intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 00:58:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: allow scratch to be optional

Browse Source 
Related-To: NEO-14130

Signed-off-by: Naklicki, Mateusz <mateusz.naklicki@intel.com>
This commit is contained in:
Naklicki, Mateusz
2025-06-23 18:31:54 +00:00
committed by Compute-Runtime-Automation
parent 408556dcd8
commit bd80531dea
15 changed files with 108 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
level_zero/core/source/cmdqueue/cmdqueue_hw.h
View File

@@ -54,7 +54,7 @@ struct CommandQueueHw : public CommandQueueImp {
uint32_t perThreadScratchSpaceSlot1Size);
bool getPreemptionCmdProgramming() override;
void patchCommands(CommandList &commandList, uint64_t scratchAddress, bool patchNewScratchAddress);
void patchCommands(CommandList &commandList, uint64_t scratchAddress, bool patchNewScratchController);
protected:
struct CommandListExecutionContext {

12
level_zero/core/source/cmdqueue/cmdqueue_hw.inl
View File

@@ -1805,7 +1805,7 @@ size_t CommandQueueHw<gfxCoreFamily>::estimateStateBaseAddressDebugTracking() {
template <GFXCORE_FAMILY gfxCoreFamily>
void CommandQueueHw<gfxCoreFamily>::patchCommands(CommandList &commandList, CommandListExecutionContext &ctx) {
bool patchNewScratchAddress = false;
bool patchNewScratchController = false;
uint64_t scratchAddress = ctx.scratchSpaceController->getScratchPatchAddress();
if (this->heaplessModeEnabled) {
@@ -1813,16 +1813,14 @@ void CommandQueueHw<gfxCoreFamily>::patchCommands(CommandList &commandList, Comm
scratchAddress += ctx.globalStatelessAllocation->getGpuAddress();
}
if (commandList.getCurrentScratchPatchAddress() != scratchAddress ||
commandList.getCommandListUsedScratchController() != ctx.scratchSpaceController) {
patchNewScratchAddress = true;
if (commandList.getCommandListUsedScratchController() != ctx.scratchSpaceController) {
patchNewScratchController = true;
}
}
patchCommands(commandList, scratchAddress, patchNewScratchAddress);
patchCommands(commandList, scratchAddress, patchNewScratchController);
if (patchNewScratchAddress) {
commandList.setCurrentScratchPatchAddress(scratchAddress);
if (patchNewScratchController) {
commandList.setCommandListUsedScratchController(ctx.scratchSpaceController);
}
}

2
level_zero/core/source/cmdqueue/cmdqueue_hw_skl_to_tgllp.inl
View File

@@ -132,7 +132,7 @@ void CommandQueueHw<gfxCoreFamily>::handleScratchSpace(NEO::HeapContainer &heapC
template <GFXCORE_FAMILY gfxCoreFamily>
void CommandQueueHw<gfxCoreFamily>::patchCommands(CommandList &commandList, uint64_t scratchAddress,
bool patchNewScratchAddress) {
bool patchNewScratchController) {
using MI_SEMAPHORE_WAIT = typename GfxFamily::MI_SEMAPHORE_WAIT;
using COMPARE_OPERATION = typename GfxFamily::MI_SEMAPHORE_WAIT::COMPARE_OPERATION;

12
level_zero/core/source/cmdqueue/cmdqueue_xe_hp_core_and_later.inl
View File

@@ -164,7 +164,7 @@ void CommandQueueHw<gfxCoreFamily>::handleScratchSpace(NEO::HeapContainer &sshHe
template <GFXCORE_FAMILY gfxCoreFamily>
void CommandQueueHw<gfxCoreFamily>::patchCommands(CommandList &commandList, uint64_t scratchAddress,
bool patchNewScratchAddress) {
bool patchNewScratchController) {
using MI_SEMAPHORE_WAIT = typename GfxFamily::MI_SEMAPHORE_WAIT;
using COMPARE_OPERATION = typename GfxFamily::MI_SEMAPHORE_WAIT::COMPARE_OPERATION;
@@ -235,21 +235,27 @@ void CommandQueueHw<gfxCoreFamily>::patchCommands(CommandList &commandList, uint
break;
}
case CommandToPatch::ComputeWalkerInlineDataScratch: {
if (!patchNewScratchAddress) {
if (NEO::isUndefined(commandToPatch.patchSize) || NEO::isUndefinedOffset(commandToPatch.offset)) {
continue;
}
if (!patchNewScratchController && commandToPatch.scratchAddressAfterPatch == scratchAddress) {
continue;
}
uint64_t fullScratchAddress = scratchAddress + commandToPatch.baseAddress;
void *scratchAddressPatch = ptrOffset(commandToPatch.pDestination, commandToPatch.offset);
std::memcpy(scratchAddressPatch, &fullScratchAddress, commandToPatch.patchSize);
commandToPatch.scratchAddressAfterPatch = scratchAddress;
break;
}
case CommandToPatch::ComputeWalkerImplicitArgsScratch: {
if (!patchNewScratchAddress) {
if (!patchNewScratchController && commandToPatch.scratchAddressAfterPatch == scratchAddress) {
continue;
}
uint64_t fullScratchAddress = scratchAddress + commandToPatch.baseAddress;
void *scratchAddressPatch = ptrOffset(commandToPatch.pDestination, commandToPatch.offset);
std::memcpy(scratchAddressPatch, &fullScratchAddress, commandToPatch.patchSize);
commandToPatch.scratchAddressAfterPatch = scratchAddress;
break;
}
case CommandToPatch::NoopSpace: {