SWDEV-505672 - Add performance test for HtoD and DtoH with kernel in-between (#27)

Author: marandje

catch/performance/memcpy/CMakeLists.txt

@@ -30,6 +30,7 @@ set(TEST_SRC
     hipMemcpyDtoHAsync.cc
     hipMemcpyHtoD.cc
     hipMemcpyHtoDAsync.cc
+    hipMemcpyWithKernel.cc
     hipMemcpyToSymbol.cc
     hipMemcpyToSymbolAsync.cc
     hipMemcpyFromSymbol.cc

catch/performance/memcpy/hipMemcpyWithKernel.cc

@@ -0,0 +1,124 @@
+/*
+Copyright (c) 2024 Advanced Micro Devices, Inc. All rights reserved.
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+*/
+#include "memcpy_performance_common.hh"
+/**
+ * @addtogroup memcpy memcpy
+ * @{
+ * @ingroup PerformanceTest
+ */
+__global__ void Sum(void* ptr, size_t size) {
+  size_t index = blockDim.x * blockIdx.x + threadIdx.x;
+  if (index != 0 && index < size) {
+    atomicAdd(&((unsigned long long*)ptr)[0], ((unsigned long long*)ptr)[index]);
+  }
+}
+class MemcpyHtoDKernelDtoHv1AsyncBenchmark : public Benchmark<MemcpyHtoDKernelDtoHv1AsyncBenchmark> {
+ public:
+  void operator()(void* host_mem, void* device_mem, size_t size, const hipStream_t& stream) {
+    size_t count = size / sizeof(size_t);
+    for (size_t i = 0; i < count; i++) {
+      ((size_t*)host_mem)[i] = i;
+    }
+    TIMED_SECTION_STREAM(kTimerTypeCpu, stream) {
+      HIP_CHECK(hipMemcpyHtoDAsync((hipDeviceptr_t)device_mem, host_mem, size, stream));
+      int threads_num = 32;
+      Sum<<<count / threads_num + 1, threads_num, 0, stream>>>(device_mem, count);
+      HIP_CHECK(hipMemcpyDtoHAsync(host_mem, (hipDeviceptr_t)device_mem, size, stream));
+      HIP_CHECK(hipStreamSynchronize(stream));
+    }
+    size_t sum = ((size_t*)host_mem)[0];
+    REQUIRE(sum == count * (count - 1) / 2);
+  }
+};
+class MemcpyHtoDKernelDtoHv2AsyncBenchmark : public Benchmark<MemcpyHtoDKernelDtoHv2AsyncBenchmark> {
+ public:
+  void operator()(void* host_mem, void* device_mem, size_t size, const hipStream_t& stream) {
+    size_t count = size / sizeof(size_t);
+    for (size_t i = 0; i < count; i++) {
+      ((size_t*)host_mem)[i] = i;
+    }
+    TIMED_SECTION_STREAM(kTimerTypeCpu, stream) {
+      HIP_CHECK(hipMemcpyAsync((hipDeviceptr_t)device_mem, host_mem, size, hipMemcpyHostToDevice,
+                               stream));
+      int threads_num = 32;
+      Sum<<<count / threads_num + 1, threads_num, 0, stream>>>(device_mem, count);
+      HIP_CHECK(hipMemcpyWithStream(host_mem, (hipDeviceptr_t)device_mem, size,
+                                    hipMemcpyDeviceToHost, stream));
+      HIP_CHECK(hipStreamSynchronize(stream));
+    }
+    size_t sum = ((size_t*)host_mem)[0];
+    REQUIRE(sum == count * (count - 1) / 2);
+  }
+};
+template<typename BenchmarkType>
+static void RunBenchmark(LinearAllocs host_allocation_type, LinearAllocs device_allocation_type,
+                         size_t size) {
+  BenchmarkType benchmark;
+  if (size < 1_KB) {
+    benchmark.AddSectionName(std::to_string(size));
+  } else if (size < 1_MB) {
+    benchmark.AddSectionName(std::to_string(size / 1024) + std::string(" KB"));
+  } else {
+    benchmark.AddSectionName(std::to_string(size / (1024 * 1024)) + std::string(" MB"));
+  }
+  benchmark.AddSectionName(GetAllocationSectionName(host_allocation_type));
+  const StreamGuard stream_guard(Streams::created);
+  const hipStream_t stream = stream_guard.stream();
+  LinearAllocGuard<size_t> device_allocation(device_allocation_type, size);
+  LinearAllocGuard<size_t> host_allocation(host_allocation_type, size);
+  benchmark.Run(host_allocation.ptr(), device_allocation.ptr(), size, stream);
+}
+/**
+ * Test Description
+ * ------------------------
+ *  - Executes `hipMemcpyHtoDAsync->Kernel->hipMemcpyDtoHAsync` from Device to Host:
+ *    -# Allocation size
+ *      - Small: 4 KB
+ *      - Medium: 4 MB
+ *      - Large: 16 MB
+ *    -# Allocation type
+ *      - Source: device malloc
+ *      - Destination: host pinned and pageable
+ * Test source
+ * ------------------------
+ *  - performance/memcpy/hipMemcpyDtoHAsync.cc
+ * Test requirements
+ * ------------------------
+ *  - HIP_VERSION >= 5.2
+ */
+TEST_CASE("Performance_hipMemcpyHtoDKernelDtoHV1Async") {
+  const auto allocation_size =
+      GENERATE(16, 128, 1_KB, 4_KB, 16_KB, 256_KB, 512_KB, 1_MB, 4_MB, 16_MB, 128_MB);
+  const auto device_allocation_type = LinearAllocs::hipMalloc;
+  const auto host_allocation_type = GENERATE(LinearAllocs::malloc, LinearAllocs::hipHostMalloc);
+  RunBenchmark<MemcpyHtoDKernelDtoHv1AsyncBenchmark>(host_allocation_type, device_allocation_type,
+                                                     allocation_size);
+}
+TEST_CASE("Performance_hipMemcpyHtoDKernelDtoHV2Async") {
+  const auto allocation_size =
+      GENERATE(16, 128, 1_KB, 4_KB, 16_KB, 256_KB, 512_KB, 1_MB, 4_MB, 16_MB, 128_MB);
+  const auto device_allocation_type = LinearAllocs::hipMalloc;
+  const auto host_allocation_type = GENERATE(LinearAllocs::malloc, LinearAllocs::hipHostMalloc);
+  RunBenchmark<MemcpyHtoDKernelDtoHv2AsyncBenchmark>(host_allocation_type, device_allocation_type,
+                                                     allocation_size);
+}
+/**
+ * End doxygen group memcpy.
+ * @}
+ */