intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add improvements to heap estimation in level zero command lists

Browse Source 
- add estimation parameter for interface descriptor data count
- add to the heap estimation alignment parameter for dynamic and surface heaps
- extend encode interface and implementations to allow child heaps

Related-To: NEO-5055

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2023-02-02 18:57:24 +00:00
committed by Compute-Runtime-Automation
parent 5a833e2c08
commit 7e0401d280
18 changed files with 154 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
shared/source/command_container/command_encoder_bdw_and_later.inl
View File

@@ -120,14 +120,14 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
auto dsHeap = args.dynamicStateHeap;
if (dsHeap == nullptr) {
if (!ApiSpecificConfig::getBindlessConfiguration()) {
auto dsHeap = container.getIndirectHeap(HeapType::DYNAMIC_STATE);
auto dshSizeRequired = NEO::EncodeDispatchKernel<Family>::getSizeRequiredDsh(kernelDescriptor);
dsHeap = container.getIndirectHeap(HeapType::DYNAMIC_STATE);
auto dshSizeRequired = NEO::EncodeDispatchKernel<Family>::getSizeRequiredDsh(kernelDescriptor, container.getNumIddPerBlock());
if (dsHeap->getAvailableSpace() <= dshSizeRequired) {
dsHeap = container.getHeapWithRequiredSizeAndAlignment(HeapType::DYNAMIC_STATE, dsHeap->getMaxAvailableSpace(), 0);
UNRECOVERABLE_IF(!dsHeap);
dsHeap = container.getHeapWithRequiredSizeAndAlignment(HeapType::DYNAMIC_STATE, dsHeap->getMaxAvailableSpace(), NEO::EncodeDispatchKernel<Family>::getDefaultDshAlignment());
}
} else {
dsHeap = args.device->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH);
}
dsHeap = ApiSpecificConfig::getBindlessConfiguration() ? args.device->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : container.getIndirectHeap(HeapType::DYNAMIC_STATE);
}
UNRECOVERABLE_IF(!dsHeap);
@@ -187,7 +187,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
}
uint32_t numIDD = 0u;
void *iddPtr = getInterfaceDescriptor(container, numIDD);
void *iddPtr = getInterfaceDescriptor(container, args.dynamicStateHeap, numIDD);
auto slmSizeNew = args.dispatchInterface->getSlmTotalSize();
bool dirtyHeaps = container.isAnyHeapDirty();
@@ -226,7 +226,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
}
if (numIDD == 0 || flush) {
EncodeMediaInterfaceDescriptorLoad<Family>::encode(container);
EncodeMediaInterfaceDescriptorLoad<Family>::encode(container, args.dynamicStateHeap);
}
cmd.setIndirectDataStartAddress(static_cast<uint32_t>(offsetThreadData));
@@ -291,10 +291,19 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
}
template <typename Family>
void EncodeMediaInterfaceDescriptorLoad<Family>::encode(CommandContainer &container) {
void EncodeMediaInterfaceDescriptorLoad<Family>::encode(CommandContainer &container, IndirectHeap *childDsh) {
using MEDIA_STATE_FLUSH = typename Family::MEDIA_STATE_FLUSH;
using MEDIA_INTERFACE_DESCRIPTOR_LOAD = typename Family::MEDIA_INTERFACE_DESCRIPTOR_LOAD;
auto heapBase = ApiSpecificConfig::getBindlessConfiguration() ? container.getDevice()->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH)->getGraphicsAllocation()->getUnderlyingBuffer() : container.getIndirectHeap(HeapType::DYNAMIC_STATE)->getCpuBase();
void *heapBase = nullptr;
if (childDsh != nullptr) {
heapBase = childDsh->getCpuBase();
} else {
if (ApiSpecificConfig::getBindlessConfiguration()) {
heapBase = container.getDevice()->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH)->getGraphicsAllocation()->getUnderlyingBuffer();
} else {
heapBase = container.getIndirectHeap(HeapType::DYNAMIC_STATE)->getCpuBase();
}
}
auto mediaStateFlush = container.getCommandStream()->getSpaceForCmd<MEDIA_STATE_FLUSH>();
*mediaStateFlush = Family::cmdInitMediaStateFlush;
@@ -570,8 +579,8 @@ template <typename Family>
void EncodeDispatchKernel<Family>::adjustWalkOrder(WALKER_TYPE &walkerCmd, uint32_t requiredWorkGroupOrder, const RootDeviceEnvironment &rootDeviceEnvironment) {}
template <typename Family>
uint32_t EncodeDispatchKernel<Family>::additionalSizeRequiredDsh() {
return sizeof(typename Family::INTERFACE_DESCRIPTOR_DATA);
size_t EncodeDispatchKernel<Family>::additionalSizeRequiredDsh(uint32_t iddCount) {
return iddCount * sizeof(typename Family::INTERFACE_DESCRIPTOR_DATA);
}
} // namespace NEO