stdexec: update

CMakeLists.txt

@@ -110,6 +110,7 @@ target_sources(
             kokkos-execution/impl/event.hpp
             kokkos-execution/impl/HIP/event.hpp
             kokkos-execution/impl/HPX/event.hpp
+            kokkos-execution/impl/immovable.hpp
             kokkos-execution/impl/sender_concepts.hpp
             kokkos-execution/impl/SYCL/event.hpp
             kokkos-execution/impl/state.hpp

docs/CMakeLists.txt

@@ -12,6 +12,7 @@ set(DOXYGEN_ENABLE_PREPROCESSING YES)
 set(DOXYGEN_EXTRACT_ALL YES)
 set(DOXYGEN_EXTRACT_STATIC YES)
 set(DOXYGEN_FILE_PATTERNS *.c *.cpp *.h *.hpp)
+set(DOXYGEN_GENERATE_BUGLIST YES)
 set(DOXYGEN_IMAGE_PATH ${CMAKE_SOURCE_DIR})
 set(DOXYGEN_MACRO_EXPANSION YES)
 set(DOXYGEN_PREDEFINED __DOXYGEN__)

kokkos-execution/execution_space.hpp

@@ -35,7 +35,7 @@ struct Scheduler {
 
     template <stdexec::receiver Rcvr>
     struct OpState {
-        using operation_state_concept = stdexec::operation_state_t;
+        using operation_state_concept = stdexec::operation_state_tag;
 
         Rcvr rcvr;
 
@@ -46,7 +46,7 @@ struct Scheduler {
     };
 
     struct Sender {
-        using sender_concept = stdexec::sender_t;
+        using sender_concept = stdexec::sender_tag;
 
         using completion_signatures = stdexec::completion_signatures<stdexec::set_value_t()>;
 

kokkos-execution/execution_space/continues_on.hpp

@@ -22,7 +22,7 @@ struct FwdWithoutExec { };
 //! Receiver for @c continues_on.
 template <stdexec::receiver Rcvr, typename FwdPolicy = FwdWithExec>
 struct ContinuesOnReceiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
 
     Rcvr rcvr;
 
@@ -60,7 +60,7 @@ struct ContinuesOnReceiver {
 //! Sender for @c continues_on.
 template <stdexec::sender Sndr>
 struct ContinuesOnSender {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     Sndr sndr; // NOLINT(cppcoreguidelines-avoid-const-or-ref-data-members)
 

kokkos-execution/execution_space/operation_state.hpp

@@ -12,6 +12,7 @@
 #include "kokkos-execution/impl/dispatch_label.hpp"
 #include "kokkos-execution/impl/env.hpp"
 #include "kokkos-execution/impl/event.hpp"
+#include "kokkos-execution/impl/immovable.hpp"
 #include "kokkos-execution/impl/sender_concepts.hpp"
 #include "kokkos-execution/impl/sync_wait.hpp"
 
@@ -100,7 +101,7 @@ struct MayDelegateCompletionWithEvent;
 
 template <stdexec::receiver Rcvr, Kokkos::ExecutionSpace Exec>
 struct WaitEventReceiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
     using event_t = Impl::Event<Exec>;
 
     MayDelegateCompletionWithEvent<Rcvr, Exec>* opstate;
@@ -162,9 +163,7 @@ struct MayDelegateCompletionWithEvent<Rcvr, Exec, true> {
 
 template <stdexec::receiver Rcvr, Closure Clsr, Closure... Clsrs>
 requires(std::same_as<typename Clsr::execution_space, typename Clsrs::execution_space> && ...)
-struct OpStateBase
-    : public stdexec::__immovable
-    , public MayDelegateCompletionWithEvent<Rcvr, typename Clsr::execution_space> {
+struct OpStateBase : public MayDelegateCompletionWithEvent<Rcvr, typename Clsr::execution_space> {
     using execution_space = typename Clsr::execution_space;
     using receiver_t = Rcvr;
     using closures_t = stdexec::__tuple<Clsr, Clsrs...>;
@@ -211,7 +210,7 @@ struct OpStateBase
 
 template <typename ParentOp>
 struct OpStateReceiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
 
     ParentOp* parent_op;
 
@@ -237,8 +236,10 @@ struct OpStateReceiver {
 
 template <stdexec::sender Sndr, stdexec::receiver Rcvr, Closure... Clsrs>
 requires(!Impl::dispatching_sender<Sndr>)
-struct OpState : public OpStateBase<Rcvr, Clsrs...> {
-    using operation_state_concept = stdexec::operation_state_t;
+struct OpState
+    : public Impl::Immovable
+    , public OpStateBase<Rcvr, Clsrs...> {
+    using operation_state_concept = stdexec::operation_state_tag;
 
     using base_t = OpStateBase<Rcvr, Clsrs...>;
 

kokkos-execution/execution_space/parallel_for.hpp

@@ -35,7 +35,7 @@ struct ParallelForClosure {
 
 template <stdexec::sender Sndr, typename Label, typename Functor, Kokkos::ExecutionPolicy ExecPolicy>
 struct ParallelForSender {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     using closure_t = ParallelForClosure<Label, Functor, ExecPolicy>;
     using execution_space = typename closure_t::execution_space;

kokkos-execution/execution_space/schedule_from.hpp

@@ -15,7 +15,7 @@ namespace Kokkos::Execution::ExecutionSpaceImpl {
 //! Receiver for @c schedule_from.
 template <stdexec::scheduler Schd, stdexec::receiver Rcvr>
 struct ScheduleFromReceiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
 
     Schd schd;
     Rcvr rcvr;
@@ -58,7 +58,7 @@ struct ScheduleFromReceiver {
 //! Sender for @c schedule_from.
 template <stdexec::scheduler Schd, stdexec::sender Sndr>
 struct ScheduleFromSender {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     KOKKOS_EXECUTION_COMPL_SIGS_KEEP(ScheduleFromSender)
 

kokkos-execution/execution_space/scoped_region.hpp

@@ -32,7 +32,7 @@ enum class Kind : std::uint8_t {
 
 template <Kind kind, stdexec::__is_instance_of<Scheduler> Schd, stdexec::receiver Rcvr>
 struct RegionReceiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
 
     std::string name;
     Schd schd;
@@ -69,7 +69,7 @@ struct RegionReceiver {
 
 template <Kind kind, stdexec::sender Sndr>
 struct RegionSender {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     KOKKOS_EXECUTION_COMPL_SIGS_KEEP(RegionSender)
 

kokkos-execution/graph.hpp

@@ -22,7 +22,7 @@ struct Scheduler {
 
     template <stdexec::receiver Rcvr>
     struct OpState {
-        using operation_state_concept = stdexec::operation_state_t;
+        using operation_state_concept = stdexec::operation_state_tag;
 
         Rcvr rcvr;
 
@@ -33,7 +33,7 @@ struct Scheduler {
     };
 
     struct Sender {
-        using sender_concept = stdexec::sender_t;
+        using sender_concept = stdexec::sender_tag;
 
         using completion_signatures = stdexec::completion_signatures<stdexec::set_value_t()>;
 

kokkos-execution/impl/immovable.hpp

@@ -0,0 +1,30 @@
+#ifndef KOKKOS_EXECUTION_IMPL_IMMOVABLE_HPP
+#define KOKKOS_EXECUTION_IMPL_IMMOVABLE_HPP
+
+namespace Kokkos::Execution::Impl {
+/**
+ * @brief Immovable type.
+ *
+ * Heavily inspired by
+ * https://github.com/NVIDIA/stdexec/blob/47bb920c84be5bdb31e6a1d0f8c47ac6e7d54588/include/stdexec/__detail/__utility.hpp#L60-L64.
+ *
+ * However, commit
+ * https://github.com/NVIDIA/stdexec/commit/aab5da8b7f7ed60053d16798a687d0b1549e2e1d
+ * changed how @c STDEXEC_IMMOVABLE is defined for @c GCC, see
+ * https://github.com/NVIDIA/stdexec/blame/47bb920c84be5bdb31e6a1d0f8c47ac6e7d54588/include/stdexec/__detail/__config.hpp#L548,
+ * that is used in the definition of @c stdexec::__immovable.
+ *
+ * This implementation avoids relying on such internals by providing a minimal, self-contained definition.
+ */
+struct Immovable {
+    Immovable() = default;
+    Immovable(Immovable&&) = delete;
+    Immovable& operator=(Immovable&&) = delete;
+    Immovable(const Immovable&) = delete;
+    Immovable& operator=(const Immovable&) = delete;
+    ~Immovable() = default;
+};
+
+} // namespace Kokkos::Execution::Impl
+
+#endif // KOKKOS_EXECUTION_IMPL_IMMOVABLE_HPP

kokkos-execution/impl/sync_wait.hpp

@@ -33,7 +33,7 @@ struct State {
 //! Receiver for @c stdexec::sync_wait.
 template <Kokkos::ExecutionSpace Exec, typename... Values>
 struct Receiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
 
     static constexpr auto label = Impl::dispatch_label<Exec, ": sync_wait">();
 

kokkos-execution/parallel_for.hpp

@@ -63,7 +63,7 @@ struct ParallelForData {
 
 template <stdexec::sender Sndr, typename Label, typename Functor, Kokkos::ExecutionPolicy ExecPolicy>
 struct ParallelForSender : stdexec::__tuple<parallel_for_t, ParallelForData<Label, Functor, ExecPolicy>, Sndr> {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     using base_t = stdexec::__tuple<parallel_for_t, ParallelForData<Label, Functor, ExecPolicy>, Sndr>;
 

tests/execution_space/CMakeLists.txt

@@ -16,3 +16,19 @@ add_one_test(NAME sync_wait)
 add_one_test(NAME then)
 add_one_test(NAME transfer_when_all)
 add_one_test(NAME when_all)
+
+if(CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND CMAKE_CXX_COMPILER_VERSION STREQUAL 14.2.0)
+  foreach(DEVICE IN LISTS Kokkos_DEVICES)
+    set(target_name tests_execution_space_any_sender.${DEVICE})
+    set_target_properties(${target_name} PROPERTIES EXCLUDE_FROM_ALL TRUE EXCLUDE_FROM_DEFAULT_BUILD TRUE)
+    set_tests_properties(${target_name} PROPERTIES DISABLED TRUE)
+    add_test(NAME ${target_name}.compiler COMMAND ${CMAKE_COMMAND} --build . --config $<CONFIGURATION> --target
+                                                  ${target_name} --verbose WORKING_DIRECTORY ${CMAKE_BINARY_DIR}
+    )
+    set_tests_properties(
+      ${target_name}.compiler
+      PROPERTIES PASS_REGULAR_EXPRESSION
+                 "error: use of deleted function \'Kokkos::Execution::Impl::Immovable::Immovable"
+    )
+  endforeach()
+endif()

tests/execution_space/test_any_sender.cpp

@@ -42,37 +42,33 @@ class AnySenderTest
         RecorderListener<EventDiscardMatcher<TEST_EXECUTION_SPACE>, BeginFenceEvent, BeginParallelForEvent>;
 };
 
-/**
- * @test In order to get the same synchronization behavior as if using fully typed senders,
- *       type erased senders must advertise the completion domain and scheduler and type erased receivers
- *       must advertise their @ref Kokkos::Execution::ExecutionSpaceImpl::get_exec_t query.
- */
+//! @test Check when synchronization happens.
 TEST_F(AnySenderTest, then) {
-    using any_sender_t = experimental::execution::any_receiver_ref<
-        stdexec::completion_signatures<stdexec::set_value_t(), stdexec::set_error_t(std::exception_ptr)>,
-        Kokkos::Execution::ExecutionSpaceImpl::get_exec
-            .signature<Kokkos::Execution::ExecutionSpaceImpl::ExecutionSpaceRef<TEST_EXECUTION_SPACE>() noexcept>
-    >::
-        template any_sender<
-            stdexec::get_completion_scheduler<stdexec::set_value_t>.signature<scheduler_t() noexcept>,
-            stdexec::get_completion_domain<stdexec::set_value_t>.signature<Kokkos::Execution::ExecutionSpaceImpl::Domain() noexcept>
-        >;
+    using completion_signatures_t =
+        stdexec::completion_signatures<stdexec::set_value_t(), stdexec::set_error_t(std::exception_ptr)>;
+    using any_receiver_t = exec::any_receiver<completion_signatures_t>;
+    using any_sender_t = exec::any_sender<any_receiver_t>;
+
+    static_assert(std::same_as<
+                  stdexec::__completion_domain_of_t<stdexec::set_value_t, any_sender_t>,
+                  stdexec::indeterminate_domain<>
+    >);
+    static_assert(std::same_as<
+                  stdexec::__completion_domain_of_t<stdexec::set_value_t, any_sender_t, stdexec::env<>>,
+                  stdexec::default_domain
+    >);
+
+    static_assert(std::same_as<
+                  std::invoke_result_t<stdexec::get_completion_signatures_t, any_sender_t, stdexec::env<>>,
+                  completion_signatures_t
+    >);
 
     const view_s_t data(Kokkos::view_alloc(exec, "data - shared space"));
 
     const context_t esc{exec};
 
     any_sender_t chain = stdexec::schedule(esc.get_scheduler()) | THEN_INCREMENT(data) | THEN_INCREMENT(data);
 
-    static_assert(std::same_as<
-                  stdexec::__completion_domain_of_t<stdexec::set_value_t, decltype(chain)>,
-                  Kokkos::Execution::ExecutionSpaceImpl::Domain
-    >);
-    static_assert(std::same_as<
-                  stdexec::__completion_scheduler_of_t<stdexec::set_value_t, decltype(chain)>,
-                  typename AnySenderTest::scheduler_t
-    >);
-
     auto continues_on = std::move(chain) | stdexec::continues_on(esc.get_scheduler());
 
     static_assert(std::same_as<
@@ -94,8 +90,9 @@ TEST_F(AnySenderTest, then) {
         testing::ElementsAre(
             MATCHER_FOR_BEGIN_PFOR(exec, dispatch_label(exec, "then")),
             MATCHER_FOR_BEGIN_PFOR(exec, dispatch_label(exec, "then")),
+            MATCHER_FOR_BEGIN_FENCE(exec, dispatch_label(exec, "after dispatch")),
             MATCHER_FOR_BEGIN_PFOR(exec, dispatch_label(exec, "then")),
-            MATCHER_FOR_BEGIN_FENCE(exec, dispatch_label(exec, "sync_wait"))));
+            MATCHER_FOR_BEGIN_FENCE(exec, dispatch_label(exec, "after dispatch"))));
 
     ASSERT_EQ(data(), 3);
 }

tests/execution_space/test_let_value.cpp

@@ -57,7 +57,7 @@ TEST_F(LetValueTest, scoped_allocation) {
     static_assert(
         Tests::Utils::has_completion_signatures<decltype(allocate), stdexec::__mset<stdexec::set_value_t(view_of_5_t)>>);
 
-//! FIXME: https://github.com/kokkos/kokkos/blob/393d4165a6c3687e78abe5e1665853f1eabc386d/core/src/Kokkos_View.hpp#L697
+//! @bug https://github.com/kokkos/kokkos/blob/393d4165a6c3687e78abe5e1665853f1eabc386d/core/src/Kokkos_View.hpp#L697
 #if defined(KOKKOS_COMPILER_CLANG) && defined(KOKKOS_ENABLE_CUDA)
     //! @c Kokkos::View is not nothrow movable, so the error channel is added.
     static_assert(Tests::Utils::has_completion_signatures<

tests/execution_space/test_operation_state.cpp

@@ -38,7 +38,8 @@ consteval bool test_op_state_traits() {
     //! Models the operation state concept.
     static_assert(stdexec::operation_state<op_state_t>);
 
-    //! By inheriting from @c stdexec::__immovable, it is neither moveable nor copyable.
+    //! By inheriting from @ref Kokkos::Execution::Impl::Immovable, it is neither moveable nor copyable.
+    static_assert(std::derived_from<op_state_t, Kokkos::Execution::Impl::Immovable>);
     static_assert(!std::move_constructible<op_state_t>);
     static_assert(!std::is_move_assignable_v<op_state_t>);
 

tests/execution_space/test_parallel_for.cpp

@@ -86,7 +86,7 @@ consteval bool test_sndr_traits() {
                   Kokkos::Execution::ExecutionSpaceImpl::ParallelForSender<schd_sndr_t, label_t, functor_t, policy_t>
     >);
 
-//! FIXME: https://github.com/kokkos/kokkos/blob/393d4165a6c3687e78abe5e1665853f1eabc386d/core/src/Kokkos_View.hpp#L697
+//! @bug https://github.com/kokkos/kokkos/blob/393d4165a6c3687e78abe5e1665853f1eabc386d/core/src/Kokkos_View.hpp#L697
 #if defined(KOKKOS_COMPILER_CLANG) && defined(KOKKOS_ENABLE_CUDA)
     /**
      * It is not no throw connectable because the @ref Kokkos::Execution::ExecutionSpaceImpl::ParallelForClosure
@@ -157,7 +157,7 @@ consteval bool test_closure_traits() {
 
     return true;
 }
-//! FIXME: https://github.com/kokkos/kokkos/blob/393d4165a6c3687e78abe5e1665853f1eabc386d/core/src/Kokkos_View.hpp#L697
+//! @bug https://github.com/kokkos/kokkos/blob/393d4165a6c3687e78abe5e1665853f1eabc386d/core/src/Kokkos_View.hpp#L697
 #if defined(KOKKOS_COMPILER_CLANG) && defined(KOKKOS_ENABLE_CUDA)
 static_assert(test_closure_traits<typename ParallelForTest::view_s_t, false>());
 #else

tests/utils/check_rcvr_env_queryable_with.hpp

@@ -36,7 +36,7 @@ struct check_rcvr_env_queryable_with_t {
 
 template <bool IsQueryable, stdexec::sender Sndr, typename... Queries>
 struct CheckRcvrEnvQueryableWithSender {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     Sndr sndr; // NOLINT(cppcoreguidelines-avoid-const-or-ref-data-members)
 

tests/utils/check_scheduler_type.hpp

@@ -36,7 +36,7 @@ struct check_scheduler_type_t {
 
 template <stdexec::sender Sndr, typename Tag, stdexec::scheduler Schd>
 struct CheckSchedulerTypeSender {
-    using sender_concept = stdexec::sender_t;
+    using sender_concept = stdexec::sender_tag;
 
     Sndr sndr; // NOLINT(cppcoreguidelines-avoid-const-or-ref-data-members)
 

tests/utils/sink_receiver.hpp

@@ -7,7 +7,7 @@ namespace Tests::Utils {
 
 //! A receiver that can handle all completions and does nothing with them.
 struct SinkReceiver {
-    using receiver_concept = stdexec::receiver_t;
+    using receiver_concept = stdexec::receiver_tag;
 
     void set_value(auto&&...) && noexcept {
     }