Merge branch 'main' into ptx-add-barrier-cluster

Author: miscco

.github/workflows/dispatch-build-and-test.yml

@@ -28,9 +28,9 @@ jobs:
         include: ${{ fromJSON(inputs.per_cuda_compiler_matrix) }}
     with:
       cpu: ${{ matrix.cpu }}
-      test_name: ${{matrix.cpu}}/${{matrix.compiler.name}}${{matrix.compiler.version}}/C++${{matrix.std}}
+      test_name: ${{matrix.cpu}}/${{matrix.compiler.name}}${{matrix.compiler.version}}/C++${{matrix.std}} ${{matrix.extra_build_args}}
       build_script: './ci/build_${{ inputs.project_name }}.sh -cxx ${{matrix.compiler.exe}} -std ${{matrix.std}} "${{matrix.extra_build_args}}"'
-      test_script:  './ci/test_${{ inputs.project_name }}.sh -cxx ${{matrix.compiler.exe}} -std ${{matrix.std}}'
+      test_script:  './ci/test_${{ inputs.project_name }}.sh -cxx ${{matrix.compiler.exe}} -std ${{matrix.std}} "${{matrix.extra_build_args}}"'
       container_image: rapidsai/devcontainers:${{inputs.devcontainer_version}}-cpp-${{matrix.compiler.name}}${{matrix.compiler.version}}-cuda${{matrix.cuda}}-${{matrix.os}}
       run_tests: ${{ contains(matrix.jobs, 'test') && !contains(github.event.head_commit.message, 'skip-tests') && matrix.os != 'windows-2022' }}
 

ci/matrix.yaml

@@ -67,6 +67,7 @@ pull_request:
     - {cuda: *cuda_newest, os: 'ubuntu20.04', cpu: 'amd64', compiler: *gcc9,     std: [11, 14, 17],     jobs: ['build']}
     - {cuda: *cuda_newest, os: 'ubuntu20.04', cpu: 'amd64', compiler: *gcc10,    std: [11, 14, 17, 20], jobs: ['build']}
     - {cuda: *cuda_newest, os: 'ubuntu22.04', cpu: 'amd64', compiler: *gcc11,    std: [11, 14, 17, 20], jobs: ['build']}
+    - {cuda: *cuda_newest, os: 'ubuntu22.04', cpu: 'amd64', compiler: *gcc12,    std: [11, 14, 17, 20], jobs: ['build'], extra_build_args: '-cmake-options -DCMAKE_CUDA_ARCHITECTURES=90a'}
     - {cuda: *cuda_newest, os: 'ubuntu22.04', cpu: 'amd64', compiler: *gcc12,    std: [11, 14, 17, 20], jobs: ['build', 'test']}
     - {cuda: *cuda_newest, os: 'ubuntu22.04', cpu: 'arm64', compiler: *gcc12,    std: [11, 14, 17, 20], jobs: ['build']}
     - {cuda: *cuda_newest, os: 'ubuntu20.04', cpu: 'amd64', compiler: *llvm9,    std: [11, 14, 17],     jobs: ['build']}

libcudacxx/include/cuda/std/detail/libcxx/include/__cccl/ptx_isa.h

@@ -97,10 +97,12 @@
 // depending on PTX ISA. This permits checking for the feature in host code.
 // When __CUDA_MINIMUM_ARCH__ is available, we only enable the feature when the
 // hardware supports it.
+#if __cccl_ptx_isa >= 800
 #if (!defined(__CUDA_MINIMUM_ARCH__)) \
-  || (defined(__CUDA_MINIMUM_ARCH__) && 900 <= __CUDA_MINIMUM_ARCH__) && __cccl_ptx_isa >= 800
+  || (defined(__CUDA_MINIMUM_ARCH__) && 900 <= __CUDA_MINIMUM_ARCH__)
 #  define __cccl_lib_local_barrier_arrive_tx
 #  define __cccl_lib_experimental_ctk12_cp_async_exposure
 #endif
+#endif // __cccl_ptx_isa >= 800
 
 #endif // __CCCL_PTX_ISA_H_

libcudacxx/include/cuda/std/detail/libcxx/include/__concepts/arithmetic.h

@@ -26,6 +26,7 @@
 #include "../__type_traits/is_arithmetic.h"
 #include "../__type_traits/is_floating_point.h"
 #include "../__type_traits/is_integral.h"
+#include "../__type_traits/is_signed.h"
 #include "../__type_traits/is_signed_integer.h"
 #include "../__type_traits/is_signed.h"
 #include "../__type_traits/is_unsigned_integer.h"
@@ -48,9 +49,6 @@ _LIBCUDACXX_CONCEPT unsigned_integral = integral<_Tp> && !signed_integral<_Tp>;
 template<class _Tp>
 _LIBCUDACXX_CONCEPT floating_point = _LIBCUDACXX_TRAIT(is_floating_point, _Tp);
 
-// Concept helpers for the internal type traits for the fundamental types.
-template <class _Tp>
-_LIBCUDACXX_CONCEPT __libcpp_unsigned_integer = __libcpp_is_unsigned_integer<_Tp>::value;
 template <class _Tp>
 _LIBCUDACXX_CONCEPT __libcpp_signed_integer = __libcpp_is_signed_integer<_Tp>::value;
 

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h

@@ -774,7 +774,6 @@ _CUDA_VSTD::uint64_t * __try_get_barrier_handle<::cuda::thread_scope_block, _CUD
 // The user is still responsible for arriving and waiting on (or otherwise
 // synchronizing with) the barrier or pipeline barrier to see the results of
 // copies from other threads participating in the synchronization object.
-extern "C" _LIBCUDACXX_HOST_DEVICE void __cuda_ptx_mbarrier_complete_tx_is_not_supported_before_SM_90__();
 struct __memcpy_completion_impl {
 
     template<typename _Group>
@@ -815,8 +814,6 @@ struct __memcpy_completion_impl {
                     if (__group.thread_rank() == 0) {
                         ::cuda::device::barrier_expect_tx(__barrier, __size);
                     }
-                ),(
-                    __cuda_ptx_mbarrier_complete_tx_is_not_supported_before_SM_90__();
                 ));
 #endif // __cccl_ptx_isa >= 800
                 return async_contract_fulfillment::async;