intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 23:03:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: copy offload signaling depends on useAdditionalBlitProperties

Browse Source 
Related-To: NEO-7067

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2025-05-23 11:26:42 +00:00
committed by Compute-Runtime-Automation
parent 86e7d5b276
commit 46a5c266b7
3 changed files with 18 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -1572,15 +1572,13 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopyBlitRegion(Ali
return ZE_RESULT_ERROR_INVALID_ARGUMENT;
}
const bool copyOnly = isCopyOnly(dualStreamCopyOffload);
auto &rootDeviceEnvironment = device->getNEODevice()->getRootDeviceEnvironmentRef();
bool copyRegionPreferred = NEO::BlitCommandsHelper<GfxFamily>::isCopyRegionPreferred(copySizeModified, rootDeviceEnvironment, blitProperties.isSystemMemoryPoolUsed);
size_t nBlits = copyRegionPreferred ? NEO::BlitCommandsHelper<GfxFamily>::getNumberOfBlitsForCopyRegion(blitProperties.copySize, rootDeviceEnvironment, blitProperties.isSystemMemoryPoolUsed) : NEO::BlitCommandsHelper<GfxFamily>::getNumberOfBlitsForCopyPerRow(blitProperties.copySize, rootDeviceEnvironment, blitProperties.isSystemMemoryPoolUsed);
bool useAdditionalTimestamp = nBlits > 1;
if (useAdditionalBlitProperties) {
setAdditionalBlitProperties(blitProperties, signalEvent, useAdditionalTimestamp);
} else if (copyOnly) {
} else {
appendEventForProfiling(signalEvent, nullptr, true, false, false, true);
}
@@ -1596,7 +1594,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopyBlitRegion(Ali
}
dummyBlitWa.isWaRequired = true;
if (!useAdditionalBlitProperties && copyOnly) {
if (!useAdditionalBlitProperties) {
appendSignalEventPostWalker(signalEvent, nullptr, nullptr, false, false, true);
}
return ZE_RESULT_SUCCESS;
@@ -1613,7 +1611,6 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendCopyImageBlit(NEO::Graph
Event *signalEvent, uint32_t numWaitEvents,
ze_event_handle_t *phWaitEvents, CmdListMemoryCopyParams &memoryCopyParams) {
const bool dualStreamCopyOffloadOperation = isDualStreamCopyOffloadOperation(memoryCopyParams.copyOffloadAllowed);
const bool isCopyOnlySignaling = isCopyOnly(dualStreamCopyOffloadOperation) && !useAdditionalBlitProperties;
auto ret = addEventsToCmdList(numWaitEvents, phWaitEvents, nullptr, memoryCopyParams.relaxedOrderingDispatch, false, true, false, dualStreamCopyOffloadOperation);
if (ret != ZE_RESULT_SUCCESS) {
@@ -1642,9 +1639,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendCopyImageBlit(NEO::Graph
bool useAdditionalTimestamp = blitProperties.copySize.z > 1;
if (useAdditionalBlitProperties) {
setAdditionalBlitProperties(blitProperties, signalEvent, useAdditionalTimestamp);
}
if (isCopyOnlySignaling) {
} else {
appendEventForProfiling(signalEvent, nullptr, true, false, false, true);
}
blitProperties.transform1DArrayTo2DArrayIfNeeded();
@@ -1656,7 +1651,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendCopyImageBlit(NEO::Graph
}
dummyBlitWa.isWaRequired = true;
if (isCopyOnlySignaling) {
if (!useAdditionalBlitProperties) {
appendSignalEventPostWalker(signalEvent, nullptr, nullptr, false, false, true);
if (this->isInOrderExecutionEnabled()) {
appendSignalInOrderDependencyCounter(signalEvent, false, false, false);
@@ -1839,10 +1834,8 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopy(void *dstptr,
launchParams.pipeControlSignalling = (signalEvent && singlePipeControlPacket) || getDcFlushRequired(dstAllocationStruct.needsFlush);
if (!isNonDualStreamCopyOffloadOperation(memoryCopyParams.copyOffloadAllowed)) {
if (!useAdditionalBlitProperties || !isCopyOnlyEnabled) {
appendEventForProfilingAllWalkers(signalEvent, nullptr, nullptr, true, singlePipeControlPacket, false, isCopyOnlyEnabled);
}
if (!useAdditionalBlitProperties || !isCopyOnlyEnabled) {
appendEventForProfilingAllWalkers(signalEvent, nullptr, nullptr, true, singlePipeControlPacket, false, isCopyOnlyEnabled);
}
if (isCopyOnlyEnabled) {
@@ -1915,10 +1908,8 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopy(void *dstptr,
appendCopyOperationFence(signalEvent, srcAllocationStruct.alloc, dstAllocationStruct.alloc, isCopyOnlyEnabled);
if (!isNonDualStreamCopyOffloadOperation(memoryCopyParams.copyOffloadAllowed)) {
if (!useAdditionalBlitProperties || !isCopyOnlyEnabled) {
appendEventForProfilingAllWalkers(signalEvent, nullptr, nullptr, false, singlePipeControlPacket, false, isCopyOnlyEnabled);
}
if (!useAdditionalBlitProperties || !isCopyOnlyEnabled) {
appendEventForProfilingAllWalkers(signalEvent, nullptr, nullptr, false, singlePipeControlPacket, false, isCopyOnlyEnabled);
}
bool l3flushInPipeControl = !l3FlushAfterPostSyncRequired || isSplitOperation;
@@ -1926,7 +1917,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopy(void *dstptr,
addToMappedEventList(signalEvent);
if (this->isInOrderExecutionEnabled() && !isNonDualStreamCopyOffloadOperation(memoryCopyParams.copyOffloadAllowed)) {
if (this->isInOrderExecutionEnabled()) {
bool emitPipeControl = !isCopyOnlyEnabled && launchParams.pipeControlSignalling;
if ((!useAdditionalBlitProperties || !isCopyOnlyEnabled) &&
@@ -2005,8 +1996,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopyRegion(void *d
memoryCopyParams.copyOffloadAllowed = isCopyOffloadAllowed(*srcAllocationStruct.alloc, *dstAllocationStruct.alloc);
const bool isCopyOnlyEnabled = isCopyOnly(memoryCopyParams.copyOffloadAllowed);
const bool inOrderCopyOnlySignalingAllowed = this->isInOrderExecutionEnabled() && !memoryCopyParams.forceDisableCopyOnlyInOrderSignaling &&
isCopyOnlyEnabled && !isNonDualStreamCopyOffloadOperation(memoryCopyParams.copyOffloadAllowed);
const bool inOrderCopyOnlySignalingAllowed = this->isInOrderExecutionEnabled() && !memoryCopyParams.forceDisableCopyOnlyInOrderSignaling && isCopyOnlyEnabled;
ze_result_t result = ZE_RESULT_SUCCESS;
if (isCopyOnlyEnabled) {

2
level_zero/core/test/unit_tests/mocks/mock_cmdlist.h
View File

@@ -126,6 +126,7 @@ struct WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>
using BaseClass::updateStreamProperties;
using BaseClass::updateStreamPropertiesForFlushTaskDispatchFlags;
using BaseClass::updateStreamPropertiesForRegularCommandLists;
using BaseClass::useAdditionalBlitProperties;
WhiteBox() : ::L0::CommandListCoreFamily<gfxCoreFamily>(BaseClass::defaultNumIddsPerBlock) {}
@@ -249,6 +250,7 @@ struct WhiteBox<L0::CommandListCoreFamilyImmediate<gfxCoreFamily>>
using BaseClass::synchronizedDispatchMode;
using BaseClass::synchronizeInOrderExecution;
using BaseClass::updateInOrderExecInfo;
using BaseClass::useAdditionalBlitProperties;
WhiteBox() : BaseClass(BaseClass::defaultNumIddsPerBlock) {}

6
level_zero/core/test/unit_tests/sources/cmdlist/test_in_order_cmdlist_2.cpp
View File

@@ -85,6 +85,7 @@ HWTEST_F(CopyOffloadInOrderTests, givenDebugFlagSetWhenCreatingCmdListThenOverri
HWTEST2_F(CopyOffloadInOrderTests, givenNonDualStreamModeWhenSubmittedThenUseDefaultQueue, IsAtLeastXeHpCore) {
debugManager.flags.OverrideCopyOffloadMode.set(nonDualStreamMode);
auto immCmdList = createImmCmdListWithOffload<FamilyType::gfxCoreFamily>();
immCmdList->useAdditionalBlitProperties = true;
auto eventPool = createEvents<FamilyType>(1, true);
auto eventHandle = events[0]->toHandle();
@@ -106,7 +107,9 @@ HWTEST2_F(CopyOffloadInOrderTests, givenNonDualStreamModeWhenSubmittedThenDontPr
using MI_LOAD_REGISTER_IMM = typename FamilyType::MI_LOAD_REGISTER_IMM;
debugManager.flags.OverrideCopyOffloadMode.set(nonDualStreamMode);
auto immCmdList0 = createImmCmdListWithOffload<FamilyType::gfxCoreFamily>();
immCmdList0->useAdditionalBlitProperties = true;
auto immCmdList1 = createImmCmdListWithOffload<FamilyType::gfxCoreFamily>();
immCmdList1->useAdditionalBlitProperties = true;
auto eventPool = createEvents<FamilyType>(2, true);
auto eventHandle0 = events[0]->toHandle();
@@ -137,6 +140,7 @@ HWTEST2_F(CopyOffloadInOrderTests, givenNonDualStreamModeAndProfilingEventWithRe
debugManager.flags.OverrideCopyOffloadMode.set(nonDualStreamMode);
auto immCmdList = createImmCmdListWithOffload<FamilyType::gfxCoreFamily>();
immCmdList->useAdditionalBlitProperties = true;
auto mainQueueCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(immCmdList->getCsr(false));
auto copyQueueCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(immCmdList->getCsr(true));
@@ -541,6 +545,7 @@ HWTEST2_F(CopyOffloadInOrderTests, givenNonDualStreamOffloadWhenImageCopyCalledT
debugManager.flags.OverrideCopyOffloadMode.set(nonDualStreamMode);
auto immCmdList = createImmCmdListWithOffload<FamilyType::gfxCoreFamily>();
immCmdList->useAdditionalBlitProperties = true;
auto eventPool = createEvents<FamilyType>(1, true);
@@ -593,6 +598,7 @@ HWTEST2_F(CopyOffloadInOrderTests, givenNonDualStreamOffloadWhenFillCalledThenSk
debugManager.flags.OverrideCopyOffloadMode.set(nonDualStreamMode);
auto immCmdList = createImmCmdListWithOffload<FamilyType::gfxCoreFamily>();
immCmdList->useAdditionalBlitProperties = true;
auto eventPool = createEvents<FamilyType>(1, true);