Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Change DG2 l1 cache policy to WB

Browse Source 
With compiler LSC WAs this gives better performance.

If debugger is active, policy will not be changed ie.
will be WBP.

Related-To: NEO-7003

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This commit is contained in:
Dominik Dabek
2022-08-23 11:48:18 +00:00
committed by Compute-Runtime-Automation
parent 6118e41741
commit 8cc0177f1c
58 changed files with 754 additions and 131 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/test/unit_test/command_stream/cache_flush_tests_dg2_and_later.cpp
View File

@ -46,7 +46,7 @@ HWTEST2_F(CacheFlushTestsDg2AndLater, WhenProgrammingCacheFlushAfterWalkerThenEx
if constexpr (FamilyType::isUsingL3Control) {
using L3_CONTROL = typename FamilyType::L3_CONTROL;
expectedCommands.push_back(new MatchHwCmd<FamilyType, L3_CONTROL>(
1, Expects{EXPECT_MEMBER(L3_CONTROL, getUnTypedDataPortCacheFlush, false)}));
1, Expects{EXPECT_MEMBER(L3_CONTROL, getUnTypedDataPortCacheFlush, true)}));
}
bool cmdBuffOk = expectCmdBuff<FamilyType>(cmdQ.getCS(0), 0, std::move(expectedCommands), &err);

2
opencl/test/unit_test/program/program_tests.cpp
View File

@ -1743,7 +1743,7 @@ TEST_F(ProgramTests, whenGetInternalOptionsThenLSCPolicyIsSet) {
MockProgram program(pContext, false, toClDeviceVector(*pClDevice));
auto internalOptions = program.getInternalOptions();
const auto &compilerHwInfoConfig = *CompilerHwInfoConfig::get(defaultHwInfo->platform.eProductFamily);
auto expectedPolicy = compilerHwInfoConfig.getCachingPolicyOptions();
auto expectedPolicy = compilerHwInfoConfig.getCachingPolicyOptions(false);
if (expectedPolicy != nullptr) {
EXPECT_TRUE(CompilerOptions::contains(internalOptions, expectedPolicy));
} else {

4
opencl/test/unit_test/xe_hpg_core/dg2/excludes_ocl_dg2.cpp
View File

@ -12,3 +12,7 @@ HWTEST_EXCLUDE_PRODUCT(HwInfoConfigTestXeHpAndLater, givenXeHPAndLaterPlatformWh
HWTEST_EXCLUDE_PRODUCT(WddmMemoryManagerSimpleTest, givenLinearStreamWhenItIsAllocatedThenItIsInLocalMemoryHasCpuPointerAndHasStandardHeap64kbAsGpuAddress, IGFX_DG2);
HWTEST_EXCLUDE_PRODUCT(HwHelperTestXeHPAndLater, GiveCcsNodeThenDefaultEngineTypeIsCcs, IGFX_DG2);
HWTEST_EXCLUDE_PRODUCT(XeHPAndLaterDeviceCapsTests, givenHwInfoWhenRequestedComputeUnitsUsedForScratchThenReturnValidValue, IGFX_DG2);
HWTEST_EXCLUDE_PRODUCT(ProgramTests, givenAtLeastXeHpgCoreWhenGetInternalOptionsThenCorrectBuildOptionIsSet_IsAtLeastXeHpgCore, IGFX_DG2);
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenL3ToL1DebugFlagWhenStatelessMocsIsProgrammedThenItHasL1CachingOn, IGFX_DG2);
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenAlignedCacheableReadOnlyBufferThenChoseOclBufferConstPolicy, IGFX_DG2);
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, whenAppendingRssThenProgramWBPL1CachePolicy, IGFX_DG2);

57
opencl/test/unit_test/xe_hpg_core/dg2/test_cmds_programming_dg2.cpp
View File

@ -37,10 +37,27 @@ DG2TEST_F(CmdsProgrammingTestsDg2, givenL3ToL1DebugFlagWhenStatelessMocsIsProgra
auto stateBaseAddress = static_cast<STATE_BASE_ADDRESS *>(hwParserCsr.cmdStateBaseAddress);
auto actualL1CachePolocy = static_cast<uint8_t>(stateBaseAddress->getL1CachePolicyL1CacheControl());
EXPECT_EQ(stateBaseAddress->getL1CachePolicyL1CacheControl(), STATE_BASE_ADDRESS::L1_CACHE_POLICY_WB);
}
const uint8_t expectedL1CachePolicy = FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WBP;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolocy);
DG2TEST_F(CmdsProgrammingTestsDg2, givenL3ToL1DebugFlagAndDebuggerActiveWhenStatelessMocsIsProgrammedThenItHasCorrectL1CachingOn) {
using STATE_BASE_ADDRESS = typename FamilyType::STATE_BASE_ADDRESS;
DebugManagerStateRestore restore;
DebugManager.flags.ForceL1Caching.set(1u);
pDevice->setDebuggerActive(true);
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
flushTask(commandStreamReceiver);
HardwareParse hwParserCsr;
hwParserCsr.parseCommands<FamilyType>(commandStreamReceiver.commandStream, 0);
hwParserCsr.findHardwareCommands<FamilyType>();
ASSERT_NE(nullptr, hwParserCsr.cmdStateBaseAddress);
auto stateBaseAddress = static_cast<STATE_BASE_ADDRESS *>(hwParserCsr.cmdStateBaseAddress);
EXPECT_EQ(stateBaseAddress->getL1CachePolicyL1CacheControl(), STATE_BASE_ADDRESS::L1_CACHE_POLICY_WBP);
}
DG2TEST_F(CmdsProgrammingTestsDg2, whenAppendingRssThenProgramWBPL1CachePolicy) {
@ -69,7 +86,10 @@ DG2TEST_F(CmdsProgrammingTestsDg2, whenAppendingRssThenProgramWBPL1CachePolicy)
args.areMultipleSubDevicesInContext = true;
EncodeSurfaceState<FamilyType>::encodeBuffer(args);
EXPECT_EQ(FamilyType::RENDER_SURFACE_STATE::L1_CACHE_POLICY_WB, rssCmd.getL1CachePolicyL1CacheControl());
args.isDebuggerActive = true;
EncodeSurfaceState<FamilyType>::encodeBuffer(args);
EXPECT_EQ(FamilyType::RENDER_SURFACE_STATE::L1_CACHE_POLICY_WBP, rssCmd.getL1CachePolicyL1CacheControl());
}
@ -95,10 +115,35 @@ DG2TEST_F(CmdsProgrammingTestsDg2, givenAlignedCacheableReadOnlyBufferThenChoseO
const auto actualMocs = surfaceState.getMemoryObjectControlState();
EXPECT_EQ(expectedMocs, actualMocs);
auto actualL1CachePolocy = static_cast<uint8_t>(surfaceState.getL1CachePolicyL1CacheControl());
EXPECT_EQ(surfaceState.getL1CachePolicyL1CacheControl(), FamilyType::RENDER_SURFACE_STATE::L1_CACHE_POLICY_WB);
const uint8_t expectedL1CachePolicy = FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WBP;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolocy);
alignedFree(ptr);
}
DG2TEST_F(CmdsProgrammingTestsDg2, givenAlignedCacheableReadOnlyBufferAndDebuggerActiveWhenBufferCreateThenChoseOclBufferConstPolicy) {
MockContext context;
const_cast<DeviceInfo &>(context.getDevice(0)->getDevice().getDeviceInfo()).debuggerActive = true;
const auto size = MemoryConstants::pageSize;
const auto ptr = (void *)alignedMalloc(size * 2, MemoryConstants::pageSize);
const auto flags = CL_MEM_USE_HOST_PTR | CL_MEM_READ_ONLY;
auto retVal = CL_SUCCESS;
auto buffer = std::unique_ptr<Buffer>(Buffer::create(
&context,
flags,
size,
ptr,
retVal));
EXPECT_EQ(CL_SUCCESS, retVal);
typename FamilyType::RENDER_SURFACE_STATE surfaceState = {};
buffer->setArgStateful(&surfaceState, false, false, false, false, context.getDevice(0)->getDevice(), false, false);
const auto expectedMocs = context.getDevice(0)->getGmmHelper()->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER_CONST);
const auto actualMocs = surfaceState.getMemoryObjectControlState();
EXPECT_EQ(expectedMocs, actualMocs);
EXPECT_EQ(surfaceState.getL1CachePolicyL1CacheControl(), FamilyType::RENDER_SURFACE_STATE::L1_CACHE_POLICY_WBP);
alignedFree(ptr);
}