/* * Copyright (C) 2017-2020 Intel Corporation * * SPDX-License-Identifier: MIT * */ #include "shared/test/unit_test/mocks/mock_device.h" #include "opencl/test/unit_test/fixtures/execution_model_kernel_fixture.h" #include "opencl/test/unit_test/mocks/mock_kernel.h" #include "opencl/test/unit_test/mocks/mock_program.h" #include "test.h" #include using namespace NEO; typedef ExecutionModelKernelFixture ParentKernelFromBinaryTest; class MockKernelWithArgumentAccess : public Kernel { public: std::vector &getKernelArguments() { return kernelArguments; } class ObjectCountsPublic : public Kernel::ObjectCounts { }; MockKernelWithArgumentAccess(Program *programArg, const KernelInfo &kernelInfoArg, const ClDevice &deviceArg) : Kernel(programArg, kernelInfoArg, deviceArg) { } void getParentObjectCountsPublic(MockKernelWithArgumentAccess::ObjectCountsPublic &objectCount) { getParentObjectCounts(objectCount); } }; TEST(ParentKernelTest, WhenArgsAddedThenObjectCountsAreIncremented) { KernelInfo info; MockClDevice *device = new MockClDevice{new MockDevice}; MockProgram program(*device->getExecutionEnvironment()); SPatchExecutionEnvironment environment = {}; environment.HasDeviceEnqueue = 1; info.patchInfo.executionEnvironment = &environment; MockKernelWithArgumentAccess kernel(&program, info, *device); std::vector &args = kernel.getKernelArguments(); Kernel::SimpleKernelArgInfo argInfo; argInfo.type = Kernel::kernelArgType::SAMPLER_OBJ; args.push_back(argInfo); argInfo.type = Kernel::kernelArgType::IMAGE_OBJ; args.push_back(argInfo); MockKernelWithArgumentAccess::ObjectCountsPublic objectCounts; kernel.getParentObjectCountsPublic(objectCounts); EXPECT_EQ(1u, objectCounts.imageCount); EXPECT_EQ(1u, objectCounts.samplerCount); delete device; } TEST(ParentKernelTest, WhenPatchingBlocksSimdSizeThenPatchIsAppliedCorrectly) { MockClDevice device{new MockDevice}; MockContext context(&device); std::unique_ptr parentKernel(MockParentKernel::create(context, true)); MockProgram *program = (MockProgram *)parentKernel->mockProgram; parentKernel->patchBlocksSimdSize(); void *blockSimdSize = ptrOffset(parentKernel->getCrossThreadData(), parentKernel->getKernelInfo().childrenKernelsIdOffset[0].second); uint32_t *simdSize = reinterpret_cast(blockSimdSize); EXPECT_EQ(program->blockKernelManager->getBlockKernelInfo(0)->getMaxSimdSize(), *simdSize); } TEST(ParentKernelTest, GivenParentKernelWhenCheckingForDeviceEnqueueThenTrueIsReturned) { MockClDevice device{new MockDevice}; MockContext context(&device); std::unique_ptr parentKernel(MockParentKernel::create(context)); EXPECT_TRUE(parentKernel->getKernelInfo().hasDeviceEnqueue()); } TEST(ParentKernelTest, GivenNormalKernelWhenCheckingForDeviceEnqueueThenFalseIsReturned) { MockClDevice device{new MockDevice}; MockKernelWithInternals kernel(device); EXPECT_FALSE(kernel.kernelInfo.hasDeviceEnqueue()); } TEST(ParentKernelTest, WhenInitializingParentKernelThenBlocksSimdSizeIsPatched) { MockClDevice device{new MockDevice}; MockContext context(&device); std::unique_ptr parentKernel(MockParentKernel::create(context, true)); MockProgram *program = (MockProgram *)parentKernel->mockProgram; parentKernel->initialize(); void *blockSimdSize = ptrOffset(parentKernel->getCrossThreadData(), parentKernel->getKernelInfo().childrenKernelsIdOffset[0].second); uint32_t *simdSize = reinterpret_cast(blockSimdSize); EXPECT_EQ(program->blockKernelManager->getBlockKernelInfo(0)->getMaxSimdSize(), *simdSize); } TEST(ParentKernelTest, WhenInitializingParentKernelThenPrivateMemoryForBlocksIsAllocated) { MockClDevice device{new MockDevice}; MockContext context(&device); std::unique_ptr parentKernel(MockParentKernel::create(context, true)); MockProgram *program = (MockProgram *)parentKernel->mockProgram; uint32_t crossThreadOffsetBlock = 0; auto infoBlock = new KernelInfo(); SPatchAllocateStatelessDefaultDeviceQueueSurface *allocateDeviceQueueBlock = new SPatchAllocateStatelessDefaultDeviceQueueSurface; allocateDeviceQueueBlock->DataParamOffset = crossThreadOffsetBlock; allocateDeviceQueueBlock->DataParamSize = 8; allocateDeviceQueueBlock->SurfaceStateHeapOffset = 0; allocateDeviceQueueBlock->Size = 8; infoBlock->patchInfo.pAllocateStatelessDefaultDeviceQueueSurface = allocateDeviceQueueBlock; crossThreadOffsetBlock += 8; SPatchAllocateStatelessEventPoolSurface *eventPoolBlock = new SPatchAllocateStatelessEventPoolSurface; eventPoolBlock->DataParamOffset = crossThreadOffsetBlock; eventPoolBlock->DataParamSize = 8; eventPoolBlock->EventPoolSurfaceIndex = 0; eventPoolBlock->Size = 8; infoBlock->patchInfo.pAllocateStatelessEventPoolSurface = eventPoolBlock; crossThreadOffsetBlock += 8; auto privateSurfaceBlock = std::make_unique(); privateSurfaceBlock->DataParamOffset = crossThreadOffsetBlock; privateSurfaceBlock->DataParamSize = 8; privateSurfaceBlock->Size = 8; privateSurfaceBlock->SurfaceStateHeapOffset = 0; privateSurfaceBlock->Token = 0; privateSurfaceBlock->PerThreadPrivateMemorySize = 1000; infoBlock->patchInfo.pAllocateStatelessPrivateSurface = privateSurfaceBlock.get(); crossThreadOffsetBlock += 8; SPatchThreadPayload *threadPayloadBlock = new SPatchThreadPayload; threadPayloadBlock->LocalIDXPresent = 0; threadPayloadBlock->LocalIDYPresent = 0; threadPayloadBlock->LocalIDZPresent = 0; threadPayloadBlock->HeaderPresent = 0; threadPayloadBlock->Size = 128; infoBlock->patchInfo.threadPayload = threadPayloadBlock; SPatchExecutionEnvironment *executionEnvironmentBlock = new SPatchExecutionEnvironment; *executionEnvironmentBlock = {}; executionEnvironmentBlock->HasDeviceEnqueue = 1; infoBlock->patchInfo.executionEnvironment = executionEnvironmentBlock; SPatchDataParameterStream *streamBlock = new SPatchDataParameterStream; streamBlock->DataParameterStreamSize = 0; streamBlock->Size = 0; infoBlock->patchInfo.dataParameterStream = streamBlock; SPatchBindingTableState *bindingTable = new SPatchBindingTableState; bindingTable->Count = 0; bindingTable->Offset = 0; bindingTable->Size = 0; bindingTable->SurfaceStateOffset = 0; infoBlock->patchInfo.bindingTableState = bindingTable; SPatchInterfaceDescriptorData *idData = new SPatchInterfaceDescriptorData; idData->BindingTableOffset = 0; idData->KernelOffset = 0; idData->Offset = 0; idData->SamplerStateOffset = 0; idData->Size = 0; infoBlock->patchInfo.interfaceDescriptorData = idData; infoBlock->patchInfo.pAllocateStatelessGlobalMemorySurfaceWithInitialization = nullptr; infoBlock->patchInfo.pAllocateStatelessConstantMemorySurfaceWithInitialization = nullptr; infoBlock->heapInfo.pDsh = (void *)new uint64_t[64]; infoBlock->crossThreadData = new char[crossThreadOffsetBlock]; program->blockKernelManager->addBlockKernelInfo(infoBlock); parentKernel->initialize(); EXPECT_NE(nullptr, program->getBlockKernelManager()->getPrivateSurface(program->getBlockKernelManager()->getCount() - 1)); } TEST_P(ParentKernelFromBinaryTest, GivenParentKernelWhenGettingInstructionHeapSizeForExecutionModelThenSizeIsGreaterThanZero) { if (std::string(pPlatform->getClDevice(0)->getDeviceInfo().clVersion).find("OpenCL 2.") != std::string::npos) { EXPECT_TRUE(pKernel->isParentKernel); EXPECT_LT(0u, pKernel->getInstructionHeapSizeForExecutionModel()); } } static const char *binaryFile = "simple_block_kernel"; static const char *KernelNames[] = {"simple_block_kernel"}; INSTANTIATE_TEST_CASE_P(ParentKernelFromBinaryTest, ParentKernelFromBinaryTest, ::testing::Combine( ::testing::Values(binaryFile), ::testing::ValuesIn(KernelNames)));