intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 09:03:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: remove local dispatch support

Browse Source 
Related-To: NEO-14561

Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
This commit is contained in:
Mateusz Hoppe
2025-07-23 16:40:40 +00:00
committed by Compute-Runtime-Automation
parent e3e1ae0def
commit 575fe7fe33
10 changed files with 2 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
level_zero/core/source/cmdlist/cmdlist.h
View File

@@ -432,10 +432,6 @@ struct CommandList : _ze_command_list_handle_t {
return engineGroupType;
}
bool getLocalDispatchSupport() const {
return localDispatchSupport;
}
bool isClosed() const {
return closedCmdList;
}
@@ -562,7 +558,6 @@ struct CommandList : _ze_command_list_handle_t {
bool scratchAddressPatchingEnabled = false;
bool taskCountUpdateFenceRequired = false;
bool statelessBuiltinsEnabled = false;
bool localDispatchSupport = false;
bool l3FlushAfterPostSyncRequired = false;
bool textureCacheFlushPending = false;
bool closedCmdList = false;

1
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -280,7 +280,6 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::initialize(Device *device, NEO
this->defaultPipelinedThreadArbitrationPolicy = NEO::debugManager.flags.OverrideThreadArbitrationPolicy.get();
}
this->statelessBuiltinsEnabled = compilerProductHelper.isForceToStatelessRequired();
this->localDispatchSupport = productHelper.getSupportedLocalDispatchSizes(hwInfo).size() > 0;
this->commandContainer.doubleSbaWaRef() = this->doubleSbaWa;
this->commandContainer.l1CachePolicyDataRef() = &this->l1CachePolicyData;

1
level_zero/core/source/kernel/kernel_imp.cpp
View File

@@ -1073,7 +1073,6 @@ ze_result_t KernelImp::initialize(const ze_kernel_desc_t *desc) {
const auto &gfxHelper = rootDeviceEnvironment.getHelper<NEO::GfxCoreHelper>();
this->heaplessEnabled = rootDeviceEnvironment.getHelper<NEO::CompilerProductHelper>().isHeaplessModeEnabled(hwInfo);
this->localDispatchSupport = productHelper.getSupportedLocalDispatchSizes(hwInfo).size() > 0;
bool platformImplicitScaling = gfxHelper.platformSupportsImplicitScaling(rootDeviceEnvironment);
this->implicitScalingEnabled = NEO::ImplicitScalingHelper::isImplicitScalingEnabled(deviceBitfield, platformImplicitScaling);

2
level_zero/core/source/mutable_cmdlist/mutable_cmdlist_imp.cpp
View File

@@ -188,7 +188,7 @@ ze_result_t MutableCommandListImp::addVariableDispatch(const NEO::KernelDescript
perThreadData = {reinterpret_cast<uint8_t *>(ptrOffset(iohCpuBase, kernelDispatch.offsets.perThreadOffset)), perThreadDataSize};
}
bool calcRegion = base->isHeaplessModeEnabled() && base->getLocalDispatchSupport();
bool calcRegion = false;
auto mutableIndirectData = std::make_unique<MutableIndirectData>(std::move(offsets), crossThreadData, perThreadData, inlineData);
kernelDispatch.varDispatch = std::make_unique<VariableDispatch>(&kernelDispatch, std::move(mutableIndirectData), mutableComputeWalker,
groupSize, groupCount, globalOffset, lastSlmArgumentVariable,

1
level_zero/core/test/unit_tests/mocks/mock_cmdlist.h
View File

@@ -321,7 +321,6 @@ struct WhiteBox<::L0::CommandListImp> : public ::L0::CommandListImp {
using BaseClass::isSyncModeQueue;
using BaseClass::isTbxMode;
using BaseClass::l3FlushAfterPostSyncRequired;
using BaseClass::localDispatchSupport;
using BaseClass::minimalSizeForBcsSplit;
using BaseClass::partitionCount;
using BaseClass::pipelineSelectStateTracking;

2
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_1.cpp
View File

@@ -3452,7 +3452,6 @@ HWTEST_F(CommandListCreateTests, givenNullEventWhenAppendEventAfterWalkerThenNot
TEST_F(CommandListCreateTests, givenCreatedCommandListWhenGettingTrackingFlagsThenDefaultValuseIsHwSupported) {
auto &rootDeviceEnvironment = device->getNEODevice()->getRootDeviceEnvironment();
auto &hwInfo = device->getNEODevice()->getHardwareInfo();
auto &l0GfxCoreHelper = rootDeviceEnvironment.getHelper<L0GfxCoreHelper>();
auto &productHelper = rootDeviceEnvironment.getHelper<NEO::ProductHelper>();
@@ -3486,7 +3485,6 @@ TEST_F(CommandListCreateTests, givenCreatedCommandListWhenGettingTrackingFlagsTh
EXPECT_EQ(commandList->heaplessModeEnabled, commandList->scratchAddressPatchingEnabled);
EXPECT_EQ(commandList->statelessBuiltinsEnabled, compilerProductHelper.isForceToStatelessRequired());
EXPECT_EQ((productHelper.getSupportedLocalDispatchSizes(hwInfo).size() > 0), commandList->getLocalDispatchSupport());
}
TEST(BuiltinTypeHelperTest, givenNonStatelessAndNonHeaplessWhenAdjustBuiltinTypeIsCalledThenCorrectBuiltinTypeIsReturned) {

5
level_zero/core/test/unit_tests/sources/mutable_cmdlist/tests/mutable_cmdlist_kernels_tests.cpp
View File

@@ -1557,7 +1557,6 @@ HWCMDTEST_F(IGFX_XE_HP_CORE,
HWCMDTEST_F(IGFX_XE_HP_CORE,
MutableCommandListKernelTest,
givenTwoMutationKernelsOneRegionBarrierWhenAppendingRegionBarrierFirstAndMutatingGroupCountThenCorrectVariableIsSet) {
static_cast<L0::ult::MockCommandList *>(this->mutableCommandList.get()->base)->localDispatchSupport = true;
mutableCommandIdDesc.flags = kernelIsaMutationFlags;
enableRegionBarrierBuffer(kernel1Bit);
@@ -1636,7 +1635,6 @@ HWCMDTEST_F(IGFX_XE_HP_CORE,
HWCMDTEST_F(IGFX_XE_HP_CORE,
MutableCommandListKernelTest,
givenTwoMutationKernelsOneRegionBarrierWhenAppendingRegionBarrierFirstAndMutatingSecondaryKernelThenCorrectVariableIsSet) {
static_cast<L0::ult::MockCommandList *>(this->mutableCommandList.get()->base)->localDispatchSupport = true;
mutableCommandIdDesc.flags = kernelIsaMutationFlags;
enableRegionBarrierBuffer(kernel1Bit);
@@ -1785,7 +1783,6 @@ HWCMDTEST_F(IGFX_XE_HP_CORE,
HWCMDTEST_F(IGFX_XE_HP_CORE,
MutableCommandListKernelTest,
givenTwoMutationKernelsOneRegionBarrierWhenAppendingRegularFirstAndMutatingRegionBarrierKernelThenCorrectVariableIsSet) {
static_cast<L0::ult::MockCommandList *>(this->mutableCommandList.get()->base)->localDispatchSupport = true;
mutableCommandIdDesc.flags = kernelIsaMutationFlags;
enableRegionBarrierBuffer(kernel1Bit);
@@ -1889,7 +1886,6 @@ HWCMDTEST_F(IGFX_XE_HP_CORE,
HWCMDTEST_F(IGFX_XE_HP_CORE,
MutableCommandListKernelTest,
givenTwoMutationRegionBarrierKernelsWhenAppendingRegionBarrierFirstAndMutatingRegionBarrierKernelWithBiggerGroupCountThenCorrectVariableIsSet) {
static_cast<L0::ult::MockCommandList *>(this->mutableCommandList.get()->base)->localDispatchSupport = true;
mutableCommandIdDesc.flags = kernelIsaMutationFlags;
enableRegionBarrierBuffer(kernelAllMask);
@@ -2038,7 +2034,6 @@ HWCMDTEST_F(IGFX_XE_HP_CORE,
HWCMDTEST_F(IGFX_XE_HP_CORE,
MutableCommandListKernelTest,
givenKernelWithSlmAndRegionBarrierKernelWhenAppendingRegularAndMutatingRegionBarrierSlmThenCorrectVariableIsSet) {
static_cast<L0::ult::MockCommandList *>(this->mutableCommandList.get()->base)->localDispatchSupport = true;
mutableCommandIdDesc.flags = kernelIsaMutationFlags;
resizeKernelArg(1);

1
level_zero/experimental/source/mutable_cmdlist/mutable_cmdlist_imp_experimental.cpp
View File

@@ -203,7 +203,6 @@ ze_result_t MutableCommandListImp::loadFromBinary(const uint8_t *pBinary, const
args.cmdListEngine = base->getEngineGroupType();
args.partitionCount = base->getPartitionCount();
args.heapless = this->getBase()->isHeaplessModeEnabled();
args.localDispatch = this->getBase()->getLocalDispatchSupport();
args.mutableWalkerCmds = &mutableWalkerCmds;
Program::Decoder::MclDecoder::decode(args);
if (base->getCmdListStateBaseAddressTracking()) {

2
level_zero/experimental/source/mutable_cmdlist/program/mcl_decoder.cpp
View File

@@ -133,7 +133,7 @@ void MclDecoder::decode(const MclDecoderArgs &args) {
dispatch.varDispatch = std::make_unique<VariableDispatch>(&dispatch, std::move(mutableIndirectData), mutableCommandWalker,
groupSize, groupCount, nullptr, nullptr, args.device->getHwInfo().capabilityTable.grfSize,
dispatchParams, args.partitionCount, args.cmdListEngine, (args.heapless && args.localDispatch));
dispatchParams, args.partitionCount, args.cmdListEngine, false);
}
}

1
level_zero/experimental/source/mutable_cmdlist/program/mcl_decoder.h
View File

@@ -52,7 +52,6 @@ struct MclDecoderArgs {
NEO::EngineGroupType cmdListEngine;
uint32_t partitionCount = 0;
bool heapless = false;
bool localDispatch = false;
};
struct VarInfo {