Merge remote-tracking branch 'upstream/develop' into staged-ci

crtrott · Oct 8, 2024 · 49e9c6a · 49e9c6a
2 parents 29cfc1a + 4e88587
commit 49e9c6a
Show file tree

Hide file tree

Showing 84 changed files with 3,021 additions and 438 deletions.
diff --git a/.git-blame-ignore-revs b/.git-blame-ignore-revs
@@ -1,3 +1,5 @@
+# Formatted CMake files with cmake-format
+0247634f35e2f9e6b9dec3c80cae567b15027554
 # Moved to clang-format-16
 60fb9cc94b40e698cbc3278c5538f58dee721276
 # Formatted the entire codebase with ClangFormat 8

diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml
@@ -24,7 +24,7 @@ jobs:
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@461ef6c76dfe95d5c364de2f431ddbd31a417628 # v3.26.9
+      uses: github/codeql-action/init@6db8d6351fd0be61f9ed8ebd12ccd35dcec51fea # v3.26.11
       with:
         languages: c-cpp
 
@@ -43,6 +43,6 @@ jobs:
         cmake --build build --parallel 2
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@461ef6c76dfe95d5c364de2f431ddbd31a417628 # v3.26.9
+      uses: github/codeql-action/analyze@6db8d6351fd0be61f9ed8ebd12ccd35dcec51fea # v3.26.11
       with:
         category: "/language:c-cpp"
diff --git a/.github/workflows/continuous-integration-linux-hpx.yml b/.github/workflows/continuous-integration-linux-hpx.yml
@@ -34,7 +34,7 @@ jobs:
           repository: STELLAR-GROUP/hpx
           ref: v1.9.0
           path: hpx
-      - uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      - uses: actions/cache@2cdf405574d6ef1f33a1d12acccd3ae82f47b3f2 # v4.1.0
         id:   cache-hpx
         with:
           path:         ./hpx/install

diff --git a/.github/workflows/continuous-integration-linux.yml b/.github/workflows/continuous-integration-linux.yml
@@ -88,7 +88,7 @@ jobs:
           sudo cmake --build . --target install --parallel 2
       - name: Checkout code
         uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
-      - uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      - uses: actions/cache@2cdf405574d6ef1f33a1d12acccd3ae82f47b3f2 # v4.1.0
         with:
           path: ~/.cache/ccache
           key: kokkos-${{ matrix.distro }}-${{ matrix.cxx }}-${{ matrix.cmake_build_type }}-${{ matrix.openmp }}-${{ github.ref }}-${{ github.sha }}

diff --git a/.github/workflows/performance-benchmark.yml b/.github/workflows/performance-benchmark.yml
@@ -23,7 +23,7 @@ jobs:
     steps:
       - name: Checkout code
         uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
-      - uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      - uses: actions/cache@2cdf405574d6ef1f33a1d12acccd3ae82f47b3f2 # v4.1.0
         with:
           path: ~/.cache/ccache
           key: kokkos-${{ matrix.distro }}-${{ matrix.cxx }}-${{ matrix.backend }}-${{ github.ref }}-${{ github.sha }}

diff --git a/.github/workflows/scorecard.yml b/.github/workflows/scorecard.yml
@@ -68,6 +68,6 @@ jobs:
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload SARIF results to code scanning"
-        uses: github/codeql-action/upload-sarif@461ef6c76dfe95d5c364de2f431ddbd31a417628 # v3.26.9
+        uses: github/codeql-action/upload-sarif@6db8d6351fd0be61f9ed8ebd12ccd35dcec51fea # v3.26.11
         with:
           sarif_file: results.sarif
diff --git a/.github/workflows/snl-ci.yml b/.github/workflows/snl-ci.yml
@@ -0,0 +1,56 @@
+name: SNL-CI
+
+on: 
+  pull_request:
+    paths-ignore:
+    - '**/*.md'
+    types: [ opened, reopened, synchronize ]
+
+permissions:
+  contents: none
+
+# Cancels any in progress 'workflow' associated with this PR
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  CUDA_12_2_CPP20:
+    name: SNL_CUDA_NVCC_12_2_CPP20
+    runs-on: [snl-kk-env-cuda-12.2.0-gcc-11.3.0-latest]
+
+    steps:
+      - name: checkout_kokkos
+        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
+        with:
+          repository: kokkos/kokkos
+          ref: ${{ github.base_ref }}
+          path: kokkos
+
+      - name: configure_kokkos
+        run: |
+          nvidia-smi
+          cd kokkos
+          cmake -B build \
+            -DCMAKE_CXX_STANDARD=20 \
+            -DKokkos_ENABLE_CUDA=ON \
+            -DKokkos_ARCH_HOPPER90=ON \
+            -DCMAKE_CXX_EXTENSIONS=OFF \
+            -DBUILD_SHARED_LIBS=OFF \
+            -DKokkos_ENABLE_DEPRECATED_CODE_4=OFF \
+            -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF \
+            -DKokkos_ENABLE_TESTS=ON \
+            -DKokkos_ENABLE_EXAMPLES=ON \
+            -DKokkos_ENABLE_BENCHMARKS=ON \
+            ./
+
+      - name: build_and_install_kokkos
+        working-directory: kokkos
+        run: |
+          cmake --build build -j36
+          cmake --install build --prefix install
+
+      - name: test_kokkos
+        working-directory: kokkos/build
+        run: ctest --output-on-failure --timeout 3600
+
diff --git a/.gitlab/hpsf-gitlab-ci.yml b/.gitlab/hpsf-gitlab-ci.yml
@@ -0,0 +1,9 @@
+test:
+  stage: test
+  tags: [nvidia-gh200]
+  image: masterleinad/kokkos-nvcc:12.6.1
+  script:
+    - cmake -B build -DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_HOPPER90=ON -DKokkos_ENABLE_IMPL_CUDA_UNIFIED_MEMORY=ON -DKokkos_ENABLE_TESTS=ON
+    - cmake --build build -j48
+    - cd build
+    - ctest -V
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -57,6 +57,7 @@
 * SIMD: Allow flexible vector width for 32 bit types [\#6802](https://github.com/kokkos/kokkos/pull/6802)
 * Updates for `Kokkos::Array`: add `kokkos_swap(Array<T, N>)` specialization [\#6943](https://github.com/kokkos/kokkos/pull/6943), add `Kokkos::to_array` [\#6375](https://github.com/kokkos/kokkos/pull/6375),  make `Kokkos::Array` equality-comparable [\#7148](https://github.com/kokkos/kokkos/pull/7148)
 * Structured binding support for `Kokkos::complex` [\#7040](https://github.com/kokkos/kokkos/pull/7040)
+* Introduce `KOKKOS_DEDUCTION_GUIDE` macro to allow for portable user-defined deduction guides [\#6954](https://github.com/kokkos/kokkos/pull/6954)
 
 ### Build System Changes
 * Do not require OpenMP support for languages other than CXX [\#6965](https://github.com/kokkos/kokkos/pull/6965)

diff --git a/Makefile.kokkos b/Makefile.kokkos
@@ -33,7 +33,7 @@ KOKKOS_STANDALONE_CMAKE ?= "no"
 # Options: force_uvm,use_ldg,rdc,enable_lambda,enable_constexpr,enable_malloc_async
 KOKKOS_CUDA_OPTIONS ?= ""
 
-# Options: rdc
+# Options: rdc,enable_malloc_async
 KOKKOS_HIP_OPTIONS ?= ""
 
 # Default settings specific options.
@@ -96,6 +96,7 @@ KOKKOS_INTERNAL_DISABLE_DEPRECATED_CODE := $(call kokkos_has_string,$(KOKKOS_OPT
 KOKKOS_INTERNAL_ENABLE_DEPRECATION_WARNINGS := $(call kokkos_has_string,$(KOKKOS_OPTIONS),enable_deprecation_warnings)
 
 KOKKOS_INTERNAL_HIP_USE_RELOC := $(call kokkos_has_string,$(KOKKOS_HIP_OPTIONS),rdc)
+KOKKOS_INTERNAL_HIP_ENABLE_MALLOC_ASYNC := $(call kokkos_has_string,$(KOKKOS_HIP_OPTIONS),enable_malloc_async)
 KOKKOS_INTERNAL_OPENACC_FORCE_HOST_AS_DEVICE := $(call kokkos_has_string,$(KOKKOS_OPENACC_OPTIONS),force_host_as_device)
 
 # Check for Kokkos Host Execution Spaces one of which must be on.
@@ -1228,6 +1229,12 @@ ifeq ($(KOKKOS_INTERNAL_USE_HIP), 1)
     KOKKOS_CXXFLAGS+=-fno-gpu-rdc
     KOKKOS_LDFLAGS+=-fno-gpu-rdc
   endif
+
+  ifeq ($(KOKKOS_INTERNAL_HIP_ENABLE_MALLOC_ASYNC), 1)
+    tmp := $(call kokkos_append_header,"$H""define KOKKOS_ENABLE_IMPL_HIP_MALLOC_ASYNC")
+  else
+    tmp := $(call kokkos_append_header,"/* $H""undef KOKKOS_ENABLE_IMPL_HIP_MALLOC_ASYNC */")
+  endif
 endif
 
 ifneq ($(KOKKOS_INTERNAL_USE_ARCH_AMD), 0)

diff --git a/algorithms/src/sorting/impl/Kokkos_SortByKeyImpl.hpp b/algorithms/src/sorting/impl/Kokkos_SortByKeyImpl.hpp
@@ -30,6 +30,7 @@
 
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wshadow"
+#pragma GCC diagnostic ignored "-Wsuggest-override"
 
 #if defined(KOKKOS_COMPILER_CLANG)
 // Some versions of Clang fail to compile Thrust, failing with errors like

diff --git a/algorithms/src/sorting/impl/Kokkos_SortImpl.hpp b/algorithms/src/sorting/impl/Kokkos_SortImpl.hpp
@@ -34,6 +34,7 @@
 
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wshadow"
+#pragma GCC diagnostic ignored "-Wsuggest-override"
 
 #if defined(KOKKOS_COMPILER_CLANG)
 // Some versions of Clang fail to compile Thrust, failing with errors like

diff --git a/algorithms/unit_tests/TestBinSortA.hpp b/algorithms/unit_tests/TestBinSortA.hpp
@@ -219,6 +219,10 @@ void test_sort_integer_overflow() {
 }  // namespace BinSortSetA
 
 TEST(TEST_CATEGORY, BinSortGenericTests) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
   using ExecutionSpace = TEST_EXECSPACE;
   using key_type       = unsigned;
   constexpr int N      = 171;

diff --git a/algorithms/unit_tests/TestBinSortB.hpp b/algorithms/unit_tests/TestBinSortB.hpp
@@ -185,6 +185,10 @@ void run_for_rank2() {
 }  // namespace BinSortSetB
 
 TEST(TEST_CATEGORY, BinSortUnsignedKeyLayoutStrideValues) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
   using ExeSpace = TEST_EXECSPACE;
   using key_type = unsigned;
   BinSortSetB::run_for_rank1<ExeSpace, key_type, int>();

diff --git a/algorithms/unit_tests/TestNestedSort.hpp b/algorithms/unit_tests/TestNestedSort.hpp
@@ -386,6 +386,11 @@ void test_nested_sort_by_key(unsigned int N, KeyType minKey, KeyType maxKey,
 }  // namespace NestedSortImpl
 
 TEST(TEST_CATEGORY, NestedSort) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
+
   using ExecutionSpace = TEST_EXECSPACE;
   NestedSortImpl::test_nested_sort<ExecutionSpace, unsigned>(171, 0U, UINT_MAX);
   NestedSortImpl::test_nested_sort<ExecutionSpace, float>(42, -1e6f, 1e6f);
@@ -394,6 +399,11 @@ TEST(TEST_CATEGORY, NestedSort) {
 }
 
 TEST(TEST_CATEGORY, NestedSortByKey) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
+
   using ExecutionSpace = TEST_EXECSPACE;
 
   // Second/third template arguments are key and value respectively.

diff --git a/algorithms/unit_tests/TestRandom.hpp b/algorithms/unit_tests/TestRandom.hpp
@@ -542,6 +542,11 @@ void test_duplicate_stream() {
 }  // namespace AlgoRandomImpl
 
 TEST(TEST_CATEGORY, Random_XorShift64) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
+
   using ExecutionSpace = TEST_EXECSPACE;
 
 #if defined(KOKKOS_ENABLE_SYCL) || defined(KOKKOS_ENABLE_CUDA) || \
@@ -562,6 +567,10 @@ TEST(TEST_CATEGORY, Random_XorShift64) {
 
 TEST(TEST_CATEGORY, Random_XorShift1024_0) {
   using ExecutionSpace = TEST_EXECSPACE;
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
 
 #if defined(KOKKOS_ENABLE_SYCL) || defined(KOKKOS_ENABLE_CUDA) || \
     defined(KOKKOS_ENABLE_HIP)

diff --git a/algorithms/unit_tests/TestRandomAccessIterator.cpp b/algorithms/unit_tests/TestRandomAccessIterator.cpp
@@ -23,7 +23,7 @@ namespace stdalgos {
 
 struct random_access_iterator_test : std_algorithms_test {
  public:
-  virtual void SetUp() {
+  void SetUp() override {
     Kokkos::parallel_for(m_static_view.extent(0),
                          AssignIndexFunctor<static_view_t>(m_static_view));
 

diff --git a/algorithms/unit_tests/TestSort.hpp b/algorithms/unit_tests/TestSort.hpp
@@ -209,6 +209,10 @@ void test_sort_integer_overflow() {
 }  // namespace SortImpl
 
 TEST(TEST_CATEGORY, SortUnsignedValueType) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
   using ExecutionSpace = TEST_EXECSPACE;
   using key_type       = unsigned;
   constexpr int N      = 171;
@@ -224,6 +228,10 @@ TEST(TEST_CATEGORY, SortUnsignedValueType) {
 }
 
 TEST(TEST_CATEGORY, SortEmptyView) {
+  // FIXME_OPENMPTARGET - causes runtime failure with CrayClang compiler
+#if defined(KOKKOS_COMPILER_CRAY_LLVM) && defined(KOKKOS_ENABLE_OPENMPTARGET)
+  GTEST_SKIP() << "known to fail with OpenMPTarget+Cray LLVM";
+#endif
   using ExecutionSpace = TEST_EXECSPACE;
 
   // does not matter if we use int or something else

diff --git a/algorithms/unit_tests/TestSortByKey.hpp b/algorithms/unit_tests/TestSortByKey.hpp
@@ -234,7 +234,9 @@ TEST(TEST_CATEGORY, SortByKeyWithStrides) {
   ASSERT_EQ(sort_fails, 0u);
 }
 
-TEST(TEST_CATEGORY, SortByKeyKeysLargerThanValues) {
+TEST(TEST_CATEGORY_DEATH, SortByKeyKeysLargerThanValues) {
+  ::testing::FLAGS_gtest_death_test_style = "threadsafe";
+
   using ExecutionSpace = TEST_EXECSPACE;
 
   // does not matter if we use int or something else

diff --git a/algorithms/unit_tests/TestStdAlgorithmsModSeqOps.cpp b/algorithms/unit_tests/TestStdAlgorithmsModSeqOps.cpp
@@ -23,7 +23,7 @@ namespace stdalgos {
 
 struct std_algorithms_mod_seq_ops_test : std_algorithms_test {
  public:
-  virtual void SetUp() {
+  void SetUp() override {
     Kokkos::parallel_for(m_static_view.extent(0),
                          AssignIndexFunctor<static_view_t>(m_static_view));
   }

diff --git a/cmake/KokkosCore_config.h.in b/cmake/KokkosCore_config.h.in
@@ -39,6 +39,7 @@
 #cmakedefine KOKKOS_ENABLE_IMPL_CUDA_UNIFIED_MEMORY
 #cmakedefine KOKKOS_ENABLE_HIP_RELOCATABLE_DEVICE_CODE
 #cmakedefine KOKKOS_ENABLE_HIP_MULTIPLE_KERNEL_INSTANTIATIONS
+#cmakedefine KOKKOS_ENABLE_IMPL_HIP_MALLOC_ASYNC
 #cmakedefine KOKKOS_ENABLE_IMPL_HIP_UNIFIED_MEMORY
 #cmakedefine KOKKOS_ENABLE_SYCL_RELOCATABLE_DEVICE_CODE
 #cmakedefine KOKKOS_IMPL_SYCL_DEVICE_GLOBAL_SUPPORTED

diff --git a/cmake/kokkos_arch.cmake b/cmake/kokkos_arch.cmake
@@ -147,6 +147,7 @@ if(KOKKOS_ENABLE_COMPILER_WARNINGS)
       "-Wsign-compare"
       "-Wtype-limits"
       "-Wuninitialized"
+      "-Wsuggest-override"
   )
 
   # NOTE KOKKOS_ prefixed variable (all uppercase) is not set yet because TPLs are processed after ARCH
@@ -762,6 +763,8 @@ if(KOKKOS_ENABLE_SYCL)
       compiler_specific_flags(DEFAULT -fsycl-device-code-split=off -DDESUL_SYCL_DEVICE_GLOBAL_SUPPORTED)
     endif()
   endif()
+
+  check_cxx_symbol_exists(SYCL_EXT_ONEAPI_GRAPH "sycl/sycl.hpp" KOKKOS_IMPL_HAVE_SYCL_EXT_ONEAPI_GRAPH)
 endif()
 
 set(CUDA_ARCH_ALREADY_SPECIFIED "")

diff --git a/cmake/kokkos_enable_options.cmake b/cmake/kokkos_enable_options.cmake
@@ -77,6 +77,7 @@ kokkos_enable_option(
   "Whether multiple kernels are instantiated at compile time - improve performance but increase compile time"
 )
 kokkos_enable_option(IMPL_HIP_UNIFIED_MEMORY OFF "Whether to leverage unified memory architectures for HIP")
+kokkos_enable_option(IMPL_HIP_MALLOC_ASYNC OFF "Whether to enable hipMallocAsync")
 kokkos_enable_option(OPENACC_FORCE_HOST_AS_DEVICE OFF "Whether to force to use host as a target device for OpenACC")
 
 # This option will go away eventually, but allows fallback to old implementation when needed.
@@ -167,7 +168,13 @@ check_device_specific_options(
   IMPL_CUDA_UNIFIED_MEMORY
 )
 check_device_specific_options(
-  DEVICE HIP OPTIONS HIP_RELOCATABLE_DEVICE_CODE HIP_MULTIPLE_KERNEL_INSTANTIATIONS IMPL_HIP_UNIFIED_MEMORY
+  DEVICE
+  HIP
+  OPTIONS
+  HIP_RELOCATABLE_DEVICE_CODE
+  HIP_MULTIPLE_KERNEL_INSTANTIATIONS
+  IMPL_HIP_UNIFIED_MEMORY
+  IMPL_HIP_MALLOC_ASYNC
 )
 check_device_specific_options(DEVICE HPX OPTIONS IMPL_HPX_ASYNC_DISPATCH)
 check_device_specific_options(DEVICE OPENACC OPTIONS OPENACC_FORCE_HOST_AS_DEVICE)

diff --git a/containers/src/Kokkos_DynRankView.hpp b/containers/src/Kokkos_DynRankView.hpp
@@ -1361,26 +1361,6 @@ struct MirrorDRViewType {
       std::conditional_t<is_same_memspace, src_view_type, dest_view_type>;
 };
 
-template <class Space, class T, class... P>
-struct MirrorDRVType {
-  // The incoming view_type
-  using src_view_type = typename Kokkos::DynRankView<T, P...>;
-  // The memory space for the mirror view
-  using memory_space = typename Space::memory_space;
-  // Check whether it is the same memory space
-  enum {
-    is_same_memspace =
-        std::is_same_v<memory_space, typename src_view_type::memory_space>
-  };
-  // The array_layout
-  using array_layout = typename src_view_type::array_layout;
-  // The data type (we probably want it non-const since otherwise we can't even
-  // deep_copy to it.
-  using data_type = typename src_view_type::non_const_data_type;
-  // The destination view type if it is not the same memory space
-  using view_type = Kokkos::DynRankView<data_type, array_layout, Space>;
-};
-
 }  // namespace Impl
 
 namespace Impl {
@@ -1397,9 +1377,9 @@ inline auto create_mirror(const DynRankView<T, P...>& src,
       arg_prop, std::string(src.label()).append("_mirror"));
 
   if constexpr (Impl::ViewCtorProp<ViewCtorArgs...>::has_memory_space) {
-    using dst_type = typename Impl::MirrorDRVType<
+    using dst_type = typename Impl::MirrorDRViewType<
         typename Impl::ViewCtorProp<ViewCtorArgs...>::memory_space, T,
-        P...>::view_type;
+        P...>::dest_view_type;
     return dst_type(prop_copy,
                     Impl::reconstructLayout(src.layout(), src.rank()));
   } else {