Merge 2d260f0 into 4b6e4c9

pika-org · May 9, 2023 · 1b5ba38 · 1b5ba38
2 parents 4b6e4c9 + 2d260f0
commit 1b5ba38
Show file tree

Hide file tree

Showing 18 changed files with 77 additions and 134 deletions.
diff --git a/.jenkins/cscs-ault/env-hipcc.sh b/.jenkins/cscs-ault/env-hipcc.sh
@@ -12,7 +12,7 @@ hip_version="5.3.3"
 hwloc_version="2.6.0"
 spack_compiler="gcc@${gcc_version}"
 spack_arch="linux-centos8-zen"
-stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
+stdexec_version="7a47a4aa411c1ca9adfcb152c28cc3dd7b156b4d"
 
 # The xnack- architectures are not supported by rocblas@5.3.3
 spack_spec="pika@main+rocm+stdexec amdgpu_target='gfx906:xnack-' arch=${spack_arch} %${spack_compiler} malloc=system cxxstd=${cxx_std} ^boost@${boost_version} ^hwloc@${hwloc_version} ^hip@${hip_version} ^whip amdgpu_target='gfx906:xnack-' ^stdexec@${stdexec_version}"

diff --git a/.jenkins/cscs/env-clang-13.sh b/.jenkins/cscs/env-clang-13.sh
@@ -10,7 +10,7 @@ boost_version="1.79.0"
 hwloc_version="2.6.0"
 spack_compiler="clang@${clang_version}"
 spack_arch="cray-cnl7-broadwell"
-stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
+stdexec_version="7a47a4aa411c1ca9adfcb152c28cc3dd7b156b4d"
 
 spack_spec="pika@main arch=${spack_arch} %${spack_compiler} cxxflags=-stdlib=libc++ malloc=system cxxstd=${cxx_std} +stdexec ^boost@${boost_version} ^hwloc@${hwloc_version} ^fmt cxxflags=-stdlib=libc++ ^stdexec@${stdexec_version}"
 

diff --git a/.jenkins/cscs/env-clang-14-cuda.sh b/.jenkins/cscs/env-clang-14-cuda.sh
@@ -11,7 +11,7 @@ hwloc_version="2.7.0"
 cuda_version="11.5.0"
 spack_compiler="clang@${clang_version}"
 spack_arch="cray-cnl7-haswell"
-stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
+stdexec_version="7a47a4aa411c1ca9adfcb152c28cc3dd7b156b4d"
 
 spack_spec="pika@main arch=${spack_arch} %${spack_compiler} +cuda malloc=system cxxstd=${cxx_std} +stdexec ^boost@${boost_version} ^cuda@${cuda_version} +allow-unsupported-compilers ^hwloc@${hwloc_version} ^stdexec@${stdexec_version}"
 

diff --git a/.jenkins/cscs/env-nvhpc.sh b/.jenkins/cscs/env-nvhpc.sh
@@ -7,7 +7,7 @@
 cxx_std="20"
 boost_version="1.78.0"
 hwloc_version="2.7.0"
-stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
+stdexec_version="7a47a4aa411c1ca9adfcb152c28cc3dd7b156b4d"
 nvhpc_version="22.11"
 spack_compiler="nvhpc@${nvhpc_version}"
 spack_arch="cray-cnl7-haswell"

diff --git a/libs/pika/async_cuda/tests/performance/synchronize.cu b/libs/pika/async_cuda/tests/performance/synchronize.cu
@@ -122,7 +122,7 @@ int pika_main(pika::program_options::variables_map& vm)
 
 // The compilation of this loop unrolling with llvm-amdgpu's
 // Clang 5.3.3 is hanging.
-#if !(defined(PIKA_HAVE_STDEXEC) || defined(PIKA_HAVE_HIP))
+#if !defined(PIKA_HAVE_HIP)
     {
         cu::enable_user_polling poll("default");
 

diff --git a/libs/pika/async_cuda/tests/unit/cublas_matmul.cpp b/libs/pika/async_cuda/tests/unit/cublas_matmul.cpp
@@ -188,11 +188,6 @@ void matrixMultiply(pika::cuda::experimental::cuda_scheduler& cuda_sched, sMatri
 
     // See https://github.com/brycelelbach/wg21_p2300_std_execution/issues/466
     // for details.
-#if defined(PIKA_HAVE_STDEXEC)
-    std::cout << "skipping remainder of test because the stdexec implementation of split does not "
-                 "yet support move-only senders"
-              << std::endl;
-#else
     pika::chrono::detail::high_resolution_timer t2;
 
     // This loop is currently inefficient. Because of the type-erasure with
@@ -259,7 +254,6 @@ void matrixMultiply(pika::cuda::experimental::cuda_scheduler& cuda_sched, sMatri
         });
 
     tt::sync_wait(std::move(all_done));
-#endif
     whip::free(d_A);
     whip::free(d_B);
     whip::free(d_C);

diff --git a/libs/pika/async_cuda/tests/unit/cuda_bulk.cu b/libs/pika/async_cuda/tests/unit/cuda_bulk.cu
@@ -76,8 +76,6 @@ int pika_main()
         }
     }
 
-    // The reference implementation does not support ranges as the shape
-#if !defined(PIKA_HAVE_STDEXEC)
     // Range
     for (element_type n : {1, 42, 10007})
     {
@@ -105,7 +103,6 @@ int pika_main()
             PIKA_TEST_EQ(host_vector[i], i);
         }
     }
-#endif
 
     return pika::finalize();
 }

diff --git a/libs/pika/execution/include/pika/execution/algorithms/sync_wait.hpp b/libs/pika/execution/include/pika/execution/algorithms/sync_wait.hpp
@@ -59,8 +59,9 @@ namespace pika::sync_wait_detail {
 #if defined(PIKA_HAVE_STDEXEC)
         // value and error_types of the predecessor sender
         template <template <typename...> class Tuple, template <typename...> class Variant>
-        using predecessor_value_types = pika::execution::experimental::value_types_of_t<Sender,
-            pika::execution::experimental::empty_env, Tuple, Variant>;
+        using predecessor_value_types =
+            pika::util::detail::unique_t<pika::execution::experimental::value_types_of_t<Sender,
+                pika::execution::experimental::empty_env, Tuple, Variant>>;
 
         template <template <typename...> class Variant>
         using predecessor_error_types = pika::execution::experimental::error_types_of_t<Sender,
@@ -70,7 +71,7 @@ namespace pika::sync_wait_detail {
         // there are multiple variants or multiple values sync_wait will
         // fail to compile.
         using result_type = std::decay_t<pika::execution::experimental::detail::single_result_t<
-            predecessor_value_types<pika::util::detail::pack, pika::util::detail::pack>>>;
+            predecessor_value_types<pika::util::detail::decayed_pack, pika::util::detail::pack>>>;
 
         // Constant to indicate if the type of the result from the
         // predecessor sender is void or not

diff --git a/libs/pika/execution/tests/unit/algorithm_ensure_started.cpp b/libs/pika/execution/tests/unit/algorithm_ensure_started.cpp
@@ -76,9 +76,6 @@ int main()
         PIKA_TEST(set_value_called);
     }
 
-    // stdexec still requires copyable senders (see issue
-    // https://github.com/NVIDIA/stdexec/issues/641).
-#if !defined(PIKA_HAVE_STDEXEC)
     {
         std::atomic<bool> set_value_called{false};
         std::atomic<bool> started{false};
@@ -107,7 +104,6 @@ int main()
         ex::start(os);
         PIKA_TEST(set_value_called);
     }
-#endif
 
     // operator| overload
     {

diff --git a/libs/pika/execution/tests/unit/algorithm_just.cpp b/libs/pika/execution/tests/unit/algorithm_just.cpp
@@ -119,10 +119,7 @@ int main()
         PIKA_TEST(set_value_called);
     }
 
-    // TODO: The reference implementation does not ADL-isolate the just-sender.
-#if !defined(PIKA_HAVE_STDEXEC)
     test_adl_isolation(ex::just(my_namespace::my_type{}));
-#endif
 
     return 0;
 }
diff --git a/libs/pika/execution/tests/unit/algorithm_start_detached.cpp b/libs/pika/execution/tests/unit/algorithm_start_detached.cpp
@@ -77,8 +77,6 @@ int main()
         PIKA_TEST(!tag_invoke_overload_called);
     }
 
-    // Piping into start_detached is not allowed according to the specification.
-#if !defined(PIKA_HAVE_STDEXEC)
     // operator| overload
     {
         std::atomic<bool> start_called{false};
@@ -89,7 +87,6 @@ int main()
         PIKA_TEST(connect_called);
         PIKA_TEST(!tag_invoke_overload_called);
     }
-#endif
 
     // tag_invoke overload
     {

diff --git a/libs/pika/execution/tests/unit/scheduler_queries.cpp b/libs/pika/execution/tests/unit/scheduler_queries.cpp
@@ -70,7 +70,11 @@ struct customized_scheduler : uncustomized_scheduler
 inline constexpr struct custom_scheduler_query_t
 {
     friend constexpr bool tag_invoke(
+#if defined(PIKA_HAVE_STDEXEC)
+        ex::forwarding_query_t, custom_scheduler_query_t const&) noexcept
+#else
         ex::forwarding_scheduler_query_t, custom_scheduler_query_t const&) noexcept
+#endif
     {
         return true;
     }
@@ -92,11 +96,19 @@ int main()
 
     // get_forward_progress_guarantee is not a forwarding query
     {
+#if defined(PIKA_HAVE_STDEXEC)
+        static_assert(!ex::forwarding_query(ex::get_forward_progress_guarantee));
+#else
         static_assert(!ex::forwarding_scheduler_query(ex::get_forward_progress_guarantee));
+#endif
     }
 
     // The custom query is a forwarding query
     {
+#if defined(PIKA_HAVE_STDEXEC)
+        static_assert(ex::forwarding_query(custom_scheduler_query));
+#else
         static_assert(ex::forwarding_scheduler_query(custom_scheduler_query));
+#endif
     }
 }
diff --git a/libs/pika/execution_base/tests/unit/any_sender.cpp b/libs/pika/execution_base/tests/unit/any_sender.cpp
@@ -625,11 +625,8 @@ struct wait_globals
 
 void test_globals()
 {
-    // TODO: No ensure_started implementation in reference implementation.
-#if !defined(PIKA_HAVE_STDEXEC)
     global_unique_any_sender = std::move(global_unique_any_sender) | ex::ensure_started();
     global_any_sender = std::move(global_any_sender) | ex::ensure_started() | ex::split();
-#endif
 }
 
 void test_empty_any_sender()

diff --git a/libs/pika/execution_base/tests/unit/basic_operation_state.cpp b/libs/pika/execution_base/tests/unit/basic_operation_state.cpp
@@ -52,12 +52,22 @@ int main()
 {
     static_assert(!pika::execution::experimental::is_operation_state_v<mylib::state_1>,
         "mylib::state_1 is not an operation state");
+#if defined(PIKA_HAVE_STDEXEC)
+    static_assert(pika::execution::experimental::is_operation_state_v<mylib::state_2>,
+        "mylib::state_2 is an operation state");
+#else
     static_assert(!pika::execution::experimental::is_operation_state_v<mylib::state_2>,
         "mylib::state_2 is not an operation state");
+#endif
     static_assert(pika::execution::experimental::is_operation_state_v<mylib::state_3>,
         "mylib::state_3 is an operation state");
+#if defined(PIKA_HAVE_STDEXEC)
+    static_assert(pika::execution::experimental::is_operation_state_v<mylib::state_4>,
+        "mylib::state_4 is an operation state");
+#else
     static_assert(!pika::execution::experimental::is_operation_state_v<mylib::state_4>,
         "mylib::state_4 is not an operation state");
+#endif
     static_assert(pika::execution::experimental::is_operation_state_v<mylib::state_5>,
         "mylib::state_5 is an operation state");
 

diff --git a/libs/pika/executors/include/pika/executors/thread_pool_scheduler_bulk.hpp b/libs/pika/executors/include/pika/executors/thread_pool_scheduler_bulk.hpp
@@ -82,12 +82,20 @@ namespace pika::thread_pool_bulk_detail {
         thread_pool_bulk_sender& operator=(thread_pool_bulk_sender&&) = default;
         thread_pool_bulk_sender& operator=(thread_pool_bulk_sender const&) = default;
 
+        template <typename Tuple>
+        struct value_types_helper
+        {
+            using type = pika::util::detail::transform_t<Tuple, std::decay>;
+        };
+
 #if defined(PIKA_HAVE_STDEXEC)
         using is_sender = void;
 
         template <template <typename...> class Tuple, template <typename...> class Variant>
-        using value_types = pika::execution::experimental::value_types_of_t<Sender,
-            pika::execution::experimental::empty_env, Tuple, Variant>;
+        using value_types = pika::util::detail::transform_t<
+            pika::execution::experimental::value_types_of_t<Sender,
+                pika::execution::experimental::empty_env, Tuple, Variant>,
+            value_types_helper>;
 
         template <template <typename...> class Variant>
         using error_types = pika::util::detail::unique_t<
@@ -102,8 +110,10 @@ namespace pika::thread_pool_bulk_detail {
                     pika::execution::experimental::set_error_t(std::exception_ptr)>>;
 #else
         template <template <typename...> class Tuple, template <typename...> class Variant>
-        using value_types = typename pika::execution::experimental::sender_traits<
-            Sender>::template value_types<Tuple, Variant>;
+        using value_types =
+            pika::util::detail::transform_t<typename pika::execution::experimental::sender_traits<
+                                                Sender>::template value_types<Tuple, Variant>,
+                value_types_helper>;
 
         template <template <typename...> class Variant>
         using error_types = pika::util::detail::unique_t<
@@ -402,7 +412,8 @@ namespace pika::thread_pool_bulk_detail {
                     auto const num_chunks = (n + chunk_size - 1) / chunk_size;
 
                     // Store sent values in the operation state
-                    r.op_state->ts.template emplace<std::tuple<Ts...>>(PIKA_FORWARD(Ts, ts)...);
+                    r.op_state->ts.template emplace<std::tuple<std::decay_t<Ts>...>>(
+                        PIKA_FORWARD(Ts, ts)...);
 
                     // Initialize the queues for all worker threads so that
                     // worker threads can start stealing immediately when