intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: engineInstancedType removed

Browse Source 
Related-To: NEO-12594

Signed-off-by: Wojciech Konior <wojciech.konior@intel.com>
This commit is contained in:
Wojciech Konior
2024-10-08 13:23:41 +00:00
committed by Compute-Runtime-Automation
parent 834f00d48b
commit 6b40f9bc5a
65 changed files with 195 additions and 966 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
shared/source/helpers/engine_node_helper.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2019-2023 Intel Corporation
* Copyright (C) 2019-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -34,14 +34,13 @@ using EngineTypeUsage = std::pair<aub_stream::EngineType, EngineUsage>;
struct EngineDescriptor {
EngineDescriptor() = delete;
constexpr EngineDescriptor(EngineTypeUsage engineTypeUsage, DeviceBitfield deviceBitfield, PreemptionMode preemptionMode, bool isRootDevice, bool isEngineInstanced)
: engineTypeUsage(engineTypeUsage), deviceBitfield(deviceBitfield), preemptionMode(preemptionMode), isRootDevice(isRootDevice), isEngineInstanced(isEngineInstanced) {}
constexpr EngineDescriptor(EngineTypeUsage engineTypeUsage, DeviceBitfield deviceBitfield, PreemptionMode preemptionMode, bool isRootDevice)
: engineTypeUsage(engineTypeUsage), deviceBitfield(deviceBitfield), preemptionMode(preemptionMode), isRootDevice(isRootDevice) {}
EngineTypeUsage engineTypeUsage;
DeviceBitfield deviceBitfield;
PreemptionMode preemptionMode;
bool isRootDevice;
bool isEngineInstanced;
};
namespace EngineHelpers {

4
shared/source/helpers/gfx_core_helper.h
View File

@@ -124,7 +124,7 @@ class GfxCoreHelper {
virtual bool isRcsAvailable(const HardwareInfo &hwInfo) const = 0;
virtual bool isCooperativeDispatchSupported(const EngineGroupType engineGroupType, const RootDeviceEnvironment &rootDeviceEnvironment) const = 0;
virtual uint32_t adjustMaxWorkGroupCount(uint32_t maxWorkGroupCount, const EngineGroupType engineGroupType,
const RootDeviceEnvironment &rootDeviceEnvironment, bool isEngineInstanced) const = 0;
const RootDeviceEnvironment &rootDeviceEnvironment) const = 0;
virtual uint32_t adjustMaxWorkGroupSize(const uint32_t grfCount, const uint32_t simd, bool isHwLocalGeneration, const uint32_t defaultMaxGroupSize, const RootDeviceEnvironment &rootDeviceEnvironment) const = 0;
virtual size_t getMaxFillPaternSizeForCopyEngine() const = 0;
virtual size_t getSipKernelMaxDbgSurfaceSize(const HardwareInfo &hwInfo) const = 0;
@@ -347,7 +347,7 @@ class GfxCoreHelperHw : public GfxCoreHelper {
bool isCooperativeDispatchSupported(const EngineGroupType engineGroupType, const RootDeviceEnvironment &rootDeviceEnvironment) const override;
uint32_t adjustMaxWorkGroupCount(uint32_t maxWorkGroupCount, const EngineGroupType engineGroupType,
const RootDeviceEnvironment &rootDeviceEnvironment, bool isEngineInstanced) const override;
const RootDeviceEnvironment &rootDeviceEnvironment) const override;
uint32_t adjustMaxWorkGroupSize(const uint32_t grfCount, const uint32_t simd, bool isHwLocalGeneration, const uint32_t defaultMaxGroupSize, const RootDeviceEnvironment &rootDeviceEnvironment) const override;
size_t getMaxFillPaternSizeForCopyEngine() const override;

4
shared/source/helpers/gfx_core_helper_bdw_to_dg2.inl
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2023 Intel Corporation
* Copyright (C) 2021-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -30,7 +30,7 @@ bool GfxCoreHelperHw<GfxFamily>::isCooperativeDispatchSupported(const EngineGrou
template <typename GfxFamily>
uint32_t GfxCoreHelperHw<GfxFamily>::adjustMaxWorkGroupCount(uint32_t maxWorkGroupCount, const EngineGroupType engineGroupType,
const RootDeviceEnvironment &rootDeviceEnvironment, bool isEngineInstanced) const {
const RootDeviceEnvironment &rootDeviceEnvironment) const {
return maxWorkGroupCount;
}

5
shared/source/helpers/gfx_core_helper_pvc_and_later.inl
View File

@@ -50,7 +50,7 @@ bool GfxCoreHelperHw<Family>::isCooperativeDispatchSupported(const EngineGroupTy
template <typename Family>
uint32_t GfxCoreHelperHw<Family>::adjustMaxWorkGroupCount(uint32_t maxWorkGroupCount, const EngineGroupType engineGroupType,
const RootDeviceEnvironment &rootDeviceEnvironment, bool isEngineInstanced) const {
const RootDeviceEnvironment &rootDeviceEnvironment) const {
if ((debugManager.flags.ForceTheoreticalMaxWorkGroupCount.get()) ||
(debugManager.flags.OverrideMaxWorkGroupCount.get() != -1)) {
return maxWorkGroupCount;
@@ -61,8 +61,7 @@ uint32_t GfxCoreHelperHw<Family>::adjustMaxWorkGroupCount(uint32_t maxWorkGroupC
auto &productHelper = rootDeviceEnvironment.getHelper<ProductHelper>();
auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
bool requiresLimitation = productHelper.isCooperativeEngineSupported(hwInfo) &&
(engineGroupType != EngineGroupType::cooperativeCompute) &&
(!isEngineInstanced);
(engineGroupType != EngineGroupType::cooperativeCompute);
auto ccsCount = hwInfo.gtSystemInfo.CCSInfo.NumberOfCCSEnabled;
auto numberOfpartsInTileForConcurrentKernels = productHelper.getNumberOfPartsInTileForConcurrentKernel(ccsCount);

8
shared/source/helpers/kernel_helpers.cpp
View File

@@ -21,13 +21,13 @@
namespace NEO {
uint32_t KernelHelper::getMaxWorkGroupCount(const RootDeviceEnvironment &rootDeviceEnvironment, const KernelDescriptor &kernelDescriptor, uint32_t numSubDevices,
uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType, bool isEngineInstanced) {
uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType) {
return KernelHelper::getMaxWorkGroupCount(rootDeviceEnvironment, kernelDescriptor.kernelAttributes.numGrfRequired, kernelDescriptor.kernelAttributes.simdSize, kernelDescriptor.kernelAttributes.barrierCount,
numSubDevices, usedSlmSize, workDim, localWorkSize, engineGroupType, isEngineInstanced);
numSubDevices, usedSlmSize, workDim, localWorkSize, engineGroupType);
}
uint32_t KernelHelper::getMaxWorkGroupCount(const RootDeviceEnvironment &rootDeviceEnvironment, uint16_t numGrfRequired, uint8_t simdSize, uint8_t barrierCount,
uint32_t numSubDevices, uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType, bool isEngineInstanced) {
uint32_t numSubDevices, uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType) {
if (debugManager.flags.OverrideMaxWorkGroupCount.get() != -1) {
return static_cast<uint32_t>(debugManager.flags.OverrideMaxWorkGroupCount.get());
}
@@ -65,7 +65,7 @@ uint32_t KernelHelper::getMaxWorkGroupCount(const RootDeviceEnvironment &rootDev
maxWorkGroupsCount = std::min(maxWorkGroupsCount, maxWorkGroupsCountDueToSlm);
}
maxWorkGroupsCount = helper.adjustMaxWorkGroupCount(maxWorkGroupsCount, engineGroupType, rootDeviceEnvironment, isEngineInstanced);
maxWorkGroupsCount = helper.adjustMaxWorkGroupCount(maxWorkGroupsCount, engineGroupType, rootDeviceEnvironment);
if (!helper.singleTileExecImplicitScalingRequired(true)) {
maxWorkGroupsCount *= numSubDevices;

4
shared/source/helpers/kernel_helpers.h
View File

@@ -24,9 +24,9 @@ struct KernelHelper {
invalidKernel = 2
};
static uint32_t getMaxWorkGroupCount(const RootDeviceEnvironment &rootDeviceEnvironment, const KernelDescriptor &kernelDescriptor,
uint32_t numSubDevices, uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType, bool isEngineInstanced);
uint32_t numSubDevices, uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType);
static uint32_t getMaxWorkGroupCount(const RootDeviceEnvironment &rootDeviceEnvironment, uint16_t numGrfRequired, uint8_t simdSize, uint8_t barrierCount,
uint32_t numSubDevices, uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType, bool isEngineInstanced);
uint32_t numSubDevices, uint32_t usedSlmSize, uint32_t workDim, const size_t *localWorkSize, EngineGroupType engineGroupType);
static inline uint64_t getPrivateSurfaceSize(uint64_t perHwThreadPrivateMemorySize, uint32_t computeUnitsUsedForScratch) {
return perHwThreadPrivateMemorySize * computeUnitsUsedForScratch;
}