GraphBLAS bug fix (GB_add_phase0), and latest CUDA draft kernels

Author: DrTimothyAldenDavis

GraphBLAS/CMakeLists.txt

@@ -203,11 +203,6 @@ configure_file ( "Config/README.md.in"
     "${PROJECT_SOURCE_DIR}/README.md"
     NEWLINE_STYLE LF )
 
-# for CUDA
-configure_file ( "CUDA/Config/GB_cuda_common_jitFactory.hpp.in"
-    "${PROJECT_SOURCE_DIR}/CUDA/GB_cuda_common_jitFactory.hpp"
-    NEWLINE_STYLE LF )
-
 #-------------------------------------------------------------------------------
 # include directories for both graphblas and the demos
 #-------------------------------------------------------------------------------
@@ -465,6 +460,7 @@ if ( GRAPHBLAS_HAS_OPENMP )
         target_link_libraries ( GraphBLAS_static PRIVATE OpenMP::OpenMP_C )
     endif ( )
     message ( STATUS "CMAKE OpenMP C flags:      ${OpenMP_C_FLAGS}" )
+    set ( GB_OPENMP_C_FLAGS "${OpenMP_C_FLAGS}" )
 else ( )
     message ( WARNING
     "WARNING:  OpenMP was not found (or was disabled with "
@@ -485,6 +481,7 @@ else ( )
         "The C compiler does not support thread-local-storage; "
         "GxB_Context_engage will return GrB_NOT_IMPLEMENTED." )
     endif ( )
+    set ( GB_OPENMP_C_FLAGS "" )
 endif ( )
 
 if ( SUITESPARSE_HAS_CUDA AND GRAPHBLAS_USE_CUDA )

GraphBLAS/CUDA/.gitignore

@@ -2,8 +2,6 @@
 *.o
 *.a
 *.so
-jitFactory
-stringify
 rmm_log.txt
 
 # Do not ignore this file

GraphBLAS/CUDA/CMakeLists.txt

@@ -2,7 +2,7 @@
 # GraphBLAS/CUDA/CMakeLists.txt:  cmake script for GraphBLAS/CUDA
 #-------------------------------------------------------------------------------
 
-# SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2023, All Rights Reserved.
+# SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2024, All Rights Reserved.
 
 # Some files in this folder are (c) NVIDIA or (c) Google.  Please refer
 # to their individual licenses (Apache, BSD, or others).
@@ -12,11 +12,6 @@
 
 cmake_minimum_required ( VERSION 3.20 ) # GraphBLAS can be built stand-alone
 
-# CMake build for generating googletest c++ files that can be compiled and
-# executed in parallel.  Build can be customized to speed up development by
-# allowing the targeting of specific specific parameters. The output of this
-# build is an executable that can be used to run the gtests.
-
 project ( GRAPHBLAS_CUDA
     VERSION "${GraphBLAS_VERSION_MAJOR}.${GraphBLAS_VERSION_MINOR}.${GraphBLAS_VERSION_SUB}"
     LANGUAGES CXX CUDA )
@@ -29,7 +24,6 @@ set ( CMAKE_CUDA_FLAGS "-cudart=static -lineinfo " )
 set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} --std=c++17 -fPIC " )
 
 add_compile_definitions ( GBNCPUFEAT )
-add_compile_definitions ( GBCUDA_CPLUSPLUS )
 
 message ( STATUS "C++ flags for CUDA: ${CMAKE_CXX_FLAGS}" )
 
@@ -160,126 +154,13 @@ if ( NOT MSVC )
 endif ( )
 
 #-------------------------------------------------------------------------------
-# test suite for the CUDA kernels
+# report
 #-------------------------------------------------------------------------------
 
-if ( 0 )
-
-# 1. Execute enumify/stringify/jitify logic to compile ptx kernels and
-# compile/link w/ relevant *.cu files.
-
-# TODO: Need to do this piece in cmake
-
-# 2. Generate test .cu files named "{semiring_operation}_test_instances.hpp"
-set ( CUDA_TEST_SUITES
-    AxB_dot3
-#    reduce_to_scalar
-)
-
-#
-set ( CUDA_TEST_MONOIDS PLUS MIN MAX) # TIMES ANY )
-set ( CUDA_TEST_BINOPS TIMES PLUS MIN MAX DIV ) #MINUS RDIV RMINUS FIRST SECOND PAIR )
-set ( CUDA_TEST_SEMIRINGS PLUS_TIMES MIN_PLUS MAX_PLUS )
-set ( CUDA_TEST_DATATYPES int32_t int64_t uint32_t uint64_t float double )
-set ( CUDA_TEST_KERNELS vsvs) # mp vsvs dndn spdn vssp )
-set ( CUDA_TEST_FORMATS sparse dense sparse_dense reduce )
-
-# TODO: Update testGen.py to accept the above CUDA_TEST_* params as arguments
-
-# Note: I don't believe there's a way to do this particular piece in parallel but
-# once all the files are written, we should be able to compile them in parallel
-
-# Separate individual kernels from larger "overview" test (e.g. 2-level testing structure)
-# We want to test all the *_cuda versions
-
-# TODO: make this a shorter test
-set(CUDA_TEST_CPP_FILES "")
-if ( FALSE ) # TODO: use a cmake option
-    foreach(var ${CUDA_TEST_SUITES})
-        foreach(semiring ${CUDA_TEST_SEMIRINGS})
-            foreach(kernel ${CUDA_TEST_KERNELS})
-                foreach(format ${CUDA_TEST_FORMATS})
-                    # TODO: Have Python script also build separate cudaTest.cpp (named something
-                    # like AxB_dot3_cuda_tests.cpp) for each suite. This way we should be able to
-                    # easily ignore them from the build
-                    add_custom_command(
-                            OUTPUT
-                            ${CMAKE_CURRENT_BINARY_DIR}/${var}_${semiring}_${format}_test_instances.hpp
-                            ${CMAKE_CURRENT_BINARY_DIR}/${var}_${semiring}_${format}_cuda_tests.cpp
-#                            DEPENDS
-#                            jitFactory.hpp
-                            COMMAND python3 ${CMAKE_CURRENT_SOURCE_DIR}/test/testGen_cmake.py "\"${CMAKE_CURRENT_SOURCE_DIR}\"" "\"${var}\"" "\"${CUDA_TEST_MONOIDS}\""
-                                "\"${CUDA_TEST_BINOPS}\"" "\"${semiring}\"" "\"${CUDA_TEST_DATATYPES}\""
-                                "\"${kernel}\""
-                    )
-                    # Construct final list of files to compile (in parallel)
-                    list(APPEND CUDA_TEST_CPP_FILES ${CMAKE_CURRENT_BINARY_DIR}/${var}_${semiring}_${format}_cuda_tests.cpp)
-                endforeach()
-            endforeach()
-        endforeach()
-    endforeach()
-endif ( )
-
-include(FetchContent)
-FetchContent_Declare(
-        googletest
-        # Specify the commit you depend on and update it regularly.
-        URL https://github.com/google/googletest/archive/e2239ee6043f73722e7aa812a459f54a28552929.zip
-)
-# For Windows: Prevent overriding the parent project's compiler/linker settings
-set(gtest_force_shared_crt ON CACHE BOOL "" FORCE)
-FetchContent_GetProperties(googletest)
-if(NOT googletest_POPULATED)
-    FetchContent_Populate(googletest)
-    add_subdirectory(${googletest_SOURCE_DIR} ${googletest_BINARY_DIR} EXCLUDE_FROM_ALL)
-endif()
-
-#FetchContent_MakeAvailable(googletest EC)
-
-
-#file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/external_includes)
-#execute_process(
-#        COMMAND git clone "https://github.com/google/googletest.git" googletest
-#        WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/external_includes)
-#
-#include_directories(${CMAKE_CURRENT_BINARY_DIR}/external_includes/googletest/googletest/include)
-
-#add_subdirectory(${CMAKE_CURRENT_BINARY_DIR}/external_includes/googletest/googletest/)
-
-# 3. Compile/link individual {test_suite_name}_cuda_tests.cpp files into a gtest executable
-set(GRAPHBLAS_CUDA_INCLUDES ${CMAKE_CURRENT_SOURCE_DIR}/test)
+message ( STATUS "CMAKE_CUDA_ARCHITECTURES: ${CMAKE_CUDA_ARCHITECTURES} ")
+message ( STATUS "CMAKE_CUDA_COMPILER: ${CMAKE_CUDA_COMPILER} ")
+message ( STATUS "CMAKE_CUDA_FLAGS: ${CMAKE_CUDA_FLAGS} ")
+message ( STATUS "CMAKE_CUDA_FLAGS_RELEASE: ${CMAKE_CUDA_FLAGS_RELEASE} ")
+message ( STATUS "CMAKE_CUDA_FLAGS_DEBUG: ${CMAKE_CUDA_FLAGS_DEBUG} ")
 
-message(STATUS "CUDA tests files: " "${CUDA_TEST_CPP_FILES}")
 
-add_executable(graphblascuda_test ${CUDA_TEST_CPP_FILES} ${CMAKE_CURRENT_SOURCE_DIR}/test/run_tests.cpp)
-
-set_target_properties(graphblascuda_test PROPERTIES POSITION_INDEPENDENT_CODE ON)
-set_target_properties(graphblascuda_test PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
-set_target_properties(graphblascuda_test PROPERTIES CUDA_ARCHITECTURES "52;75;80" )
-
-include(GoogleTest)
-
-if ( ENABLE_SHARED_LIBS )
-    target_link_libraries ( graphblascuda_test PUBLIC GraphBLAS  )
-else ( )
-    target_link_libraries ( graphblascuda_test PUBLIC GraphBLAS_static )
-endif ( )
-
-target_link_libraries ( graphblascuda_test
-    PUBLIC
-    GraphBLAS_CUDA
-    RMM_wrap
-    CUDA::cudart_static
-    CUDA::nvrtc
-    ${ADDITIONAL_DEPS}
-    PRIVATE
-    gtest_main )
-
-target_include_directories ( graphblascuda_test
-    PUBLIC
-    rmm_wrap
-    ${ADDITIONAL_INCLUDES}
-    ${CUDAToolkit_INCLUDE_DIRS}
-    ${GRAPHBLAS_CUDA_INCLUDES} )
-
-endif ( )

GraphBLAS/CUDA/Config/GB_cuda_common_jitFactory.hpp.in

@@ -1,4 +1,4 @@
-# GraphBLAS_CUDA, Copyright (c) 2017-2023, Timothy A. Davis.
+# GraphBLAS_CUDA, Copyright (c) 2017-2024, FIXME
 # All Rights Reserved.
 # SPDX-License-Identifier: Apache-2.0
 

GraphBLAS/CUDA/Config/GraphBLAS_CUDA.pc.in

@@ -4,7 +4,7 @@
 
 # The following copyright and license applies to just this file only, not to
 # the library itself:
-# GraphBLASConfig.cmake, Copyright (c) 2023, Timothy A. Davis.  All Rights Reserved.
+# GraphBLASConfig.cmake, Copyright (c) 2023-2024, FIXME
 # SPDX-License-Identifier: BSD-3-clause
 
 #-------------------------------------------------------------------------------