intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-21 01:04:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

set correct private scratch size for immediate command lists

Browse Source 
Related-To: NEO-7187

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2022-08-05 17:35:16 +00:00
committed by Compute-Runtime-Automation
parent 50046c473b
commit 505e0b7059
3 changed files with 92 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
View File

@@ -93,7 +93,7 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommandListImm
this->handleIndirectAllocationResidency();
this->csr->setRequiredScratchSizes(this->getCommandListPerThreadScratchSize(), this->getCommandListPerThreadScratchSize());
this->csr->setRequiredScratchSizes(this->getCommandListPerThreadScratchSize(), this->getCommandListPerThreadPrivateScratchSize());
if (performMigration) {
auto deviceImp = static_cast<DeviceImp *>(this->device);

1
level_zero/core/test/unit_tests/mocks/mock_cmdlist.h
View File

@@ -119,6 +119,7 @@ struct WhiteBox<L0::CommandListCoreFamilyImmediate<gfxCoreFamily>>
using BaseClass::commandsToPatch;
using BaseClass::csr;
using BaseClass::finalStreamState;
using BaseClass::isFlushTaskSubmissionEnabled;
using BaseClass::partitionCount;
using BaseClass::requiredStreamState;

90
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_7.cpp
View File

@@ -5,6 +5,7 @@
*
*/
#include "shared/test/common/libult/ult_command_stream_receiver.h"
#include "shared/test/common/mocks/ult_device_factory.h"
#include "shared/test/common/test_macros/hw_test.h"
@@ -463,5 +464,94 @@ HWTEST2_F(CommandListCreate, givenCommandListAndHostPointersWhenMemoryCopyCalled
}
}
using CmdlistAppendLaunchKernelTests = Test<ModuleImmutableDataFixture>;
using IsBetweenGen9AndGen12lp = IsWithinGfxCore<IGFX_GEN9_CORE, IGFX_GEN12LP_CORE>;
HWTEST2_F(CmdlistAppendLaunchKernelTests,
givenImmediateCommandListUsesFlushTaskWhenDispatchingKernelWithSpillScratchSpaceThenExpectCsrHasCorrectValuesSet, IsBetweenGen9AndGen12lp) {
constexpr uint32_t scratchPerThreadSize = 0x200;
std::unique_ptr<MockImmutableData> mockKernelImmData = std::make_unique<MockImmutableData>(0u);
auto kernelDescriptor = mockKernelImmData->kernelDescriptor;
kernelDescriptor->kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor->kernelAttributes.perThreadScratchSize[0] = scratchPerThreadSize;
createModuleFromBinary(0u, false, mockKernelImmData.get());
auto kernel = std::make_unique<MockKernel>(module.get());
ze_kernel_desc_t kernelDesc{ZE_STRUCTURE_TYPE_KERNEL_DESC};
kernel->initialize(&kernelDesc);
kernel->setGroupSize(4, 5, 6);
kernel->setGroupCount(3, 2, 1);
kernel->setGlobalOffsetExp(1, 2, 3);
kernel->patchGlobalOffset();
ze_result_t result;
auto commandList = std::make_unique<WhiteBox<L0::CommandListCoreFamilyImmediate<gfxCoreFamily>>>();
ASSERT_NE(nullptr, commandList);
commandList->isFlushTaskSubmissionEnabled = true;
ze_result_t ret = commandList->initialize(device, NEO::EngineGroupType::RenderCompute, 0u);
ASSERT_EQ(ZE_RESULT_SUCCESS, ret);
commandList->device = device;
commandList->cmdListType = CommandList::CommandListType::TYPE_IMMEDIATE;
commandList->csr = device->getNEODevice()->getDefaultEngine().commandStreamReceiver;
ze_group_count_t groupCount = {3, 2, 1};
CmdListKernelLaunchParams launchParams = {};
result = commandList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_EQ(scratchPerThreadSize, commandList->getCommandListPerThreadScratchSize());
auto ultCsr = reinterpret_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(device->getNEODevice()->getDefaultEngine().commandStreamReceiver);
EXPECT_EQ(scratchPerThreadSize, ultCsr->requiredScratchSize);
}
HWTEST2_F(CmdlistAppendLaunchKernelTests,
givenImmediateCommandListUsesFlushTaskWhenDispatchingKernelWithSpillAndPrivateScratchSpaceThenExpectCsrHasCorrectValuesSet, IsAtLeastXeHpCore) {
constexpr uint32_t scratchPerThreadSize = 0x200;
constexpr uint32_t privateScratchPerThreadSize = 0x100;
std::unique_ptr<MockImmutableData> mockKernelImmData = std::make_unique<MockImmutableData>(0u);
auto kernelDescriptor = mockKernelImmData->kernelDescriptor;
kernelDescriptor->kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor->kernelAttributes.perThreadScratchSize[0] = scratchPerThreadSize;
kernelDescriptor->kernelAttributes.perThreadScratchSize[1] = privateScratchPerThreadSize;
createModuleFromBinary(0u, false, mockKernelImmData.get());
auto kernel = std::make_unique<MockKernel>(module.get());
ze_kernel_desc_t kernelDesc{ZE_STRUCTURE_TYPE_KERNEL_DESC};
kernel->initialize(&kernelDesc);
kernel->setGroupSize(4, 5, 6);
kernel->setGroupCount(3, 2, 1);
kernel->setGlobalOffsetExp(1, 2, 3);
kernel->patchGlobalOffset();
ze_result_t result;
auto commandList = std::make_unique<WhiteBox<L0::CommandListCoreFamilyImmediate<gfxCoreFamily>>>();
ASSERT_NE(nullptr, commandList);
commandList->isFlushTaskSubmissionEnabled = true;
ze_result_t ret = commandList->initialize(device, NEO::EngineGroupType::RenderCompute, 0u);
ASSERT_EQ(ZE_RESULT_SUCCESS, ret);
commandList->device = device;
commandList->cmdListType = CommandList::CommandListType::TYPE_IMMEDIATE;
commandList->csr = device->getNEODevice()->getDefaultEngine().commandStreamReceiver;
ze_group_count_t groupCount = {3, 2, 1};
CmdListKernelLaunchParams launchParams = {};
result = commandList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_EQ(scratchPerThreadSize, commandList->getCommandListPerThreadScratchSize());
EXPECT_EQ(privateScratchPerThreadSize, commandList->getCommandListPerThreadPrivateScratchSize());
auto ultCsr = reinterpret_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(device->getNEODevice()->getDefaultEngine().commandStreamReceiver);
EXPECT_EQ(scratchPerThreadSize, ultCsr->requiredScratchSize);
EXPECT_EQ(privateScratchPerThreadSize, ultCsr->requiredPrivateScratchSize);
}
} // namespace ult
} // namespace L0