intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-04 23:56:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

Disable implicit scaling for cooperative kernels

Browse Source 
When implicit scaling is disabled use useSingleSubdeviceValue = true.

Resolves: NEO-5757

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
This commit is contained in:
Filip Hazubski
2021-06-21 15:24:14 +00:00
committed by Compute-Runtime-Automation
parent b5d5784b81
commit 29c64c3dd0
27 changed files with 256 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
shared/source/command_container/command_encoder_bdw_and_later.inl
View File

@@ -26,7 +26,7 @@ template <typename Family>
void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
const void *pThreadGroupDimensions, bool isIndirect, bool isPredicate, DispatchKernelEncoderI *dispatchInterface,
uint64_t eventAddress, bool isTimestampEvent, bool L3FlushEnable, Device *device, PreemptionMode preemptionMode,
bool &requiresUncachedMocs, bool useGlobalAtomics, uint32_t &partitionCount, bool isInternal) {
bool &requiresUncachedMocs, bool useGlobalAtomics, uint32_t &partitionCount, bool isInternal, bool isCooperative) {
using MEDIA_STATE_FLUSH = typename Family::MEDIA_STATE_FLUSH;
using MEDIA_INTERFACE_DESCRIPTOR_LOAD = typename Family::MEDIA_INTERFACE_DESCRIPTOR_LOAD;
@@ -49,7 +49,8 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
if (!isIndirect) {
threadDimsVec = {threadDims[0], threadDims[1], threadDims[2]};
}
size_t estimatedSizeRequired = estimateEncodeDispatchKernelCmdsSize(device, threadStartVec, threadDimsVec, isInternal);
size_t estimatedSizeRequired = estimateEncodeDispatchKernelCmdsSize(device, threadStartVec, threadDimsVec,
isInternal, isCooperative);
if (container.getCommandStream()->getAvailableSpace() < estimatedSizeRequired) {
auto bbEnd = listCmdBufferStream->getSpaceForCmd<MI_BATCH_BUFFER_END>();
*bbEnd = Family::cmdInitBatchBufferEnd;
@@ -319,7 +320,9 @@ template <typename Family>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const HardwareInfo &hwInfo, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {}
template <typename Family>
size_t EncodeDispatchKernel<Family>::estimateEncodeDispatchKernelCmdsSize(Device *device, Vec3<size_t> groupStart, Vec3<size_t> groupCount, bool isInternal) {
size_t EncodeDispatchKernel<Family>::estimateEncodeDispatchKernelCmdsSize(Device *device, Vec3<size_t> groupStart,
Vec3<size_t> groupCount, bool isInternal,
bool isCooperative) {
using MEDIA_STATE_FLUSH = typename Family::MEDIA_STATE_FLUSH;
using MEDIA_INTERFACE_DESCRIPTOR_LOAD = typename Family::MEDIA_INTERFACE_DESCRIPTOR_LOAD;
using MI_BATCH_BUFFER_END = typename Family::MI_BATCH_BUFFER_END;