Merge #330 #346 #347 #352

330: Update CUDA sender functionality to work with P2300 reference implementation r=msimberg a=msimberg

Continuation of #315. Part 2/n to fix #302.

This has now been tested with clang 14 and CUDA. ~but a CI configuration will follow in a separate PR (the llvm build requires this patch: spack/spack#31661

- Makes `get_env` customizations `noexcept` as required (and enforced) by the P2300 reference implementation.
- Add a missing `range` header to `cuda_scheduler_bulk.hpp`.
- ADL-isolate sender and receiver types in `async_cuda` module.
- Fixes the `completion_signatures` for senders in `async_cuda`. Note that the helpers for computing the types have to be guarded with a `requires is_invocable_v...` because `make_completion_signatures` instantiates the helper with non-`set_value_t` signatures as well. I have to investigate whether this is a bug or expected behaviour.
- Because of the above, a number of helper "kernels" have received trailing `decltype`s to enable SFINAE with bogus arguments.
- Removes the SFINAE for `then_with_cuda_stream` etc. since the HIP branch has been merged into DLA-Future. The SFINAE is no longer needed for choosing a `then_with_*` overload and adds unnecessary complexity.
- The reference implementation's `split` does not yet support move-only senders, so some additional tests are disabled with CUDA and the reference implementation.
- Change the `clang-14` Jenkins configuration to use CUDA and the P2300 reference implementation.

Note, this is not for 0.7.0 as it needs more work. Reviews are, however, already welcome.

346: Remove `operator|` overloads for `sync_wait` and `start_detached` r=msimberg a=msimberg

Fixes #329 .

347: Add Codacy coverage report r=msimberg a=msimberg

Part of #4. Not fully tested, but I got some reporting already to Codacy so there is hope that this works...

To do:
- [x] badge

352: Increase timeout for container algorithms tests on macOS CI configuration r=msimberg a=msimberg

I'm increasing the timeout to check if these timeouts https://github.com/pika-org/pika/runs/7692289838?check_suite_focus=true#step:6:237 are simply due to the test taking long to complete or if it really hangs. `partial_sort_copy_range` is one of the longer running tests so it's possible that it indeed just doesn't have enough time to finish (the timeout is 120 seconds and one of the later runs finished in 101 seconds: https://github.com/pika-org/pika/runs/7691710341?check_suite_focus=true#step:6:237).

Co-authored-by: Mikael Simberg <simberg@cscs.ch>
Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

.circleci/config.yml

@@ -21,7 +21,7 @@ working_dir_default: &working_dir_default
 
 docker_default: &docker_default
     docker:
-      - image: pikaorg/pika-ci-base:7
+      - image: pikaorg/pika-ci-base:8
 
 defaults: &defaults
     <<: *working_dir_default

.github/bors.toml

@@ -45,7 +45,7 @@ status = [
   "jenkins/cscs/clang-11-debug",
   "jenkins/cscs/clang-12-debug",
   "jenkins/cscs/clang-13-debug",
-  "jenkins/cscs/clang-14-debug",
+  "jenkins/cscs/clang-14-cuda-debug",
   "jenkins/cscs/gcc-9-debug",
   "jenkins/cscs/gcc-10-apex-debug",
   "jenkins/cscs/gcc-11-debug",

.github/workflows/linux_coverage.yml

@@ -0,0 +1,64 @@
+# Copyright (c) 2020-2022 ETH Zurich
+#
+# SPDX-License-Identifier: BSL-1.0
+# Distributed under the Boost Software License, Version 1.0. (See accompanying
+# file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+name: Linux CI (Coverage)
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+
+jobs:
+  build:
+    name: github/linux/coverage
+    runs-on: ubuntu-latest
+    container: pikaorg/pika-ci-base:8
+
+    steps:
+    - uses: actions/checkout@v2
+    - name: Configure
+      shell: bash
+      run: |
+          cmake \
+              . \
+              -Bbuild \
+              -GNinja \
+              -DCMAKE_BUILD_TYPE=Debug \
+              -DCMAKE_CXX_FLAGS="--coverage" \
+              -DCMAKE_EXE_LINKER_FLAGS="--coverage" \
+              -DPIKA_WITH_UNITY_BUILD=ON \
+              -DPIKA_WITH_PRECOMPILED_HEADERS=ON \
+              -DPIKA_WITH_MALLOC=system \
+              -DPIKA_WITH_EXAMPLES=ON \
+              -DPIKA_WITH_TESTS=ON \
+              -DPIKA_WITH_TESTS_HEADERS=OFF \
+              -DPIKA_WITH_TESTS_MAX_THREADS=2 \
+              -DPIKA_WITH_PARALLEL_TESTS_BIND_NONE=ON
+    - name: Build
+      shell: bash
+      run: |
+          cmake --build build --target tests
+    - name: Test
+      shell: bash
+      working-directory: build
+      continue-on-error: true
+      run: |
+          ctest -j2 --timeout 120 --output-on-failure
+    - name: Upload coverage report
+      shell: bash
+      working-directory: build
+      run: |
+          grcov . \
+              --source-dir .. \
+              --output-path lcov.info \
+              --output-type lcov \
+              --llvm \
+              --ignore-not-existing
+          bash <(curl -Ls https://coverage.codacy.com/get.sh) report \
+              --project-token ${{ secrets.CODACY_PIKA_PROJECT_TOKEN }} \
+              --language CPP \
+              --coverage-reports lcov.info

.github/workflows/linux_debug.yml

@@ -21,7 +21,7 @@ jobs:
   build:
     name: github/linux/debug/fast
     runs-on: ubuntu-latest
-    container: pikaorg/pika-ci-base:7
+    container: pikaorg/pika-ci-base:8
 
     steps:
     - uses: actions/checkout@v2

.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:7
+    container: pikaorg/pika-ci-base:8
 
     steps:
     - uses: actions/checkout@v2

.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:7
+    container: pikaorg/pika-ci-base:8
 
     steps:
     - uses: actions/checkout@v2

.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:7
+    container: pikaorg/pika-ci-base:8
 
     steps:
     - name: Check out Tracy

.github/workflows/macos_debug.yml

@@ -289,6 +289,6 @@ jobs:
           ctest \
             -j3 \
             --output-on-failure \
-            --timeout 120 \
+            --timeout 300 \
             --tests-regex tests.unit.modules.algorithms.container_algorithms \
             --exclude-regex tests.unit.modules.algorithms.container_algorithms.inplace_merge_range

.jenkins/cscs/Jenkinsfile

@@ -42,7 +42,7 @@ pipeline {
                 axes {
                     axis {
                         name 'configuration_name'
-                        values 'gcc-12', 'gcc-11', 'gcc-10-apex', 'gcc-9', 'gcc-cuda', 'clang-14', 'clang-13', 'clang-12', 'clang-11', 'cce-cuda'
+                        values 'gcc-12', 'gcc-11', 'gcc-10-apex', 'gcc-9', 'gcc-cuda', 'clang-14-cuda', 'clang-13', 'clang-12', 'clang-11', 'cce-cuda'
                     }
                     axis {
                          name 'build_type'

.jenkins/cscs/env-clang-14.sh → .jenkins/cscs/env-clang-14-cuda.sh

@@ -5,14 +5,16 @@
 # file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
 
 cxx_std="20"
-clang_version="14.0.2"
+clang_version="14.0.6"
 boost_version="1.79.0"
 hwloc_version="2.7.0"
+cuda_version="11.5.0"
 spack_compiler="clang@${clang_version}"
-spack_arch="cray-cnl7-broadwell"
+spack_arch="cray-cnl7-haswell"
 
-spack_spec="pika@main arch=${spack_arch} %${spack_compiler} malloc=system cxxstd=${cxx_std} ^boost@${boost_version} ^hwloc@${hwloc_version}"
+spack_spec="pika@main arch=${spack_arch} %${spack_compiler} +cuda malloc=system cxxstd=${cxx_std} ^boost@${boost_version} ^cuda@${cuda_version} +allow-unsupported-compilers ^hwloc@${hwloc_version}"
 
 configure_extra_options+=" -DPIKA_WITH_CXX_STANDARD=${cxx_std}"
 configure_extra_options+=" -DPIKA_WITH_MALLOC=system"
 configure_extra_options+=" -DPIKA_WITH_SPINLOCK_DEADLOCK_DETECTION=ON"
+configure_extra_options+=" -DPIKA_WITH_P2300_REFERENCE_IMPLEMENTATION=ON"

README.rst

@@ -12,6 +12,7 @@
 |github_actions_linux_sanitizers_status|
 |github_actions_macos_debug_status|
 |codacy|
+|codacy_coverage|
 
 ====
 pika
@@ -147,3 +148,7 @@ Pick your favourite meaning from the following:
 .. |codacy| image:: https://api.codacy.com/project/badge/Grade/e03f57f1c4cd40e7b514e552a723c125
      :target: https://www.codacy.com/gh/pika-org/pika
      :alt: Codacy
+
+.. |codacy_coverage| image:: https://api.codacy.com/project/badge/Coverage/e03f57f1c4cd40e7b514e552a723c125
+     :target: https://www.codacy.com/gh/pika-org/pika
+     :alt: Codacy coverage

libs/pika/algorithms/tests/unit/container_algorithms/foreach_tests.hpp

@@ -348,8 +348,8 @@ void test_for_each_sender(ExPolicy&& p, IteratorTag)
     auto rng = pika::util::make_iterator_range(
         iterator(std::begin(c)), iterator(std::end(c)));
     auto f = [](std::size_t& v) { v = 42; };
-    auto result = ex::just(rng, f) |
-        pika::ranges::for_each(std::forward<ExPolicy>(p)) | tt::sync_wait();
+    auto result = tt::sync_wait(
+        ex::just(rng, f) | pika::ranges::for_each(std::forward<ExPolicy>(p)));
     PIKA_TEST(result == iterator(std::end(c)));
 
     // verify values
@@ -380,8 +380,8 @@ void test_for_each_exception_sender(ExPolicy p, IteratorTag)
     bool caught_exception = false;
     try
     {
-        ex::just(rng, f) | pika::ranges::for_each(std::forward<ExPolicy>(p)) |
-            tt::sync_wait();
+        tt::sync_wait(ex::just(rng, f) |
+            pika::ranges::for_each(std::forward<ExPolicy>(p)));
 
         PIKA_TEST(false);
     }
@@ -417,8 +417,8 @@ void test_for_each_bad_alloc_sender(ExPolicy p, IteratorTag)
     bool caught_exception = false;
     try
     {
-        ex::just(rng, f) | pika::ranges::for_each(std::forward<ExPolicy>(p)) |
-            tt::sync_wait();
+        tt::sync_wait(ex::just(rng, f) |
+            pika::ranges::for_each(std::forward<ExPolicy>(p)));
 
         PIKA_TEST(false);
     }

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

@@ -88,6 +88,12 @@ namespace pika::cuda::experimental {
                 cuda_scheduler scheduler;
                 PIKA_NO_UNIQUE_ADDRESS std::decay_t<Receiver> receiver;
 
+                template <typename Receiver_>
+                operation_state(cuda_scheduler scheduler, Receiver_&& receiver)
+                  : scheduler(PIKA_MOVE(scheduler))
+                  , receiver(PIKA_FORWARD(Receiver_, receiver))
+                {
+                }
                 operation_state(operation_state&&) = delete;
                 operation_state(operation_state const&) = delete;
                 operation_state& operator=(operation_state&&) = delete;
@@ -148,7 +154,7 @@ namespace pika::cuda::experimental {
             friend cuda_scheduler tag_invoke(
                 pika::execution::experimental::get_completion_scheduler_t<
                     pika::execution::experimental::set_value_t>,
-                cuda_scheduler_sender const& s)
+                cuda_scheduler_sender const& s) noexcept
             {
                 return s.scheduler;
             }

libs/pika/async_cuda/include/pika/async_cuda/cuda_scheduler_bulk.hpp

@@ -19,6 +19,7 @@
 #include <pika/execution_base/operation_state.hpp>
 #include <pika/execution_base/receiver.hpp>
 #include <pika/execution_base/sender.hpp>
+#include <pika/iterator_support/range.hpp>
 
 #include <type_traits>