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.

Related-To: NEO-7003

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This commit is contained in:
Dominik Dabek
2022-07-22 08:15:16 +00:00
committed by Compute-Runtime-Automation
parent 6f77f1e468
commit 9a5e619c42
22 changed files with 181 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@ -1787,7 +1787,7 @@ HWTEST2_F(ProgramTests, givenDebugFlagSetToWbWhenGetInternalOptionsThenCorrectBu
EXPECT_TRUE(CompilerOptions::contains(internalOptions, "-cl-store-cache-default=7 -cl-load-cache-default=4"));
}
HWTEST2_F(ProgramTests, givenAtLeastDG2WhenGetInternalOptionsThenCorrectBuildOptionIsSet, IsAtLeastXeHpgCore) {
HWTEST2_F(ProgramTests, givenAtLeastXeHpgWhenGetInternalOptionsThenCorrectBuildOptionIsSet, IsAtLeastXeHpgCore) {
MockProgram program(pContext, false, toClDeviceVector(*pClDevice));
auto internalOptions = program.getInternalOptions();
EXPECT_TRUE(CompilerOptions::contains(internalOptions, "-cl-store-cache-default=2 -cl-load-cache-default=4"));

13
opencl/test/unit_test/program/xe_hpg_core/CMakeLists.txt Normal file
View File

@ -0,0 +1,13 @@
#
# Copyright (C) 2022 Intel Corporation
#
# SPDX-License-Identifier: MIT
#
if(TESTS_XE_HPG_CORE)
set(IGDRCL_SRCS_tests_program_xe_hpg_core
${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
)
target_sources(igdrcl_tests PRIVATE ${IGDRCL_SRCS_tests_program_xe_hpg_core})
add_subdirectories()
endif()

16
opencl/test/unit_test/program/xe_hpg_core/dg2/CMakeLists.txt Normal file
View File

@ -0,0 +1,16 @@
#
# Copyright (C) 2022 Intel Corporation
#
# SPDX-License-Identifier: MIT
#
if(TESTS_DG2)
set(IGDRCL_SRCS_tests_program_dg2
${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
${CMAKE_CURRENT_SOURCE_DIR}/program_tests_dg2.cpp
)
target_sources(igdrcl_tests PRIVATE ${IGDRCL_SRCS_tests_program_dg2})
endif()
add_subdirectories()

22
opencl/test/unit_test/program/xe_hpg_core/dg2/program_tests_dg2.cpp Normal file
View File

@ -0,0 +1,22 @@
/*
* Copyright (C) 2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/xe_hpg_core/hw_cmds_dg2.h"
#include "shared/test/common/test_macros/header/per_product_test_definitions.h"
#include "shared/test/common/test_macros/test.h"
#include "opencl/test/unit_test/mocks/mock_context.h"
#include "opencl/test/unit_test/mocks/mock_program.h"
#include "opencl/test/unit_test/program/program_tests.h"
using namespace NEO;
HWTEST_EXCLUDE_PRODUCT(ProgramTests, givenAtLeastXeHpgWhenGetInternalOptionsThenCorrectBuildOptionIsSet_IsAtLeastXeHpgCore, IGFX_XE_HPG_CORE);
DG2TEST_F(ProgramTests, givenDG2WhenGetInternalOptionsThenCorrectBuildOptionIsSet) {
MockProgram program(pContext, false, toClDeviceVector(*pClDevice));
auto internalOptions = program.getInternalOptions();
EXPECT_TRUE(CompilerOptions::contains(internalOptions, "-cl-store-cache-default=7 -cl-load-cache-default=4"));
}

66
opencl/test/unit_test/xe_hpg_core/dg2/test_cmds_programming_dg2.cpp
View File

@ -21,6 +21,7 @@
using namespace NEO;
using CmdsProgrammingTestsDg2 = UltCommandStreamReceiverTest;
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenL3ToL1DebugFlagWhenStatelessMocsIsProgrammedThenItHasL1CachingOn, IGFX_XE_HPG_CORE);
DG2TEST_F(CmdsProgrammingTestsDg2, givenL3ToL1DebugFlagWhenStatelessMocsIsProgrammedThenItHasL1CachingOn) {
using STATE_BASE_ADDRESS = typename FamilyType::STATE_BASE_ADDRESS;
@ -37,13 +38,14 @@ DG2TEST_F(CmdsProgrammingTestsDg2, givenL3ToL1DebugFlagWhenStatelessMocsIsProgra
auto stateBaseAddress = static_cast<STATE_BASE_ADDRESS *>(hwParserCsr.cmdStateBaseAddress);
auto actualL1CachePolocy = static_cast<uint8_t>(stateBaseAddress->getL1CachePolicyL1CacheControl());
auto actualL1CachePolicy = static_cast<uint8_t>(stateBaseAddress->getL1CachePolicyL1CacheControl());
const uint8_t expectedL1CachePolicy = FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WBP;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolocy);
const uint8_t expectedL1CachePolicy = FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WB;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolicy);
}
DG2TEST_F(CmdsProgrammingTestsDg2, givenSpecificProductFamilyWhenAppendingSbaThenProgramWtL1CachePolicy) {
auto memoryManager = pDevice->getExecutionEnvironment()->memoryManager.get();
AllocationProperties properties(pDevice->getRootDeviceIndex(), 1, AllocationType::BUFFER, pDevice->getDeviceBitfield());
auto allocation = memoryManager->allocateGraphicsMemoryWithProperties(properties);
@ -56,11 +58,12 @@ DG2TEST_F(CmdsProgrammingTestsDg2, givenSpecificProductFamilyWhenAppendingSbaThe
pDevice->getRootDeviceEnvironment().getGmmHelper(), false,
MemoryCompressionState::NotApplicable, true, false, 1u);
EXPECT_EQ(FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WBP, sbaCmd.getL1CachePolicyL1CacheControl());
EXPECT_EQ(FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WB, sbaCmd.getL1CachePolicyL1CacheControl());
memoryManager->freeGraphicsMemory(allocation);
}
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenL1CachingOverrideWhenStateBaseAddressIsProgrammedThenItMatchesTheOverrideValue, IGFX_XE_HPG_CORE);
DG2TEST_F(CmdsProgrammingTestsDg2, givenL1CachingOverrideWhenStateBaseAddressIsProgrammedThenItMatchesTheOverrideValue) {
DebugManagerStateRestore restorer;
DebugManager.flags.ForceStatelessL1CachingPolicy.set(0u);
@ -89,7 +92,8 @@ DG2TEST_F(CmdsProgrammingTestsDg2, givenL1CachingOverrideWhenStateBaseAddressIsP
memoryManager->freeGraphicsMemory(allocation);
}
DG2TEST_F(CmdsProgrammingTestsDg2, whenAppendingRssThenProgramWBPL1CachePolicy) {
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenXeHpgCoreWhenAppendingRssThenProgramWBPL1CachePolicy, IGFX_XE_HPG_CORE);
DG2TEST_F(CmdsProgrammingTestsDg2, givenDG2whenAppendingRssThenProgramWBL1CachePolicy) {
auto memoryManager = pDevice->getExecutionEnvironment()->memoryManager.get();
size_t allocationSize = MemoryConstants::pageSize;
AllocationProperties properties(pDevice->getRootDeviceIndex(), allocationSize, AllocationType::BUFFER, pDevice->getDeviceBitfield());
@ -116,39 +120,10 @@ DG2TEST_F(CmdsProgrammingTestsDg2, whenAppendingRssThenProgramWBPL1CachePolicy)
EncodeSurfaceState<FamilyType>::encodeBuffer(args);
EXPECT_EQ(FamilyType::RENDER_SURFACE_STATE::L1_CACHE_POLICY_WBP, rssCmd.getL1CachePolicyL1CacheControl());
}
DG2TEST_F(CmdsProgrammingTestsDg2, givenAlignedCacheableReadOnlyBufferThenChoseOclBufferConstPolicy) {
MockContext context;
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);
auto actualL1CachePolocy = static_cast<uint8_t>(surfaceState.getL1CachePolicyL1CacheControl());
const uint8_t expectedL1CachePolicy = FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WBP;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolocy);
alignedFree(ptr);
EXPECT_EQ(FamilyType::RENDER_SURFACE_STATE::L1_CACHE_POLICY_WB, rssCmd.getL1CachePolicyL1CacheControl());
}
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenAlignedCacheableReadOnlyBufferThenChoseOclBufferConstPolicy, IGFX_XE_HPG_CORE);
DG2TEST_F(CmdsProgrammingTestsDg2, givenDG2WithBSteppingWhenFlushingTaskThenAdditionalStateBaseAddressCommandIsPresent) {
auto &hwInfo = *pDevice->getRootDeviceEnvironment().getMutableHardwareInfo();
const auto &hwInfoConfig = *HwInfoConfig::get(hwInfo.platform.eProductFamily);
@ -168,3 +143,22 @@ DG2TEST_F(CmdsProgrammingTestsDg2, givenDG2WithBSteppingWhenFlushingTaskThenAddi
stateBaseAddressItor = find<typename FamilyType::STATE_BASE_ADDRESS *>(stateBaseAddressItor, cmdList.end());
EXPECT_NE(cmdList.end(), stateBaseAddressItor);
}
HWTEST_EXCLUDE_PRODUCT(CmdsProgrammingTestsXeHpgCore, givenSpecificProductFamilyWhenAppendingSbaThenProgramWBPL1CachePolicy, IGFX_XE_HPG_CORE);
DG2TEST_F(CmdsProgrammingTestsDg2, givenDG2WhenAppendingSbaThenProgramWBL1CachePolicy) {
auto memoryManager = pDevice->getExecutionEnvironment()->memoryManager.get();
AllocationProperties properties(pDevice->getRootDeviceIndex(), 1, AllocationType::BUFFER, pDevice->getDeviceBitfield());
auto allocation = memoryManager->allocateGraphicsMemoryWithProperties(properties);
IndirectHeap indirectHeap(allocation, 1);
DispatchFlags flags = DispatchFlagsHelper::createDefaultDispatchFlags();
auto sbaCmd = FamilyType::cmdInitStateBaseAddress;
StateBaseAddressHelper<FamilyType>::appendStateBaseAddressParameters(&sbaCmd, &indirectHeap, true, 0,
pDevice->getRootDeviceEnvironment().getGmmHelper(), false,
MemoryCompressionState::NotApplicable, true, false, 1u);
EXPECT_EQ(FamilyType::STATE_BASE_ADDRESS::L1_CACHE_POLICY_WB, sbaCmd.getL1CachePolicyL1CacheControl());
memoryManager->freeGraphicsMemory(allocation);
}

10
opencl/test/unit_test/xe_hpg_core/test_cmds_programming_xe_hpg_core.cpp
View File

@ -37,10 +37,10 @@ XE_HPG_CORETEST_F(CmdsProgrammingTestsXeHpgCore, givenL3ToL1DebugFlagWhenStatele
auto stateBaseAddress = static_cast<STATE_BASE_ADDRESS *>(hwParserCsr.cmdStateBaseAddress);
auto actualL1CachePolocy = static_cast<uint8_t>(stateBaseAddress->getL1CachePolicyL1CacheControl());
auto actualL1CachePolicy = static_cast<uint8_t>(stateBaseAddress->getL1CachePolicyL1CacheControl());
const uint8_t expectedL1CachePolicy = 0;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolocy);
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolicy);
}
XE_HPG_CORETEST_F(CmdsProgrammingTestsXeHpgCore, givenSpecificProductFamilyWhenAppendingSbaThenProgramWBPL1CachePolicy) {
@ -89,7 +89,7 @@ XE_HPG_CORETEST_F(CmdsProgrammingTestsXeHpgCore, givenL1CachingOverrideWhenState
memoryManager->freeGraphicsMemory(allocation);
}
XE_HPG_CORETEST_F(CmdsProgrammingTestsXeHpgCore, whenAppendingRssThenProgramWBPL1CachePolicy) {
XE_HPG_CORETEST_F(CmdsProgrammingTestsXeHpgCore, givenXeHpgCoreWhenAppendingRssThenProgramWBPL1CachePolicy) {
auto memoryManager = pDevice->getExecutionEnvironment()->memoryManager.get();
size_t allocationSize = MemoryConstants::pageSize;
AllocationProperties properties(pDevice->getRootDeviceIndex(), allocationSize, AllocationType::BUFFER, pDevice->getDeviceBitfield());
@ -141,10 +141,10 @@ XE_HPG_CORETEST_F(CmdsProgrammingTestsXeHpgCore, givenAlignedCacheableReadOnlyBu
const auto actualMocs = surfaceState.getMemoryObjectControlState();
EXPECT_EQ(expectedMocs, actualMocs);
auto actualL1CachePolocy = static_cast<uint8_t>(surfaceState.getL1CachePolicyL1CacheControl());
auto actualL1CachePolicy = static_cast<uint8_t>(surfaceState.getL1CachePolicyL1CacheControl());
const uint8_t expectedL1CachePolicy = 0;
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolocy);
EXPECT_EQ(expectedL1CachePolicy, actualL1CachePolicy);
alignedFree(ptr);
}