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

fix: Align thread group count to dss on all platforms 

Related-To: NEO-13263, GSD-10327
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka 2024-11-18 15:57:22 +00:00 committed by Compute-Runtime-Automation
parent 9b4bb0c9ea
commit 62d8e3e4b0
2 changed files with 26 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
shared/source/helpers/gfx_core_helper_base.inl
View File

@ -802,6 +802,8 @@ uint32_t GfxCoreHelperHw<GfxFamily>::calculateAvailableThreadCount(const Hardwar
template <typename GfxFamily>
void GfxCoreHelperHw<GfxFamily>::alignThreadGroupCountToDssSize(uint32_t &threadCount, uint32_t dssCount, uint32_t threadsPerDss, uint32_t threadGroupSize) const {
uint32_t availableTreadCount = (threadsPerDss / threadGroupSize) * dssCount;
threadCount = std::min(threadCount, availableTreadCount);
}
template <typename GfxFamily>

26
shared/test/unit_test/helpers/gfx_core_helper_tests.cpp
View File

@ -1874,10 +1874,32 @@ HWTEST_F(GfxCoreHelperTest, givenGetDeviceTimestampWidthCalledThenReturnCorrectV
EXPECT_EQ(64u, helper.getDeviceTimestampWidth());
}
HWTEST2_F(GfxCoreHelperTest, givenHwHelperWhenAligningThreadGroupCountToDssSizeThenThreadGroupCountDoesNotChange, IsAtMostXe2HpgCore) {
HWTEST_F(GfxCoreHelperTest, givenHwHelperWhenAligningThreadGroupCountToDssSizeThenThreadGroupCountChanged) {
auto &helper = getHelper<GfxCoreHelper>();
uint32_t threadGroupCountBefore = 4096;
uint32_t threadCount = threadGroupCountBefore;
helper.alignThreadGroupCountToDssSize(threadCount, 1, 1, 1);
EXPECT_EQ(threadGroupCountBefore, threadCount);
EXPECT_NE(threadGroupCountBefore, threadCount);
}
HWTEST_F(GfxCoreHelperTest, givenHwHelperWhenThreadGroupCountIsAlignedToDssThenThreadCountNotChanged) {
auto &helper = getHelper<GfxCoreHelper>();
uint32_t dssCount = 16;
uint32_t threadGroupSize = 32;
uint32_t threadsPerDss = 2 * threadGroupSize;
uint32_t maxThreadCount = (dssCount * threadsPerDss) / threadGroupSize;
uint32_t threadCount = maxThreadCount;
helper.alignThreadGroupCountToDssSize(threadCount, dssCount, threadsPerDss, threadGroupSize);
EXPECT_EQ(2 * dssCount, threadCount);
}
HWTEST_F(GfxCoreHelperTest, givenHwHelperWhenThreadGroupCountIsAlignedToDssThenThreadCountChanged) {
auto &helper = getHelper<GfxCoreHelper>();
uint32_t dssCount = 16;
uint32_t threadGroupSize = 32;
uint32_t threadsPerDss = 2 * threadGroupSize - 1;
uint32_t maxThreadCount = (dssCount * threadsPerDss) / threadGroupSize;
uint32_t threadCount = maxThreadCount;
helper.alignThreadGroupCountToDssSize(threadCount, dssCount, threadsPerDss, threadGroupSize);
EXPECT_EQ(dssCount, threadCount);
}