mirror of
https://github.com/intel/compute-runtime.git
synced 2025-12-21 01:04:57 +08:00
114 lines
5.7 KiB
C++
114 lines
5.7 KiB
C++
/*
|
|
* Copyright (C) 2021-2022 Intel Corporation
|
|
*
|
|
* SPDX-License-Identifier: MIT
|
|
*
|
|
*/
|
|
|
|
#include "shared/source/command_stream/csr_definitions.h"
|
|
#include "shared/source/debug_settings/debug_settings_manager.h"
|
|
#include "shared/source/gmm_helper/cache_settings_helper.h"
|
|
#include "shared/source/gmm_helper/client_context/gmm_client_context.h"
|
|
#include "shared/source/helpers/state_base_address_base.inl"
|
|
|
|
namespace NEO {
|
|
|
|
template <typename GfxFamily>
|
|
void setSbaStatelessCompressionParams(typename GfxFamily::STATE_BASE_ADDRESS *stateBaseAddress, MemoryCompressionState memoryCompressionState) {
|
|
using STATE_BASE_ADDRESS = typename GfxFamily::STATE_BASE_ADDRESS;
|
|
|
|
if (memoryCompressionState == MemoryCompressionState::Enabled) {
|
|
stateBaseAddress->setEnableMemoryCompressionForAllStatelessAccesses(STATE_BASE_ADDRESS::ENABLE_MEMORY_COMPRESSION_FOR_ALL_STATELESS_ACCESSES_ENABLED);
|
|
} else {
|
|
stateBaseAddress->setEnableMemoryCompressionForAllStatelessAccesses(STATE_BASE_ADDRESS::ENABLE_MEMORY_COMPRESSION_FOR_ALL_STATELESS_ACCESSES_DISABLED);
|
|
}
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
void StateBaseAddressHelper<GfxFamily>::appendStateBaseAddressParameters(
|
|
StateBaseAddressHelperArgs<GfxFamily> &args,
|
|
bool overrideBindlessSurfaceStateBase) {
|
|
using RENDER_SURFACE_STATE = typename GfxFamily::RENDER_SURFACE_STATE;
|
|
using STATE_BASE_ADDRESS = typename GfxFamily::STATE_BASE_ADDRESS;
|
|
|
|
if (args.setGeneralStateBaseAddress && is64bit) {
|
|
args.stateBaseAddressCmd->setGeneralStateBaseAddress(args.gmmHelper->decanonize(args.indirectObjectHeapBaseAddress));
|
|
}
|
|
|
|
if (overrideBindlessSurfaceStateBase && args.ssh) {
|
|
args.stateBaseAddressCmd->setBindlessSurfaceStateBaseAddress(args.ssh->getHeapGpuBase());
|
|
args.stateBaseAddressCmd->setBindlessSurfaceStateBaseAddressModifyEnable(true);
|
|
const auto surfaceStateCount = args.ssh->getMaxAvailableSpace() / sizeof(RENDER_SURFACE_STATE);
|
|
args.stateBaseAddressCmd->setBindlessSurfaceStateSize(static_cast<uint32_t>(surfaceStateCount - 1));
|
|
}
|
|
|
|
args.stateBaseAddressCmd->setBindlessSamplerStateBaseAddressModifyEnable(true);
|
|
|
|
auto heapResourceUsage = CacheSettingsHelper::getGmmUsageType(AllocationType::INTERNAL_HEAP, DebugManager.flags.DisableCachingForHeaps.get(), *args.gmmHelper->getHardwareInfo());
|
|
auto heapMocsValue = args.gmmHelper->getMOCS(heapResourceUsage);
|
|
|
|
args.stateBaseAddressCmd->setSurfaceStateMemoryObjectControlState(heapMocsValue);
|
|
args.stateBaseAddressCmd->setDynamicStateMemoryObjectControlState(heapMocsValue);
|
|
args.stateBaseAddressCmd->setGeneralStateMemoryObjectControlState(heapMocsValue);
|
|
args.stateBaseAddressCmd->setBindlessSurfaceStateMemoryObjectControlState(heapMocsValue);
|
|
args.stateBaseAddressCmd->setBindlessSamplerStateMemoryObjectControlState(heapMocsValue);
|
|
|
|
bool enableMultiGpuAtomics = args.isMultiOsContextCapable;
|
|
if (DebugManager.flags.EnableMultiGpuAtomicsOptimization.get()) {
|
|
enableMultiGpuAtomics = args.useGlobalAtomics && (args.isMultiOsContextCapable || args.areMultipleSubDevicesInContext);
|
|
}
|
|
args.stateBaseAddressCmd->setDisableSupportForMultiGpuAtomicsForStatelessAccesses(!enableMultiGpuAtomics);
|
|
|
|
args.stateBaseAddressCmd->setDisableSupportForMultiGpuPartialWritesForStatelessMessages(!args.isMultiOsContextCapable);
|
|
|
|
if (DebugManager.flags.ForceMultiGpuAtomics.get() != -1) {
|
|
args.stateBaseAddressCmd->setDisableSupportForMultiGpuAtomicsForStatelessAccesses(!!DebugManager.flags.ForceMultiGpuAtomics.get());
|
|
}
|
|
|
|
if (DebugManager.flags.ForceMultiGpuPartialWrites.get() != -1) {
|
|
args.stateBaseAddressCmd->setDisableSupportForMultiGpuPartialWritesForStatelessMessages(!!DebugManager.flags.ForceMultiGpuPartialWrites.get());
|
|
}
|
|
|
|
if (args.memoryCompressionState != MemoryCompressionState::NotApplicable) {
|
|
setSbaStatelessCompressionParams<GfxFamily>(args.stateBaseAddressCmd, args.memoryCompressionState);
|
|
}
|
|
|
|
bool l3MocsEnabled = (args.stateBaseAddressCmd->getStatelessDataPortAccessMemoryObjectControlState() >> 1) == (args.gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER) >> 1);
|
|
bool constMocsAllowed = (l3MocsEnabled && (DebugManager.flags.ForceL1Caching.get() != 0));
|
|
|
|
if (constMocsAllowed) {
|
|
auto constMocsIndex = args.gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER_CONST);
|
|
GmmHelper::applyMocsEncryptionBit(constMocsIndex);
|
|
|
|
args.stateBaseAddressCmd->setStatelessDataPortAccessMemoryObjectControlState(constMocsIndex);
|
|
}
|
|
|
|
appendExtraCacheSettings(args);
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
void StateBaseAddressHelper<GfxFamily>::programBindingTableBaseAddress(LinearStream &commandStream, uint64_t baseAddress, uint32_t sizeInPages, GmmHelper *gmmHelper) {
|
|
using _3DSTATE_BINDING_TABLE_POOL_ALLOC = typename GfxFamily::_3DSTATE_BINDING_TABLE_POOL_ALLOC;
|
|
|
|
auto bindingTablePoolAlloc = commandStream.getSpaceForCmd<_3DSTATE_BINDING_TABLE_POOL_ALLOC>();
|
|
_3DSTATE_BINDING_TABLE_POOL_ALLOC cmd = GfxFamily::cmdInitStateBindingTablePoolAlloc;
|
|
cmd.setBindingTablePoolBaseAddress(baseAddress);
|
|
cmd.setBindingTablePoolBufferSize(sizeInPages);
|
|
cmd.setSurfaceObjectControlStateIndexToMocsTables(gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_STATE_HEAP_BUFFER));
|
|
if (DebugManager.flags.DisableCachingForHeaps.get()) {
|
|
cmd.setSurfaceObjectControlStateIndexToMocsTables(gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_SYSTEM_MEMORY_BUFFER_CACHELINE_MISALIGNED));
|
|
}
|
|
|
|
*bindingTablePoolAlloc = cmd;
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
void StateBaseAddressHelper<GfxFamily>::appendIohParameters(StateBaseAddressHelperArgs<GfxFamily> &args) {
|
|
}
|
|
|
|
template <typename GfxFamily>
|
|
uint32_t StateBaseAddressHelper<GfxFamily>::getMaxBindlessSurfaceStates() {
|
|
return std::numeric_limits<uint32_t>::max();
|
|
}
|
|
} // namespace NEO
|