intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: correct suggested number of work groups for concurrent kernels on PVC

Browse Source 
value depends on CCS count:
- single CCS mode (default) - no limitations
- two CCS mode - 25% available
- four CCS mode - 12.5% available

Related-To: NEO-8377
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2023-09-14 13:53:58 +00:00
committed by Compute-Runtime-Automation
parent 513045bd90
commit 6fc673b0fe
8 changed files with 31 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
shared/test/unit_test/xe_hpc_core/pvc/gfx_core_helper_tests_pvc.cpp
View File

@@ -172,35 +172,27 @@ PVCTEST_F(GfxCoreHelperTestsPvc, GivenCooperativeEngineSupportedAndNotUsedWhenAd
auto tilePartsForConcurrentKernels = PVC::numberOfpartsInTileForConcurrentKernels;
auto passedMaxWorkGroupCount = 1024;
uint32_t revisions[] = {REVISION_A0, REVISION_B};
uint32_t revisions[] = {REVISION_A0, REVISION_B, REVISION_C};
for (auto &revision : revisions) {
auto hwRevId = productHelper.getHwRevIdFromStepping(revision, hwInfo);
if (hwRevId == CommonConstants::invalidStepping) {
continue;
}
hwInfo.platform.usRevId = hwRevId;
for (auto isEngineInstanced : ::testing::Bool()) {
for (auto engineGroupType : {EngineGroupType::RenderCompute, EngineGroupType::Compute, EngineGroupType::CooperativeCompute}) {
for (auto isRcsEnabled : ::testing::Bool()) {
hwInfo.featureTable.flags.ftrRcsNode = isRcsEnabled;
for (auto engineGroupType : {EngineGroupType::RenderCompute, EngineGroupType::Compute, EngineGroupType::CooperativeCompute}) {
if (productHelper.isCooperativeEngineSupported(hwInfo)) {
bool disallowDispatch = (engineGroupType == EngineGroupType::RenderCompute) ||
((engineGroupType == EngineGroupType::Compute) && isRcsEnabled);
bool applyLimitation = !isEngineInstanced &&
(engineGroupType != EngineGroupType::CooperativeCompute);
if (disallowDispatch) {
EXPECT_EQ(1u, gfxCoreHelper.adjustMaxWorkGroupCount(passedMaxWorkGroupCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced));
} else if (applyLimitation) {
hwInfo.gtSystemInfo.CCSInfo.NumberOfCCSEnabled = 4;
EXPECT_EQ(passedMaxWorkGroupCount / tilePartsForConcurrentKernels, gfxCoreHelper.adjustMaxWorkGroupCount(passedMaxWorkGroupCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced));
hwInfo.gtSystemInfo.CCSInfo.NumberOfCCSEnabled = 16;
EXPECT_EQ(passedMaxWorkGroupCount / hwInfo.gtSystemInfo.CCSInfo.NumberOfCCSEnabled, gfxCoreHelper.adjustMaxWorkGroupCount(passedMaxWorkGroupCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced));
} else {
bool disallowDispatch = (engineGroupType == EngineGroupType::RenderCompute ||
(engineGroupType == EngineGroupType::Compute && isRcsEnabled)) &&
productHelper.isCooperativeEngineSupported(hwInfo);
for (auto isEngineInstanced : ::testing::Bool()) {
if (disallowDispatch) {
EXPECT_EQ(1u, gfxCoreHelper.adjustMaxWorkGroupCount(passedMaxWorkGroupCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced));
} else {
for (uint32_t ccsCount : {1, 2, 4}) {
hwInfo.gtSystemInfo.CCSInfo.NumberOfCCSEnabled = ccsCount;
tilePartsForConcurrentKernels = ccsCount == 1 ? 1
: ccsCount == 2 ? 4
: 8;
EXPECT_EQ(passedMaxWorkGroupCount / tilePartsForConcurrentKernels, gfxCoreHelper.adjustMaxWorkGroupCount(passedMaxWorkGroupCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced));
}
} else {
EXPECT_EQ(passedMaxWorkGroupCount / tilePartsForConcurrentKernels, gfxCoreHelper.adjustMaxWorkGroupCount(passedMaxWorkGroupCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced));
}
}
}