Add a nvrtc configuration to the libcu++ build matrix

NVIDIA · Jul 11, 2023 · cc12a71 · cc12a71
1 parent 25b7a06
commit cc12a71
Show file tree

Hide file tree

Showing 4 changed files with 64 additions and 10 deletions.
diff --git a/.github/workflows/pr.yml b/.github/workflows/pr.yml
@@ -30,8 +30,8 @@ jobs:
     uses: ./.github/workflows/dispatch-build-and-test.yml
     strategy:
       fail-fast: false
-      matrix: 
-        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }} 
+      matrix:
+        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }}
         compiler: ${{ fromJSON(needs.compute-matrix.outputs.COMPILERS) }}
     with:
       per_cuda_compiler_matrix: ${{ toJSON(fromJSON(needs.compute-matrix.outputs.PER_CUDA_COMPILER_MATRIX)[ format('{0}-{1}', matrix.cuda_version, matrix.compiler) ]) }}
@@ -44,36 +44,54 @@ jobs:
     uses: ./.github/workflows/dispatch-build-and-test.yml
     strategy:
       fail-fast: false
-      matrix: 
-        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }} 
+      matrix:
+        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }}
         compiler: ${{ fromJSON(needs.compute-matrix.outputs.COMPILERS) }}
     with:
       per_cuda_compiler_matrix: ${{ toJSON(fromJSON(needs.compute-matrix.outputs.PER_CUDA_COMPILER_MATRIX)[ format('{0}-{1}', matrix.cuda_version, matrix.compiler) ]) }}
       build_script: "./ci/build_cub.sh"
       test_script: "./ci/test_cub.sh"
-  
+
   libcudacxx:
     name: libcudacxx CUDA${{ matrix.cuda_version }} ${{ matrix.compiler }}
     needs: compute-matrix
     uses: ./.github/workflows/dispatch-build-and-test.yml
     strategy:
       fail-fast: false
-      matrix: 
-        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }} 
+      matrix:
+        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }}
         compiler: ${{ fromJSON(needs.compute-matrix.outputs.COMPILERS) }}
     with:
       per_cuda_compiler_matrix: ${{ toJSON(fromJSON(needs.compute-matrix.outputs.PER_CUDA_COMPILER_MATRIX)[ format('{0}-{1}', matrix.cuda_version, matrix.compiler) ]) }}
       build_script: "./ci/build_libcudacxx.sh"
-      test_script: "./ci/test_libcudacxx.sh" 
+      test_script: "./ci/test_libcudacxx.sh"
+
+  nvrtc:
+    name: libcudacxx nvrtc
+    needs: compute-matrix
+    uses: ./.github/workflows/dispatch-build-and-test.yml
+    strategy:
+      fail-fast: false
+      matrix:
+        cuda_version: ${{ fromJSON(needs.compute-matrix.outputs.CUDA_VERSIONS) }}
+        compiler: ${{ fromJSON(needs.compute-matrix.outputs.COMPILERS) }}
+        exclude:
+          - cuda_version: "11.1"
+          - compiler: clang
+    with:
+      per_cuda_compiler_matrix: ${{ toJSON(fromJSON(needs.compute-matrix.outputs.PER_CUDA_COMPILER_MATRIX)[ format('{0}-{1}', matrix.cuda_version, matrix.compiler) ]) }}
+      build_script: "./ci/build_libcudacxx_nvrtc.sh"
+      test_script: "./ci/test_libcudacxx_nvrtc.sh"
 
   # This job is the final job that runs after all other jobs and is used for branch protection status checks.
   # See: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/collaborating-on-repositories-with-code-quality-features/about-status-checks
   ci:
     runs-on: ubuntu-latest
     name: CI
     needs:
-      - libcudacxx
       - cub
+      - libcudacxx
+      - nvrtc
       - thrust
     steps:
       - run: echo "CI success"
diff --git a/ci/build_libcudacxx_nvrtc.sh b/ci/build_libcudacxx_nvrtc.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+
+source "$(dirname "$0")/build_common.sh"
+
+CMAKE_OPTIONS="
+    -DCCCL_ENABLE_THRUST=OFF \
+    -DCCCL_ENABLE_LIBCUDACXX=ON \
+    -DCCCL_ENABLE_CUB=OFF \
+    -DCCCL_ENABLE_TESTING=OFF \
+    -DLIBCUDACXX_ENABLE_LIBCUDACXX_TESTS=ON \
+    -DLIBCUDACXX_TEST_WITH_NVRTC=ON \
+"
+configure "$CMAKE_OPTIONS"
+
+source "./sccache_stats.sh" "start"
+LIBCUDACXX_SITE_CONFIG="${BUILD_DIR}/libcudacxx/test/lit.site.cfg" lit -v --no-progress-bar -Dexecutor="NoopExecutor()" -Dcompute_archs=${GPU_ARCHS} -Dstd="c++${CXX_STANDARD}" ../libcudacxx/.upstream-tests/test
+source "./sccache_stats.sh" "end"
diff --git a/ci/test_libcudacxx_nvrtc.sh b/ci/test_libcudacxx_nvrtc.sh
@@ -0,0 +1,19 @@
+#!/bin/bash
+
+source "$(dirname "$0")/build_common.sh"
+
+CMAKE_OPTIONS="
+    -DCCCL_ENABLE_THRUST=OFF \
+    -DCCCL_ENABLE_LIBCUDACXX=ON \
+    -DCCCL_ENABLE_CUB=OFF \
+    -DCCCL_ENABLE_TESTING=OFF \
+    -DLIBCUDACXX_ENABLE_LIBCUDACXX_TESTS=ON \
+    -DLIBCUDACXX_TEST_WITH_NVRTC=ON \
+"
+configure "$CMAKE_OPTIONS"
+
+readonly TEST_PARALLEL_LEVEL=8
+
+source "./sccache_stats.sh" "start"
+LIBCUDACXX_SITE_CONFIG="${BUILD_DIR}/libcudacxx/test/lit.site.cfg" lit -v -j ${TEST_PARALLEL_LEVEL} --no-progress-bar -Dcompute_archs=${GPU_ARCHS} -Dstd="c++${CXX_STANDARD}" ../libcudacxx/.upstream-tests/test
+source "./sccache_stats.sh" "end"
diff --git a/libcudacxx/.upstream-tests/utils/libcudacxx/test/config.py b/libcudacxx/.upstream-tests/utils/libcudacxx/test/config.py
@@ -423,7 +423,7 @@ def configure_no_execute(self):
     def configure_ccache(self):
         use_ccache_default = os.environ.get('CMAKE_CUDA_COMPILER_LAUNCHER') is not None
         use_ccache = self.get_lit_bool('use_ccache', use_ccache_default)
-        if use_ccache:
+        if use_ccache and not self.cxx.is_nvrtc:
             self.cxx.use_ccache = True
             self.lit_config.note('enabling ccache')