intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-04 15:53:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improve calculateAvailableThreadCount implementation

Browse Source 
Signed-off-by: Rafal Maziejuk <rafal.maziejuk@intel.com>
This commit is contained in:
Rafal Maziejuk
2022-07-26 11:10:43 +00:00
committed by Compute-Runtime-Automation
parent fe0c857f1a
commit af91f94098
10 changed files with 77 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
shared/source/helpers/hw_helper.h
View File

@@ -89,8 +89,7 @@ class HwHelper {
virtual uint32_t getMocsIndex(const GmmHelper &gmmHelper, bool l3enabled, bool l1enabled) const = 0;
virtual bool isLinearStoragePreferred(bool isSharedContext, bool isImage1d, bool forceLinearStorage) = 0;
virtual uint8_t getBarriersCountFromHasBarriers(uint8_t hasBarriers) const = 0;
virtual uint32_t calculateAvailableThreadCount(PRODUCT_FAMILY family, uint32_t grfCount, uint32_t euCount,
uint32_t threadsPerEu) = 0;
virtual uint32_t calculateAvailableThreadCount(const HardwareInfo &hwInfo, uint32_t grfCount) = 0;
virtual uint32_t alignSlmSize(uint32_t slmSize) = 0;
virtual uint32_t computeSlmValues(const HardwareInfo &hwInfo, uint32_t slmSize) = 0;
@@ -292,7 +291,7 @@ class HwHelperHw : public HwHelper {
uint8_t getBarriersCountFromHasBarriers(uint8_t hasBarriers) const override;
uint32_t calculateAvailableThreadCount(PRODUCT_FAMILY family, uint32_t grfCount, uint32_t euCount, uint32_t threadsPerEu) override;
uint32_t calculateAvailableThreadCount(const HardwareInfo &hwInfo, uint32_t grfCount) override;
uint32_t alignSlmSize(uint32_t slmSize) override;

5
shared/source/helpers/hw_helper_bdw_and_later.inl
View File

@@ -93,9 +93,8 @@ uint32_t HwHelperHw<GfxFamily>::getMocsIndex(const GmmHelper &gmmHelper, bool l3
}
template <typename GfxFamily>
uint32_t HwHelperHw<GfxFamily>::calculateAvailableThreadCount(PRODUCT_FAMILY family, uint32_t grfCount, uint32_t euCount,
uint32_t threadsPerEu) {
return threadsPerEu * euCount;
uint32_t HwHelperHw<GfxFamily>::calculateAvailableThreadCount(const HardwareInfo &hwInfo, uint32_t grfCount) {
return hwInfo.gtSystemInfo.ThreadCount;
}
template <typename GfxFamily>

6
shared/source/helpers/hw_helper_pvc_and_later.inl
View File

@@ -72,4 +72,10 @@ size_t HwHelperHw<Family>::getPaddingForISAAllocation() const {
return 0xE00;
}
template <>
uint32_t HwHelperHw<Family>::calculateAvailableThreadCount(const HardwareInfo &hwInfo, uint32_t grfCount) {
auto maxThreadsPerEuCount = 1024u / grfCount;
return maxThreadsPerEuCount * hwInfo.gtSystemInfo.EUCount;
}
} // namespace NEO

7
shared/source/helpers/hw_helper_xehp_and_later.inl
View File

@@ -129,12 +129,11 @@ uint32_t HwHelperHw<GfxFamily>::getMocsIndex(const GmmHelper &gmmHelper, bool l3
}
template <typename GfxFamily>
uint32_t HwHelperHw<GfxFamily>::calculateAvailableThreadCount(PRODUCT_FAMILY family, uint32_t grfCount, uint32_t euCount,
uint32_t threadsPerEu) {
uint32_t HwHelperHw<GfxFamily>::calculateAvailableThreadCount(const HardwareInfo &hwInfo, uint32_t grfCount) {
if (grfCount > GrfConfig::DefaultGrfNumber) {
return threadsPerEu / 2u * euCount;
return hwInfo.gtSystemInfo.ThreadCount / 2u;
}
return threadsPerEu * euCount;
return hwInfo.gtSystemInfo.ThreadCount;
}
template <typename GfxFamily>