intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 21:18:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: align thread group to dss size if kernel uses slm

Browse Source 
Related-To: NEO-12133
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka
2024-10-17 10:58:39 +00:00
committed by Compute-Runtime-Automation
parent b17fabb120
commit 9d6d6e85f1
9 changed files with 59 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
shared/source/helpers/kernel_helpers.cpp
View File

@@ -54,16 +54,16 @@ uint32_t KernelHelper::getMaxWorkGroupCount(const RootDeviceEnvironment &rootDev
UNRECOVERABLE_IF(workGroupSize == 0);
auto numThreadsPerThreadGroup = static_cast<uint32_t>(Math::divideAndRoundUp(workGroupSize, simdSize));
auto maxWorkGroupsCount = availableThreadCount / numThreadsPerThreadGroup;
if (barrierCount > 0) {
auto maxWorkGroupsCountDueToBarrierUsage = dssCount * (maxBarrierCount / barrierCount);
if (barrierCount > 0 || usedSlmSize > 0) {
helper.alignThreadGroupCountToDssSize(maxWorkGroupsCount, dssCount, availableThreadCount / dssCount, numThreadsPerThreadGroup);
maxWorkGroupsCount = std::min(maxWorkGroupsCount, maxWorkGroupsCountDueToBarrierUsage);
}
if (usedSlmSize > 0) {
auto maxWorkGroupsCountDueToSlm = availableSlmSize / usedSlmSize;
maxWorkGroupsCount = std::min(maxWorkGroupsCount, maxWorkGroupsCountDueToSlm);
if (barrierCount > 0) {
auto maxWorkGroupsCountDueToBarrierUsage = dssCount * (maxBarrierCount / barrierCount);
maxWorkGroupsCount = std::min(maxWorkGroupsCount, maxWorkGroupsCountDueToBarrierUsage);
}
if (usedSlmSize > 0) {
auto maxWorkGroupsCountDueToSlm = availableSlmSize / usedSlmSize;
maxWorkGroupsCount = std::min(maxWorkGroupsCount, maxWorkGroupsCountDueToSlm);
}
}
maxWorkGroupsCount = helper.adjustMaxWorkGroupCount(maxWorkGroupsCount, engineGroupType, rootDeviceEnvironment);