Media
/
compute-runtime
mirror of https://github.com/intel/compute-runtime.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Revert "performance: Remove global fence from command stream on BMG" 

This reverts commit 4e823f9e6e.

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk 2025-05-12 06:23:29 +00:00 committed by Compute-Runtime-Automation
parent 9c9a0a87b9
commit 4fe3bc40e3
5 changed files with 19 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
shared/source/xe2_hpg_core/os_agnostic_product_helper_xe2_hpg_core.inl
View File

@ -20,7 +20,7 @@ bool ProductHelperHw<gfxProduct>::isBlitterForImagesSupported() const {
template <>
bool ProductHelperHw<gfxProduct>::isGlobalFenceInCommandStreamRequired(const HardwareInfo &hwInfo) const {
return false;
return !hwInfo.capabilityTable.isIntegratedDevice;
}
template <>

2
shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp
View File

@ -3075,7 +3075,7 @@ HWTEST2_F(CommandStreamReceiverHwTest, givenDeviceToHostCopyWhenFenceIsRequiredT
fenceExpected &= getHelper<ProductHelper>().isDeviceToHostCopySignalingFenceRequired();
size_t expectedFenceCount = fenceExpected ? 1 : 0;
if (getHelper<ProductHelper>().isGlobalFenceInCommandStreamRequired(pDevice->getHardwareInfo())) {
if (!pDevice->getHardwareInfo().capabilityTable.isIntegratedDevice) {
expectedFenceCount += 2;
}

27
shared/test/unit_test/direct_submission/direct_submission_tests_2.cpp
View File

@ -48,7 +48,7 @@ struct DirectSubmissionDispatchMiMemFenceTest : public DirectSubmissionDispatchB
DirectSubmissionDispatchBufferTest::SetUp();
auto &productHelper = pDevice->getProductHelper();
miMemFenceSupported = productHelper.isGlobalFenceInDirectSubmissionRequired(pDevice->getHardwareInfo());
miMemFenceSupported = pDevice->getHardwareInfo().capabilityTable.isIntegratedDevice ? false : productHelper.isGlobalFenceInDirectSubmissionRequired(pDevice->getHardwareInfo());
auto &compilerProductHelper = pDevice->getCompilerProductHelper();
heaplessStateInit = compilerProductHelper.isHeaplessStateInitEnabled(compilerProductHelper.isHeaplessModeEnabled(*defaultHwInfo));
@ -108,18 +108,9 @@ struct DirectSubmissionDispatchMiMemFenceTest : public DirectSubmissionDispatchB
EXPECT_EQ(expectedFenceCount, fenceCount);
EXPECT_EQ(expectedSysMemFenceCount, sysMemFenceCount);
} else {
if (directSubmission.globalFenceAllocation) {
if (expectedSysMemFenceCount > 0) {
EXPECT_NE(-1, systemMemoryFenceId);
} else {
EXPECT_EQ(-1, systemMemoryFenceId);
}
EXPECT_EQ(expectedSysMemFenceCount, sysMemFenceCount);
} else {
EXPECT_EQ(-1, systemMemoryFenceId);
EXPECT_EQ(0u, sysMemFenceCount);
}
EXPECT_EQ(-1, systemMemoryFenceId);
EXPECT_EQ(0u, fenceCount);
EXPECT_EQ(0u, sysMemFenceCount);
}
}
@ -269,7 +260,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, DirectSubmissionDispatchBufferTest,
directSubmission.getSizeSemaphoreSection(false) +
sizeof(MI_LOAD_REGISTER_IMM) +
sizeof(MI_LOAD_REGISTER_MEM);
if (directSubmission.globalFenceAllocation && !heaplessStateInit) {
if (directSubmission.miMemFenceRequired && !heaplessStateInit) {
submitSize += directSubmission.getSizeSystemMemoryFenceAddress();
}
if (directSubmission.isRelaxedOrderingEnabled()) {
@ -566,7 +557,7 @@ HWTEST_F(DirectSubmissionDispatchBufferTest,
EXPECT_EQ(1u, directSubmission.submitCount);
size_t submitSize = RenderDispatcher<FamilyType>::getSizePreemption() +
directSubmission.getSizeSemaphoreSection(false);
if (directSubmission.globalFenceAllocation && !heaplessStateInit) {
if (directSubmission.miMemFenceRequired && !heaplessStateInit) {
submitSize += directSubmission.getSizeSystemMemoryFenceAddress();
}
if (directSubmission.isRelaxedOrderingEnabled()) {
@ -620,7 +611,7 @@ HWTEST_F(DirectSubmissionDispatchBufferTest,
EXPECT_EQ(1u, directSubmission.handleResidencyCount);
size_t submitSize = directSubmission.getSizeDispatch(false, false, directSubmission.dispatchMonitorFenceRequired(false)) - directSubmission.getSizeNewResourceHandler();
if (directSubmission.globalFenceAllocation && !heaplessStateInit) {
if (directSubmission.miMemFenceRequired && !heaplessStateInit) {
submitSize += directSubmission.getSizeSystemMemoryFenceAddress();
}
if (directSubmission.isRelaxedOrderingEnabled()) {
@ -646,7 +637,7 @@ HWTEST_F(DirectSubmissionDispatchBufferTest,
EXPECT_EQ(1u, directSubmission.submitCount);
size_t submitSize = RenderDispatcher<FamilyType>::getSizePreemption() +
directSubmission.getSizeSemaphoreSection(false);
if (directSubmission.globalFenceAllocation && !heaplessStateInit) {
if (directSubmission.miMemFenceRequired && !heaplessStateInit) {
submitSize += directSubmission.getSizeSystemMemoryFenceAddress();
}
if (directSubmission.isRelaxedOrderingEnabled()) {
@ -697,7 +688,7 @@ HWTEST_F(DirectSubmissionDispatchBufferTest,
EXPECT_EQ(1u, directSubmission.handleResidencyCount);
size_t submitSize = directSubmission.getSizeDispatch(false, false, directSubmission.dispatchMonitorFenceRequired(false)) - directSubmission.getSizeNewResourceHandler();
if (directSubmission.globalFenceAllocation && !heaplessStateInit) {
if (directSubmission.miMemFenceRequired && !heaplessStateInit) {
submitSize += directSubmission.getSizeSystemMemoryFenceAddress();
}
if (directSubmission.isRelaxedOrderingEnabled()) {
@ -790,7 +781,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, DirectSubmissionDispatchBufferTest,
directSubmission.getSizeSemaphoreSection(false) +
sizeof(MI_LOAD_REGISTER_IMM) +
sizeof(MI_LOAD_REGISTER_MEM);
if (directSubmission.globalFenceAllocation && !heaplessStateInit) {
if (directSubmission.miMemFenceRequired && !heaplessStateInit) {
submitSize += directSubmission.getSizeSystemMemoryFenceAddress();
}

2
shared/test/unit_test/os_interface/windows/device_command_stream_tests.cpp
View File

@ -1368,7 +1368,7 @@ HWTEST_TEMPLATED_F(WddmCommandStreamMockGdiTest, givenDirectSubmissionEnabledOnB
auto &compilerProductHelper = device->getCompilerProductHelper();
auto heaplessStateInit = compilerProductHelper.isHeaplessStateInitEnabled(compilerProductHelper.isHeaplessModeEnabled(*defaultHwInfo));
if (directSubmission->globalFenceAllocation && !heaplessStateInit) {
if (directSubmission->miMemFenceRequired && !heaplessStateInit) {
expectedSize += directSubmission->getSizeSystemMemoryFenceAddress();
}
if (directSubmission->isRelaxedOrderingEnabled()) {

11
shared/test/unit_test/xe2_hpg_core/gfx_core_helper_tests_xe2_hpg_core.cpp
View File

@ -520,14 +520,17 @@ XE2_HPG_CORETEST_F(GfxCoreHelperTestsXe2HpgCore, givenGfxCoreHelperWhenAskedIfFe
}
XE2_HPG_CORETEST_F(GfxCoreHelperTestsXe2HpgCore, givenDefaultMemorySynchronizationCommandsWhenGettingSizeForAdditionalSynchronizationThenCorrectValueIsReturned) {
EXPECT_EQ(0u, MemorySynchronizationCommands<FamilyType>::getSizeForAdditonalSynchronization(pDevice->getRootDeviceEnvironment()));
using MI_MEM_FENCE = typename FamilyType::MI_MEM_FENCE;
EXPECT_EQ(!pDevice->getHardwareInfo().capabilityTable.isIntegratedDevice * sizeof(MI_MEM_FENCE), MemorySynchronizationCommands<FamilyType>::getSizeForAdditonalSynchronization(pDevice->getRootDeviceEnvironment()));
}
XE2_HPG_CORETEST_F(GfxCoreHelperTestsXe2HpgCore, givenDebugMemorySynchronizationCommandsWhenGettingSizeForAdditionalSynchronizationThenCorrectValueIsReturned) {
DebugManagerStateRestore restorer;
debugManager.flags.DisablePipeControlPrecedingPostSyncCommand.set(1);
using MI_MEM_FENCE = typename FamilyType::MI_MEM_FENCE;
EXPECT_EQ(0u, MemorySynchronizationCommands<FamilyType>::getSizeForAdditonalSynchronization(pDevice->getRootDeviceEnvironment()));
EXPECT_EQ(!pDevice->getHardwareInfo().capabilityTable.isIntegratedDevice * 2 * sizeof(MI_MEM_FENCE), MemorySynchronizationCommands<FamilyType>::getSizeForAdditonalSynchronization(pDevice->getRootDeviceEnvironment()));
}
XE2_HPG_CORETEST_F(GfxCoreHelperTestsXe2HpgCore, givenDontProgramGlobalFenceAsMiMemFenceCommandInCommandStreamWhenGettingSizeForAdditionalSynchronizationThenCorrectValueIsReturned) {
@ -677,9 +680,9 @@ XE2_HPG_CORETEST_F(ProductHelperTestXe2HpgCore, givenProductHelperWhenCallUseGem
EXPECT_TRUE(productHelper.useGemCreateExtInAllocateMemoryByKMD());
}
XE2_HPG_CORETEST_F(ProductHelperTestXe2HpgCore, givenProductHelperWhenAskingForGlobalFenceSupportThenReturnFalse) {
XE2_HPG_CORETEST_F(ProductHelperTestXe2HpgCore, givenProductHelperWhenAskingForGlobalFenceSupportThenReturnTrue) {
const auto &productHelper = getHelper<ProductHelper>();
EXPECT_FALSE(productHelper.isGlobalFenceInCommandStreamRequired(*defaultHwInfo));
EXPECT_EQ(productHelper.isGlobalFenceInCommandStreamRequired(*defaultHwInfo), !defaultHwInfo->capabilityTable.isIntegratedDevice);
}
XE2_HPG_CORETEST_F(ProductHelperTestXe2HpgCore, givenProductHelperWhenAskingForCooperativeEngineSupportThenReturnTrue) {