compute-runtime/shared/source/gen12lp/preamble_gen12lp.cpp

107 lines
4.0 KiB
C++

/*
* Copyright (C) 2019-2023 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_stream/csr_definitions.h"
#include "shared/source/execution_environment/root_device_environment.h"
#include "shared/source/gen12lp/hw_cmds_base.h"
#include "shared/source/helpers/hw_info.h"
#include "shared/source/helpers/pipe_control_args.h"
#include "shared/source/helpers/pipeline_select_helper.h"
#include "shared/source/helpers/preamble_bdw_and_later.inl"
namespace NEO {
using Family = Gen12LpFamily;
template <>
uint32_t PreambleHelper<Family>::getL3Config(const HardwareInfo &hwInfo, bool useSLM) {
uint32_t l3Config = 0;
switch (hwInfo.platform.eProductFamily) {
case IGFX_TIGERLAKE_LP:
l3Config = getL3ConfigHelper<IGFX_TIGERLAKE_LP>(useSLM);
break;
default:
l3Config = getL3ConfigHelper<IGFX_TIGERLAKE_LP>(true);
}
return l3Config;
}
template <>
void PreambleHelper<Family>::programPipelineSelect(LinearStream *pCommandStream,
const PipelineSelectArgs &pipelineSelectArgs,
const RootDeviceEnvironment &rootDeviceEnvironment) {
using PIPELINE_SELECT = typename Family::PIPELINE_SELECT;
if (MemorySynchronizationCommands<Family>::isBarrierPriorToPipelineSelectWaRequired(rootDeviceEnvironment)) {
PipeControlArgs args;
args.renderTargetCacheFlushEnable = true;
MemorySynchronizationCommands<Family>::addSingleBarrier(*pCommandStream, args);
}
auto cmdSpace = pCommandStream->getSpaceForCmd<PIPELINE_SELECT>();
PIPELINE_SELECT pipelineSelectCmd = Family::cmdInitPipelineSelect;
auto mask = pipelineSelectEnablePipelineSelectMaskBits | pipelineSelectMediaSamplerDopClockGateMaskBits;
auto pipeline = pipelineSelectArgs.is3DPipelineRequired ? PIPELINE_SELECT::PIPELINE_SELECTION_3D : PIPELINE_SELECT::PIPELINE_SELECTION_GPGPU;
pipelineSelectCmd.setPipelineSelection(pipeline);
pipelineSelectCmd.setMediaSamplerDopClockGateEnable(!pipelineSelectArgs.mediaSamplerRequired);
if (pipelineSelectArgs.systolicPipelineSelectSupport) {
mask |= pipelineSelectSystolicModeEnableMaskBits;
pipelineSelectCmd.setSpecialModeEnable(pipelineSelectArgs.systolicPipelineSelectMode);
}
pipelineSelectCmd.setMaskBits(mask);
*cmdSpace = pipelineSelectCmd;
}
template <>
void PreambleHelper<Family>::addPipeControlBeforeVfeCmd(LinearStream *pCommandStream, const HardwareInfo *hwInfo, EngineGroupType engineGroupType) {
PipeControlArgs args = {};
if (hwInfo->workaroundTable.flags.waSendMIFLUSHBeforeVFE) {
if (engineGroupType != EngineGroupType::Compute) {
args.renderTargetCacheFlushEnable = true;
args.depthCacheFlushEnable = true;
args.depthStallEnable = true;
}
args.dcFlushEnable = true;
}
MemorySynchronizationCommands<Family>::addSingleBarrier(*pCommandStream, args);
}
template <>
void PreambleHelper<Family>::programL3(LinearStream *pCommandStream, uint32_t l3Config) {
}
template <>
uint32_t PreambleHelper<Family>::getUrbEntryAllocationSize() {
return 1024u;
}
template <>
void PreambleHelper<Family>::appendProgramVFEState(const RootDeviceEnvironment &rootDeviceEnvironment, const StreamProperties &streamProperties, void *cmd) {
VFE_STATE_TYPE *mediaVfeState = static_cast<VFE_STATE_TYPE *>(cmd);
bool disableEUFusion = streamProperties.frontEndState.disableEUFusion.value == 1;
auto &gfxCoreHelper = rootDeviceEnvironment.getHelper<GfxCoreHelper>();
auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
if (!gfxCoreHelper.isFusedEuDispatchEnabled(hwInfo, disableEUFusion)) {
mediaVfeState->setDisableSlice0Subslice2(true);
}
if (DebugManager.flags.MediaVfeStateMaxSubSlices.get() != -1) {
mediaVfeState->setMaximumNumberOfDualSubslices(DebugManager.flags.MediaVfeStateMaxSubSlices.get());
}
}
// Explicitly instantiate PreambleHelper for Gen12Lp device family
template struct PreambleHelper<Family>;
} // namespace NEO