intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 05:56:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: Skip fence after PC when no post sync write

Browse Source 
Related-To: NEO-14642

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2025-05-22 15:59:42 +00:00
committed by Compute-Runtime-Automation
parent 8e5b29f55e
commit 86e7d5b276
32 changed files with 58 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
level_zero/core/source/cmdqueue/cmdqueue_hw.inl
View File

@@ -262,7 +262,7 @@ size_t CommandQueueHw<gfxCoreFamily>::estimateStreamSizeForExecuteCommandListsRe
}
if (ctx.isDispatchTaskCountPostSyncRequired) {
linearStreamSizeEstimate += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(this->device->getNEODevice()->getRootDeviceEnvironment());
linearStreamSizeEstimate += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(this->device->getNEODevice()->getRootDeviceEnvironment(), true);
}
if (instructionCacheFlushRequired) {
@@ -327,7 +327,7 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandListsRegular(
linearStreamSizeEstimate += this->computeDebuggerCmdsSize(ctx);
if (ctx.isDispatchTaskCountPostSyncRequired) {
linearStreamSizeEstimate += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(neoDevice->getRootDeviceEnvironment());
linearStreamSizeEstimate += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(neoDevice->getRootDeviceEnvironment(), true);
}
NEO::LinearStream child(nullptr);

2
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_barrier.cpp
View File

@@ -240,7 +240,7 @@ struct MultiTileCommandListAppendBarrierFixture : public MultiTileCommandListFix
sizeof(MI_STORE_DATA_IMM) +
sizeof(MI_ATOMIC) + NEO::EncodeSemaphore<FamilyType>::getSizeMiSemaphoreWait();
size_t postSyncSize = NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment());
size_t postSyncSize = NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true);
auto useSizeBefore = cmdListStream->getUsed();
auto result = commandList->appendBarrier(eventHandle, 0, nullptr, false);

2
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_event_reset.cpp
View File

@@ -390,7 +390,7 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent,
auto gpuAddress = event->getGpuAddress(device) + event->getContextEndOffset();
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment()) +
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true) +
commandList->estimateBufferSizeMultiTileBarrier(device->getNEODevice()->getRootDeviceEnvironment());
size_t usedSize = cmdStream->getUsed();
EXPECT_EQ(expectedSize, usedSize);

8
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_signal_event.cpp
View File

@@ -520,7 +520,7 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent,
auto gpuAddress = event->getGpuAddress(device) + event->getContextEndOffset();
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment());
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true);
size_t usedSize = cmdStream->getUsed();
EXPECT_EQ(expectedSize, usedSize);
@@ -561,7 +561,7 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent, givenMultiTileAndDynamicPostSy
commandList->partitionCount = 2;
EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendSignalEvent(event->toHandle(), false));
size_t expectedSize = NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment());
size_t expectedSize = NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true);
auto unifiedPostSyncLayout = device->getL0GfxCoreHelper().hasUnifiedPostSyncAllocationLayout();
@@ -713,7 +713,7 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent,
auto gpuAddress = event->getCompletionFieldGpuAddress(device);
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment());
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true);
size_t usedSize = cmdStream->getUsed();
EXPECT_EQ(expectedSize, usedSize);
@@ -777,7 +777,7 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent,
auto gpuAddress = event->getCompletionFieldGpuAddress(device);
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment());
size_t expectedSize = NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true);
size_t usedSize = cmdStream->getUsed();
EXPECT_EQ(expectedSize, usedSize);

2
level_zero/core/test/unit_tests/sources/cmdqueue/test_cmdqueue_1.cpp
View File

@@ -494,7 +494,7 @@ HWTEST_F(CommandQueueCreate, GivenDispatchTaskCountPostSyncRequiredWhenExecuteCo
commandQueue->executeCommandLists(1, &cmdListHandle, nullptr, false, nullptr, nullptr);
auto estimatedSizeWithtBarrier = commandQueue->requiredSizeCalled;
auto sizeForBarrier = NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment());
auto sizeForBarrier = NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(device->getNEODevice()->getRootDeviceEnvironment(), true);
EXPECT_GT(sizeForBarrier, 0u);
EXPECT_EQ(estimatedSizeWithtBarrier, estimatedSizeWithoutBarrier + sizeForBarrier);

2
level_zero/core/test/unit_tests/sources/cmdqueue/test_cmdqueue_2.cpp
View File

@@ -485,7 +485,7 @@ HWTEST_F(CommandQueueSynchronizeTest, givenSynchronousCommandQueueWhenTagUpdateF
} else {
expectedSize += sizeof(MI_BATCH_BUFFER_END);
}
expectedSize += NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(neoDevice->getRootDeviceEnvironment());
expectedSize += NEO::MemorySynchronizationCommands<FamilyType>::getSizeForBarrierWithPostSyncOperation(neoDevice->getRootDeviceEnvironment(), true);
expectedSize = alignUp(expectedSize, 8);
const ze_command_queue_desc_t desc{ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC, nullptr, 0, 0, 0, ZE_COMMAND_QUEUE_MODE_SYNCHRONOUS, ZE_COMMAND_QUEUE_PRIORITY_NORMAL};