuda-pkgs: "nvidia-cuda-dev nvidia-cuda-toolkit"
-            cuda-cmake-flags:
-              -DSUITESPARSE_USE_CUDA=ON
-              -DSUITESPARSE_USE_STRICT=ON
-              -DCUDAToolkit_INCLUDE_DIRS="/usr/include"
-              -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"
             openmp: with
             link: static
             # "Fake" a cross-compilation to exercise that build system path
-            link-cmake-flags:
-              -DBUILD_SHARED_LIBS=OFF
-              -DBUILD_STATIC_LIBS=ON
+            extra-cmake-flags:
               -DCMAKE_SYSTEM_NAME="Linux"
 
     env:
-      CC: ${{ matrix.cc }}
-      CXX: ${{ matrix.cxx }}
+      CC: ${{ matrix.compiler == 'gcc' && 'gcc' || 'clang' }}
+      CXX: ${{ matrix.compiler == 'gcc' && 'g++' || 'clang++' }}
 
     steps:
       - name: get CPU information
@@ -97,14 +77,15 @@ jobs:
         uses: actions/checkout@v4
 
       - name: install dependencies
-        env:
-          COMPILER_PKGS: ${{ matrix.compiler-pkgs }}
-          CUDA_PKGS: ${{ matrix.cuda-pkgs }}
         run: |
           sudo apt -qq update
-          sudo apt install -y ${COMPILER_PKGS} autoconf automake ccache cmake \
+          sudo apt install -y \
+            ${{ matrix.compiler == 'gcc' && 'g++ gcc' || 'clang' }} \
+            ${{ matrix.compiler == 'clang' &&  matrix.openmp == 'with' && 'libomp-dev' || '' }} \
+            autoconf automake ccache cmake \
             dvipng gfortran libgmp-dev liblapack-dev libmpfr-dev valgrind \
-            libopenblas-dev ${CUDA_PKGS}
+            libopenblas-dev \
+            ${{ matrix.cuda == 'with' && 'nvidia-cuda-dev nvidia-cuda-toolkit' || '' }}
 
       - name: prepare ccache
         # create key with human readable timestamp
@@ -144,7 +125,8 @@ jobs:
           CCACHE_MAX: ${{ matrix.ccache-max }}
         run: |
           test -d ~/.ccache || mkdir ~/.ccache
-          echo "max_size = $CCACHE_MAX" >> ~/.ccache/ccache.conf
+          # Clang seems to generally require less cache size (smaller object files?).
+          echo "max_size = ${{ matrix.compiler == 'gcc' && '600M' || '500M' }}" >> ~/.ccache/ccache.conf
           echo "compression = true" >> ~/.ccache/ccache.conf
           ccache -s
           echo "/usr/lib/ccache" >> $GITHUB_PATH
@@ -156,21 +138,29 @@ jobs:
             printf "   \033[0;32m==>\033[0m Building library \033[0;32m${lib}\033[0m\n"
             echo "::group::Configure $lib"
             cd ${GITHUB_WORKSPACE}/${lib}/build
-            cmake -DCMAKE_BUILD_TYPE="Release" \
+            cmake -DCMAKE_BUILD_TYPE=${{ matrix.compiler == 'clang' && 'Debug' || 'Release' }} \
                   -DCMAKE_INSTALL_PREFIX="${GITHUB_WORKSPACE}" \
                   -DCMAKE_C_COMPILER_LAUNCHER="ccache" \
                   -DCMAKE_CXX_COMPILER_LAUNCHER="ccache" \
                   -DCMAKE_Fortran_COMPILER_LAUNCHER="ccache" \
                   -DBLA_VENDOR="OpenBLAS" \
                   -DSUITESPARSE_DEMOS=OFF \
                   -DBUILD_TESTING=OFF \
-                  ${{ matrix.cuda-cmake-flags }} \
-                  ${{ matrix.openmp-cmake-flags }} \
-                  ${{ matrix.link-cmake-flags }} \
+                  ${{ matrix.cuda == 'with'
+                    && '-DSUITESPARSE_USE_CUDA=ON \
+                        -DSUITESPARSE_USE_STRICT=ON \
+                        -DCUDAToolkit_INCLUDE_DIRS="/usr/include" \
+                        -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"'
+                    || '-DSUITESPARSE_USE_CUDA=OFF' }} \
+                  -DSUITESPARSE_USE_OPENMP=${{ matrix.openmp == 'without' && 'OFF' || 'ON' }} \
+                  ${{ matrix.link == 'static'
+                    && '-DBUILD_SHARED_LIBS=OFF -DBUILD_STATIC_LIBS=ON'
+                    || '' }} \
+                  ${{ matrix.extra-cmake-flags }} \
                   ..
             echo "::endgroup::"
             echo "::group::Build $lib"
-            cmake --build . --config Release
+            cmake --build .
             echo "::endgroup::"
           done
 
@@ -214,8 +204,15 @@ jobs:
           cmake \
             -DCMAKE_PREFIX_PATH="${GITHUB_WORKSPACE}/lib/cmake" \
             -DBLA_VENDOR="OpenBLAS" \
-            ${{ matrix.cuda-cmake-flags }} \
-            ${{ matrix.link-cmake-flags }} \
+            ${{ matrix.cuda == 'with'
+              && '-DSUITESPARSE_USE_CUDA=ON \
+                  -DSUITESPARSE_USE_STRICT=ON \
+                  -DCUDAToolkit_INCLUDE_DIRS="/usr/include" \
+                  -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"'
+              || '' }} \
+            ${{ matrix.link == 'static'
+              && '-DBUILD_SHARED_LIBS=OFF -DBUILD_STATIC_LIBS=ON'
+              || '' }} \
             ..
           echo "::endgroup::"
           printf "::group::\033[0;32m==>\033[0m Building example\n"

.github/workflows/root-cmakelists.yaml

@@ -30,50 +30,29 @@ jobs:
       fail-fast: false
 
       matrix:
-        compiler: [gcc, clang]
-        cuda: [with, without]
+        compiler: [gcc]
+        cuda: [with]
         link: [both]
         include:
           - compiler: gcc
-            compiler-pkgs: "g++ gcc"
-            cc: "gcc"
-            cxx: "g++"
-          - compiler: clang
-            compiler-pkgs: "clang libomp-dev"
-            cc: "clang"
-            cxx: "clang++"
-          # Clang seems to generally require less cache size (smaller object files?).
+            cuda: with
+            link: both
           - compiler: gcc
-            ccache-max: 600M
+            cuda: without
+            link: both
           - compiler: clang
-            ccache-max: 500M
-          - cuda: with
-            cuda-pkgs: "nvidia-cuda-dev nvidia-cuda-toolkit"
-            cuda-cmake-flags:
-              -DSUITESPARSE_USE_CUDA=ON
-              -DSUITESPARSE_USE_STRICT=ON
-              -DCUDAToolkit_INCLUDE_DIRS="/usr/include"
-              -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"
+            cuda: with
+            link: both
+          - compiler: clang
+            cuda: without
+            link: both
           - compiler: gcc
-            compiler-pkgs: "g++ gcc"
-            cc: "gcc"
-            cxx: "g++"
-            ccache-max: 600M
             cuda: with
-            cuda-pkgs: "nvidia-cuda-dev nvidia-cuda-toolkit"
-            cuda-cmake-flags:
-              -DSUITESPARSE_USE_CUDA=ON
-              -DSUITESPARSE_USE_STRICT=ON
-              -DCUDAToolkit_INCLUDE_DIRS="/usr/include"
-              -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"
             link: static
-            link-cmake-flags:
-              -DBUILD_SHARED_LIBS=OFF
-              -DBUILD_STATIC_LIBS=ON
 
     env:
-      CC: ${{ matrix.cc }}
-      CXX: ${{ matrix.cxx }}
+      CC: ${{ matrix.compiler == 'gcc' && 'gcc' || 'clang' }}
+      CXX: ${{ matrix.compiler == 'gcc' && 'g++' || 'clang++' }}
 
     steps:
       - name: get CPU information
@@ -83,14 +62,15 @@ jobs:
         uses: actions/checkout@v4
 
       - name: install dependencies
-        env:
-          COMPILER_PKGS: ${{ matrix.compiler-pkgs }}
-          CUDA_PKGS: ${{ matrix.cuda-pkgs }}
         run: |
           sudo apt -qq update
-          sudo apt install -y ${COMPILER_PKGS} autoconf automake ccache cmake \
+          sudo apt install -y \
+            ${{ matrix.compiler == 'gcc' && 'g++ gcc' || 'clang' }} \
+            ${{ matrix.compiler == 'clang' && 'libomp-dev' || '' }} \
+            autoconf automake ccache cmake \
             dvipng gfortran libgmp-dev liblapack-dev libmpfr-dev \
-            libopenblas-dev ${CUDA_PKGS}
+            libopenblas-dev \
+            ${{ matrix.cuda == 'with' && 'nvidia-cuda-dev nvidia-cuda-toolkit' || '' }}
 
       - name: prepare ccache
         # create key with human readable timestamp
@@ -126,11 +106,10 @@ jobs:
           sudo mv ./libpthread.a /usr/lib/x86_64-linux-gnu/libpthread.a
 
       - name: configure ccache
-        env:
-          CCACHE_MAX: ${{ matrix.ccache-max }}
         run: |
           test -d ~/.ccache || mkdir ~/.ccache
-          echo "max_size = $CCACHE_MAX" >> ~/.ccache/ccache.conf
+          # Clang seems to generally require less cache size (smaller object files?).
+          echo "max_size = ${{ matrix.compiler == 'gcc' && '600M' || '500M' }}" >> ~/.ccache/ccache.conf
           echo "compression = true" >> ~/.ccache/ccache.conf
           ccache -s
           echo "/usr/lib/ccache" >> $GITHUB_PATH
@@ -154,16 +133,23 @@ jobs:
       - name: configure
         run: |
           mkdir -p ${GITHUB_WORKSPACE}/build && cd ${GITHUB_WORKSPACE}/build
-          cmake -DCMAKE_BUILD_TYPE="Release" \
+          cmake -DCMAKE_BUILD_TYPE=${{ matrix.compiler == 'clang' && 'Debug' || 'Release' }} \
                 -DCMAKE_INSTALL_PREFIX=".." \
                 -DCMAKE_C_COMPILER_LAUNCHER="ccache" \
                 -DCMAKE_CXX_COMPILER_LAUNCHER="ccache" \
                 -DCMAKE_Fortran_COMPILER_LAUNCHER="ccache" \
                 -DBLA_VENDOR="OpenBLAS" \
                 -DSUITESPARSE_DEMOS=OFF \
                 -DBUILD_TESTING=OFF \
-                ${{ matrix.cuda-cmake-flags }} \
-                ${{ matrix.link-cmake-flags }} \
+                ${{ matrix.cuda == 'with'
+                  && '-DSUITESPARSE_USE_CUDA=ON \
+                      -DSUITESPARSE_USE_STRICT=ON \
+                      -DCUDAToolkit_INCLUDE_DIRS="/usr/include" \
+                      -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"'
+                  || '' }} \
+                ${{ matrix.link == 'static'
+                  && '-DBUILD_SHARED_LIBS=OFF -DBUILD_STATIC_LIBS=ON'
+                  || '' }} \
                 ..
 
       - name: build libraries
@@ -213,8 +199,15 @@ jobs:
           cmake \
             -DCMAKE_PREFIX_PATH="${GITHUB_WORKSPACE}/lib/cmake" \
             -DBLA_VENDOR="OpenBLAS" \
-            ${{ matrix.cuda-cmake-flags }} \
-            ${{ matrix.link-cmake-flags }} \
+            ${{ matrix.cuda == 'with'
+              && '-DSUITESPARSE_USE_CUDA=ON \
+                  -DSUITESPARSE_USE_STRICT=ON \
+                  -DCUDAToolkit_INCLUDE_DIRS="/usr/include" \
+                  -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"'
+              || '' }} \
+            ${{ matrix.link == 'static'
+              && '-DBUILD_SHARED_LIBS=OFF -DBUILD_STATIC_LIBS=ON'
+              || '' }} \
             ..
           echo "::endgroup::"
           printf "::group::\033[0;32m==>\033[0m Building example\n"

CHOLMOD/CMakeLists.txt

@@ -573,15 +573,9 @@ endif ( )
 
 # CHOLMOD_CUDA
 if ( CHOLMOD_HAS_CUDA )
-#   cmake now configures cholmod.h with "#define CHOLMOD_HAS_CUDA" if CHOLMOD is
-#   being compiled with CUDA, so the -DCHOLMOD_HAS_CUDA flag is no longer needed.
-#   if ( BUILD_SHARED_LIBS )
-#       target_compile_definitions ( CHOLMOD PUBLIC "CHOLMOD_HAS_CUDA" )
-#   endif ( )
-#   set ( CHOLMOD_CFLAGS "${CHOLMOD_CFLAGS} -DCHOLMOD_HAS_CUDA" )
-#   if ( BUILD_STATIC_LIBS )
-#       target_compile_definitions ( CHOLMOD_static PUBLIC "CHOLMOD_HAS_CUDA" )
-#   endif ( )
+    if ( NOT CHOLMOD_HAS_OPENMP )
+      message ( FATAL_ERROR "CHOLMOD_CUDA requires OpenMP. But it has been disabled, or no working OpenMP could be found." )
+    endif ()
     if ( BUILD_SHARED_LIBS )
         target_link_libraries ( CHOLMOD PRIVATE CUDA::nvrtc CUDA::cudart_static CUDA::cublas )
         target_include_directories ( CHOLMOD INTERFACE

ChangeLog

@@ -1,3 +1,29 @@
+Dec 20, 2024: version 7.9.0
+
+    * GraphBLAS v9.4.3: Added the GxB_IndexBinaryOp.  Added new JIT kernels.
+        Disabled more FactoryKernels to reduce compiled library size.
+    * Package versions in this release: (* denotes a new version)
+        SuiteSparse_config 7.9.0    *
+        AMD 3.3.3
+        BTF 2.3.2
+        CAMD 3.3.3
+        CCOLAMD 3.3.4
+        CHOLMOD 5.3.0
+        COLAMD 3.3.4
+        CSparse 4.3.2
+        CXSparse 4.4.1
+        Example 1.8.4               *
+        GraphBLAS 9.4.3             *
+        KLU 2.3.5
+        LDL 3.3.2
+        LAGraph 1.1.4
+        SuiteSparse_Mongoose 3.3.4
+        ParU 1.0.0
+        RBio 4.3.4
+        SPEX 3.2.1
+        SPQR 4.3.4
+        UMFPACK 6.3.5
+
 Oct 10, 2024: version 7.8.3
 
     * ParU 1.0.0: first stable release. No change since last version

Example/CMakeLists.txt

@@ -53,10 +53,10 @@ message ( STATUS "MY prefix path:    ${CMAKE_PREFIX_PATH}" )
 #-------------------------------------------------------------------------------
 
 # cmake inserts the date and version number into Include/my.h:
-set ( MY_DATE "Oct 10, 2024" )
+set ( MY_DATE "Dec 20, 2024" )
 set ( MY_VERSION_MAJOR 1 )
 set ( MY_VERSION_MINOR 8 )
-set ( MY_VERSION_PATCH 3 )
+set ( MY_VERSION_PATCH 4 )
 
 message ( STATUS "Building MY library version: v"
     ${MY_VERSION_MAJOR}.
@@ -87,15 +87,15 @@ project ( my
 #-------------------------------------------------------------------------------
 
 # look for all SuiteSparse packages:
-find_package ( SuiteSparse_config 7.8.3 REQUIRED )
+find_package ( SuiteSparse_config 7.9.0 REQUIRED )
 find_package ( AMD 3.3.3 REQUIRED )
 find_package ( BTF 2.3.2 REQUIRED )
 find_package ( CAMD 3.3.3 REQUIRED )
 find_package ( CCOLAMD 3.3.4 REQUIRED )
 find_package ( CHOLMOD 5.3.0 REQUIRED )
 find_package ( COLAMD 3.3.4 REQUIRED )
 find_package ( CXSparse 4.4.1 REQUIRED )
-find_package ( GraphBLAS 9.3.1 )
+find_package ( GraphBLAS 9.4.3 )
 find_package ( KLU 2.3.5 REQUIRED )
 find_package ( KLU_CHOLMOD 2.3.5 REQUIRED )
 find_package ( LDL 3.3.2 REQUIRED )