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

feature: optimize local ids generation 

- only emit local ids for required dimensions

Related-To: NEO-15007

Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
This commit is contained in:
Mateusz Hoppe 2025-05-23 08:23:29 +00:00 committed by Compute-Runtime-Automation
parent 0d57dcfdf0
commit 8e5b29f55e
2 changed files with 68 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@ -653,8 +653,15 @@ void EncodeDispatchKernel<Family>::encodeThreadData(WalkerType &walkerCmd,
// so whenever local ids are driver or hw generated, reserve space by setting right values for emitLocalIds
// 2) Auto-generation of local ids should be possible, when in fact local ids are used
if (!localIdsGenerationByRuntime && localIdDimensions > 0) {
UNRECOVERABLE_IF(localIdDimensions != 3);
uint32_t emitLocalIdsForDim = (1 << 0) | (1 << 1) | (1 << 2);
UNRECOVERABLE_IF(localIdDimensions > 3);
uint32_t emitLocalIdsForDim = (1 << 0);
if (localIdDimensions > 1) {
emitLocalIdsForDim |= (1 << 1);
}
if (localIdDimensions > 2) {
emitLocalIdsForDim |= (1 << 2);
}
walkerCmd.setEmitLocalId(emitLocalIdsForDim);
walkerCmd.setLocalXMaximum(static_cast<uint32_t>(workGroupSizes[0] - 1));

67
shared/test/unit_test/encoders/test_encode_dispatch_kernel_xehp_and_later.cpp
View File

@ -673,6 +673,9 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, CommandEncodeStatesTest, givenInlineDataRequiredAnd
dispatchInterface->requiredWalkGroupOrder = 2u;
dispatchInterface->kernelDescriptor.kernelAttributes.flags.passInlineData = true;
dispatchInterface->kernelDescriptor.kernelAttributes.numLocalIdChannels = 3u;
dispatchInterface->kernelDescriptor.kernelAttributes.localId[0] = 1;
dispatchInterface->kernelDescriptor.kernelAttributes.localId[1] = 1;
dispatchInterface->kernelDescriptor.kernelAttributes.localId[2] = 1;
dispatchInterface->kernelDescriptor.kernelAttributes.simdSize = 32u;
dispatchInterface->getCrossThreadDataSizeResult = 32u;
@ -943,7 +946,57 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerThreadTestXeHPAndLater, givenLocalIdGeneratio
workGroupSizes[1] = workGroupSizes[2] = 2u;
MockExecutionEnvironment executionEnvironment{};
auto &rootDeviceEnvironment = *executionEnvironment.rootDeviceEnvironments[0];
EncodeDispatchKernel<FamilyType>::encodeThreadData(walkerCmd, nullptr, numWorkGroups, workGroupSizes, simd, localIdDimensions,
uint8_t localIdDims[3] = {2,
1,
3};
uint32_t expectedEmitLocalIds[3] = {(1 << 0) | (1 << 1),
(1 << 0),
(1 << 0) | (1 << 1) | (1 << 2)};
for (int i = 0; i < 3; i++) {
EncodeDispatchKernel<FamilyType>::encodeThreadData(walkerCmd, nullptr, numWorkGroups, workGroupSizes, simd, localIdDims[i],
0, 0, false, false, false, requiredWorkGroupOrder, rootDeviceEnvironment);
EXPECT_FALSE(walkerCmd.getIndirectParameterEnable());
EXPECT_EQ(1u, walkerCmd.getThreadGroupIdXDimension());
EXPECT_EQ(1u, walkerCmd.getThreadGroupIdYDimension());
EXPECT_EQ(1u, walkerCmd.getThreadGroupIdZDimension());
EXPECT_EQ(0u, walkerCmd.getThreadGroupIdStartingX());
EXPECT_EQ(0u, walkerCmd.getThreadGroupIdStartingY());
EXPECT_EQ(0u, walkerCmd.getThreadGroupIdStartingZ());
auto expectedSimd = getSimdConfig<DefaultWalkerType>(simd);
EXPECT_EQ(expectedSimd, walkerCmd.getSimdSize());
EXPECT_EQ(expectedSimd, walkerCmd.getMessageSimd());
EXPECT_EQ(0xffffffffu, walkerCmd.getExecutionMask());
EXPECT_EQ(expectedEmitLocalIds[i], walkerCmd.getEmitLocalId());
EXPECT_EQ(31u, walkerCmd.getLocalXMaximum());
EXPECT_EQ(1u, walkerCmd.getLocalYMaximum());
EXPECT_EQ(1u, walkerCmd.getLocalZMaximum());
EXPECT_EQ(2u, walkerCmd.getWalkOrder());
EXPECT_TRUE(walkerCmd.getGenerateLocalId());
EXPECT_FALSE(walkerCmd.getEmitInlineParameter());
}
}
HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerThreadTestXeHPAndLater, givenLocalIdGenerationByHwWhenLocalIdsNotPresentThenEmitLocalIdsIsNotSet) {
using DefaultWalkerType = typename FamilyType::DefaultWalkerType;
DefaultWalkerType walkerCmd = FamilyType::template getInitGpuWalker<DefaultWalkerType>();
requiredWorkGroupOrder = 2u;
workGroupSizes[1] = workGroupSizes[2] = 2u;
MockExecutionEnvironment executionEnvironment{};
auto &rootDeviceEnvironment = *executionEnvironment.rootDeviceEnvironments[0];
uint8_t localIdDims = 0;
uint32_t expectedEmitLocalIds = 0;
EncodeDispatchKernel<FamilyType>::encodeThreadData(walkerCmd, nullptr, numWorkGroups, workGroupSizes, simd, localIdDims,
0, 0, false, false, false, requiredWorkGroupOrder, rootDeviceEnvironment);
EXPECT_FALSE(walkerCmd.getIndirectParameterEnable());
EXPECT_EQ(1u, walkerCmd.getThreadGroupIdXDimension());
@ -960,14 +1013,12 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerThreadTestXeHPAndLater, givenLocalIdGeneratio
EXPECT_EQ(0xffffffffu, walkerCmd.getExecutionMask());
uint32_t expectedEmitLocalIds = (1 << 0) | (1 << 1) | (1 << 2);
EXPECT_EQ(expectedEmitLocalIds, walkerCmd.getEmitLocalId());
EXPECT_EQ(31u, walkerCmd.getLocalXMaximum());
EXPECT_EQ(1u, walkerCmd.getLocalYMaximum());
EXPECT_EQ(1u, walkerCmd.getLocalZMaximum());
EXPECT_EQ(2u, walkerCmd.getWalkOrder());
EXPECT_TRUE(walkerCmd.getGenerateLocalId());
EXPECT_EQ(0u, walkerCmd.getLocalXMaximum());
EXPECT_EQ(0u, walkerCmd.getLocalYMaximum());
EXPECT_EQ(0u, walkerCmd.getLocalZMaximum());
EXPECT_EQ(0u, walkerCmd.getWalkOrder());
EXPECT_FALSE(walkerCmd.getGenerateLocalId());
EXPECT_FALSE(walkerCmd.getEmitInlineParameter());
}