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: respect number of tiles when getting total private memory size of module

Browse Source 
Related-To: GSD-8374
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2024-04-23 14:59:24 +00:00
committed by Compute-Runtime-Automation
parent 8bef74133f
commit 959333d9bf
4 changed files with 30 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
level_zero/core/source/module/module_imp.cpp
View File

@@ -1286,19 +1286,22 @@ void ModuleImp::verifyDebugCapabilities() {
void ModuleImp::checkIfPrivateMemoryPerDispatchIsNeeded() {
size_t modulePrivateMemorySize = 0;
auto neoDevice = this->device->getNEODevice();
for (auto &kernelImmData : this->kernelImmDatas) {
if (0 == kernelImmData->getDescriptor().kernelAttributes.perHwThreadPrivateMemorySize) {
continue;
}
auto kernelPrivateMemorySize = NEO::KernelHelper::getPrivateSurfaceSize(kernelImmData->getDescriptor().kernelAttributes.perHwThreadPrivateMemorySize,
this->device->getNEODevice()->getDeviceInfo().computeUnitsUsedForScratch);
neoDevice->getDeviceInfo().computeUnitsUsedForScratch);
modulePrivateMemorySize += kernelPrivateMemorySize;
}
this->allocatePrivateMemoryPerDispatch = false;
if (modulePrivateMemorySize > 0U) {
auto globalMemorySize = device->getNEODevice()->getRootDevice()->getGlobalMemorySize(static_cast<uint32_t>(device->getNEODevice()->getDeviceBitfield().to_ulong()));
this->allocatePrivateMemoryPerDispatch = modulePrivateMemorySize > globalMemorySize;
auto deviceBitfield = neoDevice->getDeviceBitfield();
auto globalMemorySize = neoDevice->getRootDevice()->getGlobalMemorySize(static_cast<uint32_t>(deviceBitfield.to_ulong()));
auto numSubDevices = deviceBitfield.count();
this->allocatePrivateMemoryPerDispatch = modulePrivateMemorySize * numSubDevices > globalMemorySize;
}
}