intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-20 08:53:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: Set dispatch all for small TG

Browse Source 
Resolves: NEO-11814

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2024-09-27 09:54:41 +00:00
committed by Compute-Runtime-Automation
parent ef1b569a85
commit 0dc2870513
19 changed files with 110 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/source/command_queue/hardware_interface_bdw_and_later.inl
View File

@@ -84,7 +84,7 @@ inline void HardwareInterface<GfxFamily>::programWalker(
commandQueue.getDevice());
EncodeWalkerArgs encodeWalkerArgs{kernel.getExecutionType(), false, kernel.getKernelInfo().kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0, 0};
EncodeDispatchKernel<GfxFamily>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, encodeWalkerArgs);
EncodeDispatchKernel<GfxFamily>::template encodeAdditionalWalkerFields<WalkerType, INTERFACE_DESCRIPTOR_DATA>(rootDeviceEnvironment, walkerCmd, nullptr, encodeWalkerArgs);
*walkerCmdBuf = walkerCmd;
}
} // namespace NEO

2
opencl/source/command_queue/hardware_interface_xehp_and_later.inl
View File

@@ -143,7 +143,7 @@ inline void HardwareInterface<GfxFamily>::programWalker(
bool requiredSystemFence = kernelSystemAllocation && walkerArgs.event != nullptr;
auto maxFrontEndThreads = device.getDeviceInfo().maxFrontEndThreads;
EncodeWalkerArgs encodeWalkerArgs{kernel.getExecutionType(), requiredSystemFence, kernelInfo.kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0, maxFrontEndThreads};
EncodeDispatchKernel<GfxFamily>::template encodeAdditionalWalkerFields<WalkerType>(rootDeviceEnvironment, walkerCmd, encodeWalkerArgs);
EncodeDispatchKernel<GfxFamily>::template encodeAdditionalWalkerFields<WalkerType>(rootDeviceEnvironment, walkerCmd, interfaceDescriptor, encodeWalkerArgs);
auto devices = queueCsr.getOsContext().getDeviceBitfield();
auto partitionWalker = ImplicitScalingHelper::isImplicitScalingEnabled(devices, true);

2
opencl/test/unit_test/command_queue/dispatch_walker_tests_dg2_and_later.cpp
View File

@@ -88,7 +88,7 @@ HWTEST2_F(WalkerDispatchTestDg2AndLater, givenDebugVariableSetWhenProgramCompute
EncodeWalkerArgs walkerArgs{KernelExecutionType::defaultType, true, kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0, 0};
for (auto forceL3PrefetchForComputeWalker : {false, true}) {
debugManager.flags.ForceL3PrefetchForComputeWalker.set(forceL3PrefetchForComputeWalker);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_EQ(!forceL3PrefetchForComputeWalker, walkerCmd.getL3PrefetchDisable());
}
}

1
opencl/test/unit_test/xe2_hpg_core/CMakeLists.txt
View File

@@ -17,7 +17,6 @@ if(TESTS_XE2_HPG_CORE)
${CMAKE_CURRENT_SOURCE_DIR}/cmd_parse_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/command_stream_receiver_hw_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/copy_engine_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/dispatch_walker_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/enqueue_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/test_cmds_programming_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/test_cl_device_caps_xe2_hpg_core.cpp

44
opencl/test/unit_test/xe2_hpg_core/dispatch_walker_tests_xe2_hpg_core.cpp
View File

@@ -1,44 +0,0 @@
/*
* Copyright (C) 2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_container/command_encoder.h"
#include "shared/source/kernel/kernel_descriptor.h"
#include "shared/source/os_interface/product_helper.h"
#include "shared/source/xe2_hpg_core/hw_cmds.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/helpers/default_hw_info.h"
#include "shared/test/common/mocks/mock_execution_environment.h"
#include "shared/test/common/test_macros/header/per_product_test_definitions.h"
#include "shared/test/common/test_macros/test.h"
using namespace NEO;
using WalkerDispatchTestsXe2HpGCore = ::testing::Test;
XE2_HPG_CORETEST_F(WalkerDispatchTestsXe2HpGCore, whenEncodeAdditionalWalkerFieldsIsCalledThenComputeDispatchAllIsCorrectlySet) {
using COMPUTE_WALKER = typename FamilyType::COMPUTE_WALKER;
DebugManagerStateRestore debugRestorer;
auto walkerCmd = FamilyType::cmdInitGpgpuWalker;
MockExecutionEnvironment mockExecutionEnvironment{};
auto &rootDeviceEnvironment = *mockExecutionEnvironment.rootDeviceEnvironments[0];
KernelDescriptor kernelDescriptor;
EncodeWalkerArgs walkerArgs{KernelExecutionType::concurrent, true, kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0, 0};
{
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EXPECT_TRUE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
walkerArgs.kernelExecutionType = KernelExecutionType::defaultType;
debugManager.flags.ComputeDispatchAllWalkerEnableInComputeWalker.set(1);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EXPECT_TRUE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
}

4
shared/source/command_container/command_encoder.h
View File

@@ -125,8 +125,8 @@ struct EncodeDispatchKernel {
template <typename WalkerType>
static void encode(CommandContainer &container, EncodeDispatchKernelArgs &args);
template <typename WalkerType>
static void encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs);
template <typename WalkerType, typename InterfaceDescriptorType>
static void encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const InterfaceDescriptorType *idd, const EncodeWalkerArgs &walkerArgs);
template <typename InterfaceDescriptorType>
static void appendAdditionalIDDFields(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment,

4
shared/source/command_container/command_encoder_bdw_and_later.inl
View File

@@ -406,8 +406,8 @@ void EncodeDispatchKernel<Family>::programBarrierEnable(InterfaceDescriptorType
}
template <typename Family>
template <typename WalkerType>
inline void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs) {}
template <typename WalkerType, typename InterfaceDescriptorType>
inline void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const InterfaceDescriptorType *idd, const EncodeWalkerArgs &walkerArgs) {}
template <typename Family>
template <typename InterfaceDescriptorType>

2
shared/source/command_container/command_encoder_enablers.inl
View File

@@ -10,7 +10,7 @@
#include "shared/source/command_container/command_encoder.h"
template struct NEO::EncodeDispatchKernel<Family>;
template void NEO::EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields<Family::DefaultWalkerType>(const RootDeviceEnvironment &rootDeviceEnvironment, Family::DefaultWalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs);
template void NEO::EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields<Family::DefaultWalkerType, Family::INTERFACE_DESCRIPTOR_DATA>(const RootDeviceEnvironment &rootDeviceEnvironment, Family::DefaultWalkerType &walkerCmd, const Family::INTERFACE_DESCRIPTOR_DATA *idd, const EncodeWalkerArgs &walkerArgs);
template void NEO::EncodeDispatchKernel<Family>::adjustTimestampPacket<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const EncodeDispatchKernelArgs &args);
template void NEO::EncodeDispatchKernel<Family>::setupPostSyncForRegularEvent<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const EncodeDispatchKernelArgs &args);
template void NEO::EncodeDispatchKernel<Family>::setupPostSyncForInOrderExec<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const EncodeDispatchKernelArgs &args);

2
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -408,7 +408,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
args.requiredDispatchWalkOrder,
args.additionalSizeParam,
args.device->getDeviceInfo().maxFrontEndThreads};
EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &idd, walkerArgs);
uint32_t workgroupSize = args.dispatchInterface->getGroupSize()[0] * args.dispatchInterface->getGroupSize()[1] * args.dispatchInterface->getGroupSize()[2];
bool isRequiredWorkGroupOrder = args.requiredDispatchWalkOrder != NEO::RequiredDispatchWalkOrder::none;

1
shared/source/execution_environment/execution_environment.cpp
View File

@@ -359,6 +359,7 @@ void ExecutionEnvironment::setDeviceHierarchy(const GfxCoreHelper &gfxCoreHelper
void ExecutionEnvironment::adjustCcsCountImpl(RootDeviceEnvironment *rootDeviceEnvironment) const {
auto hwInfo = rootDeviceEnvironment->getMutableHardwareInfo();
auto &productHelper = rootDeviceEnvironment->getHelper<ProductHelper>();
rootDeviceEnvironment->setNonLimitedNumberOfCcs(hwInfo->gtSystemInfo.CCSInfo.NumberOfCCSEnabled);
productHelper.adjustNumberOfCcs(*hwInfo);
}

3
shared/source/execution_environment/root_device_environment.h
View File

@@ -88,6 +88,8 @@ struct RootDeviceEnvironment : NonCopyableClass {
void initCompilerProductHelper();
void initReleaseHelper();
void initAilConfigurationHelper();
void setNonLimitedNumberOfCcs(uint32_t numberOfCss) { this->nonLimitedNumberOfCcs = numberOfCss; };
uint32_t getNonLimitedNumberOfCcs() const { return this->nonLimitedNumberOfCcs; };
ReleaseHelper *getReleaseHelper() const;
AILConfiguration *getAILConfigurationHelper() const;
template <typename HelperType>
@@ -124,6 +126,7 @@ struct RootDeviceEnvironment : NonCopyableClass {
protected:
using GraphicsAllocationUniquePtrType = std::unique_ptr<GraphicsAllocation, std::function<void(GraphicsAllocation *)>>;
GraphicsAllocationUniquePtrType dummyAllocation = nullptr;
uint32_t nonLimitedNumberOfCcs = 0u;
bool limitedNumberOfCcs = false;
bool isWddmOnLinuxEnable = false;

10
shared/source/xe2_hpg_core/command_encoder_xe2_hpg_core.cpp
View File

@@ -230,8 +230,8 @@ void EncodeSurfaceState<Family>::disableCompressionFlags(R_SURFACE_STATE *surfac
}
template <>
template <typename WalkerType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs) {
template <typename WalkerType, typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const InterfaceDescriptorType *idd, const EncodeWalkerArgs &walkerArgs) {
auto programGlobalFenceAsPostSyncOperationInComputeWalker = walkerArgs.requiredSystemFence;
int32_t overrideProgramSystemMemoryFence = debugManager.flags.ProgramGlobalFenceAsPostSyncOperationInComputeWalker.get();
if (overrideProgramSystemMemoryFence != -1) {
@@ -245,7 +245,11 @@ void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDevice
walkerCmd.setL3PrefetchDisable(!forceL3PrefetchForComputeWalker);
}
bool computeDispatchAllWalkerEnable = walkerArgs.kernelExecutionType == KernelExecutionType::concurrent;
bool computeDispatchAllWalkerEnable = walkerArgs.kernelExecutionType == KernelExecutionType::concurrent || (rootDeviceEnvironment.getNonLimitedNumberOfCcs() > 1 &&
idd &&
idd->getThreadGroupDispatchSize() == InterfaceDescriptorType::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1 &&
walkerCmd.getThreadGroupIdXDimension() * walkerCmd.getThreadGroupIdYDimension() * walkerCmd.getThreadGroupIdZDimension() * idd->getNumberOfThreadsInGpgpuThreadGroup() < walkerArgs.maxFrontEndThreads);
int32_t overrideComputeDispatchAllWalkerEnable = debugManager.flags.ComputeDispatchAllWalkerEnableInComputeWalker.get();
if (overrideComputeDispatchAllWalkerEnable != -1) {
computeDispatchAllWalkerEnable = !!overrideComputeDispatchAllWalkerEnable;

4
shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp
View File

@@ -161,8 +161,8 @@ void EncodeDispatchKernel<Family>::programBarrierEnable(INTERFACE_DESCRIPTOR_DAT
}
template <>
template <typename WalkerType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs) {
template <typename WalkerType, typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const InterfaceDescriptorType *idd, const EncodeWalkerArgs &walkerArgs) {
const auto &productHelper = rootDeviceEnvironment.getHelper<ProductHelper>();
auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
auto programGlobalFenceAsPostSyncOperationInComputeWalker = productHelper.isGlobalFenceInCommandStreamRequired(hwInfo) &&

4
shared/source/xe_hpg_core/command_encoder_xe_hpg_core.cpp
View File

@@ -116,8 +116,8 @@ void EncodeDispatchKernel<Family>::programBarrierEnable(INTERFACE_DESCRIPTOR_DAT
}
template <>
template <typename WalkerType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs) {
template <typename WalkerType, typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const InterfaceDescriptorType *idd, const EncodeWalkerArgs &walkerArgs) {
auto *releaseHelper = rootDeviceEnvironment.getReleaseHelper();
bool l3PrefetchDisable = releaseHelper->isPrefetchDisablingRequired();
int32_t overrideL3PrefetchDisable = debugManager.flags.ForceL3PrefetchForComputeWalker.get();

1
shared/test/unit_test/xe2_hpg_core/CMakeLists.txt
View File

@@ -13,6 +13,7 @@ if(TESTS_XE2_HPG_CORE)
${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
${CMAKE_CURRENT_SOURCE_DIR}/aub_command_stream_receiver_2_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/compute_mode_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/dispatch_walker_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/gfx_core_helper_tests_xe2_hpg_core.cpp
${CMAKE_CURRENT_SOURCE_DIR}/hw_cmds_xe2_hpg_core_tests.cpp
${CMAKE_CURRENT_SOURCE_DIR}/image_surface_state_tests_xe2_hpg_core.cpp

77
shared/test/unit_test/xe2_hpg_core/dispatch_walker_tests_xe2_hpg_core.cpp Normal file
View File

@@ -0,0 +1,77 @@
/*
* Copyright (C) 2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_container/command_encoder.h"
#include "shared/source/kernel/kernel_descriptor.h"
#include "shared/source/os_interface/product_helper.h"
#include "shared/source/xe2_hpg_core/hw_cmds.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/helpers/default_hw_info.h"
#include "shared/test/common/mocks/mock_execution_environment.h"
#include "shared/test/common/test_macros/header/per_product_test_definitions.h"
#include "shared/test/common/test_macros/test.h"
using namespace NEO;
using WalkerDispatchTestsXe2HpGCore = ::testing::Test;
XE2_HPG_CORETEST_F(WalkerDispatchTestsXe2HpGCore, whenEncodeAdditionalWalkerFieldsIsCalledThenComputeDispatchAllIsCorrectlySet) {
using COMPUTE_WALKER = typename FamilyType::COMPUTE_WALKER;
DebugManagerStateRestore debugRestorer;
auto walkerCmd = FamilyType::cmdInitGpgpuWalker;
MockExecutionEnvironment mockExecutionEnvironment{};
auto &rootDeviceEnvironment = *mockExecutionEnvironment.rootDeviceEnvironments[0];
KernelDescriptor kernelDescriptor;
EncodeWalkerArgs walkerArgs{KernelExecutionType::concurrent, true, kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0, 113};
{
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_TRUE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
auto backupCcsNumber = rootDeviceEnvironment.getNonLimitedNumberOfCcs();
rootDeviceEnvironment.setNonLimitedNumberOfCcs(2);
walkerArgs.kernelExecutionType = KernelExecutionType::defaultType;
{
walkerCmd.getInterfaceDescriptor().setThreadGroupDispatchSize(FamilyType::INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_TRUE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
walkerArgs.maxFrontEndThreads = 0u;
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_FALSE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
walkerCmd.getInterfaceDescriptor().setThreadGroupDispatchSize(FamilyType::INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_2);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_FALSE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
walkerCmd.getInterfaceDescriptor().setThreadGroupDispatchSize(FamilyType::INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1);
EncodeDispatchKernel<FamilyType>::template encodeAdditionalWalkerFields<COMPUTE_WALKER, typename FamilyType::INTERFACE_DESCRIPTOR_DATA>(rootDeviceEnvironment, walkerCmd, nullptr, walkerArgs);
EXPECT_FALSE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
rootDeviceEnvironment.setNonLimitedNumberOfCcs(1);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_FALSE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
rootDeviceEnvironment.setNonLimitedNumberOfCcs(backupCcsNumber);
debugManager.flags.ComputeDispatchAllWalkerEnableInComputeWalker.set(1);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_TRUE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
}

6
shared/test/unit_test/xe_hpc_core/dispatch_walker_tests_xe_hpc_core.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2023 Intel Corporation
* Copyright (C) 2021-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -28,13 +28,13 @@ XE_HPC_CORETEST_F(WalkerDispatchTestsXeHpcCore, givenXeHpcWhenEncodeAdditionalWa
KernelDescriptor kernelDescriptor;
EncodeWalkerArgs walkerArgs{KernelExecutionType::defaultType, true, kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0};
{
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_FALSE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
{
debugManager.flags.ComputeDispatchAllWalkerEnableInComputeWalker.set(1);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_TRUE(walkerCmd.getComputeDispatchAllWalkerEnable());
}
}

6
shared/test/unit_test/xe_hpc_core/pvc/dispatch_walker_tests_pvc.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2022-2023 Intel Corporation
* Copyright (C) 2022-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -53,7 +53,7 @@ PVCTEST_F(WalkerDispatchTestsPvc, givenPvcWhenEncodeAdditionalWalkerFieldsThenPo
testInput.programGlobalFenceAsPostSyncOperationInComputeWalker);
postSyncData.setSystemMemoryFenceRequest(false);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_EQ(testInput.expectSystemMemoryFenceRequest, postSyncData.getSystemMemoryFenceRequest());
}
}
@@ -75,7 +75,7 @@ PVCTEST_F(WalkerDispatchTestsPvc, givenPvcSupportsSystemMemoryFenceWhenNoSystemF
hwInfo.platform.usDeviceID = deviceId;
postSyncData.setSystemMemoryFenceRequest(true);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
EXPECT_FALSE(postSyncData.getSystemMemoryFenceRequest());
}
}

4
shared/test/unit_test/xe_hpg_core/dg2/test_encode_dg2.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2023 Intel Corporation
* Copyright (C) 2021-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -86,7 +86,7 @@ DG2TEST_F(CommandEncodeDG2Test, whenProgramComputeWalkerThenApplyL3WAForDg2G10A0
hwInfo.ipVersion = compilerProductHelper.getHwIpVersion(hwInfo);
rootDeviceEnvironment.releaseHelper = ReleaseHelper::create(hwInfo.ipVersion);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
EncodeDispatchKernel<FamilyType>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, &walkerCmd.getInterfaceDescriptor(), walkerArgs);
if (DG2::isG10(hwInfo) && revisionID < revIdB0) {
EXPECT_TRUE(walkerCmd.getL3PrefetchDisable());