From 0ff0779a753888f53d1cce905dc1d0767ff35868 Mon Sep 17 00:00:00 2001
From: Michael Varvarin <55709728+MichaelVarvarin@users.noreply.github.com>
Date: Sun, 17 Nov 2024 18:14:58 +0200
Subject: [PATCH] Add gridSyncTest to test syncGridThreads()

---
 .../alpaka/test/KernelExecutionFixture.hpp    | 12 ++-
 test/unit/CMakeLists.txt                      |  1 +
 test/unit/grid/CMakeLists.txt                 | 20 +++++
 test/unit/grid/src/GridSync.cpp               | 86 +++++++++++++++++++
 4 files changed, 117 insertions(+), 2 deletions(-)
 create mode 100644 test/unit/grid/CMakeLists.txt
 create mode 100644 test/unit/grid/src/GridSync.cpp

diff --git a/include/alpaka/test/KernelExecutionFixture.hpp b/include/alpaka/test/KernelExecutionFixture.hpp
index 0e59344497e..893f85bab94 100644
--- a/include/alpaka/test/KernelExecutionFixture.hpp
+++ b/include/alpaka/test/KernelExecutionFixture.hpp
@@ -22,7 +22,7 @@
 namespace alpaka::test
 {
     //! The fixture for executing a kernel on a given accelerator.
-    template<typename TAcc>
+    template<typename TAcc, bool TCooperative = false>
     class KernelExecutionFixture
     {
     public:
@@ -80,7 +80,15 @@ namespace alpaka::test
                     getPtrNative(bufAccResult),
                     std::forward<TArgs>(args)...);
 
-            exec<Acc>(m_queue, m_workDiv, kernelFnObj, getPtrNative(bufAccResult), std::forward<TArgs>(args)...);
+            if constexpr(TCooperative)
+                execCooperative<Acc>(
+                    m_queue,
+                    m_workDiv,
+                    kernelFnObj,
+                    getPtrNative(bufAccResult),
+                    std::forward<TArgs>(args)...);
+            else
+                exec<Acc>(m_queue, m_workDiv, kernelFnObj, getPtrNative(bufAccResult), std::forward<TArgs>(args)...);
 
             // Copy the result value to the host
             auto bufHostResult = allocBuf<bool, Idx>(m_devHost, static_cast<Idx>(1u));
diff --git a/test/unit/CMakeLists.txt b/test/unit/CMakeLists.txt
index f3ea6672d1a..9094dbf3406 100644
--- a/test/unit/CMakeLists.txt
+++ b/test/unit/CMakeLists.txt
@@ -21,6 +21,7 @@ add_subdirectory("block/sync/")
 add_subdirectory("core/")
 add_subdirectory("dev/")
 add_subdirectory("event/")
+add_subdirectory("grid/")
 add_subdirectory("idx/")
 add_subdirectory("intrinsic/")
 add_subdirectory("kernel/")
diff --git a/test/unit/grid/CMakeLists.txt b/test/unit/grid/CMakeLists.txt
new file mode 100644
index 00000000000..5c7bcfb8f7b
--- /dev/null
+++ b/test/unit/grid/CMakeLists.txt
@@ -0,0 +1,20 @@
+#
+# Copyright 2024 Mykhailo Varvarin
+# SPDX-License-Identifier: MPL-2.0
+#
+
+set(_TARGET_NAME "gridSyncTest")
+
+append_recursive_files_add_to_src_group("src/" "src/" "cpp" _FILES_SOURCE)
+
+alpaka_add_executable(
+        ${_TARGET_NAME}
+        ${_FILES_SOURCE})
+target_link_libraries(
+        ${_TARGET_NAME}
+        PRIVATE common)
+
+set_target_properties(${_TARGET_NAME} PROPERTIES FOLDER "test/unit")
+target_compile_definitions(${_TARGET_NAME} PRIVATE "-DTEST_UNIT_GRID_SYNC")
+
+add_test(NAME ${_TARGET_NAME} COMMAND ${_TARGET_NAME} ${_alpaka_TEST_OPTIONS})
diff --git a/test/unit/grid/src/GridSync.cpp b/test/unit/grid/src/GridSync.cpp
new file mode 100644
index 00000000000..13f5107f308
--- /dev/null
+++ b/test/unit/grid/src/GridSync.cpp
@@ -0,0 +1,86 @@
+/* Copyright 2024 Mykhailo Varvarin
+ * SPDX-License-Identifier: MPL-2.0
+ */
+
+#include <alpaka/grid/Traits.hpp>
+#include <alpaka/test/KernelExecutionFixture.hpp>
+#include <alpaka/test/acc/TestAccs.hpp>
+
+#include <catch2/catch_template_test_macros.hpp>
+#include <catch2/catch_test_macros.hpp>
+
+class GridSyncTestKernel
+{
+public:
+    static constexpr std::uint8_t blockThreadExtentPerDim()
+    {
+        return 2u;
+    }
+
+    ALPAKA_NO_HOST_ACC_WARNING
+    template<typename TAcc, typename T>
+    ALPAKA_FN_ACC auto operator()(TAcc const& acc, bool* success, T* array) const -> void
+    {
+        using Idx = alpaka::Idx<TAcc>;
+
+        // Get the index of the current thread within the grid and the grid extent and map them to 1D.
+        auto const gridThreadIdx = alpaka::getIdx<alpaka::Grid, alpaka::Threads>(acc);
+        auto const gridThreadExtent = alpaka::getWorkDiv<alpaka::Grid, alpaka::Threads>(acc);
+        auto const gridThreadIdx1D = alpaka::mapIdx<1u>(gridThreadIdx, gridThreadExtent)[0u];
+        auto const gridThreadExtent1D = gridThreadExtent.prod();
+
+
+        // Write the thread index into the shared array.
+        array[gridThreadIdx1D] = static_cast<T>(gridThreadIdx1D);
+
+        // Synchronize the threads in the block.
+        alpaka::syncGridThreads(acc);
+
+        // All other threads within the block should now have written their index into the shared memory.
+        for(auto i = static_cast<Idx>(0u); i < gridThreadExtent1D; ++i)
+        {
+            ALPAKA_CHECK(*success, static_cast<Idx>(array[i]) == i);
+        }
+    }
+};
+
+TEMPLATE_LIST_TEST_CASE("synchronize", "[gridSync]", alpaka::test::TestAccs)
+{
+    using Acc = TestType;
+    using Dim = alpaka::Dim<Acc>;
+    using Idx = alpaka::Idx<Acc>;
+
+    // Select the first device available on a system, for the chosen accelerator
+    auto const platformAcc = alpaka::Platform<Acc>{};
+    auto const devAcc = getDevByIdx(platformAcc, 0u);
+
+
+    auto const blockThreadExtentMax = alpaka::getAccDevProps<Acc>(devAcc).m_blockThreadExtentMax;
+    auto threadsPerBlock = alpaka::elementwise_min(
+        blockThreadExtentMax,
+        alpaka::Vec<Dim, Idx>::all(static_cast<Idx>(GridSyncTestKernel::blockThreadExtentPerDim())));
+
+    auto elementsPerThread = alpaka::Vec<Dim, Idx>::all(1);
+    auto blocksPerGrid = alpaka::Vec<Dim, Idx>::all(1);
+    blocksPerGrid[0] = 200;
+
+    // Allocate memory on the device.
+    alpaka::Vec<alpaka::DimInt<1>, Idx> bufferExtent{
+        blocksPerGrid.prod() * threadsPerBlock.prod() * elementsPerThread.prod()};
+    auto deviceMemory = alpaka::allocBuf<Idx, Idx>(devAcc, bufferExtent);
+
+    GridSyncTestKernel kernel;
+    int maxBlocks = alpaka::getMaxActiveBlocks<Acc>(
+        devAcc,
+        kernel,
+        threadsPerBlock,
+        elementsPerThread,
+        alpaka::getPtrNative(deviceMemory));
+
+    blocksPerGrid[0] = std::min(static_cast<Idx>(maxBlocks), blocksPerGrid[0]);
+    constexpr bool IsCooperative = true;
+    alpaka::test::KernelExecutionFixture<Acc, IsCooperative> fixture(
+        alpaka::WorkDivMembers<Dim, Idx>{blocksPerGrid, threadsPerBlock, elementsPerThread});
+
+    REQUIRE(fixture(kernel, alpaka::getPtrNative(deviceMemory)));
+}