fixes: new clang-tidy suggestions

tests/kokkos_ext/execution_space/test_operation_state.cpp

@@ -1,3 +1,11 @@
+#include "tests/IgnoreWarnings.hpp"
+PRAGMA_DIAGNOSTIC_PUSH
+PRAGMA_DIAGNOSTIC_IGNORED_DEPRECATED_ATTRIBUTES
+PRAGMA_DIAGNOSTIC_IGNORED("-Wshadow")
+PRAGMA_DIAGNOSTIC_IGNORED("-Wsuggest-override")
+#include "stdexec/execution.hpp"
+PRAGMA_DIAGNOSTIC_POP
+
 #include "kokkos_ext/impl/execution_space/operation_state.hpp"
 #include "kokkos_ext/impl/execution_space/parallel_for.hpp"
 

tests/kokkos_ext/execution_space/test_parallel_for.cpp

@@ -1,5 +1,13 @@
 #include <bit>
 
+#include "tests/IgnoreWarnings.hpp"
+PRAGMA_DIAGNOSTIC_PUSH
+PRAGMA_DIAGNOSTIC_IGNORED_DEPRECATED_ATTRIBUTES
+PRAGMA_DIAGNOSTIC_IGNORED("-Wshadow")
+PRAGMA_DIAGNOSTIC_IGNORED("-Wsuggest-override")
+#include "stdexec/execution.hpp"
+PRAGMA_DIAGNOSTIC_POP
+
 #include "kokkos_ext/impl/execution_space/parallel_for.hpp"
 
 #include "kokkos-utils/callbacks/RecorderListener.hpp"

tests/kokkos_ext/execution_space/test_scoped_region.cpp

@@ -1,3 +1,11 @@
+#include "tests/IgnoreWarnings.hpp"
+PRAGMA_DIAGNOSTIC_PUSH
+PRAGMA_DIAGNOSTIC_IGNORED_DEPRECATED_ATTRIBUTES
+PRAGMA_DIAGNOSTIC_IGNORED("-Wshadow")
+PRAGMA_DIAGNOSTIC_IGNORED("-Wsuggest-override")
+#include "stdexec/execution.hpp"
+PRAGMA_DIAGNOSTIC_POP
+
 #include "kokkos-utils/callbacks/RecorderListener.hpp"
 #include "kokkos-utils/tests/scoped/callbacks/Manager.hpp"
 

tests/kokkos_ext/graph/test_inter_op.cpp

@@ -1,5 +1,6 @@
 #include "tests/IgnoreWarnings.hpp"
 PRAGMA_DIAGNOSTIC_PUSH
+PRAGMA_DIAGNOSTIC_IGNORED_DEPRECATED_ATTRIBUTES
 PRAGMA_DIAGNOSTIC_IGNORED("-Wdeprecated-copy")
 PRAGMA_DIAGNOSTIC_IGNORED("-Wshadow")
 PRAGMA_DIAGNOSTIC_IGNORED("-Wsuggest-override")

tests/native/APIWrappers.hpp

@@ -155,7 +155,7 @@ template <typename Functor>
 struct GraphNodeKernel : public GraphNode
 {
     PREFIXED_API(KernelNodeParams) params;
-    std::vector<void*> inputs;
+    std::array<void*, 1> inputs {};
 
     //! @warning The user is responsible for keeping the @c functor alive.
     GraphNodeKernel(const Functor& functor, const size_t shape);

tests/native/APIWrappers_def.hpp

@@ -226,8 +226,7 @@ GraphNodeKernel<Functor>::GraphNodeKernel(const Functor& functor, const size_t s
 
     params = {};
 
-    inputs.resize(1);
-    inputs[0] = (void*)&functor;
+    inputs.at(0) = (void * )&functor;
 
     constexpr size_t max_block_size = 1024;
 

tests/native/CMakeLists.txt

@@ -8,13 +8,22 @@ set(launcher)
 
 if(CUDA IN_LIST Kokkos_DEVICES)
     set(enabled_backend GRAPH_DISPATCHING_ENABLE_CUDA)
-    set(launcher compute-sanitizer --leak-check=full)
+    # Using 'compute-sanitizer' may trigger a weird supposedly false positive error with nvcc on blackwell120.
+    # See https://forums.developer.nvidia.com/t/compute-sanitizer-and-cuda-graph-false-positives.
+    if(Kokkos_CXX_COMPILER_ID STREQUAL "NVIDIA" AND BLACKWELL120 IN_LIST Kokkos_ARCH)
+        set(launcher)
+    else()
+        set(launcher compute-sanitizer --leak-check=full)
+    endif()
 elseif(HIP IN_LIST Kokkos_DEVICES)
     set(enabled_backend GRAPH_DISPATCHING_ENABLE_HIP)
     set(launcher)
 elseif(NOT DEFINED ENABLE_DOXYGEN)
     message(FATAL_ERROR "Unsupported device backend.")
 endif()
+if(launcher)
+    message(STATUS "Launcher for native tests set to '${launcher}'")
+endif()
 
 # Tests that work for both Cuda and HIP.
 add_one_test(NAME APIWrappers COMPILE_DEFINITIONS ${enabled_backend} LAUNCHER ${launcher})

tests/native/test_APIWrappers.cpp

@@ -66,6 +66,7 @@ TEST(APIWrappers, view_get_host_copy)
     const View<double> data(stream, 2<<6);
 
     const auto mirror = data.get_host_copy(stream);
+    stream.fence();
 
     ASSERT_EQ(mirror.size(), 2<<6);
 }

tests/native/test_graph_capture.cpp

@@ -125,6 +125,7 @@ PRAGMA_DIAGNOSTIC_POP
 
         const auto mirror = dense.get_host_copy(stream);
         stream.fence();
+
         ASSERT_EQ(mirror, (std::vector<value_t>{2., 3., 4., 5., 6.}));
     }
 

tests/native/test_node_enable.cpp

@@ -54,6 +54,7 @@ TEST_P(cuda, node_enable)
     }
 
     const auto mirror = data.get_host_copy(stream);
+    stream.fence();
 
     ASSERT_EQ(mirror.at(0), expected);
 

tests/nvexec/adaptors/test_stream_context.cpp

@@ -157,7 +157,7 @@ TEST_F(StreamContextTest, move_to_static_thread_pool) {
 
     const view_t witness(Kokkos::view_alloc("witness"));
 
-    exec::static_thread_pool pool{1};
+    exec::static_thread_pool pool{1}; // NOLINT(misc-const-correctness)
 
     auto chain =
         ::stdexec::schedule(stream_ctx.get_scheduler())

tests/nvexec/adaptors/test_when_all.cpp

@@ -48,7 +48,7 @@ TEST_F(StreamContextTest, same_type_children) {
 
 //! @test Check @c ::stdexec::when_all with one branch on stream scheduler and one on static thread pool scheduler.
 TEST_F(StreamContextTest, different_type_children) {
-    ::exec::static_thread_pool thread_ctx{1};
+    ::exec::static_thread_pool thread_ctx{1}; // NOLINT(misc-const-correctness)
 
     auto schd_thread = thread_ctx.get_scheduler();
     auto schd_stream = stream_ctx.get_scheduler();

tests/stdexec/adaptors/test_customization.cpp

@@ -367,7 +367,7 @@ TEST_F(CustomizationTest, on_in_the_middle_otherwise_custom_scheduler) {
 TEST_F(CustomizationTest, on) {
     SETUP_TRACE(8, {'W', 'L'}, {'W', 'L'}, DEFAULT_PAIR, DEFAULT_PAIR, {'Y', 'L'}, DEFAULT_PAIR, {'Z', 'L'}, {'Z', 'L'})
 
-    exec::static_thread_pool pool{1};
+    exec::static_thread_pool pool{1}; // NOLINT(misc-const-correctness)
 
     ::stdexec::sender auto chain = ::stdexec::schedule(DomainSpecificScheduler<Domain<'W'>>{}) | ADD_THEN(0)
                                  | ADD_THEN(1) | ::stdexec::continues_on(pool.get_scheduler()) | ADD_THEN(2)

tests/stdexec/adaptors/test_starts_on.cpp

@@ -50,8 +50,8 @@ TEST_F(StartsOnTest, twice_with_just_a_bulk) {
     ::stdexec::sender auto chain =
         ::stdexec::just(std::vector<size_t>(size, 0))
         | ::stdexec::bulk(::stdexec::par, size, [](const auto index, auto& data) {
-              data[index] =
-                  ::utils::get_thread_id(); // NOLINT(cppcoreguidelines-pro-bounds-avoid-unchecked-container-access)
+              data[index] = // NOLINT(cppcoreguidelines-pro-bounds-avoid-unchecked-container-access)
+                  ::utils::get_thread_id();
           });
 
     //! Run on pool A.

tests/stdexec/adaptors/test_when_all.cpp

@@ -31,7 +31,7 @@ namespace tests::stdexec::adaptors {
 
 //! @test Same as https://github.com/NVIDIA/stdexec/blob/970dbace4ad52a38c9b18665d077f14159792b23/test/stdexec/algos/adaptors/test_when_all.cpp#L253.
 TEST(when_all, propagates_completion_domain_from_same_type_children) {
-    ::exec::static_thread_pool pool_A{1}, pool_B{1};
+    ::exec::static_thread_pool pool_A{1}, pool_B{1}; // NOLINT(misc-const-correctness)
 
     ::stdexec::sender auto sndr = ::stdexec::when_all(
         ::stdexec::starts_on(pool_A.get_scheduler(), ::stdexec::just(13)),
@@ -52,7 +52,7 @@ TEST(when_all, propagates_completion_domain_from_same_type_children) {
  * @warning As stated in @cite P2999R3 (Section 2.1.1, "Dispatching via execution domain tags"), @c stdexec::when_all only accepts a set of senders when they all share a common domain.
  */
 TEST(when_all, propagates_completion_domain_from_different_type_children) {
-    ::exec::static_thread_pool pool{1};
+    ::exec::static_thread_pool pool{1}; // NOLINT(misc-const-correctness)
 
     ::stdexec::sender auto sndr = ::stdexec::when_all(
         ::stdexec::starts_on(::stdexec::inline_scheduler{}, ::stdexec::just(13)),

tests/stdexec/factories/test_read_env.cpp

@@ -40,7 +40,7 @@ TEST(read_env, get_scheduler_default) {
 
 //! @test Use @c stdexec::read_env to retrieve the static thread pool scheduler type.
 TEST(read_env, get_scheduler_static_thread_pool) {
-    ::exec::static_thread_pool pool{1};
+    ::exec::static_thread_pool pool{1}; // NOLINT(misc-const-correctness)
 
     ::stdexec::sync_wait(::stdexec::schedule(pool.get_scheduler()) | ::stdexec::let_value([] {
                              return ::stdexec::read_env(::stdexec::get_scheduler)

tests/stdexec/test_HelloWorld.cpp

@@ -59,7 +59,7 @@ ::stdexec::sender auto tester(Sender&& start) {
 TEST(stdexec, hello_world) {
     //! Retrieve the NUMA configuration, and get the number of CPUs on the first node for sizing the thread pool.
     const exec::numa_policy numa(exec::no_numa_policy{});
-    exec::static_thread_pool context(numa.num_cpus(0));
+    exec::static_thread_pool context(numa.num_cpus(0)); // NOLINT(misc-const-correctness)
 
     //! Get a scheduler from the thread pool.
     ::stdexec::scheduler auto scheduler = context.get_scheduler();