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

fix: program partition offset for secondary csr

Browse Source 
Related-To: HSD-18039998563

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2024-09-18 16:22:31 +00:00
committed by Compute-Runtime-Automation
parent 08cc1a6286
commit 7a56424c2c
1 changed files with 8 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
level_zero/core/source/cmdqueue/cmdqueue_hw.inl
View File

@@ -116,7 +116,6 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandLists(
if (this->isCopyOnlyCommandQueue) { if (this->isCopyOnlyCommandQueue) {
ret = this->executeCommandListsCopyOnly(ctx, numCommandLists, phCommandLists, hFence, parentImmediateCommandlistLinearStream); ret = this->executeCommandListsCopyOnly(ctx, numCommandLists, phCommandLists, hFence, parentImmediateCommandlistLinearStream);
} else if (this->heaplessStateInitEnabled) { } else if (this->heaplessStateInitEnabled) {
ctx.globalInit = false;
ret = this->executeCommandListsRegularHeapless(ctx, numCommandLists, phCommandLists, hFence, parentImmediateCommandlistLinearStream); ret = this->executeCommandListsRegularHeapless(ctx, numCommandLists, phCommandLists, hFence, parentImmediateCommandlistLinearStream);
} else { } else {
ret = this->executeCommandListsRegular(ctx, numCommandLists, phCommandLists, hFence, parentImmediateCommandlistLinearStream); ret = this->executeCommandListsRegular(ctx, numCommandLists, phCommandLists, hFence, parentImmediateCommandlistLinearStream);
@@ -176,6 +175,8 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandListsRegularHeapless(
this->makeCsrTagAllocationResident(); this->makeCsrTagAllocationResident();
this->programActivePartitionConfig(ctx.isProgramActivePartitionConfigRequired, *streamForDispatch);
if (instructionCacheFlushRequired) { if (instructionCacheFlushRequired) {
NEO::MemorySynchronizationCommands<GfxFamily>::addInstructionCacheFlush(*streamForDispatch); NEO::MemorySynchronizationCommands<GfxFamily>::addInstructionCacheFlush(*streamForDispatch);
this->csr->setInstructionCacheFlushed(); this->csr->setInstructionCacheFlushed();
@@ -253,6 +254,7 @@ size_t CommandQueueHw<gfxCoreFamily>::estimateStreamSizeForExecuteCommandListsRe
for (uint32_t i = 0; i < numCommandLists; i++) { for (uint32_t i = 0; i < numCommandLists; i++) {
auto cmdList = CommandList::fromHandle(commandListHandles[i]); auto cmdList = CommandList::fromHandle(commandListHandles[i]);
linearStreamSizeEstimate += estimateCommandListSecondaryStart(cmdList); linearStreamSizeEstimate += estimateCommandListSecondaryStart(cmdList);
linearStreamSizeEstimate += this->estimateCommandListPrimaryStart(ctx.globalInit);
ctx.spaceForResidency += estimateCommandListResidencySize(cmdList); ctx.spaceForResidency += estimateCommandListResidencySize(cmdList);
} }
@@ -268,6 +270,11 @@ size_t CommandQueueHw<gfxCoreFamily>::estimateStreamSizeForExecuteCommandListsRe
linearStreamSizeEstimate += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForFullCacheFlush(); linearStreamSizeEstimate += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForFullCacheFlush();
} }
auto csrHw = reinterpret_cast<NEO::CommandStreamReceiverHw<GfxFamily> *>(this->csr);
if (ctx.isProgramActivePartitionConfigRequired) {
linearStreamSizeEstimate += csrHw->getCmdSizeForActivePartitionConfig();
}
return linearStreamSizeEstimate; return linearStreamSizeEstimate;
} }