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

Change dispatch kernel interface to provide already prepared heap objects

Browse Source 
Related-To: NEO-5055

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2023-01-26 14:33:36 +00:00
committed by Compute-Runtime-Automation
parent 401344137c
commit 5097ef4825
10 changed files with 106 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
shared/source/command_container/command_encoder_bdw_and_later.inl
View File

@@ -98,7 +98,10 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
if (!isBindlessKernel) {
container.prepareBindfulSsh();
if (bindingTableStateCount > 0u) {
auto ssh = container.getHeapWithRequiredSizeAndAlignment(HeapType::SURFACE_STATE, args.dispatchInterface->getSurfaceStateHeapDataSize(), BINDING_TABLE_STATE::SURFACESTATEPOINTER_ALIGN_SIZE);
auto ssh = args.surfaceStateHeap;
if (ssh == nullptr) {
ssh = container.getHeapWithRequiredSizeAndAlignment(HeapType::SURFACE_STATE, args.dispatchInterface->getSurfaceStateHeapDataSize(), BINDING_TABLE_STATE::SURFACESTATEPOINTER_ALIGN_SIZE);
}
bindingTablePointer = static_cast<uint32_t>(EncodeSurfaceState<Family>::pushBindingTableAndSurfaceStates(
*ssh, bindingTableStateCount,
args.dispatchInterface->getSurfaceStateHeapData(),
@@ -114,20 +117,22 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
uint32_t samplerCount = 0;
if (kernelDescriptor.payloadMappings.samplerTable.numSamplers > 0) {
if (!ApiSpecificConfig::getBindlessConfiguration()) {
auto heap = container.getIndirectHeap(HeapType::DYNAMIC_STATE);
auto dshSizeRequired = NEO::EncodeDispatchKernel<Family>::getSizeRequiredDsh(kernelDescriptor);
if (heap->getAvailableSpace() <= dshSizeRequired) {
heap = container.getHeapWithRequiredSizeAndAlignment(HeapType::DYNAMIC_STATE, heap->getMaxAvailableSpace(), 0);
UNRECOVERABLE_IF(!heap);
auto dsHeap = args.dynamicStateHeap;
if (dsHeap == nullptr) {
if (!ApiSpecificConfig::getBindlessConfiguration()) {
auto dsHeap = container.getIndirectHeap(HeapType::DYNAMIC_STATE);
auto dshSizeRequired = NEO::EncodeDispatchKernel<Family>::getSizeRequiredDsh(kernelDescriptor);
if (dsHeap->getAvailableSpace() <= dshSizeRequired) {
dsHeap = container.getHeapWithRequiredSizeAndAlignment(HeapType::DYNAMIC_STATE, dsHeap->getMaxAvailableSpace(), 0);
UNRECOVERABLE_IF(!dsHeap);
}
}
dsHeap = ApiSpecificConfig::getBindlessConfiguration() ? args.device->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : container.getIndirectHeap(HeapType::DYNAMIC_STATE);
}
auto heap = ApiSpecificConfig::getBindlessConfiguration() ? args.device->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : container.getIndirectHeap(HeapType::DYNAMIC_STATE);
UNRECOVERABLE_IF(!heap);
UNRECOVERABLE_IF(!dsHeap);
samplerCount = kernelDescriptor.payloadMappings.samplerTable.numSamplers;
samplerStateOffset = EncodeStates<Family>::copySamplerState(heap, kernelDescriptor.payloadMappings.samplerTable.tableOffset,
samplerStateOffset = EncodeStates<Family>::copySamplerState(dsHeap, kernelDescriptor.payloadMappings.samplerTable.tableOffset,
kernelDescriptor.payloadMappings.samplerTable.numSamplers,
kernelDescriptor.payloadMappings.samplerTable.borderColor,
args.dispatchInterface->getDynamicStateHeapData(),
@@ -382,6 +387,11 @@ inline void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const Roo
template <typename Family>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const RootDeviceEnvironment &rootDeviceEnvironment, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {}
template <typename Family>
inline bool EncodeDispatchKernel<Family>::isDshNeeded(const DeviceInfo &deviceInfo) {
return true;
}
template <typename Family>
inline void EncodeComputeMode<Family>::adjustPipelineSelect(CommandContainer &container, const NEO::KernelDescriptor &kernelDescriptor) {
}