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

fix: add scratch ptr in implicit args patching for L0 immediate cmdlists 

Related-To: NEO-11874
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
This commit is contained in:
Kamil Kopryk 2024-07-04 16:28:20 +00:00 committed by Compute-Runtime-Automation
parent 922286633b
commit 880aaee16c
4 changed files with 25 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
shared/source/command_container/command_encoder.h
View File

@ -43,6 +43,7 @@ struct PipelineSelectArgs;
struct RootDeviceEnvironment; struct RootDeviceEnvironment;
struct StateBaseAddressProperties; struct StateBaseAddressProperties;
struct StateComputeModeProperties; struct StateComputeModeProperties;
struct ImplicitArgs;
struct EncodeDispatchKernelArgs { struct EncodeDispatchKernelArgs {
uint64_t eventAddress = 0; uint64_t eventAddress = 0;
@ -187,7 +188,7 @@ struct EncodeDispatchKernel {
static void adjustWalkOrder(WalkerType &walkerCmd, uint32_t requiredWorkGroupOrder, const RootDeviceEnvironment &rootDeviceEnvironment); static void adjustWalkOrder(WalkerType &walkerCmd, uint32_t requiredWorkGroupOrder, const RootDeviceEnvironment &rootDeviceEnvironment);
template <bool heaplessModeEnabled> template <bool heaplessModeEnabled>
static void programInlineDataHeapless(uint8_t *inlineDataPtr, EncodeDispatchKernelArgs &args, CommandContainer &container, uint64_t offsetThreadData); static void programInlineDataHeapless(uint8_t *inlineDataPtr, EncodeDispatchKernelArgs &args, CommandContainer &container, uint64_t offsetThreadData, uint64_t scratchPtr);
static size_t getSizeRequiredDsh(const KernelDescriptor &kernelDescriptor, uint32_t iddCount); static size_t getSizeRequiredDsh(const KernelDescriptor &kernelDescriptor, uint32_t iddCount);
static size_t getSizeRequiredSsh(const KernelInfo &kernelInfo); static size_t getSizeRequiredSsh(const KernelInfo &kernelInfo);
@ -201,6 +202,10 @@ struct EncodeDispatchKernel {
template <bool isHeapless> template <bool isHeapless>
static void setScratchAddress(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr); static void setScratchAddress(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr);
template <bool isHeapless>
static uint64_t getScratchAddressForImmediatePatching(CommandContainer &container, EncodeDispatchKernelArgs &args);
template <bool isHeapless>
static void patchScratchAddressInImplicitArgs(ImplicitArgs &implicitArgs, uint64_t scratchAddress, bool scratchPtrPatchingRequired);
static size_t getInlineDataOffset(EncodeDispatchKernelArgs &args); static size_t getInlineDataOffset(EncodeDispatchKernelArgs &args);
}; };

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

@ -24,8 +24,10 @@ template void NEO::EncodeDispatchKernel<Family>::adjustWalkOrder<Family::Default
template void NEO::EncodeDispatchKernel<Family>::programBarrierEnable<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, uint32_t value, const HardwareInfo &hwInfo); template void NEO::EncodeDispatchKernel<Family>::programBarrierEnable<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, uint32_t value, const HardwareInfo &hwInfo);
template void NEO::EncodeDispatchKernel<Family>::setScratchAddress<false>(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr); template void NEO::EncodeDispatchKernel<Family>::setScratchAddress<false>(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr);
template void NEO::EncodeDispatchKernel<Family>::setScratchAddress<true>(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr); template void NEO::EncodeDispatchKernel<Family>::setScratchAddress<true>(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr);
template void NEO::EncodeDispatchKernel<Family>::programInlineDataHeapless<false>(uint8_t *inlineDataPtr, EncodeDispatchKernelArgs &args, CommandContainer &container, uint64_t offsetThreadData); template void NEO::EncodeDispatchKernel<Family>::programInlineDataHeapless<false>(uint8_t *inlineDataPtr, EncodeDispatchKernelArgs &args, CommandContainer &container, uint64_t offsetThreadData, uint64_t scratchPtr);
template void NEO::EncodeDispatchKernel<Family>::encodeEuSchedulingPolicy<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const KernelDescriptor &kernelDesc, int32_t defaultPipelinedThreadArbitrationPolicy); template void NEO::EncodeDispatchKernel<Family>::encodeEuSchedulingPolicy<Family::INTERFACE_DESCRIPTOR_DATA>(Family::INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const KernelDescriptor &kernelDesc, int32_t defaultPipelinedThreadArbitrationPolicy);
template uint64_t NEO::EncodeDispatchKernel<Family>::getScratchAddressForImmediatePatching<false>(CommandContainer &container, EncodeDispatchKernelArgs &args);
template void NEO::EncodeDispatchKernel<Family>::patchScratchAddressInImplicitArgs<false>(ImplicitArgs &implicitArgs, uint64_t scratchAddress, bool scratchPtrPatchingRequired);
template struct NEO::EncodeStates<Family>; template struct NEO::EncodeStates<Family>;
template struct NEO::EncodeMath<Family>; template struct NEO::EncodeMath<Family>;

13
shared/source/command_container/command_encoder_heap_addressing.inl
View File

@ -10,7 +10,18 @@
namespace NEO { namespace NEO {
template <typename Family> template <typename Family>
template <bool heaplessModeEnabled> template <bool heaplessModeEnabled>
void EncodeDispatchKernel<Family>::programInlineDataHeapless(uint8_t *inlineDataPtr, EncodeDispatchKernelArgs &args, CommandContainer &container, uint64_t offsetThreadData) { void EncodeDispatchKernel<Family>::programInlineDataHeapless(uint8_t *inlineDataPtr, EncodeDispatchKernelArgs &args, CommandContainer &container, uint64_t offsetThreadData, uint64_t scratchPtr) {
}
template <typename Family>
template <bool heaplessModeEnabled>
uint64_t EncodeDispatchKernel<Family>::getScratchAddressForImmediatePatching(CommandContainer &container, EncodeDispatchKernelArgs &args) {
return 0u;
}
template <typename Family>
template <bool heaplessModeEnabled>
void EncodeDispatchKernel<Family>::patchScratchAddressInImplicitArgs(ImplicitArgs &implicitArgs, uint64_t scratchAddress, bool scratchPtrRequired) {
} }
} // namespace NEO } // namespace NEO

5
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@ -239,6 +239,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
inlineDataProgramming = inlineDataProgrammingOffset != 0; inlineDataProgramming = inlineDataProgrammingOffset != 0;
} }
auto scratchAddressForImmediatePatching = EncodeDispatchKernel<Family>::getScratchAddressForImmediatePatching<heaplessModeEnabled>(container, args);
uint32_t sizeThreadData = sizePerThreadDataForWholeGroup + sizeCrossThreadData; uint32_t sizeThreadData = sizePerThreadDataForWholeGroup + sizeCrossThreadData;
uint32_t sizeForImplicitArgsPatching = NEO::ImplicitArgsHelper::getSizeForImplicitArgsPatching(pImplicitArgs, kernelDescriptor, !localIdsGenerationByRuntime, rootDeviceEnvironment); uint32_t sizeForImplicitArgsPatching = NEO::ImplicitArgsHelper::getSizeForImplicitArgsPatching(pImplicitArgs, kernelDescriptor, !localIdsGenerationByRuntime, rootDeviceEnvironment);
uint32_t iohRequiredSize = sizeThreadData + sizeForImplicitArgsPatching + args.reserveExtraPayloadSpace; uint32_t iohRequiredSize = sizeThreadData + sizeForImplicitArgsPatching + args.reserveExtraPayloadSpace;
@ -258,6 +259,8 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
if (pImplicitArgs) { if (pImplicitArgs) {
offsetThreadData -= ImplicitArgs::getSize(); offsetThreadData -= ImplicitArgs::getSize();
pImplicitArgs->localIdTablePtr = heap->getGraphicsAllocation()->getGpuAddress() + heap->getUsed() - iohRequiredSize; pImplicitArgs->localIdTablePtr = heap->getGraphicsAllocation()->getGpuAddress() + heap->getUsed() - iohRequiredSize;
EncodeDispatchKernel<Family>::patchScratchAddressInImplicitArgs<heaplessModeEnabled>(*pImplicitArgs, scratchAddressForImmediatePatching, args.immediateScratchAddressPatching);
ptr = NEO::ImplicitArgsHelper::patchImplicitArgs(ptr, *pImplicitArgs, kernelDescriptor, std::make_pair(localIdsGenerationByRuntime, requiredWorkgroupOrder), rootDeviceEnvironment); ptr = NEO::ImplicitArgsHelper::patchImplicitArgs(ptr, *pImplicitArgs, kernelDescriptor, std::make_pair(localIdsGenerationByRuntime, requiredWorkgroupOrder), rootDeviceEnvironment);
} }
@ -322,7 +325,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
} }
uint8_t *inlineData = reinterpret_cast<uint8_t *>(walkerCmd.getInlineDataPointer()); uint8_t *inlineData = reinterpret_cast<uint8_t *>(walkerCmd.getInlineDataPointer());
EncodeDispatchKernel<Family>::programInlineDataHeapless<heaplessModeEnabled>(inlineData, args, container, offsetThreadData); EncodeDispatchKernel<Family>::programInlineDataHeapless<heaplessModeEnabled>(inlineData, args, container, offsetThreadData, scratchAddressForImmediatePatching);
if constexpr (heaplessModeEnabled == false) { if constexpr (heaplessModeEnabled == false) {
walkerCmd.setIndirectDataStartAddress(static_cast<uint32_t>(offsetThreadData)); walkerCmd.setIndirectDataStartAddress(static_cast<uint32_t>(offsetThreadData));