intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-19 06:24:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: change parameter names into more meaningful

Browse Source 
- change additional size into local region size
- change walk order into dispatch walk order to distinguish for local id walk

Related-To: NEO-13350

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-11-27 00:13:50 +00:00
committed by Compute-Runtime-Automation
parent b56d4324c5
commit 6b7235cd6c
26 changed files with 168 additions and 168 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
shared/source/command_container/command_encoder.h
View File

@@ -63,7 +63,7 @@ struct EncodeDispatchKernelArgs {
PreemptionMode preemptionMode = PreemptionMode::Initial;
NEO::RequiredPartitionDim requiredPartitionDim = NEO::RequiredPartitionDim::none;
NEO::RequiredDispatchWalkOrder requiredDispatchWalkOrder = NEO::RequiredDispatchWalkOrder::none;
uint32_t additionalSizeParam = NEO::additionalKernelLaunchSizeParamNotSet;
uint32_t localRegionSize = NEO::localRegionSizeParamNotSet;
uint32_t partitionCount = 0u;
uint32_t reserveExtraPayloadSpace = 0;
uint32_t maxWgCountPerTile = 0;
@@ -109,7 +109,7 @@ struct EncodeWalkerArgs {
const KernelDescriptor &kernelDescriptor;
KernelExecutionType kernelExecutionType = KernelExecutionType::defaultType;
NEO::RequiredDispatchWalkOrder requiredDispatchWalkOrder = NEO::RequiredDispatchWalkOrder::none;
uint32_t additionalSizeParam = NEO::additionalKernelLaunchSizeParamNotSet;
uint32_t localRegionSize = NEO::localRegionSizeParamNotSet;
uint32_t maxFrontEndThreads = 0;
bool requiredSystemFence = false;
};
@@ -188,7 +188,7 @@ struct EncodeDispatchKernel {
static void setupPostSyncForRegularEvent(WalkerType &walkerCmd, const EncodeDispatchKernelArgs &args);
template <typename WalkerType>
static void setWalkerRegionSettings(WalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount, uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredWalkOrder);
static void setWalkerRegionSettings(WalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount, uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredDispatchWalkOrder);
template <typename WalkerType>
static void setupPostSyncForInOrderExec(WalkerType &walkerCmd, const EncodeDispatchKernelArgs &args);

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

@@ -283,7 +283,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
kernelDescriptor, // kernelDescriptor
KernelExecutionType::defaultType, // kernelExecutionType
args.requiredDispatchWalkOrder, // requiredDispatchWalkOrder
args.additionalSizeParam, // additionalSizeParam
args.localRegionSize, // localRegionSize
args.device->getDeviceInfo().maxFrontEndThreads, // maxFrontEndThreads
args.requiresSystemMemoryFence()}; // requiredSystemFence
EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(rootDeviceEnvironment, cmd, walkerArgs);
@@ -684,7 +684,7 @@ void EncodeDispatchKernel<Family>::encodeEuSchedulingPolicy(InterfaceDescriptorT
template <typename Family>
template <typename WalkerType>
void EncodeDispatchKernel<Family>::setWalkerRegionSettings(WalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount, uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredWalkOrder) {}
void EncodeDispatchKernel<Family>::setWalkerRegionSettings(WalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount, uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredDispatchWalkOrder) {}
template <typename Family>
template <typename WalkerType>

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

@@ -30,7 +30,7 @@ template uint64_t NEO::EncodeDispatchKernel<Family>::getScratchAddressForImmedia
template void NEO::EncodeDispatchKernel<Family>::patchScratchAddressInImplicitArgs<false>(ImplicitArgs &implicitArgs, uint64_t scratchAddress, bool scratchPtrPatchingRequired);
template void NEO::EncodeDispatchKernel<Family>::forceComputeWalkerPostSyncFlushWithWrite<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd);
template void NEO::EncodeDispatchKernel<Family>::setWalkerRegionSettings<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount,
uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredWalkOrder);
uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredDispatchWalkOrder);
template void NEO::EncodeDispatchKernel<Family>::overrideDefaultValues<Family::DefaultWalkerType, Family::DefaultWalkerType::InterfaceDescriptorType>(Family::DefaultWalkerType &walkerCmd, Family::DefaultWalkerType::InterfaceDescriptorType &interfaceDescriptor);
template void NEO::EncodeDispatchKernel<Family>::encodeWalkerPostSyncFields<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs);
template void NEO::EncodeDispatchKernel<Family>::encodeComputeDispatchAllWalker<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs);

2
shared/source/command_container/command_encoder_from_xe_hpg_core_to_xe3_core.inl
View File

@@ -90,6 +90,6 @@ void EncodeDispatchKernel<Family>::adjustTimestampPacket(WalkerType &walkerCmd,
template <typename Family>
template <typename WalkerType>
void EncodeDispatchKernel<Family>::setWalkerRegionSettings(WalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount, uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredWalkOrder) {}
void EncodeDispatchKernel<Family>::setWalkerRegionSettings(WalkerType &walkerCmd, const NEO::Device &device, uint32_t partitionCount, uint32_t workgroupSize, uint32_t maxWgCountPerTile, bool requiredDispatchWalkOrder) {}
} // namespace NEO

8
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -402,7 +402,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
kernelDescriptor, // kernelDescriptor
kernelExecutionType, // kernelExecutionType
args.requiredDispatchWalkOrder, // requiredDispatchWalkOrder
args.additionalSizeParam, // additionalSizeParam
args.localRegionSize, // localRegionSize
args.device->getDeviceInfo().maxFrontEndThreads, // maxFrontEndThreads
args.requiresSystemMemoryFence()}; // requiresMemoryFence
EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, walkerArgs);
@@ -412,7 +412,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
EncodeDispatchKernel<Family>::overrideDefaultValues(walkerCmd, idd);
uint32_t workgroupSize = args.dispatchInterface->getGroupSize()[0] * args.dispatchInterface->getGroupSize()[1] * args.dispatchInterface->getGroupSize()[2];
bool isRequiredWorkGroupOrder = args.requiredDispatchWalkOrder != NEO::RequiredDispatchWalkOrder::none;
bool isRequiredDispatchWorkGroupOrder = args.requiredDispatchWalkOrder != NEO::RequiredDispatchWalkOrder::none;
if (args.partitionCount > 1 && !args.isInternal) {
const uint64_t workPartitionAllocationGpuVa = args.device->getDefaultEngine().commandStreamReceiver->getWorkPartitionAllocationGpuAddress();
@@ -429,7 +429,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
args.dcFlushEnable, // dcFlush
EncodeDispatchKernel<Family>::singleTileExecImplicitScalingRequired(args.isCooperative), // forceExecutionOnSingleTile
args.makeCommandView, // blockDispatchToCommandBuffer
isRequiredWorkGroupOrder}; // isRequiredWorkGroupOrder
isRequiredDispatchWorkGroupOrder}; // isRequiredDispatchWorkGroupOrder
ImplicitScalingDispatch<Family>::dispatchCommands(*listCmdBufferStream,
walkerCmd,
@@ -438,7 +438,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
args.partitionCount = implicitScalingArgs.partitionCount;
} else {
args.partitionCount = 1;
EncodeDispatchKernel<Family>::setWalkerRegionSettings(walkerCmd, *args.device, args.partitionCount, workgroupSize, args.maxWgCountPerTile, isRequiredWorkGroupOrder);
EncodeDispatchKernel<Family>::setWalkerRegionSettings(walkerCmd, *args.device, args.partitionCount, workgroupSize, args.maxWgCountPerTile, isRequiredDispatchWorkGroupOrder);
if (!args.makeCommandView) {
auto buffer = listCmdBufferStream->getSpaceForCmd<WalkerType>();

2
shared/source/command_container/implicit_scaling.h
View File

@@ -53,7 +53,7 @@ struct ImplicitScalingDispatchCommandArgs {
bool dcFlush = false;
bool forceExecutionOnSingleTile = false;
bool blockDispatchToCommandBuffer = false;
bool isRequiredWorkGroupOrder = false;
bool isRequiredDispatchWorkGroupOrder = false;
};
template <typename GfxFamily>

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

@@ -51,7 +51,7 @@ WalkerPartition::WalkerPartitionArgs prepareWalkerPartitionArgs(ImplicitScalingD
args.workgroupSize = dispatchCommandArgs.workgroupSize;
args.maxWgCountPerTile = dispatchCommandArgs.maxWgCountPerTile;
args.isRequiredWorkGroupOrder = dispatchCommandArgs.isRequiredWorkGroupOrder;
args.isRequiredDispatchWorkGroupOrder = dispatchCommandArgs.isRequiredDispatchWorkGroupOrder;
return args;
}

2
shared/source/command_container/walker_partition_interface.h
View File

@@ -35,7 +35,7 @@ struct WalkerPartitionArgs {
bool dcFlushEnable = false;
bool forceExecutionOnSingleTile = false;
bool blockDispatchToCommandBuffer = false;
bool isRequiredWorkGroupOrder = false;
bool isRequiredDispatchWorkGroupOrder = false;
};
inline constexpr uint32_t wparidCCSOffset = 0x221C;

2
shared/source/command_container/walker_partition_xehp_and_later.h
View File

@@ -531,7 +531,7 @@ void *programPartitionedWalker(void *&inputAddress, uint32_t &totalBytesProgramm
args.partitionCount,
args.workgroupSize,
args.maxWgCountPerTile,
args.isRequiredWorkGroupOrder);
args.isRequiredDispatchWorkGroupOrder);
appendWalkerFields<GfxFamily, WalkerType>(*inputWalker, args.tileCount, workgroupCount);
}