intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-05 09:09:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add debug flag to control programming of thread arbitration policy with SCM

Browse Source 
Related-To: NEO-6801

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
This commit is contained in:
Filip Hazubski
2022-05-16 20:34:53 +00:00
committed by Compute-Runtime-Automation
parent ec6d580a27
commit 35d1f2e341
28 changed files with 96 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -39,9 +39,6 @@ template <typename Family>
void EncodeDispatchKernel<Family>::setGrfInfo(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, uint32_t numGrf,
const size_t &sizeCrossThreadData, const size_t &sizePerThreadData,
const HardwareInfo &hwInfo) {
auto &hwInfoConfig = *HwInfoConfig::get(hwInfo.platform.eProductFamily);
hwInfoConfig.updateIddCommand(pInterfaceDescriptor, numGrf);
}
template <typename Family>
@@ -79,6 +76,10 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
EncodeDispatchKernel<Family>::setGrfInfo(&idd, kernelDescriptor.kernelAttributes.numGrfRequired, sizeCrossThreadData,
sizePerThreadData, hwInfo);
auto &hwInfoConfig = *HwInfoConfig::get(hwInfo.platform.eProductFamily);
hwInfoConfig.updateIddCommand(&idd, kernelDescriptor.kernelAttributes.numGrfRequired,
args.dispatchInterface->getSchedulingHintExp());
bool localIdsGenerationByRuntime = args.dispatchInterface->requiresGenerationOfLocalIdsByRuntime();
auto requiredWorkgroupOrder = args.dispatchInterface->getRequiredWorkgroupOrder();
bool inlineDataProgramming = EncodeDispatchKernel<Family>::inlineDataProgrammingRequired(kernelDescriptor);

2
shared/source/command_stream/definitions/stream_properties.inl
View File

@@ -24,7 +24,7 @@ struct StateComputeModeProperties {
void clearIsDirty();
bool isDirtyExtra() const;
void setPropertiesExtra(bool reportNumGrf);
void setPropertiesExtra(bool reportNumGrf, bool reportThreadArbitrationPolicy);
void setPropertiesExtra(const StateComputeModeProperties &properties);
void clearIsDirtyExtra();
};

7
shared/source/command_stream/stream_properties.cpp
View File

@@ -51,9 +51,12 @@ void StateComputeModeProperties::setProperties(bool requiresCoherency, uint32_t
if (DebugManager.flags.OverrideThreadArbitrationPolicy.get() != -1) {
threadArbitrationPolicy = DebugManager.flags.OverrideThreadArbitrationPolicy.get();
}
this->threadArbitrationPolicy.set(threadArbitrationPolicy);
bool reportThreadArbitrationPolicy = hwInfoConfig.isThreadArbitrationPolicyReportedWithScm();
if (reportThreadArbitrationPolicy) {
this->threadArbitrationPolicy.set(threadArbitrationPolicy);
}
setPropertiesExtra(reportNumGrf);
setPropertiesExtra(reportNumGrf, reportThreadArbitrationPolicy);
}
void StateComputeModeProperties::setProperties(const StateComputeModeProperties &properties) {

2
shared/source/command_stream/stream_properties_extra.cpp
View File

@@ -9,7 +9,7 @@
using namespace NEO;
void StateComputeModeProperties::setPropertiesExtra(bool reportNumGrf) {
void StateComputeModeProperties::setPropertiesExtra(bool reportNumGrf, bool reportThreadArbitrationPolicy) {
}
void StateComputeModeProperties::setPropertiesExtra(const StateComputeModeProperties &properties) {
}

1
shared/source/debug_settings/debug_variables_base.inl
View File

@@ -161,6 +161,7 @@ DECLARE_DEBUG_VARIABLE(int32_t, OverrideL1CacheControlInSurfaceState, -1, "-1: f
DECLARE_DEBUG_VARIABLE(int32_t, OverrideL1CacheControlInSurfaceStateForScratchSpace, -1, "-1: feature inactive, >=0 : following L1 cache control value will be programmed in render surface state for scratch space")
DECLARE_DEBUG_VARIABLE(int32_t, OverridePreferredSlmAllocationSizePerDss, -1, "-1: default, >=0: program value for preferred SLM allocation size per dss")
DECLARE_DEBUG_VARIABLE(int32_t, ForceGrfNumProgrammingWithScm, -1, "-1: default, 0: do not program grf num with SCM, 1: program grf num with SCM")
DECLARE_DEBUG_VARIABLE(int32_t, ForceThreadArbitrationPolicyProgrammingWithScm, -1, "-1: default, 0: do not program thread arbitration policy with SCM, 1: program thread arbitration policy with SCM")
DECLARE_DEBUG_VARIABLE(int32_t, ForceL3PrefetchForComputeWalker, -1, "-1: default, 0: disable, 1: enable")
DECLARE_DEBUG_VARIABLE(int32_t, ForceZPassAsyncComputeThreadLimit, -1, "-1: default, >0: Limit value in STATE_COMPUTE_MODE")
DECLARE_DEBUG_VARIABLE(int32_t, ForcePixelAsyncComputeThreadLimit, -1, "-1: default, >0: Limit value in STATE_COMPUTE_MODE")

3
shared/source/generated/xe_hpc_core/hw_cmds_generated_xe_hpc_core.inl
View File

@@ -5172,7 +5172,8 @@ typedef struct tagINTERFACE_DESCRIPTOR_DATA {
// DWORD 1
uint64_t Reserved_32 : BITFIELD_RANGE(32, 63);
// DWORD 2
uint32_t Reserved_64 : BITFIELD_RANGE(0, 6);
uint32_t Reserved_64 : BITFIELD_RANGE(0, 1);
uint32_t Reserved_66 : BITFIELD_RANGE(2, 6);
uint32_t SoftwareExceptionEnable : BITFIELD_RANGE(7, 7);
uint32_t Reserved_72 : BITFIELD_RANGE(8, 10);
uint32_t MaskStackExceptionEnable : BITFIELD_RANGE(11, 11);

2
shared/source/kernel/dispatch_kernel_encoder_interface.h
View File

@@ -46,6 +46,8 @@ struct DispatchKernelEncoderI {
virtual uint32_t getRequiredWorkgroupOrder() const = 0;
virtual bool requiresGenerationOfLocalIdsByRuntime() const = 0;
virtual int32_t getSchedulingHintExp() const = 0;
virtual ImplicitArgs *getImplicitArgs() const = 0;
};
} // namespace NEO

6
shared/source/os_interface/hw_info_config.h
View File

@@ -53,7 +53,7 @@ class HwInfoConfig {
virtual uint32_t getMaxThreadsForWorkgroup(const HardwareInfo &hwInfo, uint32_t maxNumEUsPerSubSlice) const = 0;
virtual void setForceNonCoherent(void *const commandPtr, const StateComputeModeProperties &properties) = 0;
virtual void updateScmCommand(void *const commandPtr, const StateComputeModeProperties &properties) = 0;
virtual void updateIddCommand(void *const commandPtr, uint32_t numGrf) = 0;
virtual void updateIddCommand(void *const commandPtr, uint32_t numGrf, int32_t threadArbitrationPolicy) = 0;
virtual bool obtainBlitterPreference(const HardwareInfo &hwInfo) const = 0;
virtual bool isBlitterFullySupported(const HardwareInfo &hwInfo) const = 0;
virtual bool isPageTableManagerSupported(const HardwareInfo &hwInfo) const = 0;
@@ -104,6 +104,7 @@ class HwInfoConfig {
virtual bool useChannelRedForUnusedShaderChannels() const = 0;
virtual bool isIpSamplingSupported(const NEO::HardwareInfo &hwInfo) const = 0;
virtual bool isGrfNumReportedWithScm() const = 0;
virtual bool isThreadArbitrationPolicyReportedWithScm() const = 0;
virtual bool isCooperativeEngineSupported(const HardwareInfo &hwInfo) const = 0;
virtual bool isTimestampWaitSupportedForEvents() const = 0;
virtual bool isTilePlacementResourceWaRequired(const HardwareInfo &hwInfo) const = 0;
@@ -146,7 +147,7 @@ class HwInfoConfigHw : public HwInfoConfig {
uint32_t getMaxThreadsForWorkgroup(const HardwareInfo &hwInfo, uint32_t maxNumEUsPerSubSlice) const override;
void setForceNonCoherent(void *const commandPtr, const StateComputeModeProperties &properties) override;
void updateScmCommand(void *const commandPtr, const StateComputeModeProperties &properties) override;
void updateIddCommand(void *const commandPtr, uint32_t numGrf) override;
void updateIddCommand(void *const commandPtr, uint32_t numGrf, int32_t threadArbitrationPolicy) override;
bool obtainBlitterPreference(const HardwareInfo &hwInfo) const override;
bool isBlitterFullySupported(const HardwareInfo &hwInfo) const override;
bool isPageTableManagerSupported(const HardwareInfo &hwInfo) const override;
@@ -197,6 +198,7 @@ class HwInfoConfigHw : public HwInfoConfig {
bool useChannelRedForUnusedShaderChannels() const override;
bool isIpSamplingSupported(const NEO::HardwareInfo &hwInfo) const override;
bool isGrfNumReportedWithScm() const override;
bool isThreadArbitrationPolicyReportedWithScm() const override;
bool isCooperativeEngineSupported(const HardwareInfo &hwInfo) const override;
bool isTimestampWaitSupportedForEvents() const override;
bool isTilePlacementResourceWaRequired(const HardwareInfo &hwInfo) const override;

10
shared/source/os_interface/hw_info_config.inl
View File

@@ -120,7 +120,7 @@ template <PRODUCT_FAMILY gfxProduct>
void HwInfoConfigHw<gfxProduct>::updateScmCommand(void *const commandPtr, const StateComputeModeProperties &properties) {}
template <PRODUCT_FAMILY gfxProduct>
void HwInfoConfigHw<gfxProduct>::updateIddCommand(void *const commandPtr, uint32_t numGrf) {}
void HwInfoConfigHw<gfxProduct>::updateIddCommand(void *const commandPtr, uint32_t numGrf, int32_t threadArbitrationPolicy) {}
template <PRODUCT_FAMILY gfxProduct>
bool HwInfoConfigHw<gfxProduct>::isPageTableManagerSupported(const HardwareInfo &hwInfo) const {
@@ -359,6 +359,14 @@ bool HwInfoConfigHw<gfxProduct>::isGrfNumReportedWithScm() const {
return true;
}
template <PRODUCT_FAMILY gfxProduct>
bool HwInfoConfigHw<gfxProduct>::isThreadArbitrationPolicyReportedWithScm() const {
if (DebugManager.flags.ForceThreadArbitrationPolicyProgrammingWithScm.get() != -1) {
return DebugManager.flags.ForceThreadArbitrationPolicyProgrammingWithScm.get();
}
return true;
}
template <PRODUCT_FAMILY gfxProduct>
bool HwInfoConfigHw<gfxProduct>::isCooperativeEngineSupported(const HardwareInfo &hwInfo) const {
return false;