intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-04 15:53:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Remove redundant value CsrSizeRequestFlags::numGrfRequiredChanged

Browse Source 
Related-To: NEO-5995

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
This commit is contained in:
Filip Hazubski
2022-03-14 16:40:32 +00:00
committed by Compute-Runtime-Automation
parent adc5fb30d7
commit 32b0f7b014
34 changed files with 72 additions and 207 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
shared/source/command_container/cmdcontainer.cpp
View File

@@ -154,7 +154,6 @@ void CommandContainer::reset() {
iddBlock = nullptr;
nextIddInBlock = this->getNumIddPerBlock();
lastSentNumGrfRequired = 0;
lastPipelineSelectModeRequired = false;
lastSentUseGlobalAtomics = false;
}

1
shared/source/command_container/cmdcontainer.h
View File

@@ -77,7 +77,6 @@ class CommandContainer : public NonCopyableOrMovableClass {
uint32_t slmSize = std::numeric_limits<uint32_t>::max();
uint32_t nextIddInBlock = 0;
uint32_t lastSentNumGrfRequired = 0;
bool lastPipelineSelectModeRequired = false;
bool lastSentUseGlobalAtomics = false;

1
shared/source/command_stream/command_stream_receiver.h
View File

@@ -384,7 +384,6 @@ class CommandStreamReceiver {
std::chrono::microseconds gpuHangCheckPeriod{500'000};
uint32_t lastSentL3Config = 0;
uint32_t latestSentStatelessMocsConfig = 0;
uint32_t lastSentNumGrfRequired = GrfConfig::DefaultGrfNumber;
uint64_t lastSentSliceCount = QueueSliceCount::defaultSliceCount;
uint32_t requiredScratchSize = 0;

1
shared/source/command_stream/command_stream_receiver_hw.h
View File

@@ -75,7 +75,6 @@ class CommandStreamReceiverHw : public CommandStreamReceiver {
size_t getCmdSizeForComputeMode();
MOCKABLE_VIRTUAL bool hasSharedHandles();
bool isComputeModeNeeded() const;
bool isPipelineSelectAlreadyProgrammed() const;
void programComputeMode(LinearStream &csr, DispatchFlags &dispatchFlags, const HardwareInfo &hwInfo);

11
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@@ -268,10 +268,6 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
dispatchFlags.pipelineSelectArgs.specialPipelineSelectMode,
hwInfo);
if (dispatchFlags.numGrfRequired == GrfConfig::NotApplicable) {
dispatchFlags.numGrfRequired = lastSentNumGrfRequired;
}
auto requiresCoherency = hwHelper.forceNonGpuCoherencyWA(dispatchFlags.requiresCoherency);
this->streamProperties.stateComputeMode.setProperties(requiresCoherency, dispatchFlags.numGrfRequired,
dispatchFlags.threadArbitrationPolicy, hwInfo);
@@ -281,9 +277,6 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
csrSizeRequestFlags.mediaSamplerConfigChanged = this->lastMediaSamplerConfig != static_cast<int8_t>(dispatchFlags.pipelineSelectArgs.mediaSamplerRequired);
csrSizeRequestFlags.specialPipelineSelectModeChanged = isSpecialPipelineSelectModeChanged;
csrSizeRequestFlags.numGrfRequiredChanged = this->lastSentNumGrfRequired != dispatchFlags.numGrfRequired;
lastSentNumGrfRequired = dispatchFlags.numGrfRequired;
csrSizeRequestFlags.activePartitionsChanged = isProgramActivePartitionConfigRequired();
auto force32BitAllocations = getMemoryManager()->peekForce32BitAllocations();
@@ -809,7 +802,7 @@ size_t CommandStreamReceiverHw<GfxFamily>::getRequiredCmdStreamSize(const Dispat
size += sizeof(typename GfxFamily::MI_BATCH_BUFFER_START);
size += getCmdSizeForL3Config();
if (isComputeModeNeeded()) {
if (this->streamProperties.stateComputeMode.isDirty()) {
size += getCmdSizeForComputeMode();
}
size += getCmdSizeForMediaSampler(dispatchFlags.pipelineSelectArgs.mediaSamplerRequired);
@@ -1373,7 +1366,7 @@ inline MemoryCompressionState CommandStreamReceiverHw<GfxFamily>::getMemoryCompr
template <typename GfxFamily>
inline bool CommandStreamReceiverHw<GfxFamily>::isPipelineSelectAlreadyProgrammed() const {
const auto &hwInfoConfig = *HwInfoConfig::get(peekHwInfo().platform.eProductFamily);
return isComputeModeNeeded() && hwInfoConfig.is3DPipelineSelectWARequired() && isRcs();
return this->streamProperties.stateComputeMode.isDirty() && hwInfoConfig.is3DPipelineSelectWARequired() && isRcs();
}
template <typename GfxFamily>

9
shared/source/command_stream/command_stream_receiver_hw_tgllp_and_later.inl
View File

@@ -9,25 +9,18 @@
#include "shared/source/command_stream/command_stream_receiver_hw.h"
#include "shared/source/device/device.h"
#include "shared/source/helpers/pipe_control_args.h"
#include "shared/source/helpers/state_compute_mode_helper.h"
#include "shared/source/os_interface/hw_info_config.h"
namespace NEO {
template <typename GfxFamily>
void CommandStreamReceiverHw<GfxFamily>::programComputeMode(LinearStream &stream, DispatchFlags &dispatchFlags, const HardwareInfo &hwInfo) {
if (isComputeModeNeeded()) {
if (this->streamProperties.stateComputeMode.isDirty()) {
EncodeComputeMode<GfxFamily>::programComputeModeCommandWithSynchronization(
stream, this->streamProperties.stateComputeMode, dispatchFlags.pipelineSelectArgs,
hasSharedHandles(), hwInfo, isRcs());
}
}
template <typename GfxFamily>
inline bool CommandStreamReceiverHw<GfxFamily>::isComputeModeNeeded() const {
return StateComputeModeHelper<Family>::isStateComputeModeRequired(csrSizeRequestFlags, false) ||
this->streamProperties.stateComputeMode.isDirty();
}
template <>
inline void CommandStreamReceiverHw<Family>::addPipeControlBeforeStateBaseAddress(LinearStream &commandStream) {
PipeControlArgs args;

1
shared/source/command_stream/csr_definitions.h
View File

@@ -123,7 +123,6 @@ struct CsrSizeRequestFlags {
bool preemptionRequestChanged = false;
bool mediaSamplerConfigChanged = false;
bool hasSharedHandles = false;
bool numGrfRequiredChanged = false;
bool specialPipelineSelectModeChanged = false;
bool activePartitionsChanged = false;
};

6
shared/source/command_stream/stream_properties.cpp
View File

@@ -19,8 +19,10 @@ void StateComputeModeProperties::setProperties(bool requiresCoherency, uint32_t
int32_t isCoherencyRequired = (requiresCoherency ? 1 : 0);
this->isCoherencyRequired.set(isCoherencyRequired);
int32_t largeGrfMode = (numGrfRequired == GrfConfig::LargeGrfNumber ? 1 : 0);
this->largeGrfMode.set(largeGrfMode);
if (this->largeGrfMode.value == -1 || numGrfRequired != GrfConfig::NotApplicable) {
int32_t largeGrfMode = (numGrfRequired == GrfConfig::LargeGrfNumber ? 1 : 0);
this->largeGrfMode.set(largeGrfMode);
}
int32_t zPassAsyncComputeThreadLimit = -1;
if (DebugManager.flags.ForceZPassAsyncComputeThreadLimit.get() != -1) {

4
shared/source/enable_cores.cmake
View File

@@ -160,10 +160,6 @@ macro(macro_for_each_core_type)
endif()
endforeach()
set(SRC_FILE "${CORE_COREX_PREFIX}/state_compute_mode_helper_${CORE_TYPE_LOWER}.cpp")
if(EXISTS ${SRC_FILE})
list(APPEND CORE_SRCS_${CORE_TYPE}_CPP_BASE ${SRC_FILE})
endif()
if(EXISTS ${GENERATED_COREX_PREFIX}/hw_cmds_generated_${CORE_TYPE_LOWER}.inl)
list(APPEND CORE_SRCS_${CORE_TYPE}_H_BASE "${GENERATED_COREX_PREFIX}/hw_cmds_generated_${CORE_TYPE_LOWER}.inl")
endif()

7
shared/source/gen11/command_stream_receiver_hw_gen11.cpp
View File

@@ -19,18 +19,13 @@ static auto gfxCore = IGFX_GEN11_CORE;
template <>
void CommandStreamReceiverHw<Family>::programComputeMode(LinearStream &stream, DispatchFlags &dispatchFlags, const HardwareInfo &hwInfo) {
if (this->isComputeModeNeeded()) {
if (this->streamProperties.stateComputeMode.isCoherencyRequired.isDirty) {
EncodeComputeMode<Family>::programComputeModeCommandWithSynchronization(
stream, this->streamProperties.stateComputeMode, dispatchFlags.pipelineSelectArgs,
hasSharedHandles(), hwInfo, isRcs());
}
}
template <typename GfxFamily>
inline bool CommandStreamReceiverHw<GfxFamily>::isComputeModeNeeded() const {
return this->streamProperties.stateComputeMode.isCoherencyRequired.isDirty;
}
template <>
void CommandStreamReceiverHw<Family>::programMediaSampler(LinearStream &stream, DispatchFlags &dispatchFlags) {
using PWR_CLK_STATE_REGISTER = Family::PWR_CLK_STATE_REGISTER;

17
shared/source/gen12lp/state_compute_mode_helper_gen12lp.cpp
View File

@@ -1,17 +0,0 @@
/*
* Copyright (C) 2019-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/helpers/state_compute_mode_helper.h"
namespace NEO {
template <>
bool StateComputeModeHelper<TGLLPFamily>::isStateComputeModeRequired(const CsrSizeRequestFlags &csrSizeRequestFlags, bool isThreadArbitionPolicyProgrammed) {
return csrSizeRequestFlags.numGrfRequiredChanged;
}
} // namespace NEO

5
shared/source/gen8/command_stream_receiver_hw_gen8.cpp
View File

@@ -19,11 +19,6 @@ template <>
void CommandStreamReceiverHw<Family>::programComputeMode(LinearStream &stream, DispatchFlags &dispatchFlags, const HardwareInfo &hwInfo) {
}
template <typename GfxFamily>
inline bool CommandStreamReceiverHw<GfxFamily>::isComputeModeNeeded() const {
return false;
}
template <>
void populateFactoryTable<CommandStreamReceiverHw<Family>>() {
extern CommandStreamReceiverCreateFunc commandStreamReceiverFactory[2 * IGFX_MAX_CORE];

5
shared/source/gen9/command_stream_receiver_hw_gen9.cpp
View File

@@ -19,11 +19,6 @@ template <>
void CommandStreamReceiverHw<Family>::programComputeMode(LinearStream &stream, DispatchFlags &dispatchFlags, const HardwareInfo &hwInfo) {
}
template <typename GfxFamily>
inline bool CommandStreamReceiverHw<GfxFamily>::isComputeModeNeeded() const {
return false;
}
template <>
void populateFactoryTable<CommandStreamReceiverHw<Family>>() {
extern CommandStreamReceiverCreateFunc commandStreamReceiverFactory[2 * IGFX_MAX_CORE];

1
shared/source/helpers/CMakeLists.txt
View File

@@ -116,7 +116,6 @@ set(NEO_CORE_HELPERS
${CMAKE_CURRENT_SOURCE_DIR}/state_base_address_bdw_and_later.inl
${CMAKE_CURRENT_SOURCE_DIR}/state_base_address_icllp_and_later.inl
${CMAKE_CURRENT_SOURCE_DIR}/state_base_address_skl.inl
${CMAKE_CURRENT_SOURCE_DIR}/state_compute_mode_helper.h
${CMAKE_CURRENT_SOURCE_DIR}/stdio.h
${CMAKE_CURRENT_SOURCE_DIR}/string.h
${CMAKE_CURRENT_SOURCE_DIR}/string_helpers.h

18
shared/source/helpers/state_compute_mode_helper.h
View File

@@ -1,18 +0,0 @@
/*
* Copyright (C) 2019-2020 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/command_stream/csr_definitions.h"
#include "hw_cmds.h"
namespace NEO {
template <typename GfxFamily>
struct StateComputeModeHelper {
static bool isStateComputeModeRequired(const CsrSizeRequestFlags &csrSizeRequestFlags, bool isThreadArbitionPolicyProgrammed);
};
} // namespace NEO

17
shared/source/xe_hp_core/state_compute_mode_helper_xe_hp_core.cpp
View File

@@ -1,17 +0,0 @@
/*
* Copyright (C) 2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/helpers/state_compute_mode_helper.h"
namespace NEO {
template <>
bool StateComputeModeHelper<XeHpFamily>::isStateComputeModeRequired(const CsrSizeRequestFlags &csrSizeRequestFlags, bool isThreadArbitionPolicyProgrammed) {
return csrSizeRequestFlags.numGrfRequiredChanged;
}
} // namespace NEO

1
shared/source/xe_hpc_core/command_stream_receiver_hw_xe_hpc_core.cpp
View File

@@ -16,7 +16,6 @@ using Family = NEO::XE_HPC_COREFamily;
#include "shared/source/command_stream/command_stream_receiver_hw_xehp_and_later.inl"
#include "shared/source/helpers/blit_commands_helper_xehp_and_later.inl"
#include "shared/source/helpers/populate_factory.h"
#include "shared/source/helpers/state_compute_mode_helper.h"
namespace NEO {
static auto gfxCore = IGFX_XE_HPC_CORE;

21
shared/source/xe_hpc_core/state_compute_mode_helper_xe_hpc_core.cpp
View File

@@ -1,21 +0,0 @@
/*
* Copyright (C) 2020-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/helpers/state_compute_mode_helper.h"
namespace NEO {
using Family = XE_HPC_COREFamily;
template <>
bool StateComputeModeHelper<Family>::isStateComputeModeRequired(const CsrSizeRequestFlags &csrSizeRequestFlags, bool isThreadArbitionPolicyProgrammed) {
return csrSizeRequestFlags.numGrfRequiredChanged;
}
template struct StateComputeModeHelper<Family>;
} // namespace NEO

17
shared/source/xe_hpg_core/state_compute_mode_helper_xe_hpg_core.cpp
View File

@@ -1,17 +0,0 @@
/*
* Copyright (C) 2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/helpers/state_compute_mode_helper.h"
namespace NEO {
template <>
bool StateComputeModeHelper<XE_HPG_COREFamily>::isStateComputeModeRequired(const CsrSizeRequestFlags &csrSizeRequestFlags, bool isThreadArbitionPolicyProgrammed) {
return csrSizeRequestFlags.numGrfRequiredChanged;
}
} // namespace NEO