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

Adjust minWorkGroupSize when EUFusion is enabled

Browse Source 
Related-To: NEO-5260

Co-authored-by: Bartlomiej Wolny <bartlomiej.wolny@intel.com>
Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>
This commit is contained in:
Bartlomiej Wolny
2021-03-23 17:54:16 +00:00
committed by Compute-Runtime-Automation
parent 3c2a7ee49a
commit b98148fb3f
6 changed files with 131 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
opencl/source/program/kernel_info.cpp
View File

@@ -117,18 +117,21 @@ std::map<std::string, size_t> typeSizeMap = {
{"double8", sizeof(cl_double8)},
{"double16", sizeof(cl_double16)},
};
WorkSizeInfo::WorkSizeInfo(uint32_t maxWorkGroupSize, bool hasBarriers, uint32_t simdSize, uint32_t slmTotalSize, GFXCORE_FAMILY coreFamily, uint32_t numThreadsPerSubSlice, uint32_t localMemSize, bool imgUsed, bool yTiledSurface) {
WorkSizeInfo::WorkSizeInfo(uint32_t maxWorkGroupSize, bool hasBarriers, uint32_t simdSize, uint32_t slmTotalSize, const HardwareInfo *hwInfo, uint32_t numThreadsPerSubSlice, uint32_t localMemSize, bool imgUsed, bool yTiledSurface) {
this->maxWorkGroupSize = maxWorkGroupSize;
this->hasBarriers = hasBarriers;
this->simdSize = simdSize;
this->slmTotalSize = slmTotalSize;
this->coreFamily = coreFamily;
this->coreFamily = hwInfo->platform.eRenderCoreFamily;
this->numThreadsPerSubSlice = numThreadsPerSubSlice;
this->localMemSize = localMemSize;
this->imgUsed = imgUsed;
this->yTiledSurfaces = yTiledSurface;
setMinWorkGroupSize();
setMinWorkGroupSize(hwInfo);
}
WorkSizeInfo::WorkSizeInfo(const DispatchInfo &dispatchInfo) {
auto &device = dispatchInfo.getClDevice();
const auto &kernelInfo = dispatchInfo.getKernel()->getKernelInfo();
@@ -140,8 +143,9 @@ WorkSizeInfo::WorkSizeInfo(const DispatchInfo &dispatchInfo) {
this->numThreadsPerSubSlice = static_cast<uint32_t>(device.getSharedDeviceInfo().maxNumEUsPerSubSlice) *
device.getSharedDeviceInfo().numThreadsPerEU;
this->localMemSize = static_cast<uint32_t>(device.getSharedDeviceInfo().localMemSize);
setIfUseImg(kernelInfo);
setMinWorkGroupSize();
setMinWorkGroupSize(&device.getHardwareInfo());
}
void WorkSizeInfo::setIfUseImg(const KernelInfo &kernelInfo) {
for (const auto &arg : kernelInfo.kernelDescriptor.payloadMappings.explicitArgs) {
@@ -152,7 +156,7 @@ void WorkSizeInfo::setIfUseImg(const KernelInfo &kernelInfo) {
}
}
}
void WorkSizeInfo::setMinWorkGroupSize() {
void WorkSizeInfo::setMinWorkGroupSize(const HardwareInfo *hwInfo) {
minWorkGroupSize = 0;
if (hasBarriers) {
uint32_t maxBarriersPerHSlice = (coreFamily >= IGFX_GEN9_CORE) ? 32 : 16;
@@ -161,6 +165,11 @@ void WorkSizeInfo::setMinWorkGroupSize() {
if (slmTotalSize > 0) {
minWorkGroupSize = std::max(maxWorkGroupSize / ((localMemSize / slmTotalSize)), minWorkGroupSize);
}
const auto &hwHelper = HwHelper::get(hwInfo->platform.eRenderCoreFamily);
if (hwHelper.isFusedEuDispatchEnabled(*hwInfo)) {
minWorkGroupSize *= 2;
}
}
void WorkSizeInfo::checkRatio(const size_t workItems[3]) {
if (slmTotalSize > 0) {

4
opencl/source/program/kernel_info.h
View File

@@ -55,10 +55,10 @@ struct WorkSizeInfo {
bool useStrictRatio = false;
float targetRatio = 0;
WorkSizeInfo(uint32_t maxWorkGroupSize, bool hasBarriers, uint32_t simdSize, uint32_t slmTotalSize, GFXCORE_FAMILY coreFamily, uint32_t numThreadsPerSubSlice, uint32_t localMemSize, bool imgUsed, bool yTiledSurface);
WorkSizeInfo(uint32_t maxWorkGroupSize, bool hasBarriers, uint32_t simdSize, uint32_t slmTotalSize, const HardwareInfo *hwInfo, uint32_t numThreadsPerSubSlice, uint32_t localMemSize, bool imgUsed, bool yTiledSurface);
WorkSizeInfo(const DispatchInfo &dispatchInfo);
void setIfUseImg(const KernelInfo &kernelInfo);
void setMinWorkGroupSize();
void setMinWorkGroupSize(const HardwareInfo *hwInfo);
void checkRatio(const size_t workItems[3]);
};