Media
/
compute-runtime
mirror of https://github.com/intel/compute-runtime.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

test: refactor aub tests to add fixture into separate file 

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz 2024-12-13 12:31:57 +00:00 committed by Compute-Runtime-Automation
parent 9ba671fde3
commit 98fd7c9432
6 changed files with 302 additions and 239 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
level_zero/core/test/aub_tests/debugger/debugger_aub_tests.cpp
View File

@ -9,6 +9,7 @@
#include "shared/source/gmm_helper/gmm_helper.h" #include "shared/source/gmm_helper/gmm_helper.h"
#include "shared/source/helpers/array_count.h" #include "shared/source/helpers/array_count.h"
#include "shared/source/helpers/bindless_heaps_helper.h" #include "shared/source/helpers/bindless_heaps_helper.h"
#include "shared/source/helpers/compiler_product_helper.h"
#include "shared/source/helpers/file_io.h" #include "shared/source/helpers/file_io.h"
#include "shared/source/helpers/register_offsets.h" #include "shared/source/helpers/register_offsets.h"
#include "shared/source/indirect_heap/indirect_heap.h" #include "shared/source/indirect_heap/indirect_heap.h"
@ -36,7 +37,9 @@ struct DebuggerAubFixture : AUBFixtureL0 {
AUBFixtureL0::setUp(NEO::defaultHwInfo.get(), true); AUBFixtureL0::setUp(NEO::defaultHwInfo.get(), true);
} }
void tearDown() { void tearDown() {
module->destroy(); if (module != nullptr) {
module->destroy();
}
AUBFixtureL0::tearDown(); AUBFixtureL0::tearDown();
} }
@ -63,6 +66,9 @@ using DebuggerSingleAddressSpaceAub = Test<DebuggerSingleAddressSpaceAubFixture>
using PlatformsSupportingSingleAddressSpace = MatchAny; using PlatformsSupportingSingleAddressSpace = MatchAny;
HWTEST2_F(DebuggerSingleAddressSpaceAub, GivenSingleAddressSpaceWhenCmdListIsExecutedThenSbaAddressesAreTracked, PlatformsSupportingSingleAddressSpace) { HWTEST2_F(DebuggerSingleAddressSpaceAub, GivenSingleAddressSpaceWhenCmdListIsExecutedThenSbaAddressesAreTracked, PlatformsSupportingSingleAddressSpace) {
if (neoDevice->getCompilerProductHelper().isHeaplessModeEnabled()) {
GTEST_SKIP();
}
constexpr size_t bufferSize = MemoryConstants::pageSize; constexpr size_t bufferSize = MemoryConstants::pageSize;
const uint32_t groupSize[] = {32, 1, 1}; const uint32_t groupSize[] = {32, 1, 1};
const uint32_t groupCount[] = {bufferSize / 32, 1, 1}; const uint32_t groupCount[] = {bufferSize / 32, 1, 1};

243
opencl/test/unit_test/aub_tests/command_stream/aub_walker_partition_tests_xehp_and_later.cpp
View File

@ -5,262 +5,29 @@
* *
*/ */
#include "shared/source/command_container/walker_partition_xehp_and_later.h"
#include "shared/source/helpers/array_count.h" #include "shared/source/helpers/array_count.h"
#include "shared/source/helpers/basic_math.h" #include "shared/source/helpers/basic_math.h"
#include "shared/source/helpers/timestamp_packet.h"
#include "shared/source/utilities/io_functions.h" #include "shared/source/utilities/io_functions.h"
#include "shared/source/utilities/tag_allocator.h" #include "shared/source/utilities/tag_allocator.h"
#include "shared/test/common/cmd_parse/hw_parse.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/helpers/dispatch_flags_helper.h"
#include "shared/test/common/mocks/mock_execution_environment.h" #include "shared/test/common/mocks/mock_execution_environment.h"
#include "shared/test/common/test_macros/hw_test.h"
#include "opencl/source/command_queue/command_queue.h" #include "opencl/source/command_queue/command_queue.h"
#include "opencl/source/event/event.h" #include "opencl/source/event/event.h"
#include "opencl/source/mem_obj/buffer.h" #include "opencl/test/unit_test/aub_tests/fixtures/aub_walker_partition_fixture.h"
#include "opencl/test/unit_test/aub_tests/command_stream/aub_command_stream_fixture.h"
#include "opencl/test/unit_test/aub_tests/fixtures/aub_fixture.h"
#include "opencl/test/unit_test/command_queue/command_queue_fixture.h"
#include "opencl/test/unit_test/fixtures/cl_device_fixture.h"
#include "opencl/test/unit_test/fixtures/simple_arg_kernel_fixture.h"
#include "opencl/test/unit_test/indirect_heap/indirect_heap_fixture.h"
using namespace NEO; using namespace NEO;
using namespace WalkerPartition; using namespace WalkerPartition;
static int32_t testPartitionCount[] = {1, 2, 4, 8, 16}; int32_t testPartitionCount[] = {1, 2, 4, 8, 16};
static int32_t testPartitionType[] = {1, 2, 3}; int32_t testPartitionType[] = {1, 2, 3};
static uint32_t testWorkingDimensions[] = {3}; uint32_t testWorkingDimensions[] = {3};
extern bool generateRandomInput; DispatchParameters dispatchParametersForTests[] = {
struct DispatchParameters {
size_t globalWorkSize[3];
size_t localWorkSize[3];
} dispatchParametersForTests[] = {
{{12, 25, 21}, {3, 5, 7}}, {{12, 25, 21}, {3, 5, 7}},
{{8, 16, 20}, {8, 4, 2}}, {{8, 16, 20}, {8, 4, 2}},
{{7, 13, 17}, {1, 1, 1}}, {{7, 13, 17}, {1, 1, 1}},
}; };
struct AubWalkerPartitionFixture : public KernelAUBFixture<SimpleKernelFixture> {
void setUp() {
debugRestorer = std::make_unique<DebugManagerStateRestore>();
debugManager.flags.EnableTimestampPacket.set(1);
kernelIds |= (1 << 5);
KernelAUBFixture<SimpleKernelFixture>::setUp();
size_t userMemorySize = 16 * MemoryConstants::kiloByte;
if (generateRandomInput) {
userMemorySize = 16000 * MemoryConstants::kiloByte;
}
sizeUserMemory = userMemorySize;
auto destMemory = alignedMalloc(sizeUserMemory, 4096);
ASSERT_NE(nullptr, destMemory);
memset(destMemory, 0x0, sizeUserMemory);
dstBuffer.reset(Buffer::create(context, CL_MEM_COPY_HOST_PTR, sizeUserMemory, destMemory, retVal));
ASSERT_NE(nullptr, dstBuffer);
alignedFree(destMemory);
kernels[5]->setArg(0, dstBuffer.get());
}
void tearDown() {
pCmdQ->flush();
KernelAUBFixture<SimpleKernelFixture>::tearDown();
}
template <typename FamilyType>
void validatePartitionProgramming(uint64_t postSyncAddress, int32_t partitionCount) {
using WalkerVariant = typename FamilyType::WalkerVariant;
uint32_t totalWorkgroupCount = 1u;
uint32_t totalWorkItemsInWorkgroup = 1u;
uint32_t totalWorkItemsCount = 1;
for (auto dimension = 0u; dimension < workingDimensions; dimension++) {
totalWorkgroupCount *= static_cast<uint32_t>(dispatchParamters.globalWorkSize[dimension] / dispatchParamters.localWorkSize[dimension]);
totalWorkItemsInWorkgroup *= static_cast<uint32_t>(dispatchParamters.localWorkSize[dimension]);
totalWorkItemsCount *= static_cast<uint32_t>(dispatchParamters.globalWorkSize[dimension]);
}
const uint32_t workgroupCount = static_cast<uint32_t>(dispatchParamters.globalWorkSize[partitionType - 1] / dispatchParamters.localWorkSize[partitionType - 1]);
auto partitionSize = Math::divideAndRoundUp(workgroupCount, partitionCount);
if (static_cast<uint32_t>(partitionType) > workingDimensions) {
partitionSize = 1;
}
hwParser.parseCommands<FamilyType>(pCmdQ->getCS(0), 0);
hwParser.findHardwareCommands<FamilyType>();
WalkerVariant walkerVariant = NEO::UnitTestHelper<FamilyType>::getWalkerVariant(*hwParser.itorWalker);
std::visit([&](auto &&walkerCmd) {
using WalkerType = std::decay_t<decltype(*walkerCmd)>;
using PostSyncType = decltype(FamilyType::template getPostSyncType<WalkerType>());
EXPECT_EQ(0u, walkerCmd->getPartitionId());
if (partitionCount > 1) {
EXPECT_TRUE(walkerCmd->getWorkloadPartitionEnable());
EXPECT_EQ(partitionSize, walkerCmd->getPartitionSize());
EXPECT_EQ(partitionType, walkerCmd->getPartitionType());
} else {
EXPECT_FALSE(walkerCmd->getWorkloadPartitionEnable());
EXPECT_EQ(0u, walkerCmd->getPartitionSize());
EXPECT_EQ(0u, walkerCmd->getPartitionType());
}
EXPECT_EQ(PostSyncType::OPERATION::OPERATION_WRITE_TIMESTAMP, walkerCmd->getPostSync().getOperation());
EXPECT_EQ(postSyncAddress, walkerCmd->getPostSync().getDestinationAddress());
int notExpectedValue[] = {1, 1, 1, 1};
for (auto partitionId = 0; partitionId < debugManager.flags.ExperimentalSetWalkerPartitionCount.get(); partitionId++) {
expectNotEqualMemory<FamilyType>(reinterpret_cast<void *>(postSyncAddress), &notExpectedValue, sizeof(notExpectedValue));
postSyncAddress += 16; // next post sync needs to be right after the previous one
}
},
walkerVariant);
auto dstGpuAddress = addrToPtr(ptrOffset(dstBuffer->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress(), dstBuffer->getOffset()));
expectMemory<FamilyType>(dstGpuAddress, &totalWorkItemsCount, sizeof(uint32_t));
auto groupSpecificWorkCounts = ptrOffset(dstGpuAddress, 4);
StackVec<uint32_t, 8> workgroupCounts;
workgroupCounts.resize(totalWorkgroupCount);
for (uint32_t workgroupId = 0u; workgroupId < totalWorkgroupCount; workgroupId++) {
workgroupCounts[workgroupId] = totalWorkItemsInWorkgroup;
}
expectMemory<FamilyType>(groupSpecificWorkCounts, workgroupCounts.begin(), workgroupCounts.size() * sizeof(uint32_t));
}
template <typename FamilyType>
typename FamilyType::PIPE_CONTROL *retrieveSyncPipeControl(void *startAddress,
const RootDeviceEnvironment &rootDeviceEnvironment) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
uint8_t buffer[256];
LinearStream stream(buffer, 256);
MemorySynchronizationCommands<FamilyType>::addBarrierWa(stream, 0ull, rootDeviceEnvironment);
void *syncPipeControlAddress = reinterpret_cast<void *>(reinterpret_cast<size_t>(startAddress) + stream.getUsed());
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(syncPipeControlAddress);
return pipeControl;
}
std::unique_ptr<DebugManagerStateRestore> debugRestorer;
std::unique_ptr<Buffer> dstBuffer;
size_t sizeUserMemory = 0;
cl_uint workingDimensions = 1;
int32_t partitionCount;
int32_t partitionType;
HardwareParse hwParser;
DispatchParameters dispatchParamters;
};
struct AubWalkerPartitionTest : public AubWalkerPartitionFixture,
public ::testing::TestWithParam<std::tuple<int32_t, int32_t, DispatchParameters, uint32_t>> {
void SetUp() override {
AubWalkerPartitionFixture::setUp();
std::tie(partitionCount, partitionType, dispatchParamters, workingDimensions) = GetParam();
if (generateRandomInput) {
workingDimensions = (rand() % 3 + 1);
partitionType = (rand() % 3 + 1);
partitionCount = rand() % 16 + 1;
// now generate dimensions that makes sense
auto goodWorkingSizeGenerated = false;
while (!goodWorkingSizeGenerated) {
dispatchParamters.localWorkSize[0] = rand() % 128 + 1;
dispatchParamters.localWorkSize[1] = rand() % 128 + 1;
dispatchParamters.localWorkSize[2] = rand() % 128 + 1;
auto totalWorkItemsInWorkgroup = 1;
for (auto dimension = 0u; dimension < workingDimensions; dimension++) {
totalWorkItemsInWorkgroup *= static_cast<uint32_t>(dispatchParamters.localWorkSize[dimension]);
}
if (totalWorkItemsInWorkgroup <= 1024) {
dispatchParamters.globalWorkSize[0] = dispatchParamters.localWorkSize[0] * (rand() % 32 + 1);
dispatchParamters.globalWorkSize[1] = dispatchParamters.localWorkSize[1] * (rand() % 32 + 1);
dispatchParamters.globalWorkSize[2] = dispatchParamters.localWorkSize[2] * (rand() % 32 + 1);
printf("\n generated following dispatch paramters work dim %u gws %zu %zu %zu lws %zu %zu %zu, partition type %d partitionCount %d",
workingDimensions,
dispatchParamters.globalWorkSize[0],
dispatchParamters.globalWorkSize[1],
dispatchParamters.globalWorkSize[2],
dispatchParamters.localWorkSize[0],
dispatchParamters.localWorkSize[1],
dispatchParamters.localWorkSize[2],
partitionType,
partitionCount);
IoFunctions::fflushPtr(stdout);
goodWorkingSizeGenerated = true;
}
};
}
debugManager.flags.ExperimentalSetWalkerPartitionCount.set(partitionCount);
debugManager.flags.ExperimentalSetWalkerPartitionType.set(partitionType);
debugManager.flags.EnableWalkerPartition.set(1u);
}
void TearDown() override {
AubWalkerPartitionFixture::tearDown();
}
};
struct AubWalkerPartitionZeroFixture : public AubWalkerPartitionFixture {
void setUp() {
AubWalkerPartitionFixture::setUp();
partitionCount = 0;
partitionType = 0;
workingDimensions = 1;
debugManager.flags.ExperimentalSetWalkerPartitionCount.set(0);
debugManager.flags.ExperimentalSetWalkerPartitionType.set(0);
commandBufferProperties = std::make_unique<AllocationProperties>(device->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::commandBuffer, false, device->getDeviceBitfield());
auto memoryManager = this->device->getMemoryManager();
streamAllocation = memoryManager->allocateGraphicsMemoryWithProperties(*commandBufferProperties);
helperSurface = memoryManager->allocateGraphicsMemoryWithProperties(*commandBufferProperties);
memset(helperSurface->getUnderlyingBuffer(), 0, MemoryConstants::pageSize);
taskStream = std::make_unique<LinearStream>(streamAllocation);
}
void tearDown() {
auto memoryManager = this->device->getMemoryManager();
memoryManager->freeGraphicsMemory(streamAllocation);
memoryManager->freeGraphicsMemory(helperSurface);
AubWalkerPartitionFixture::tearDown();
}
void flushStream() {
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.guardCommandBufferWithPipeControl = true;
csr->makeResident(*helperSurface);
csr->flushTask(*taskStream, 0,
&csr->getIndirectHeap(IndirectHeap::Type::dynamicState, 0u),
&csr->getIndirectHeap(IndirectHeap::Type::indirectObject, 0u),
&csr->getIndirectHeap(IndirectHeap::Type::surfaceState, 0u),
0u, dispatchFlags, device->getDevice());
csr->flushBatchedSubmissions();
}
std::unique_ptr<LinearStream> taskStream;
GraphicsAllocation *streamAllocation = nullptr;
GraphicsAllocation *helperSurface = nullptr;
std::unique_ptr<AllocationProperties> commandBufferProperties;
};
using AubWalkerPartitionZeroTest = Test<AubWalkerPartitionZeroFixture>; using AubWalkerPartitionZeroTest = Test<AubWalkerPartitionZeroFixture>;
HWCMDTEST_F(IGFX_XE_HP_CORE, AubWalkerPartitionZeroTest, whenPartitionCountSetToZeroThenProvideEqualSingleWalker) { HWCMDTEST_F(IGFX_XE_HP_CORE, AubWalkerPartitionZeroTest, whenPartitionCountSetToZeroThenProvideEqualSingleWalker) {

2
opencl/test/unit_test/aub_tests/fixtures/CMakeLists.txt
View File

@ -8,6 +8,8 @@ target_sources(igdrcl_aub_tests PRIVATE
${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt ${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
${CMAKE_CURRENT_SOURCE_DIR}/aub_fixture.cpp ${CMAKE_CURRENT_SOURCE_DIR}/aub_fixture.cpp
${CMAKE_CURRENT_SOURCE_DIR}/aub_fixture.h ${CMAKE_CURRENT_SOURCE_DIR}/aub_fixture.h
${CMAKE_CURRENT_SOURCE_DIR}/aub_walker_partition_fixture.cpp
${CMAKE_CURRENT_SOURCE_DIR}/aub_walker_partition_fixture.h
${CMAKE_CURRENT_SOURCE_DIR}/hello_world_fixture.h ${CMAKE_CURRENT_SOURCE_DIR}/hello_world_fixture.h
${CMAKE_CURRENT_SOURCE_DIR}/image_aub_fixture.h ${CMAKE_CURRENT_SOURCE_DIR}/image_aub_fixture.h
${CMAKE_CURRENT_SOURCE_DIR}/multicontext_ocl_aub_fixture.cpp ${CMAKE_CURRENT_SOURCE_DIR}/multicontext_ocl_aub_fixture.cpp

143
opencl/test/unit_test/aub_tests/fixtures/aub_walker_partition_fixture.cpp Normal file
View File

@ -0,0 +1,143 @@
/*
* Copyright (C) 2022-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "opencl/test/unit_test/aub_tests/fixtures/aub_walker_partition_fixture.h"
#include "shared/source/command_container/walker_partition_xehp_and_later.h"
#include "shared/source/helpers/array_count.h"
#include "shared/source/helpers/basic_math.h"
#include "shared/source/helpers/timestamp_packet.h"
#include "shared/source/utilities/io_functions.h"
#include "shared/source/utilities/tag_allocator.h"
#include "shared/test/common/cmd_parse/hw_parse.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/helpers/dispatch_flags_helper.h"
#include "shared/test/common/mocks/mock_execution_environment.h"
#include "shared/test/common/test_macros/hw_test.h"
#include "opencl/source/command_queue/command_queue.h"
#include "opencl/source/event/event.h"
using namespace NEO;
using namespace WalkerPartition;
void AubWalkerPartitionFixture::setUp() {
debugRestorer = std::make_unique<DebugManagerStateRestore>();
debugManager.flags.EnableTimestampPacket.set(1);
kernelIds |= (1 << 5);
KernelAUBFixture<SimpleKernelFixture>::setUp();
size_t userMemorySize = 16 * MemoryConstants::kiloByte;
if (generateRandomInput) {
userMemorySize = 16000 * MemoryConstants::kiloByte;
}
sizeUserMemory = userMemorySize;
auto destMemory = alignedMalloc(sizeUserMemory, 4096);
ASSERT_NE(nullptr, destMemory);
memset(destMemory, 0x0, sizeUserMemory);
dstBuffer.reset(Buffer::create(context, CL_MEM_COPY_HOST_PTR, sizeUserMemory, destMemory, retVal));
ASSERT_NE(nullptr, dstBuffer);
alignedFree(destMemory);
kernels[5]->setArg(0, dstBuffer.get());
}
void AubWalkerPartitionFixture::tearDown() {
pCmdQ->flush();
KernelAUBFixture<SimpleKernelFixture>::tearDown();
}
void AubWalkerPartitionTest::SetUp() {
AubWalkerPartitionFixture::setUp();
std::tie(partitionCount, partitionType, dispatchParamters, workingDimensions) = GetParam();
if (generateRandomInput) {
workingDimensions = (rand() % 3 + 1);
partitionType = (rand() % 3 + 1);
partitionCount = rand() % 16 + 1;
// now generate dimensions that makes sense
auto goodWorkingSizeGenerated = false;
while (!goodWorkingSizeGenerated) {
dispatchParamters.localWorkSize[0] = rand() % 128 + 1;
dispatchParamters.localWorkSize[1] = rand() % 128 + 1;
dispatchParamters.localWorkSize[2] = rand() % 128 + 1;
auto totalWorkItemsInWorkgroup = 1;
for (auto dimension = 0u; dimension < workingDimensions; dimension++) {
totalWorkItemsInWorkgroup *= static_cast<uint32_t>(dispatchParamters.localWorkSize[dimension]);
}
if (totalWorkItemsInWorkgroup <= 1024) {
dispatchParamters.globalWorkSize[0] = dispatchParamters.localWorkSize[0] * (rand() % 32 + 1);
dispatchParamters.globalWorkSize[1] = dispatchParamters.localWorkSize[1] * (rand() % 32 + 1);
dispatchParamters.globalWorkSize[2] = dispatchParamters.localWorkSize[2] * (rand() % 32 + 1);
printf("\n generated following dispatch paramters work dim %u gws %zu %zu %zu lws %zu %zu %zu, partition type %d partitionCount %d",
workingDimensions,
dispatchParamters.globalWorkSize[0],
dispatchParamters.globalWorkSize[1],
dispatchParamters.globalWorkSize[2],
dispatchParamters.localWorkSize[0],
dispatchParamters.localWorkSize[1],
dispatchParamters.localWorkSize[2],
partitionType,
partitionCount);
IoFunctions::fflushPtr(stdout);
goodWorkingSizeGenerated = true;
}
};
}
debugManager.flags.ExperimentalSetWalkerPartitionCount.set(partitionCount);
debugManager.flags.ExperimentalSetWalkerPartitionType.set(partitionType);
debugManager.flags.EnableWalkerPartition.set(1u);
}
void AubWalkerPartitionTest::TearDown() {
AubWalkerPartitionFixture::tearDown();
}
void AubWalkerPartitionZeroFixture::setUp() {
AubWalkerPartitionFixture::setUp();
partitionCount = 0;
partitionType = 0;
workingDimensions = 1;
debugManager.flags.ExperimentalSetWalkerPartitionCount.set(0);
debugManager.flags.ExperimentalSetWalkerPartitionType.set(0);
commandBufferProperties = std::make_unique<AllocationProperties>(device->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::commandBuffer, false, device->getDeviceBitfield());
auto memoryManager = this->device->getMemoryManager();
streamAllocation = memoryManager->allocateGraphicsMemoryWithProperties(*commandBufferProperties);
helperSurface = memoryManager->allocateGraphicsMemoryWithProperties(*commandBufferProperties);
memset(helperSurface->getUnderlyingBuffer(), 0, MemoryConstants::pageSize);
taskStream = std::make_unique<LinearStream>(streamAllocation);
}
void AubWalkerPartitionZeroFixture::tearDown() {
auto memoryManager = this->device->getMemoryManager();
memoryManager->freeGraphicsMemory(streamAllocation);
memoryManager->freeGraphicsMemory(helperSurface);
AubWalkerPartitionFixture::tearDown();
}
void AubWalkerPartitionZeroFixture::flushStream() {
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.guardCommandBufferWithPipeControl = true;
csr->makeResident(*helperSurface);
csr->flushTask(*taskStream, 0,
&csr->getIndirectHeap(IndirectHeap::Type::dynamicState, 0u),
&csr->getIndirectHeap(IndirectHeap::Type::indirectObject, 0u),
&csr->getIndirectHeap(IndirectHeap::Type::surfaceState, 0u),
0u, dispatchFlags, device->getDevice());
csr->flushBatchedSubmissions();
}

141
opencl/test/unit_test/aub_tests/fixtures/aub_walker_partition_fixture.h Normal file
View File

@ -0,0 +1,141 @@
/*
* Copyright (C) 2022-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_container/walker_partition_xehp_and_later.h"
#include "shared/source/helpers/timestamp_packet.h"
#include "shared/test/common/cmd_parse/hw_parse.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/helpers/dispatch_flags_helper.h"
#include "shared/test/common/test_macros/hw_test.h"
#include "opencl/source/mem_obj/buffer.h"
#include "opencl/test/unit_test/aub_tests/command_stream/aub_command_stream_fixture.h"
#include "opencl/test/unit_test/aub_tests/fixtures/aub_fixture.h"
#include "opencl/test/unit_test/fixtures/simple_arg_kernel_fixture.h"
extern bool generateRandomInput;
struct DispatchParameters {
size_t globalWorkSize[3];
size_t localWorkSize[3];
};
extern DispatchParameters dispatchParametersForTests[];
struct AubWalkerPartitionFixture : public KernelAUBFixture<SimpleKernelFixture> {
void setUp();
void tearDown();
template <typename FamilyType>
void validatePartitionProgramming(uint64_t postSyncAddress, int32_t partitionCount) {
using WalkerVariant = typename FamilyType::WalkerVariant;
uint32_t totalWorkgroupCount = 1u;
uint32_t totalWorkItemsInWorkgroup = 1u;
uint32_t totalWorkItemsCount = 1;
for (auto dimension = 0u; dimension < workingDimensions; dimension++) {
totalWorkgroupCount *= static_cast<uint32_t>(dispatchParamters.globalWorkSize[dimension] / dispatchParamters.localWorkSize[dimension]);
totalWorkItemsInWorkgroup *= static_cast<uint32_t>(dispatchParamters.localWorkSize[dimension]);
totalWorkItemsCount *= static_cast<uint32_t>(dispatchParamters.globalWorkSize[dimension]);
}
const uint32_t workgroupCount = static_cast<uint32_t>(dispatchParamters.globalWorkSize[partitionType - 1] / dispatchParamters.localWorkSize[partitionType - 1]);
auto partitionSize = Math::divideAndRoundUp(workgroupCount, partitionCount);
if (static_cast<uint32_t>(partitionType) > workingDimensions) {
partitionSize = 1;
}
hwParser.parseCommands<FamilyType>(pCmdQ->getCS(0), 0);
hwParser.findHardwareCommands<FamilyType>();
WalkerVariant walkerVariant = NEO::UnitTestHelper<FamilyType>::getWalkerVariant(*hwParser.itorWalker);
std::visit([&](auto &&walkerCmd) {
using WalkerType = std::decay_t<decltype(*walkerCmd)>;
using PostSyncType = decltype(FamilyType::template getPostSyncType<WalkerType>());
EXPECT_EQ(0u, walkerCmd->getPartitionId());
if (partitionCount > 1) {
EXPECT_TRUE(walkerCmd->getWorkloadPartitionEnable());
EXPECT_EQ(partitionSize, walkerCmd->getPartitionSize());
EXPECT_EQ(partitionType, walkerCmd->getPartitionType());
} else {
EXPECT_FALSE(walkerCmd->getWorkloadPartitionEnable());
EXPECT_EQ(0u, walkerCmd->getPartitionSize());
EXPECT_EQ(0u, walkerCmd->getPartitionType());
}
EXPECT_EQ(PostSyncType::OPERATION::OPERATION_WRITE_TIMESTAMP, walkerCmd->getPostSync().getOperation());
EXPECT_EQ(postSyncAddress, walkerCmd->getPostSync().getDestinationAddress());
int notExpectedValue[] = {1, 1, 1, 1};
for (auto partitionId = 0; partitionId < debugManager.flags.ExperimentalSetWalkerPartitionCount.get(); partitionId++) {
expectNotEqualMemory<FamilyType>(reinterpret_cast<void *>(postSyncAddress), &notExpectedValue, sizeof(notExpectedValue));
postSyncAddress += 16; // next post sync needs to be right after the previous one
}
},
walkerVariant);
auto dstGpuAddress = addrToPtr(ptrOffset(dstBuffer->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress(), dstBuffer->getOffset()));
expectMemory<FamilyType>(dstGpuAddress, &totalWorkItemsCount, sizeof(uint32_t));
auto groupSpecificWorkCounts = ptrOffset(dstGpuAddress, 4);
StackVec<uint32_t, 8> workgroupCounts;
workgroupCounts.resize(totalWorkgroupCount);
for (uint32_t workgroupId = 0u; workgroupId < totalWorkgroupCount; workgroupId++) {
workgroupCounts[workgroupId] = totalWorkItemsInWorkgroup;
}
expectMemory<FamilyType>(groupSpecificWorkCounts, workgroupCounts.begin(), workgroupCounts.size() * sizeof(uint32_t));
}
template <typename FamilyType>
typename FamilyType::PIPE_CONTROL *retrieveSyncPipeControl(void *startAddress,
const RootDeviceEnvironment &rootDeviceEnvironment) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
uint8_t buffer[256];
LinearStream stream(buffer, 256);
MemorySynchronizationCommands<FamilyType>::addBarrierWa(stream, 0ull, rootDeviceEnvironment);
void *syncPipeControlAddress = reinterpret_cast<void *>(reinterpret_cast<size_t>(startAddress) + stream.getUsed());
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(syncPipeControlAddress);
return pipeControl;
}
std::unique_ptr<DebugManagerStateRestore> debugRestorer;
std::unique_ptr<Buffer> dstBuffer;
size_t sizeUserMemory = 0;
cl_uint workingDimensions = 1;
int32_t partitionCount;
int32_t partitionType;
HardwareParse hwParser;
DispatchParameters dispatchParamters;
};
struct AubWalkerPartitionTest : public AubWalkerPartitionFixture,
public ::testing::TestWithParam<std::tuple<int32_t, int32_t, DispatchParameters, uint32_t>> {
void SetUp();
void TearDown();
};
struct AubWalkerPartitionZeroFixture : public AubWalkerPartitionFixture {
void setUp();
void tearDown();
void flushStream();
std::unique_ptr<LinearStream> taskStream;
GraphicsAllocation *streamAllocation = nullptr;
GraphicsAllocation *helperSurface = nullptr;
std::unique_ptr<AllocationProperties> commandBufferProperties;
};

4
shared/test/common/fixtures/aub_fixtures/multicontext_aub_fixture.cpp
View File

@ -27,6 +27,10 @@ void MulticontextAubFixture::setUp(uint32_t numberOfTiles, EnabledCommandStreame
HardwareInfo localHwInfo = *defaultHwInfo; HardwareInfo localHwInfo = *defaultHwInfo;
if (debugManager.flags.BlitterEnableMaskOverride.get() > 0) {
localHwInfo.featureTable.ftrBcsInfo = debugManager.flags.BlitterEnableMaskOverride.get();
}
if (numberOfEnabledTiles > 1 && localHwInfo.gtSystemInfo.MultiTileArchInfo.IsValid == 0) { if (numberOfEnabledTiles > 1 && localHwInfo.gtSystemInfo.MultiTileArchInfo.IsValid == 0) {
skipped = true; skipped = true;
GTEST_SKIP(); GTEST_SKIP();