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

refactor: unify programming of preferred slm size 1/n 

- rename function to reflect actual task function does

Related-To: NEO-12639

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz 2024-10-03 16:26:19 +00:00 committed by Compute-Runtime-Automation
parent e5658e5327
commit 49371cb13e
11 changed files with 25 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
opencl/source/helpers/hardware_commands_helper_base.inl
View File

@ -162,8 +162,8 @@ size_t HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(
EncodeDispatchKernel<GfxFamily>::setGrfInfo(&interfaceDescriptor, kernelDescriptor.kernelAttributes.numGrfRequired,
sizeCrossThreadData, sizePerThreadData, device.getRootDeviceEnvironment());
EncodeDispatchKernel<GfxFamily>::appendAdditionalIDDFields(&interfaceDescriptor, device.getRootDeviceEnvironment(),
threadsPerThreadGroup, slmTotalSize, SlmPolicy::slmPolicyNone);
EncodeDispatchKernel<GfxFamily>::setupPreferredSlmSize(&interfaceDescriptor, device.getRootDeviceEnvironment(),
threadsPerThreadGroup, slmTotalSize, SlmPolicy::slmPolicyNone);
if constexpr (heaplessModeEnabled == false) {
interfaceDescriptor.setBindingTablePointer(static_cast<uint32_t>(bindingTablePointer));

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

@ -129,8 +129,8 @@ struct EncodeDispatchKernel {
static void encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs);
template <typename InterfaceDescriptorType>
static void appendAdditionalIDDFields(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment,
const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy);
static void setupPreferredSlmSize(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment,
const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy);
template <typename InterfaceDescriptorType>
static void encodeEuSchedulingPolicy(InterfaceDescriptorType *pInterfaceDescriptor, const KernelDescriptor &kernelDesc, int32_t defaultPipelinedThreadArbitrationPolicy);

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

@ -411,7 +411,7 @@ inline void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const Roo
template <typename Family>
template <typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {}
void EncodeDispatchKernel<Family>::setupPreferredSlmSize(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {}
template <typename Family>
inline bool EncodeDispatchKernel<Family>::isDshNeeded(const DeviceInfo &deviceInfo) {

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

@ -15,7 +15,7 @@ template void NEO::EncodeDispatchKernel<Family>::adjustTimestampPacket<Family::D
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);
template void NEO::EncodeDispatchKernel<Family>::setGrfInfo<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, uint32_t grfCount, const size_t &sizeCrossThreadData, const size_t &sizePerThreadData, const RootDeviceEnvironment &rootDeviceEnvironment);
template void NEO::EncodeDispatchKernel<Family>::appendAdditionalIDDFields<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy);
template void NEO::EncodeDispatchKernel<Family>::setupPreferredSlmSize<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy);
template void NEO::EncodeDispatchKernel<Family>::adjustInterfaceDescriptorData<Family::DefaultWalkerType, Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, const Device &device, const HardwareInfo &hwInfo, const uint32_t threadGroupCount, const uint32_t grfCount, Family::DefaultWalkerType &walkerCmd);
template void NEO::EncodeDispatchKernel<Family>::setupPostSyncMocs<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const RootDeviceEnvironment &rootDeviceEnvironment, bool dcFlush);
template void NEO::EncodeDispatchKernel<Family>::encode<Family::DefaultWalkerType>(CommandContainer &container, EncodeDispatchKernelArgs &args);

6
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@ -397,9 +397,9 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
idd.getThreadGroupDispatchSize());
}
EncodeDispatchKernel<Family>::appendAdditionalIDDFields(&idd, rootDeviceEnvironment, threadsPerThreadGroup,
args.dispatchInterface->getSlmTotalSize(),
args.dispatchInterface->getSlmPolicy());
EncodeDispatchKernel<Family>::setupPreferredSlmSize(&idd, rootDeviceEnvironment, threadsPerThreadGroup,
args.dispatchInterface->getSlmTotalSize(),
args.dispatchInterface->getSlmPolicy());
EncodeWalkerArgs walkerArgs{
args.isCooperative ? KernelExecutionType::concurrent : KernelExecutionType::defaultType,

2
shared/source/xe2_hpg_core/command_encoder_xe2_hpg_core.cpp
View File

@ -255,7 +255,7 @@ void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDevice
template <>
template <typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {
void EncodeDispatchKernel<Family>::setupPreferredSlmSize(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {
using PREFERRED_SLM_ALLOCATION_SIZE = typename InterfaceDescriptorType::PREFERRED_SLM_ALLOCATION_SIZE;
auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();

2
shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp
View File

@ -187,7 +187,7 @@ void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDevice
template <>
template <typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {
void EncodeDispatchKernel<Family>::setupPreferredSlmSize(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {
using PREFERRED_SLM_ALLOCATION_SIZE = typename InterfaceDescriptorType::PREFERRED_SLM_ALLOCATION_SIZE;
auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
const uint32_t threadsPerDssCount = hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.DualSubSliceCount;

2
shared/source/xe_hpg_core/command_encoder_xe_hpg_core.cpp
View File

@ -29,7 +29,7 @@ namespace NEO {
template <>
template <typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {
void EncodeDispatchKernel<Family>::setupPreferredSlmSize(InterfaceDescriptorType *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {
using PREFERRED_SLM_ALLOCATION_SIZE = typename InterfaceDescriptorType::PREFERRED_SLM_ALLOCATION_SIZE;
auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
const uint32_t threadsPerDssCount = hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.DualSubSliceCount;

12
shared/test/unit_test/encoders/test_encode_dispatch_kernel_dg2_and_later.h
View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2023 Intel Corporation
* Copyright (C) 2021-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@ -44,11 +44,11 @@ void verifyPreferredSlmValues(std::vector<PreferredSlmTestValues<FamilyType>> va
? valueToTest.preferredSlmAllocationSizePerDss
: valueToTest.preferredSlmAllocationSizePerDss / localWorkGroupsPerDssCount;
NEO::EncodeDispatchKernel<FamilyType>::appendAdditionalIDDFields(&idd,
rootDeviceEnvironment,
threadsPerThreadGroup,
slmTotalSize,
slmPolicy);
NEO::EncodeDispatchKernel<FamilyType>::setupPreferredSlmSize(&idd,
rootDeviceEnvironment,
threadsPerThreadGroup,
slmTotalSize,
slmPolicy);
EXPECT_EQ(valueToTest.expectedValueInIdd, idd.getPreferredSlmAllocationSize());
}

4
shared/test/unit_test/xe_hpc_core/pvc/test_encode_dispatch_kernel_pvc.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2023 Intel Corporation
* Copyright (C) 2021-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@ -50,7 +50,7 @@ PVCTEST_F(CommandEncodeStatesPvcTest, GivenSmallSlmTotalSizesWhenSetAdditionalIn
hwInfo.platform.usDeviceID = deviceId;
hwInfo.platform.usRevId = revisionToTest.revisionId;
INTERFACE_DESCRIPTOR_DATA idd = FamilyType::cmdInitInterfaceDescriptorData;
EncodeDispatchKernel<FamilyType>::appendAdditionalIDDFields(&idd, rootDeviceEnvironment, threadsCount, slmTotalSize, SlmPolicy::slmPolicyNone);
EncodeDispatchKernel<FamilyType>::setupPreferredSlmSize(&idd, rootDeviceEnvironment, threadsCount, slmTotalSize, SlmPolicy::slmPolicyNone);
if (revisionToTest.isWaRequired) {
EXPECT_EQ(PREFERRED_SLM_ALLOCATION_SIZE::PREFERRED_SLM_ALLOCATION_SIZE_16K, idd.getPreferredSlmAllocationSize());
} else {

10
shared/test/unit_test/xe_hpg_core/dg2/test_encode_dispatch_kernel_dg2.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2023 Intel Corporation
* Copyright (C) 2021-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@ -41,28 +41,28 @@ DG2TEST_F(CommandEncodeStatesDg2Test, givenNoWorkaroundNeededWhenSelectingPrefer
const uint32_t threadsPerThreadGroup = 7; // 18 groups will fit in one DSS
const uint32_t slmSizePerThreadGroup = 2 * MemoryConstants::kiloByte;
INTERFACE_DESCRIPTOR_DATA idd = FamilyType::cmdInitInterfaceDescriptorData;
EncodeDispatchKernel<FamilyType>::appendAdditionalIDDFields(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EncodeDispatchKernel<FamilyType>::setupPreferredSlmSize(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EXPECT_EQ(PREFERRED_SLM_ALLOCATION_SIZE::PREFERRED_SLM_ALLOCATION_SIZE_64K, idd.getPreferredSlmAllocationSize());
}
{
const uint32_t threadsPerThreadGroup = 8; // 16 groups will fit in one DSS
const uint32_t slmSizePerThreadGroup = 2 * MemoryConstants::kiloByte;
INTERFACE_DESCRIPTOR_DATA idd = FamilyType::cmdInitInterfaceDescriptorData;
EncodeDispatchKernel<FamilyType>::appendAdditionalIDDFields(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EncodeDispatchKernel<FamilyType>::setupPreferredSlmSize(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EXPECT_EQ(PREFERRED_SLM_ALLOCATION_SIZE::PREFERRED_SLM_ALLOCATION_SIZE_32K, idd.getPreferredSlmAllocationSize());
}
{
const uint32_t threadsPerThreadGroup = 9; // 14 groups will fit in one DSS
const uint32_t slmSizePerThreadGroup = 2 * MemoryConstants::kiloByte;
INTERFACE_DESCRIPTOR_DATA idd = FamilyType::cmdInitInterfaceDescriptorData;
EncodeDispatchKernel<FamilyType>::appendAdditionalIDDFields(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EncodeDispatchKernel<FamilyType>::setupPreferredSlmSize(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EXPECT_EQ(PREFERRED_SLM_ALLOCATION_SIZE::PREFERRED_SLM_ALLOCATION_SIZE_32K, idd.getPreferredSlmAllocationSize());
}
{
const uint32_t threadsPerThreadGroup = 50; // 2 groups will fit in one DSS
const uint32_t slmSizePerThreadGroup = 16 * MemoryConstants::kiloByte;
INTERFACE_DESCRIPTOR_DATA idd = FamilyType::cmdInitInterfaceDescriptorData;
EncodeDispatchKernel<FamilyType>::appendAdditionalIDDFields(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EncodeDispatchKernel<FamilyType>::setupPreferredSlmSize(&idd, rootDeviceEnvironment, threadsPerThreadGroup, slmSizePerThreadGroup, SlmPolicy::slmPolicyLargeSlm);
EXPECT_EQ(PREFERRED_SLM_ALLOCATION_SIZE::PREFERRED_SLM_ALLOCATION_SIZE_32K, idd.getPreferredSlmAllocationSize());
}
}