intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-10 15:12:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: unify further calculation to get max work group count

Browse Source 
- move available device calculcation into common helper
- change interface to have code available where no descriptor is available
- expand unit test for implementation of new inteface

Related-To: NEO-13350

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-11-29 15:42:31 +00:00
committed by Compute-Runtime-Automation
parent 99a353a15a
commit 56b15f17f7
5 changed files with 60 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
shared/test/unit_test/helpers/kernel_helpers_tests.cpp
View File

@@ -24,40 +24,50 @@
using namespace NEO;
struct KernelHelperMaxWorkGroupsTests : ::testing::Test {
struct KernelHelperMaxWorkGroupsFixture : public DeviceFixture {
size_t lws[3] = {10, 10, 10};
EngineGroupType engineType = EngineGroupType::compute;
uint32_t simd = 8;
uint32_t dssCount = 16;
uint32_t availableSlm = 64 * MemoryConstants::kiloByte;
uint32_t usedSlm = 0;
uint32_t numberOfBarriers = 0;
uint32_t workDim = 3;
uint32_t grf = 128;
uint32_t numSubdevices = 1;
size_t lws[3] = {10, 10, 10};
void SetUp() override {
executionEnvironment = std::make_unique<MockExecutionEnvironment>(defaultHwInfo.get(), false, 1u);
rootDeviceEnvironment = executionEnvironment->rootDeviceEnvironments[0].get();
uint16_t grf = 128;
uint8_t simd = 8;
uint8_t numberOfBarriers = 0;
bool implicitScalingEnabled = false;
bool forceSingleTileQuery = true;
void setUp() {
DeviceFixture::setUp();
rootDeviceEnvironment = &pDevice->getRootDeviceEnvironmentRef();
}
uint32_t getMaxWorkGroupCount() {
KernelDescriptor descriptor = {};
descriptor.kernelAttributes.simdSize = simd;
descriptor.kernelAttributes.barrierCount = numberOfBarriers;
descriptor.kernelAttributes.numGrfRequired = grf;
auto hwInfo = rootDeviceEnvironment->getMutableHardwareInfo();
hwInfo->gtSystemInfo.DualSubSliceCount = dssCount;
hwInfo->capabilityTable.slmSize = (availableSlm / MemoryConstants::kiloByte) / dssCount;
return KernelHelper::getMaxWorkGroupCount(*rootDeviceEnvironment, descriptor, numSubdevices, usedSlm, workDim, lws, engineType);
if (numSubdevices > 1) {
forceSingleTileQuery = false;
implicitScalingEnabled = true;
for (uint32_t pos = 0; pos < numSubdevices; pos++) {
pDevice->deviceBitfield.set(pos);
}
}
return KernelHelper::getMaxWorkGroupCount(*pDevice, grf, simd, numberOfBarriers, usedSlm, workDim, lws, engineType, implicitScalingEnabled, forceSingleTileQuery);
}
std::unique_ptr<MockExecutionEnvironment> executionEnvironment;
RootDeviceEnvironment *rootDeviceEnvironment = nullptr;
};
using KernelHelperMaxWorkGroupsTests = Test<KernelHelperMaxWorkGroupsFixture>;
TEST_F(KernelHelperMaxWorkGroupsTests, GivenNoBarriersOrSlmUsedWhenCalculatingMaxWorkGroupsCountThenResultIsCalculatedWithSimd) {
auto &helper = rootDeviceEnvironment->getHelper<NEO::GfxCoreHelper>();
@@ -72,6 +82,8 @@ TEST_F(KernelHelperMaxWorkGroupsTests, GivenDebugFlagSetWhenGetMaxWorkGroupCount
DebugManagerStateRestore restore;
debugManager.flags.OverrideMaxWorkGroupCount.set(123);
forceSingleTileQuery = false;
EXPECT_EQ(123u, getMaxWorkGroupCount());
}
@@ -160,7 +172,7 @@ TEST_F(KernelHelperMaxWorkGroupsTests, GivenVariousValuesWhenCalculatingMaxWorkG
hwInfo->gtSystemInfo.ThreadCount = 1024;
EXPECT_NE(1u, getMaxWorkGroupCount());
numberOfBarriers = static_cast<uint32_t>(helper.getMaxBarrierRegisterPerSlice());
numberOfBarriers = static_cast<uint8_t>(helper.getMaxBarrierRegisterPerSlice());
EXPECT_EQ(1u, getMaxWorkGroupCount());
numberOfBarriers = 1;