Merge branch 'develop' into fix/2433

kokkos · Dec 2, 2024 · 5933e67 · 5933e67
2 parents 735c72a + 78f4efd
commit 5933e67
Show file tree

Hide file tree

Showing 5 changed files with 97 additions and 138 deletions.
diff --git a/.github/workflows/at2.yml b/.github/workflows/at2.yml
@@ -38,7 +38,8 @@ jobs:
     uses: ./.github/workflows/spr.yml
     with:
       kokkos_version: 4.5.00
-  volta70:
-    uses: ./.github/workflows/volta70.yml
-    with:
-      kokkos_version: 4.5.00
+  # All these ran on sogpu01, which has been repurposed
+  # volta70:
+  #   uses: ./.github/workflows/volta70.yml
+  #   with:
+  #     kokkos_version: 4.5.00
diff --git a/.github/workflows/bdw.yml b/.github/workflows/bdw.yml
@@ -261,86 +261,87 @@ jobs:
         working-directory: kokkos-kernels/build
         run: ctest --output-on-failure -V --timeout 3600
 
-  PR_BDW_CLANG1001_THREADS_SERIAL_LEFT_REL:
-    name: PR_BDW_CLANG1001_THREADS_SERIAL_LEFT_REL
-    runs-on: [kk-env-llvm-10.0.1-latest]
+  # this job ran on sogpu01 which is now used for another purpose
+  # PR_BDW_CLANG1001_THREADS_SERIAL_LEFT_REL:
+  #   name: PR_BDW_CLANG1001_THREADS_SERIAL_LEFT_REL
+  #   runs-on: [kk-env-llvm-10.0.1-latest]
 
-    steps:
-      - name: checkout_kokkos_kernels
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
-        with:
-          path: kokkos-kernels
+  #   steps:
+  #     - name: checkout_kokkos_kernels
+  #       uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+  #       with:
+  #         path: kokkos-kernels
 
-      - name: checkout_kokkos
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
-        with:
-          repository: kokkos/kokkos
-          ref: ${{ inputs.kokkos_version }}
-          path: kokkos
+  #     - name: checkout_kokkos
+  #       uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+  #       with:
+  #         repository: kokkos/kokkos
+  #         ref: ${{ inputs.kokkos_version }}
+  #         path: kokkos
 
-      - name: configure_kokkos
-        run: |
-          mkdir -p kokkos/{build,install}
-          cd kokkos/build
-          cmake \
-            -DCMAKE_CXX_COMPILER=clang++ \
-            -DCMAKE_CXX_FLAGS="-O3 -Wall -Wunused-parameter -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wuninitialized" \
-            -DCMAKE_EXE_LINKER_FLAGS= \
-            -DCMAKE_INSTALL_PREFIX=$PWD/../install \
-            -DKokkos_ENABLE_SERIAL=ON \
-            -DKokkos_ENABLE_THREADS=ON \
-            -DKokkos_ARCH_BDW=ON \
-            -DKokkos_ENABLE_TESTS=OFF \
-            -DKokkos_ENABLE_EXAMPLES=OFF \
-            -DCMAKE_VERBOSE_MAKEFILE=ON \
-            -DCMAKE_CXX_EXTENSIONS=OFF \
-            -DCMAKE_CXX_STANDARD=17 \
-            -DBUILD_SHARED_LIBS=OFF \
-            -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF \
-            -DKokkos_ENABLE_DEPRECATED_CODE_4=OFF \
-            -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF \
-            ..
+  #     - name: configure_kokkos
+  #       run: |
+  #         mkdir -p kokkos/{build,install}
+  #         cd kokkos/build
+  #         cmake \
+  #           -DCMAKE_CXX_COMPILER=clang++ \
+  #           -DCMAKE_CXX_FLAGS="-O3 -Wall -Wunused-parameter -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wuninitialized" \
+  #           -DCMAKE_EXE_LINKER_FLAGS= \
+  #           -DCMAKE_INSTALL_PREFIX=$PWD/../install \
+  #           -DKokkos_ENABLE_SERIAL=ON \
+  #           -DKokkos_ENABLE_THREADS=ON \
+  #           -DKokkos_ARCH_BDW=ON \
+  #           -DKokkos_ENABLE_TESTS=OFF \
+  #           -DKokkos_ENABLE_EXAMPLES=OFF \
+  #           -DCMAKE_VERBOSE_MAKEFILE=ON \
+  #           -DCMAKE_CXX_EXTENSIONS=OFF \
+  #           -DCMAKE_CXX_STANDARD=17 \
+  #           -DBUILD_SHARED_LIBS=OFF \
+  #           -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF \
+  #           -DKokkos_ENABLE_DEPRECATED_CODE_4=OFF \
+  #           -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF \
+  #           ..
 
-      - name: build_and_install_kokkos
-        working-directory: kokkos/build
-        run: make -j8 install
+  #     - name: build_and_install_kokkos
+  #       working-directory: kokkos/build
+  #       run: make -j8 install
 
-      - name: configure_kokkos_kernels
-        run: |
-          mkdir -p kokkos-kernels/{build,install}
-          cd kokkos-kernels/build
-          cmake \
-            -DCMAKE_CXX_COMPILER=clang++ \
-            -DKokkos_DIR=$PWD/../../kokkos/install/lib64/cmake/Kokkos \
-            -DCMAKE_CXX_FLAGS="-O3 -Wall -Wunused-parameter -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wuninitialized" \
-            -DCMAKE_INSTALL_PREFIX= \
-            -DKokkosKernels_ENABLE_TESTS_AND_PERFSUITE=OFF \
-            -DKokkosKernels_ENABLE_TESTS=ON \
-            -DKokkosKernels_ENABLE_PERFTESTS=ON \
-            -DKokkosKernels_ENABLE_EXAMPLES:BOOL=ON \
-            -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=OFF \
-            -DKokkosKernels_INST_COMPLEX_DOUBLE=ON \
-            -DKokkosKernels_INST_DOUBLE=ON \
-            -DKokkosKernels_INST_ORDINAL_INT=ON \
-            -DKokkosKernels_INST_OFFSET_SIZE_T=ON \
-            -DKokkosKernels_INST_OFFSET_INT=ON \
-            -DKokkosKernels_INST_LAYOUTLEFT=ON \
-            -DKokkosKernels_ENABLE_TPL_ROCSPARSE=OFF \
-            -DKokkosKernels_ENABLE_TPL_ROCBLAS=OFF \
-            -DKokkosKernels_ENABLE_TPL_CUSOLVER=OFF \
-            -DKokkosKernels_ENABLE_TPL_CUSPARSE=OFF \
-            -DKokkosKernels_ENABLE_TPL_CUBLAS=OFF \
-            -DCMAKE_EXE_LINKER_FLAGS="" \
-            -DBUILD_SHARED_LIBS=OFF \
-            -DKokkosKernels_ENABLE_DOCS=OFF \
-            ..
+  #     - name: configure_kokkos_kernels
+  #       run: |
+  #         mkdir -p kokkos-kernels/{build,install}
+  #         cd kokkos-kernels/build
+  #         cmake \
+  #           -DCMAKE_CXX_COMPILER=clang++ \
+  #           -DKokkos_DIR=$PWD/../../kokkos/install/lib64/cmake/Kokkos \
+  #           -DCMAKE_CXX_FLAGS="-O3 -Wall -Wunused-parameter -Wshadow -pedantic -Werror -Wsign-compare -Wtype-limits -Wuninitialized" \
+  #           -DCMAKE_INSTALL_PREFIX= \
+  #           -DKokkosKernels_ENABLE_TESTS_AND_PERFSUITE=OFF \
+  #           -DKokkosKernels_ENABLE_TESTS=ON \
+  #           -DKokkosKernels_ENABLE_PERFTESTS=ON \
+  #           -DKokkosKernels_ENABLE_EXAMPLES:BOOL=ON \
+  #           -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=OFF \
+  #           -DKokkosKernels_INST_COMPLEX_DOUBLE=ON \
+  #           -DKokkosKernels_INST_DOUBLE=ON \
+  #           -DKokkosKernels_INST_ORDINAL_INT=ON \
+  #           -DKokkosKernels_INST_OFFSET_SIZE_T=ON \
+  #           -DKokkosKernels_INST_OFFSET_INT=ON \
+  #           -DKokkosKernels_INST_LAYOUTLEFT=ON \
+  #           -DKokkosKernels_ENABLE_TPL_ROCSPARSE=OFF \
+  #           -DKokkosKernels_ENABLE_TPL_ROCBLAS=OFF \
+  #           -DKokkosKernels_ENABLE_TPL_CUSOLVER=OFF \
+  #           -DKokkosKernels_ENABLE_TPL_CUSPARSE=OFF \
+  #           -DKokkosKernels_ENABLE_TPL_CUBLAS=OFF \
+  #           -DCMAKE_EXE_LINKER_FLAGS="" \
+  #           -DBUILD_SHARED_LIBS=OFF \
+  #           -DKokkosKernels_ENABLE_DOCS=OFF \
+  #           ..
 
-      - name: build_kokkos_kernels
-        working-directory: kokkos-kernels/build
-        run: make -j8 all
+  #     - name: build_kokkos_kernels
+  #       working-directory: kokkos-kernels/build
+  #       run: make -j8 all
 
-      - name: test
-        working-directory: kokkos-kernels/build
-        run: ctest --output-on-failure -V --timeout 3600
+  #     - name: test
+  #       working-directory: kokkos-kernels/build
+  #       run: ctest --output-on-failure -V --timeout 3600
 
 
diff --git a/sparse/impl/KokkosSparse_sptrsv_solve_impl.hpp b/sparse/impl/KokkosSparse_sptrsv_solve_impl.hpp
@@ -2083,7 +2083,6 @@ struct SptrsvWrap {
                                 const std::vector<RowMapType> &row_map_v, const std::vector<EntriesType> &entries_v,
                                 const std::vector<ValuesType> &values_v, const std::vector<RHSType> &rhs_v,
                                 std::vector<LHSType> &lhs_v) {
-    // NOTE: Only support SEQLVLSCHD_RP and SEQLVLSCHD_TP1 at this moment
     using nodes_per_level_type        = typename TriSolveHandle::hostspace_nnz_lno_view_t;
     using nodes_grouped_by_level_type = typename TriSolveHandle::nnz_lno_view_t;
     using RPFunctor                   = FunctorTypeMacro(TriLvlSchedRPSolverFunctor, IsLower, BlockEnabled);
@@ -2131,6 +2130,10 @@ struct SptrsvWrap {
               const int scratch_size = TPFunctor::SBlock::shmem_size(block_size, block_size);
               tp                     = tp.set_scratch_size(0, Kokkos::PerTeam(scratch_size));
               Kokkos::parallel_for("parfor_l_team", tp, tstf);
+            } else {
+              // NOTE: Only support SEQLVLSCHD_RP and SEQLVLSCHD_TP1 at this moment
+              auto alg_name = thandle_v[i]->return_algorithm_string();
+              KK_REQUIRE_MSG(false, "Algorithm " << alg_name << " does not support streams");
             }
             node_count_v[i] += lvl_nodes;
           }  // end if (lvl_nodes != 0)

diff --git a/sparse/src/KokkosSparse_sptrsv_handle.hpp b/sparse/src/KokkosSparse_sptrsv_handle.hpp
@@ -959,69 +959,26 @@ class SPTRSVHandle {
   int get_num_chain_entries() const { return this->num_chain_entries; }
   void set_num_chain_entries(const int nce) { this->num_chain_entries = nce; }
 
-  void print_algorithm() {
-    if (algm == SPTRSVAlgorithm::SEQLVLSCHD_RP) std::cout << "SEQLVLSCHD_RP" << std::endl;
-    ;
-
-    if (algm == SPTRSVAlgorithm::SEQLVLSCHD_TP1) std::cout << "SEQLVLSCHD_TP1" << std::endl;
-    ;
-    /*
-        if ( algm == SPTRSVAlgorithm::SEQLVLSCHED_TP2 ) {
-          std::cout << "SEQLVLSCHED_TP2" << std::endl;;
-          std::cout << "WARNING: With CUDA this is currently only reliable with
-       int-int ordinal-offset pair" << std::endl;
-        }
-    */
-    if (algm == SPTRSVAlgorithm::SEQLVLSCHD_TP1CHAIN) std::cout << "SEQLVLSCHD_TP1CHAIN" << std::endl;
-    ;
-
-    if (algm == SPTRSVAlgorithm::SPTRSV_CUSPARSE) std::cout << "SPTRSV_CUSPARSE" << std::endl;
-    ;
-
-    if (algm == SPTRSVAlgorithm::SUPERNODAL_NAIVE) std::cout << "SUPERNODAL_NAIVE" << std::endl;
-
-    if (algm == SPTRSVAlgorithm::SUPERNODAL_ETREE) std::cout << "SUPERNODAL_ETREE" << std::endl;
-
-    if (algm == SPTRSVAlgorithm::SUPERNODAL_DAG) std::cout << "SUPERNODAL_DAG" << std::endl;
-
-    if (algm == SPTRSVAlgorithm::SUPERNODAL_SPMV) std::cout << "SUPERNODAL_SPMV" << std::endl;
-
-    if (algm == SPTRSVAlgorithm::SUPERNODAL_SPMV_DAG) std::cout << "SUPERNODAL_SPMV_DAG" << std::endl;
-  }
+  inline void print_algorithm() { std::cout << return_algorithm_string() << std::endl; }
 
   std::string return_algorithm_string() {
     std::string ret_string;
 
-    if (algm == SPTRSVAlgorithm::SEQLVLSCHD_RP) ret_string = "SEQLVLSCHD_RP";
-
-    if (algm == SPTRSVAlgorithm::SEQLVLSCHD_TP1) ret_string = "SEQLVLSCHD_TP1";
-    /*
-        if ( algm == SPTRSVAlgorithm::SEQLVLSCHED_TP2 )
-          ret_string = "SEQLVLSCHED_TP2";
-    */
-    if (algm == SPTRSVAlgorithm::SEQLVLSCHD_TP1CHAIN) ret_string = "SEQLVLSCHD_TP1CHAIN";
-
-    if (algm == SPTRSVAlgorithm::SPTRSV_CUSPARSE) ret_string = "SPTRSV_CUSPARSE";
+    switch (algm) {
+      case SPTRSVAlgorithm::SEQLVLSCHD_RP: ret_string = "SEQLVLSCHD_RP"; break;
+      case SPTRSVAlgorithm::SEQLVLSCHD_TP1: ret_string = "SEQLVLSCHD_TP1"; break;
+      case SPTRSVAlgorithm::SEQLVLSCHD_TP1CHAIN: ret_string = "SEQLVLSCHD_TP1CHAIN"; break;
+      case SPTRSVAlgorithm::SPTRSV_CUSPARSE: ret_string = "SPTRSV_CUSPARSE"; break;
+      case SPTRSVAlgorithm::SUPERNODAL_NAIVE: ret_string = "SUPERNODAL_NAIVE"; break;
+      case SPTRSVAlgorithm::SUPERNODAL_ETREE: ret_string = "SUPERNODAL_ETREE"; break;
+      case SPTRSVAlgorithm::SUPERNODAL_DAG: ret_string = "SUPERNODAL_DAG"; break;
+      case SPTRSVAlgorithm::SUPERNODAL_SPMV: ret_string = "SUPERNODAL_SPMV"; break;
+      case SPTRSVAlgorithm::SUPERNODAL_SPMV_DAG: ret_string = "SUPERNODAL_SPMV_DAG"; break;
+      default: KK_REQUIRE_MSG(false, "Unhandled sptrsv algorithm: " << static_cast<int>(algm));
+    }
 
     return ret_string;
   }
-
-  inline SPTRSVAlgorithm StringToSPTRSVAlgorithm(std::string &name) {
-    if (name == "SPTRSV_DEFAULT")
-      return SPTRSVAlgorithm::SEQLVLSCHD_RP;
-    else if (name == "SPTRSV_RANGEPOLICY")
-      return SPTRSVAlgorithm::SEQLVLSCHD_RP;
-    else if (name == "SPTRSV_TEAMPOLICY1")
-      return SPTRSVAlgorithm::SEQLVLSCHD_TP1;
-    /*else if(name=="SPTRSV_TEAMPOLICY2")       return
-     * SPTRSVAlgorithm::SEQLVLSCHED_TP2;*/
-    else if (name == "SPTRSV_TEAMPOLICY1CHAIN")
-      return SPTRSVAlgorithm::SEQLVLSCHD_TP1CHAIN;
-    else if (name == "SPTRSV_CUSPARSE")
-      return SPTRSVAlgorithm::SPTRSV_CUSPARSE;
-    else
-      throw std::runtime_error("Invalid SPTRSVAlgorithm name");
-  }
 };
 
 }  // namespace Experimental

diff --git a/sparse/unit_test/Test_Sparse_sptrsv.hpp b/sparse/unit_test/Test_Sparse_sptrsv.hpp
@@ -620,9 +620,6 @@ template <typename scalar_t, typename lno_t, typename size_type, typename device
 void test_sptrsv_streams() {
   using TestStruct                  = Test::SptrsvTest<scalar_t, lno_t, size_type, device>;
   std::vector<SPTRSVAlgorithm> algs = {SPTRSVAlgorithm::SEQLVLSCHD_RP, SPTRSVAlgorithm::SEQLVLSCHD_TP1};
-  if (TestStruct::do_cusparse()) {
-    algs.push_back(SPTRSVAlgorithm::SPTRSV_CUSPARSE);
-  }
 
   for (auto alg : algs) {
     for (int nstreams = 1; nstreams <= 4; ++nstreams) {