/* Copyright (c) 2022 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. */ #pragma once #include "hip_module_common.hh" #include #include #include inline ModuleGuard InitModule() { HIP_CHECK(hipFree(nullptr)); return ModuleGuard::LoadModule("launch_kernel_module.code"); } inline ModuleGuard mg{InitModule()}; using ExtModuleLaunchKernelSig = hipError_t(hipFunction_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, size_t, hipStream_t, void**, void**, hipEvent_t, hipEvent_t, uint32_t); template void ModuleLaunchKernelPositiveBasic() { SECTION("Kernel with no arguments") { hipFunction_t f = GetKernel(mg.module(), "NOPKernel"); HIP_CHECK(func(f, 1, 1, 1, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u)); HIP_CHECK(hipDeviceSynchronize()); } SECTION("Kernel with arguments using kernelParams") { hipFunction_t f = GetKernel(mg.module(), "Kernel42"); LinearAllocGuard result_dev(LinearAllocs::hipMalloc, sizeof(int)); HIP_CHECK(hipMemset(result_dev.ptr(), 0, sizeof(*result_dev.ptr()))); int* result_ptr = result_dev.ptr(); void* kernel_args[1] = {&result_ptr}; HIP_CHECK(func(f, 1, 1, 1, 1, 1, 1, 0, nullptr, kernel_args, nullptr, nullptr, nullptr, 0u)); int result = 0; HIP_CHECK(hipMemcpy(&result, result_dev.ptr(), sizeof(result), hipMemcpyDefault)); REQUIRE(result == 42); } SECTION("Kernel with arguments using extra") { hipFunction_t f = GetKernel(mg.module(), "Kernel42"); LinearAllocGuard result_dev(LinearAllocs::hipMalloc, sizeof(int)); HIP_CHECK(hipMemset(result_dev.ptr(), 0, sizeof(*result_dev.ptr()))); int* result_ptr = result_dev.ptr(); size_t size = sizeof(result_ptr); // clang-format off void *extra[] = { HIP_LAUNCH_PARAM_BUFFER_POINTER, &result_ptr, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size, HIP_LAUNCH_PARAM_END }; // clang-format on HIP_CHECK(func(f, 1, 1, 1, 1, 1, 1, 0, nullptr, nullptr, extra, nullptr, nullptr, 0u)); int result = 0; HIP_CHECK(hipMemcpy(&result, result_dev.ptr(), sizeof(result), hipMemcpyDefault)); REQUIRE(result == 42); } } template void ModuleLaunchKernelPositiveParameters() { const auto LaunchNOPKernel = [=](unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ) { hipFunction_t f = GetKernel(mg.module(), "NOPKernel"); HIP_CHECK(func(f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY, blockDimZ, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u)); HIP_CHECK(hipDeviceSynchronize()); }; SECTION("gridDimX == maxGridDimX") { const unsigned int x = GetDeviceAttribute(hipDeviceAttributeMaxGridDimX, 0); LaunchNOPKernel(x, 1, 1, 1, 1, 1); } SECTION("gridDimY == maxGridDimY") { const unsigned int y = GetDeviceAttribute(hipDeviceAttributeMaxGridDimY, 0); LaunchNOPKernel(1, y, 1, 1, 1, 1); } SECTION("gridDimZ == maxGridDimZ") { const unsigned int z = GetDeviceAttribute(hipDeviceAttributeMaxGridDimZ, 0); LaunchNOPKernel(1, 1, z, 1, 1, 1); } SECTION("blockDimX == maxBlockDimX") { const unsigned int x = GetDeviceAttribute(hipDeviceAttributeMaxBlockDimX, 0); LaunchNOPKernel(1, 1, 1, x, 1, 1); } SECTION("blockDimY == maxBlockDimY") { const unsigned int y = GetDeviceAttribute(hipDeviceAttributeMaxBlockDimY, 0); LaunchNOPKernel(1, 1, 1, 1, y, 1); } SECTION("blockDimZ == maxBlockDimZ") { const unsigned int z = GetDeviceAttribute(hipDeviceAttributeMaxBlockDimZ, 0); LaunchNOPKernel(1, 1, 1, 1, 1, z); } } template void ModuleLaunchKernelNegativeParameters() { hipFunction_t f = GetKernel(mg.module(), "NOPKernel"); // Disabled on AMD due to defect - EXSWHTEC-157 #if HT_NVIDIA SECTION("f == nullptr") { HIP_CHECK_ERROR( func(nullptr, 1, 1, 1, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidResourceHandle); } #endif SECTION("gridDimX == 0") { HIP_CHECK_ERROR(func(f, 0, 1, 1, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("gridDimY == 0") { HIP_CHECK_ERROR(func(f, 1, 0, 1, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("gridDimZ == 0") { HIP_CHECK_ERROR(func(f, 1, 1, 0, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("blockDimX == 0") { HIP_CHECK_ERROR(func(f, 1, 1, 1, 0, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("blockDimY == 0") { HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 0, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("blockDimZ == 0") { HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 1, 0, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } // Disabled on AMD due to defect - EXSWHTEC-158 #if HT_NVIDIA SECTION("gridDimX > maxGridDimX") { const unsigned int x = GetDeviceAttribute(hipDeviceAttributeMaxGridDimX, 0) + 1u; HIP_CHECK_ERROR(func(f, x, 1, 1, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("gridDimY > maxGridDimY") { const unsigned int y = GetDeviceAttribute(hipDeviceAttributeMaxGridDimY, 0) + 1u; HIP_CHECK_ERROR(func(f, 1, y, 1, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("gridDimZ > maxGridDimZ") { const unsigned int z = GetDeviceAttribute(hipDeviceAttributeMaxGridDimZ, 0) + 1u; HIP_CHECK_ERROR(func(f, 1, 1, z, 1, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } #endif // Disabled on AMD due to defect - EXSWHTEC-156 #if HT_NVIDIA SECTION("blockDimX > maxBlockDimX") { const unsigned int x = GetDeviceAttribute(hipDeviceAttributeMaxBlockDimX, 0) + 1u; HIP_CHECK_ERROR(func(f, 1, 1, 1, x, 1, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("blockDimY > maxBlockDimY") { const unsigned int y = GetDeviceAttribute(hipDeviceAttributeMaxBlockDimY, 0) + 1u; HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, y, 1, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } SECTION("blockDimZ > maxBlockDimZ") { const unsigned int z = GetDeviceAttribute(hipDeviceAttributeMaxBlockDimZ, 0) + 1u; HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 1, z, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } #endif // Disabled on AMD due to defect - EXSWHTEC-162 #if HT_NVIDIA SECTION("blockDimX * blockDimY * blockDimZ > MaxThreadsPerBlock") { const unsigned int max = GetDeviceAttribute(hipDeviceAttributeMaxThreadsPerBlock, 0); const unsigned int dim = std::ceil(std::cbrt(max)) + 1; HIP_CHECK_ERROR( func(f, 1, 1, 1, dim, dim, dim, 0, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } #endif // Disabled on AMD due to defect - EXSWHTEC-159 #if HT_NVIDIA SECTION("sharedMemBytes > max shared memory per block") { const unsigned int max = GetDeviceAttribute(hipDeviceAttributeMaxSharedMemoryPerBlock, 0) + 1u; HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 1, 1, max, nullptr, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorInvalidValue); } #endif // Disabled on AMD due to defect - EXSWHTEC-160 #if HT_NVIDIA SECTION("Invalid stream") { hipStream_t stream = nullptr; HIP_CHECK(hipStreamCreate(&stream)); HIP_CHECK(hipStreamDestroy(stream)); HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 1, 0, 0, stream, nullptr, nullptr, nullptr, nullptr, 0u), hipErrorContextIsDestroyed); } #endif SECTION("Passing kernel_args and extra simultaneously") { hipFunction_t f = GetKernel(mg.module(), "Kernel42"); LinearAllocGuard result_dev(LinearAllocs::hipMalloc, sizeof(int)); int* result_ptr = result_dev.ptr(); size_t size = sizeof(result_ptr); void* kernel_args[1] = {&result_ptr}; // clang-format off void *extra[] = { HIP_LAUNCH_PARAM_BUFFER_POINTER, &result_ptr, HIP_LAUNCH_PARAM_BUFFER_SIZE, &size, HIP_LAUNCH_PARAM_END }; // clang-format on HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 1, 1, 0, nullptr, kernel_args, extra, nullptr, nullptr, 0u), hipErrorInvalidValue); } // Disabled on AMD due to defect - EXSWHTEC-161 #if HT_NVIDIA SECTION("Invalid extra") { hipFunction_t f = GetKernel(mg.module(), "Kernel42"); void* extra[0] = {}; HIP_CHECK_ERROR(func(f, 1, 1, 1, 1, 1, 1, 0, nullptr, nullptr, extra, nullptr, nullptr, 0u), hipErrorInvalidValue); } #endif }