Try #597:

.circleci/config.yml

@@ -26,7 +26,7 @@ setup_arm64_machine: &setup_arm64_machine
 
 docker_default: &docker_default
     docker:
-      - image: pikaorg/pika-ci-base:14
+      - image: pikaorg/pika-ci-base:15
 
 defaults: &defaults
     <<: *working_dir_default

.github/workflows/linux_coverage.yml

@@ -15,7 +15,7 @@ jobs:
   build:
     name: github/linux/coverage
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-base:14
+    container: pikaorg/pika-ci-base:15
 
     steps:
     - uses: actions/checkout@v3

.github/workflows/linux_debug.yml

@@ -20,7 +20,7 @@ jobs:
   build:
     name: github/linux/debug/fast
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-base:14
+    container: pikaorg/pika-ci-base:15
 
     steps:
     - uses: actions/checkout@v3

.github/workflows/linux_hip.yml

@@ -20,7 +20,7 @@ jobs:
   build:
     name: github/linux/hip/fast
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-hip:7
+    container: pikaorg/pika-ci-hip:8
     steps:
     - uses: actions/checkout@v3
     - name: Update apt repositories for ccache

.github/workflows/linux_release_fetchcontent.yml

@@ -20,7 +20,7 @@ jobs:
   build:
     name: github/linux/fetchcontent/fast
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-base:14
+    container: pikaorg/pika-ci-base:15
 
     steps:
     - uses: actions/checkout@v3

.github/workflows/linux_sanitizers.yml

@@ -20,7 +20,7 @@ jobs:
   build:
     name: github/linux/sanitizers/fast
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-base:14
+    container: pikaorg/pika-ci-base:15
 
     steps:
     - uses: actions/checkout@v3

.github/workflows/linux_tracy.yml

@@ -20,7 +20,7 @@ jobs:
   build:
     name: github/linux/tracy/fast
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-base:14
+    container: pikaorg/pika-ci-base:15
 
     steps:
     - name: Check out Tracy

.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="c0fe16f451137b00ceb85d912ec3a288990ce766"
+stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
 
 # The xnack- architectures are not supported by rocblas@5.3.3
 spack_spec="pika@main+rocm+p2300 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}"

.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="c0fe16f451137b00ceb85d912ec3a288990ce766"
+stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
 
 spack_spec="pika@main arch=${spack_arch} %${spack_compiler} cxxflags=-stdlib=libc++ malloc=system cxxstd=${cxx_std} +p2300 ^boost@${boost_version} ^hwloc@${hwloc_version} ^fmt cxxflags=-stdlib=libc++ ^stdexec@${stdexec_version}"
 

.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="c0fe16f451137b00ceb85d912ec3a288990ce766"
+stdexec_version="6510f5bd69cc03b24668f26eda3dd3cca7e81bb2"
 
 spack_spec="pika@main arch=${spack_arch} %${spack_compiler} +cuda malloc=system cxxstd=${cxx_std} +p2300 ^boost@${boost_version} ^cuda@${cuda_version} +allow-unsupported-compilers ^hwloc@${hwloc_version} ^stdexec@${stdexec_version}"
 

README.rst

@@ -49,8 +49,8 @@ pika optionally requires:
 * `Boost.Context, Thread, and Chrono <https://boost.org>`_ on macOS
 * `stdexec <https://github.com/NVIDIA/stdexec>`_ when
   ``PIKA_WITH_P2300_REFERENCE_IMPLEMENTATION`` is enabled (currently tested with commit
-  `c0fe16f451137b00ceb85d912ec3a288990ce766
-  <https://github.com/NVIDIA/stdexec/commit/c0fe16f451137b00ceb85d912ec3a288990ce766>`).
+  `6510f5bd69cc03b24668f26eda3dd3cca7e81bb2
+  <https://github.com/NVIDIA/stdexec/commit/6510f5bd69cc03b24668f26eda3dd3cca7e81bb2>`).
   The integration is experimental.
 
 

libs/pika/async_cuda/include/pika/async_cuda/cuda_scheduler.hpp

@@ -136,12 +136,23 @@ namespace pika::cuda::experimental {
                 return {s.scheduler, PIKA_FORWARD(Receiver, receiver)};
             }
 
-            friend cuda_scheduler tag_invoke(
-                pika::execution::experimental::get_completion_scheduler_t<
-                    pika::execution::experimental::set_value_t>,
-                cuda_scheduler_sender const& s) noexcept
+            struct env
             {
-                return s.scheduler;
+                cuda_scheduler scheduler;
+
+                friend cuda_scheduler tag_invoke(
+                    pika::execution::experimental::get_completion_scheduler_t<
+                        pika::execution::experimental::set_value_t>,
+                    env const& e) noexcept
+                {
+                    return e.scheduler;
+                }
+            };
+
+            friend env tag_invoke(
+                pika::execution::experimental::get_env_t, cuda_scheduler_sender const& s)
+            {
+                return {s.scheduler};
             }
         };
     }    // namespace detail

libs/pika/async_cuda/include/pika/async_cuda/then_on_host.hpp

@@ -37,6 +37,8 @@ namespace pika::cuda::experimental {
         template <typename Receiver, typename F>
         struct then_on_host_receiver_impl<Receiver, F>::then_on_host_receiver_type
         {
+            using is_receiver = void;
+
             PIKA_NO_UNIQUE_ADDRESS std::decay_t<Receiver> receiver;
             PIKA_NO_UNIQUE_ADDRESS std::decay_t<F> f;
             cuda_scheduler sched;
@@ -104,7 +106,7 @@ namespace pika::cuda::experimental {
                     });
             }
 
-            friend constexpr pika::execution::experimental::detail::empty_env tag_invoke(
+            friend constexpr pika::execution::experimental::empty_env tag_invoke(
                 pika::execution::experimental::get_env_t,
                 then_on_host_receiver_type const&) noexcept
             {
@@ -125,6 +127,8 @@ namespace pika::cuda::experimental {
         template <typename Sender, typename F>
         struct then_on_host_sender_impl<Sender, F>::then_on_host_sender_type
         {
+            using is_sender = void;
+
             std::decay_t<Sender> sender;
             std::decay_t<F> f;
             cuda_scheduler sched;
@@ -151,7 +155,7 @@ namespace pika::cuda::experimental {
 
             using completion_signatures =
                 pika::execution::experimental::make_completion_signatures<Sender,
-                    pika::execution::experimental::detail::empty_env,
+                    pika::execution::experimental::empty_env,
                     pika::execution::experimental::completion_signatures<
                         pika::execution::experimental::set_error_t(std::exception_ptr)>,
                     invoke_result_helper>;
@@ -200,12 +204,10 @@ namespace pika::cuda::experimental {
                         PIKA_FORWARD(Receiver, receiver), s.f, s.sched});
             }
 
-            friend cuda_scheduler tag_invoke(
-                pika::execution::experimental::get_completion_scheduler_t<
-                    pika::execution::experimental::set_value_t>,
-                then_on_host_sender_type const& s) noexcept
+            friend auto tag_invoke(
+                pika::execution::experimental::get_env_t, then_on_host_sender_type const& s)
             {
-                return s.sched;
+                return pika::execution::experimental::get_env(s.sender);
             }
         };
     }    // namespace then_on_host_detail
@@ -227,7 +229,8 @@ namespace pika::cuda::experimental {
         friend PIKA_FORCEINLINE auto tag_fallback_invoke(then_on_host_t, Sender&& sender, F&& f)
         {
             auto completion_sched = pika::execution::experimental::get_completion_scheduler<
-                pika::execution::experimental::set_value_t>(sender);
+                pika::execution::experimental::set_value_t>(
+                pika::execution::experimental::get_env(sender));
             static_assert(std::is_same_v<std::decay_t<decltype(completion_sched)>, cuda_scheduler>,
                 "then_on_host can only be used with senders whose "
                 "completion scheduler is cuda_scheduler");

libs/pika/async_cuda/include/pika/async_cuda/then_with_stream.hpp

@@ -136,6 +136,8 @@ namespace pika::cuda::experimental::then_with_stream_detail {
     template <typename Sender, typename F>
     struct then_with_cuda_stream_sender_impl<Sender, F>::then_with_cuda_stream_sender_type
     {
+        using is_sender = void;
+
         PIKA_NO_UNIQUE_ADDRESS std::decay_t<Sender> sender;
         PIKA_NO_UNIQUE_ADDRESS std::decay_t<F> f;
         cuda_scheduler sched;
@@ -165,7 +167,7 @@ namespace pika::cuda::experimental::then_with_stream_detail {
 
         using completion_signatures =
             pika::execution::experimental::make_completion_signatures<std::decay_t<Sender>,
-                pika::execution::experimental::detail::empty_env,
+                pika::execution::experimental::empty_env,
                 pika::execution::experimental::completion_signatures<
                     pika::execution::experimental::set_error_t(std::exception_ptr)>,
                 invoke_result_helper>;
@@ -223,6 +225,7 @@ namespace pika::cuda::experimental::then_with_stream_detail {
 
             struct then_with_cuda_stream_receiver
             {
+                using is_receiver = void;
                 using then_with_cuda_stream_receiver_tag = void;
 
                 operation_state& op_state;
@@ -411,7 +414,7 @@ namespace pika::cuda::experimental::then_with_stream_detail {
                         });
                 }
 
-                friend constexpr pika::execution::experimental::detail::empty_env tag_invoke(
+                friend constexpr pika::execution::experimental::empty_env tag_invoke(
                     pika::execution::experimental::get_env_t,
                     then_with_cuda_stream_receiver const&) noexcept
                 {
@@ -446,7 +449,7 @@ namespace pika::cuda::experimental::then_with_stream_detail {
             using ts_type = pika::util::detail::prepend_t<
                 pika::util::detail::transform_t<
                     pika::execution::experimental::value_types_of_t<std::decay_t<Sender>,
-                        pika::execution::experimental::detail::empty_env, std::tuple,
+                        pika::execution::experimental::empty_env, std::tuple,
                         pika::detail::variant>,
                     value_types_helper>,
                 pika::detail::monostate>;
@@ -493,8 +496,8 @@ namespace pika::cuda::experimental::then_with_stream_detail {
                     pika::util::detail::transform_t<
                         pika::execution::experimental::value_types_of_t<
                             then_with_cuda_stream_sender_type,
-                            pika::execution::experimental::detail::empty_env,
-                            pika::util::detail::pack, pika::util::detail::pack>,
+                            pika::execution::experimental::empty_env, pika::util::detail::pack,
+                            pika::util::detail::pack>,
                         result_types_helper>,
                     pika::detail::monostate>>>;
 #else
@@ -541,11 +544,10 @@ namespace pika::cuda::experimental::then_with_stream_detail {
                 PIKA_FORWARD(Receiver, receiver), s.f, s.sched, s.sender);
         }
 
-        friend cuda_scheduler tag_invoke(pika::execution::experimental::get_completion_scheduler_t<
-                                             pika::execution::experimental::set_value_t>,
-            then_with_cuda_stream_sender_type const& s) noexcept
+        friend auto tag_invoke(
+            pika::execution::experimental::get_env_t, then_with_cuda_stream_sender_type const& s)
         {
-            return s.sched;
+            return pika::execution::experimental::get_env(s.sender);
         }
     };
 
@@ -555,7 +557,8 @@ namespace pika::cuda::experimental::then_with_stream_detail {
     auto then_with_cuda_stream(Sender&& sender, F&& f)
     {
         auto completion_sched = pika::execution::experimental::get_completion_scheduler<
-            pika::execution::experimental::set_value_t>(sender);
+            pika::execution::experimental::set_value_t>(
+            pika::execution::experimental::get_env(sender));
         static_assert(std::is_same_v<std::decay_t<decltype(completion_sched)>, cuda_scheduler>,
             "then_with_cuda_stream can only be used with senders whose "
             "completion scheduler is cuda_scheduler");

libs/pika/async_cuda/tests/unit/cuda_scheduler.cu

@@ -30,12 +30,12 @@ inline constexpr bool is_cuda_scheduler_v =
     std::is_same_v<std::decay_t<Scheduler>, cu::cuda_scheduler>;
 
 #define CHECK_CUDA_COMPLETION_SCHEDULER(...)                                                       \
- static_assert(                                                                                    \
-     is_cuda_scheduler_v<decltype(ex::get_completion_scheduler<ex::set_value_t>(__VA_ARGS__))>)
+ static_assert(is_cuda_scheduler_v<decltype(ex::get_completion_scheduler<ex::set_value_t>(         \
+         ex::get_env(__VA_ARGS__)))>)
 
 #define CHECK_NOT_CUDA_COMPLETION_SCHEDULER(...)                                                   \
- static_assert(                                                                                    \
-     !is_cuda_scheduler_v<decltype(ex::get_completion_scheduler<ex::set_value_t>(__VA_ARGS__))>)
+ static_assert(!is_cuda_scheduler_v<decltype(ex::get_completion_scheduler<ex::set_value_t>(        \
+                   ex::get_env(__VA_ARGS__)))>)
 
 int pika_main()
 {

libs/pika/async_cuda/tests/unit/then_with_stream.cu

@@ -63,11 +63,23 @@ struct const_reference_cuda_sender
         return operation_state<R>{std::move(s.x), std::forward<R>(r)};
     }
 
-    friend cu::cuda_scheduler tag_invoke(pika::execution::experimental::get_completion_scheduler_t<
-                                             pika::execution::experimental::set_value_t>,
-        const_reference_cuda_sender const& s) noexcept
+    struct env
     {
-        return s.sched;
+        cu::cuda_scheduler sched;
+
+        friend cu::cuda_scheduler tag_invoke(
+            pika::execution::experimental::get_completion_scheduler_t<
+                pika::execution::experimental::set_value_t>,
+            env const& e) noexcept
+        {
+            return e.sched;
+        }
+    };
+
+    friend env tag_invoke(
+        pika::execution::experimental::get_env_t, const_reference_cuda_sender const& s)
+    {
+        return {s.sched};
     }
 };
 
@@ -108,11 +120,23 @@ struct const_reference_error_cuda_sender
         return {std::forward<R>(r)};
     }
 
-    friend cu::cuda_scheduler tag_invoke(pika::execution::experimental::get_completion_scheduler_t<
-                                             pika::execution::experimental::set_value_t>,
-        const_reference_error_cuda_sender const& s) noexcept
+    struct env
+    {
+        cu::cuda_scheduler sched;
+
+        friend cu::cuda_scheduler tag_invoke(
+            pika::execution::experimental::get_completion_scheduler_t<
+                pika::execution::experimental::set_value_t>,
+            env const& e) noexcept
+        {
+            return e.sched;
+        }
+    };
+
+    friend env tag_invoke(
+        pika::execution::experimental::get_env_t, const_reference_error_cuda_sender const& s)
     {
-        return s.sched;
+        return {s.sched};
     }
 };