Merge pull request #559 from mmuetzel/ci

CI: Add runner using MSVC with CUDA

.github/workflows/root-cmakelists.yaml

@@ -212,7 +212,7 @@ jobs:
     # https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners
     runs-on: windows-latest
 
-    name: msvc (${{ matrix.openmp }} OpenMP)
+    name: msvc (${{ matrix.openmp }} OpenMP ${{ matrix.cuda }} CUDA)
 
     defaults:
       run:
@@ -225,9 +225,15 @@ jobs:
 
       matrix:
         openmp: [with, without]
+        cuda: [without]
         include:
           - openmp: without
             openmp-cmake-flags: "-DNOPENMP=ON"
+          - openmp: with
+            cuda: with
+            cuda-cmake-flags:
+              -DENABLE_CUDA=ON
+              -DCMAKE_CUDA_COMPILER_LAUNCHER="ccache"
 
     env:
       CHERE_INVOKING: 1
@@ -283,6 +289,18 @@ jobs:
 
           msystem: UCRT64
 
+      - uses: Jimver/[email protected]
+        name: install CUDA toolkit
+        if: matrix.cuda == 'with'
+        id: cuda-toolkit
+        with:
+          cuda: '12.2.0'
+          #See https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#install-the-cuda-software
+          method: 'local'
+          # Do not cache the installer (~3 GiB). It doesn't speed up the
+          # installation significantly. And we need the space for ccache.
+          use-github-cache: 'false'
+
       - name: setup build environment
         # get packages from MSYS2
         # Copy only relevant parts to avoid picking up headers and libraries
@@ -311,7 +329,7 @@ jobs:
         shell: msys2 {0}
         run: |
           echo "ccachedir=$(cygpath -m $(${CCACHE} -k cache_dir))" >> $GITHUB_OUTPUT
-          echo "key=ccache:msvc:root:${{ matrix.openmp }}:${{ github.ref }}:$(date +"%Y-%m-%d_%H-%M-%S"):${{ github.sha }}" >> $GITHUB_OUTPUT
+          echo "key=ccache:msvc:root:${{ matrix.openmp }}:${{ matrix.cuda }}:${{ github.ref }}:$(date +"%Y-%m-%d_%H-%M-%S"):${{ github.sha }}" >> $GITHUB_OUTPUT
 
       - name: restore ccache
         # Setup the GitHub cache used to maintain the ccache from one job to the next
@@ -321,8 +339,8 @@ jobs:
           key: ${{ steps.ccache-prepare.outputs.key }}
           # Prefer caches from the same branch. Fall back to caches from the dev branch.
           restore-keys: |
-            ccache:msvc:root:${{ matrix.openmp }}:${{ github.ref }}
-            ccache:msvc:root:${{ matrix.openmp }}:
+            ccache:msvc:root:${{ matrix.openmp }}:${{ matrix.cuda }}:${{ github.ref }}
+            ccache:msvc:root:${{ matrix.openmp }}:${{ matrix.cuda }}:
 
       - name: configure ccache
         # Limit the maximum size and switch on compression to avoid exceeding the total disk or cache quota.
@@ -338,6 +356,11 @@ jobs:
 
       - name: configure
         run: |
+          declare -a _extra_config
+          if [ ${{ matrix.cuda }} = 'with' ]; then
+            _extra_config+=(-DCUDAToolkit_ROOT="$(cygpath -m "${{ steps.cuda-toolkit.outputs.CUDA_PATH }}")")
+            _extra_config+=(-DCMAKE_CUDA_COMPILER="$(cygpath -m "${{ steps.cuda-toolkit.outputs.CUDA_PATH }}")/bin/nvcc.exe")
+          fi
           mkdir -p ${GITHUB_WORKSPACE}/build && cd ${GITHUB_WORKSPACE}/build
           cmake -G"Ninja Multi-Config" \
                 -DCMAKE_BUILD_TYPE="Release" \
@@ -350,6 +373,8 @@ jobs:
                 -DBLA_VENDOR="All" \
                 -DPython_EXECUTABLE="C:/msys64/ucrt64/bin/python.exe" \
                 ${{ matrix.openmp-cmake-flags }} \
+                ${{ matrix.cuda-cmake-flags }} \
+                "${_extra_config[@]}" \
                 ..
 
       - name: build libraries

CHOLMOD/CMakeLists.txt

@@ -748,7 +748,7 @@ add_test ( NAME CHOLMOD_int64_single_supernodal
 if ( WIN32 AND BUILD_SHARED_LIBS )
     # Set PATH to pick up the necessary libraries for all tests
 
-    set ( CHOLMOD_CTEST_NAMES
+    set ( CHOLMOD_CTEST_NAMES 
         CHOLMOD_int32_double_bcsstk01 CHOLMOD_int64_double_bcsstk01 CHOLMOD_int32_single_bcsstk01 CHOLMOD_int64_single_bcsstk01
         CHOLMOD_int32_double_lp_afiro CHOLMOD_int64_double_lp_afiro CHOLMOD_int32_single_lp_afiro CHOLMOD_int64_single_lp_afiro
         CHOLMOD_int32_double_can24 CHOLMOD_int64_double_can24 CHOLMOD_int32_single_can24 CHOLMOD_int64_single_can24

GraphBLAS/CUDA/CMakeLists.txt

@@ -73,7 +73,12 @@ set_target_properties(GraphBLAS_CUDA PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
 # FIXME: use SUITESPARSE_CUDA_ARCHITECTURES
 set_target_properties(GraphBLAS_CUDA PROPERTIES CUDA_ARCHITECTURES "52;75;80" )
 
-target_link_libraries(GraphBLAS_CUDA CUDA::nvrtc CUDA::cudart_static CUDA::cuda_driver CUDA::nvToolsExt )
+target_link_libraries ( GraphBLAS_CUDA CUDA::nvrtc CUDA::cudart_static CUDA::cuda_driver )
+
+if ( TARGET CUDA::nvtx3 )
+    target_link_libraries ( GraphBLAS_CUDA CUDA::nvtx3 )
+    target_compile_definitions ( GraphBLAS_CUDA PRIVATE GBNVTX )
+endif ( )
 
 target_compile_definitions ( GraphBLAS_CUDA PUBLIC "SUITESPARSE_CUDA" )
 

GraphBLAS/cmake_modules/SuiteSparsePolicy.cmake

@@ -254,30 +254,30 @@ if ( ENABLE_CUDA )
         # with CUDA:
         message ( STATUS "Find CUDA tool kit:" )
         # FindCUDAToolKit needs to have C or CXX enabled first (see above)
-        include ( FindCUDAToolkit )
+        find_package ( CUDAToolkit )
         message ( STATUS "CUDA toolkit found:   " ${CUDAToolkit_FOUND} )
         message ( STATUS "CUDA toolkit version: " ${CUDAToolkit_VERSION} )
         message ( STATUS "CUDA toolkit include: " ${CUDAToolkit_INCLUDE_DIRS} )
         message ( STATUS "CUDA toolkit lib dir: " ${CUDAToolkit_LIBRARY_DIR} )
         if ( CUDAToolkit_VERSION VERSION_LESS "11.2" )
             # CUDA is present but too old
             message ( STATUS "CUDA: not enabled (CUDA 11.2 or later required)" )
-            set ( SUITESPARSE_CUDA off )
+            set ( SUITESPARSE_CUDA OFF )
         else ( )
             # CUDA 11.2 or later present
             enable_language ( CUDA )
-            set ( SUITESPARSE_CUDA on )
+            set ( SUITESPARSE_CUDA ON )
         endif ( )
     else ( )
         # without CUDA:
         message ( STATUS "CUDA: not found" )
-        set ( SUITESPARSE_CUDA off )
+        set ( SUITESPARSE_CUDA OFF )
     endif ( )
 
 else ( )
 
     # CUDA is disabled
-    set ( SUITESPARSE_CUDA off )
+    set ( SUITESPARSE_CUDA OFF )
 
 endif ( )
 

SPQR/Demo/qrdemo_gpu2.cpp

@@ -61,7 +61,7 @@ int main (int argc, char **argv)
     m = A->nrow ;
     n = A->ncol ;
 
-    long ordering = (argc < 3 ? SPQR_ORDERING_DEFAULT : atoi(argv[2]));
+    int ordering = (argc < 3 ? SPQR_ORDERING_DEFAULT : atoi(argv[2]));
 
     printf ("Matrix %6ld-by-%-6ld nnz: %6ld\n",
         m, n, cholmod_l_nnz (A, cc)) ;
@@ -70,7 +70,7 @@ int main (int argc, char **argv)
     B = cholmod_l_ones (m, 1, A->xtype, cc) ;
 
     double tol = SPQR_NO_TOL ;
-    long econ = 0 ;
+    int64_t econ = 0 ;
 
     // [Q,R,E] = qr (A), but discard Q
     // SuiteSparseQR <double> (ordering, tol, econ, A, &R, &E, cc) ;

SPQR/Demo/qrdemo_gpu3.cpp

@@ -63,7 +63,7 @@ int main (int argc, char **argv)
     m = A->nrow ;
     n = A->ncol ;
 
-    long ordering = (argc < 3 ? SPQR_ORDERING_DEFAULT : atoi(argv[2]));
+    int ordering = (argc < 3 ? SPQR_ORDERING_DEFAULT : atoi(argv[2]));
 
     printf ("Matrix %6ld-by-%-6ld nnz: %6ld\n",
         m, n, cholmod_l_nnz (A, cc)) ;
@@ -72,7 +72,7 @@ int main (int argc, char **argv)
     B = cholmod_l_ones (m, 1, A->xtype, cc) ;
 
     double tol = SPQR_NO_TOL ;
-    long econ = 0 ;
+    int64_t econ = 0 ;
 
     // [Q,R,E] = qr (A), but discard Q
     // SuiteSparseQR <double> (ordering, tol, econ, A, &R, &E, cc) ;

SPQR/GPUQREngine/CMakeLists.txt

@@ -21,7 +21,7 @@ message ( STATUS "Building SPQR/GPUQRENGINE version: v"
 # define the project
 #-------------------------------------------------------------------------------
 
-project ( gpuqrengine
+project ( gpuqrengine 
     VERSION "${SPQR_VERSION_MAJOR}.${SPQR_VERSION_MINOR}.${SPQR_VERSION_SUB}"
     LANGUAGES CXX CUDA )
 
@@ -90,9 +90,10 @@ if ( BUILD_SHARED_LIBS )
         WINDOWS_EXPORT_ALL_SYMBOLS ON )
 
     target_include_directories ( GPUQREngine PRIVATE
-        ${GPUQRENGINE_INCLUDES}
-        "$<TARGET_PROPERTY:SuiteSparse::SuiteSparseConfig,INTERFACE_INCLUDE_DIRECTORIES>"
-        "$<TARGET_PROPERTY:GPURuntime,INTERFACE_INCLUDE_DIRECTORIES>" )
+        ${GPUQRENGINE_INCLUDES} )
+
+    target_link_libraries ( GPUQREngine PRIVATE
+        GPURuntime SuiteSparse::SuiteSparseConfig )
 
     set_target_properties ( GPUQREngine PROPERTIES
         POSITION_INDEPENDENT_CODE ON
@@ -128,9 +129,15 @@ if ( BUILD_STATIC_LIBS )
 
     target_include_directories ( GPUQREngine_static PRIVATE
         ${CUDAToolkit_INCLUDE_DIRS}
-        ${GPUQRENGINE_INCLUDES}
-        $<TARGET_PROPERTY:SuiteSparse::SuiteSparseConfig,INTERFACE_INCLUDE_DIRECTORIES>
-        $<TARGET_PROPERTY:GPURuntime_static,INTERFACE_INCLUDE_DIRECTORIES> )
+        ${GPUQRENGINE_INCLUDES} )
+
+    target_link_libraries ( GPUQREngine_static PUBLIC GPURuntime_static )
+
+    if ( TARGET SuiteSparse::SuiteSparseConfig_static )
+        target_link_libraries ( GPUQREngine_static PUBLIC SuiteSparse::SuiteSparseConfig_static )
+    else ( )
+        target_link_libraries ( GPUQREngine_static PUBLIC SuiteSparse::SuiteSparseConfig )
+    endif ( )
 
     set_target_properties ( GPUQREngine_static PROPERTIES
         POSITION_INDEPENDENT_CODE ON

SPQR/GPUQREngine/Include/GPUQREngine_BucketList.hpp

@@ -20,6 +20,8 @@
 #ifndef GPUQRENGINE_BUCKETLIST_HPP
 #define GPUQRENGINE_BUCKETLIST_HPP
 
+#include <cstddef>
+
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_TaskDescriptor.hpp"
 #include "GPUQREngine_LLBundle.hpp"
@@ -72,7 +74,7 @@ template <typename Int = int64_t> class BucketList
     int VThead;              // Index of the first available entry in VTlist
 
     // Constructors
-    void *operator new(size_t, BucketList <Int>* p)
+    void *operator new(std::size_t, BucketList <Int>* p)
     {
         return p;
     }

SPQR/GPUQREngine/Include/GPUQREngine_Front.hpp

@@ -19,6 +19,8 @@
 #ifndef GPUQRENGINE_FRONT_HPP
 #define GPUQRENGINE_FRONT_HPP
 
+#include <cstddef>
+
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_SparseMeta.hpp"
 #include "GPUQREngine_FrontState.hpp"
@@ -53,7 +55,7 @@ template <typename Int = int64_t> class Front
     /* Debug Fields */
     bool printMe;
 
-    void* operator new(size_t reqMem, Front* ptr){ return ptr; }
+    void* operator new(std::size_t reqMem, Front* ptr){ return ptr; }
 
     Front(
         Int fids_arg,                   // the front identifier

SPQR/GPUQREngine/Include/GPUQREngine_LLBundle.hpp

@@ -24,6 +24,8 @@
 #ifndef GPUQRENGINE_LLBUNDLE_HPP
 #define GPUQRENGINE_LLBUNDLE_HPP
 
+#include <cstddef>
+
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_TaskDescriptor.hpp"
 
@@ -76,7 +78,7 @@ template <typename Int = int64_t> class LLBundle
 
     TaskType CurrentTask;
 
-    void *operator new(size_t, LLBundle <Int>* p){ return p; }
+    void *operator new(std::size_t, LLBundle <Int>* p){ return p; }
     //------------------------------------------------------------------------------
     //
     // This file contains the constructor and destructor for the LLBundle class.

SPQR/GPUQREngine/Include/GPUQREngine_Scheduler.hpp

@@ -19,6 +19,8 @@
 #ifndef GPUQRENGINE_SCHEDULER_HPP
 #define GPUQRENGINE_SCHEDULER_HPP
 
+#include <cstddef>
+
 #include "GPUQREngine_Common.hpp"
 #include "GPUQREngine_FrontState.hpp"
 #include "GPUQREngine_TaskDescriptor.hpp"
@@ -97,7 +99,7 @@ class Scheduler
     cudaStream_t memoryStreamD2H;
 
     /* Scheduler.cpp */
-    void *operator new(size_t, Scheduler <Int>* p){ return p; }
+    void *operator new(std::size_t, Scheduler <Int>* p){ return p; }
     Scheduler(Front <Int> *fronts, Int numFronts, size_t gpuMemorySize);
     ~Scheduler();
 

SPQR/GPURuntime/CMakeLists.txt

@@ -21,7 +21,7 @@ message ( STATUS "Building SPQR/GPURUNTIME version: v"
 # define the project
 #-------------------------------------------------------------------------------
 
-project ( suitesparse_gpuruntime
+project ( suitesparse_gpuruntime 
     VERSION "${SPQR_VERSION_MAJOR}.${SPQR_VERSION_MINOR}.${SPQR_VERSION_SUB}"
     LANGUAGES C CXX CUDA )
 
@@ -70,8 +70,9 @@ if ( BUILD_SHARED_LIBS )
 
     target_include_directories ( GPURuntime PRIVATE
         ${CUDAToolkit_INCLUDE_DIRS}
-        ${SUITESPARSE_GPURUNTIME_INCLUDES}
-        $<TARGET_PROPERTY:SuiteSparse::SuiteSparseConfig,INTERFACE_INCLUDE_DIRECTORIES> )
+        ${SUITESPARSE_GPURUNTIME_INCLUDES} )
+
+    target_link_libraries ( GPURuntime PRIVATE SuiteSparse::SuiteSparseConfig )
 
     set_target_properties ( GPURuntime PROPERTIES
         POSITION_INDEPENDENT_CODE ON
@@ -81,7 +82,7 @@ if ( BUILD_SHARED_LIBS )
         CUDA::nvrtc CUDA::cudart_static CUDA::cublas )
     target_compile_definitions ( GPURuntime PRIVATE "SUITESPARSE_CUDA" )
 
-    target_include_directories ( GPURuntime
+    target_include_directories ( GPURuntime 
         INTERFACE $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/Include>
                   $<INSTALL_INTERFACE:${SUITESPARSE_INCLUDEDIR}> )
 endif ( )
@@ -106,8 +107,13 @@ if ( BUILD_STATIC_LIBS )
 
     target_include_directories ( GPURuntime_static PRIVATE
         ${CUDAToolkit_INCLUDE_DIRS}
-        ${SUITESPARSE_GPURUNTIME_INCLUDES}
-        $<TARGET_PROPERTY:SuiteSparse::SuiteSparseConfig,INTERFACE_INCLUDE_DIRECTORIES> )
+        ${SUITESPARSE_GPURUNTIME_INCLUDES} )
+
+    if ( TARGET SuiteSparse::SuiteSparseConfig_static )
+        target_link_libraries ( GPURuntime_static PUBLIC SuiteSparse::SuiteSparseConfig_static )
+    else ( )
+        target_link_libraries ( GPURuntime_static PUBLIC SuiteSparse::SuiteSparseConfig )
+    endif ( )
 
     set_target_properties ( GPURuntime_static PROPERTIES
         POSITION_INDEPENDENT_CODE ON
@@ -117,7 +123,7 @@ if ( BUILD_STATIC_LIBS )
         CUDA::nvrtc CUDA::cudart_static CUDA::cublas )
     target_compile_definitions ( GPURuntime_static PRIVATE "SUITESPARSE_CUDA" )
 
-    target_include_directories ( GPURuntime_static
+    target_include_directories ( GPURuntime_static 
         INTERFACE $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/Include>
                   $<INSTALL_INTERFACE:${SUITESPARSE_INCLUDEDIR}> )
 endif ( )

SuiteSparse_config/cmake_modules/SuiteSparsePolicy.cmake

@@ -255,30 +255,30 @@ if ( ENABLE_CUDA )
         # with CUDA:
         message ( STATUS "Find CUDA tool kit:" )
         # FindCUDAToolKit needs to have C or CXX enabled first (see above)
-        include ( FindCUDAToolkit )
+        find_package ( CUDAToolkit )
         message ( STATUS "CUDA toolkit found:   " ${CUDAToolkit_FOUND} )
         message ( STATUS "CUDA toolkit version: " ${CUDAToolkit_VERSION} )
         message ( STATUS "CUDA toolkit include: " ${CUDAToolkit_INCLUDE_DIRS} )
         message ( STATUS "CUDA toolkit lib dir: " ${CUDAToolkit_LIBRARY_DIR} )
         if ( CUDAToolkit_VERSION VERSION_LESS "11.2" )
             # CUDA is present but too old
             message ( STATUS "CUDA: not enabled (CUDA 11.2 or later required)" )
-            set ( SUITESPARSE_CUDA off )
+            set ( SUITESPARSE_CUDA OFF )
         else ( )
             # CUDA 11.2 or later present
             enable_language ( CUDA )
-            set ( SUITESPARSE_CUDA on )
+            set ( SUITESPARSE_CUDA ON )
         endif ( )
     else ( )
         # without CUDA:
         message ( STATUS "CUDA: not found" )
-        set ( SUITESPARSE_CUDA off )
+        set ( SUITESPARSE_CUDA OFF )
     endif ( )
 
 else ( )
 
     # CUDA is disabled
-    set ( SUITESPARSE_CUDA off )
+    set ( SUITESPARSE_CUDA OFF )
 
 endif ( )