intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Simplify Context method

Browse Source 
return if context has multiple sub devices related to a given root device

Related-To: NEO-3691
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2021-03-29 17:06:29 +00:00
committed by Compute-Runtime-Automation
parent 21e89b121a
commit 8215395401
37 changed files with 162 additions and 170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
shared/source/command_container/command_encoder.h
View File

@@ -233,9 +233,9 @@ struct EncodeSurfaceState {
static void encodeBuffer(void *dst, uint64_t address, size_t size, uint32_t mocs,
bool cpuCoherent, bool forceNonAuxMode, bool isReadOnly, uint32_t numAvailableDevices,
GraphicsAllocation *allocation, GmmHelper *gmmHelper, bool useGlobalAtomics, size_t numDevicesInContext);
GraphicsAllocation *allocation, GmmHelper *gmmHelper, bool useGlobalAtomics, bool areMultipleSubDevicesInContext);
static void encodeExtraBufferParams(R_SURFACE_STATE *surfaceState, GraphicsAllocation *allocation, GmmHelper *gmmHelper,
bool isReadOnly, uint32_t numAvailableDevices, bool useGlobalAtomics, size_t numDevicesInContext);
bool isReadOnly, uint32_t numAvailableDevices, bool useGlobalAtomics, bool areMultipleSubDevicesInContext);
static void encodeExtraCacheSettings(R_SURFACE_STATE *surfaceState, const HardwareInfo &hwInfo);
static constexpr uintptr_t getSurfaceBaseAddressAlignmentMask() {

4
shared/source/command_container/command_encoder.inl
View File

@@ -318,7 +318,7 @@ void EncodeStoreMMIO<Family>::encode(LinearStream &csr, uint32_t offset, uint64_
template <typename Family>
void EncodeSurfaceState<Family>::encodeBuffer(void *dst, uint64_t address, size_t size, uint32_t mocs,
bool cpuCoherent, bool forceNonAuxMode, bool isReadOnly, uint32_t numAvailableDevices,
GraphicsAllocation *allocation, GmmHelper *gmmHelper, bool useGlobalAtomics, size_t numDevicesInContext) {
GraphicsAllocation *allocation, GmmHelper *gmmHelper, bool useGlobalAtomics, bool areMultipleSubDevicesInContext) {
auto surfaceState = reinterpret_cast<R_SURFACE_STATE *>(dst);
UNRECOVERABLE_IF(!isAligned<getSurfaceBaseAddressMinimumAlignment()>(size));
@@ -356,7 +356,7 @@ void EncodeSurfaceState<Family>::encodeBuffer(void *dst, uint64_t address, size_
surfaceState->setMemoryObjectControlState(gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER_CACHELINE_MISALIGNED));
}
EncodeSurfaceState<Family>::encodeExtraBufferParams(surfaceState, allocation, gmmHelper, isReadOnly, numAvailableDevices, useGlobalAtomics, numDevicesInContext);
EncodeSurfaceState<Family>::encodeExtraBufferParams(surfaceState, allocation, gmmHelper, isReadOnly, numAvailableDevices, useGlobalAtomics, areMultipleSubDevicesInContext);
}
template <typename Family>

2
shared/source/command_container/command_encoder_bdw_plus.inl
View File

@@ -409,7 +409,7 @@ inline size_t EncodeWA<GfxFamily>::getAdditionalPipelineSelectSize(Device &devic
template <typename GfxFamily>
void EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(R_SURFACE_STATE *surfaceState, GraphicsAllocation *allocation, GmmHelper *gmmHelper,
bool isReadOnly, uint32_t numAvailableDevices, bool useGlobalAtomics, size_t numDevicesInContext) {
bool isReadOnly, uint32_t numAvailableDevices, bool useGlobalAtomics, bool areMultipleSubDevicesInContext) {
encodeExtraCacheSettings(surfaceState, *gmmHelper->getHardwareInfo());
}

4
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@@ -375,7 +375,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
latestSentStatelessMocsConfig = mocsIndex;
}
if ((isMultiOsContextCapable() || (dispatchFlags.numDevicesInContext > 1)) && (dispatchFlags.useGlobalAtomics != lastSentUseGlobalAtomics)) {
if ((isMultiOsContextCapable() || dispatchFlags.areMultipleSubDevicesInContext) && (dispatchFlags.useGlobalAtomics != lastSentUseGlobalAtomics)) {
isStateBaseAddressDirty = true;
lastSentUseGlobalAtomics = dispatchFlags.useGlobalAtomics;
}
@@ -427,7 +427,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
isMultiOsContextCapable(),
memoryCompressionState,
dispatchFlags.useGlobalAtomics,
dispatchFlags.numDevicesInContext);
dispatchFlags.areMultipleSubDevicesInContext);
*pCmd = cmd;
if (sshDirty) {

56
shared/source/command_stream/csr_definitions.h
View File

@@ -57,33 +57,33 @@ struct DispatchFlags {
KernelExecutionType kernelExecutionTypeP, MemoryCompressionState memoryCompressionStateP,
uint64_t sliceCountP, bool blockingP, bool dcFlushP, bool useSLMP, bool guardCommandBufferWithPipeControlP, bool gsba32BitRequiredP,
bool requiresCoherencyP, bool lowPriorityP, bool implicitFlushP, bool outOfOrderExecutionAllowedP, bool epilogueRequiredP,
bool usePerDSSbackedBufferP, bool useSingleSubdeviceP, bool useGlobalAtomicsP, size_t numDevicesInContextP) : csrDependencies(csrDependenciesP),
barrierTimestampPacketNodes(barrierTimestampPacketNodesP),
pipelineSelectArgs(pipelineSelectArgsP),
flushStampReference(flushStampReferenceP),
throttle(throttleP),
preemptionMode(preemptionModeP),
numGrfRequired(numGrfRequiredP),
l3CacheSettings(l3CacheSettingsP),
threadArbitrationPolicy(threadArbitrationPolicyP),
additionalKernelExecInfo(additionalKernelExecInfoP),
kernelExecutionType(kernelExecutionTypeP),
memoryCompressionState(memoryCompressionStateP),
sliceCount(sliceCountP),
blocking(blockingP),
dcFlush(dcFlushP),
useSLM(useSLMP),
guardCommandBufferWithPipeControl(guardCommandBufferWithPipeControlP),
gsba32BitRequired(gsba32BitRequiredP),
requiresCoherency(requiresCoherencyP),
lowPriority(lowPriorityP),
implicitFlush(implicitFlushP),
outOfOrderExecutionAllowed(outOfOrderExecutionAllowedP),
epilogueRequired(epilogueRequiredP),
usePerDssBackedBuffer(usePerDSSbackedBufferP),
useSingleSubdevice(useSingleSubdeviceP),
useGlobalAtomics(useGlobalAtomicsP),
numDevicesInContext(numDevicesInContextP){};
bool usePerDSSbackedBufferP, bool useSingleSubdeviceP, bool useGlobalAtomicsP, size_t areMultipleSubDevicesInContextP) : csrDependencies(csrDependenciesP),
barrierTimestampPacketNodes(barrierTimestampPacketNodesP),
pipelineSelectArgs(pipelineSelectArgsP),
flushStampReference(flushStampReferenceP),
throttle(throttleP),
preemptionMode(preemptionModeP),
numGrfRequired(numGrfRequiredP),
l3CacheSettings(l3CacheSettingsP),
threadArbitrationPolicy(threadArbitrationPolicyP),
additionalKernelExecInfo(additionalKernelExecInfoP),
kernelExecutionType(kernelExecutionTypeP),
memoryCompressionState(memoryCompressionStateP),
sliceCount(sliceCountP),
blocking(blockingP),
dcFlush(dcFlushP),
useSLM(useSLMP),
guardCommandBufferWithPipeControl(guardCommandBufferWithPipeControlP),
gsba32BitRequired(gsba32BitRequiredP),
requiresCoherency(requiresCoherencyP),
lowPriority(lowPriorityP),
implicitFlush(implicitFlushP),
outOfOrderExecutionAllowed(outOfOrderExecutionAllowedP),
epilogueRequired(epilogueRequiredP),
usePerDssBackedBuffer(usePerDSSbackedBufferP),
useSingleSubdevice(useSingleSubdeviceP),
useGlobalAtomics(useGlobalAtomicsP),
areMultipleSubDevicesInContext(areMultipleSubDevicesInContextP){};
CsrDependencies csrDependencies;
TimestampPacketContainer *barrierTimestampPacketNodes = nullptr;
@@ -112,7 +112,7 @@ struct DispatchFlags {
bool usePerDssBackedBuffer = false;
bool useSingleSubdevice = false;
bool useGlobalAtomics = false;
size_t numDevicesInContext = 1u;
bool areMultipleSubDevicesInContext = false;
};
struct CsrSizeRequestFlags {

2
shared/source/gen12lp/command_encoder_gen12lp.cpp
View File

@@ -52,7 +52,7 @@ void EncodeWA<Family>::encodeAdditionalPipelineSelect(Device &device, LinearStre
template <>
void EncodeSurfaceState<Family>::encodeExtraBufferParams(R_SURFACE_STATE *surfaceState, GraphicsAllocation *allocation, GmmHelper *gmmHelper,
bool isReadOnly, uint32_t numAvailableDevices, bool useGlobalAtomics, size_t numDevicesInContext) {
bool isReadOnly, uint32_t numAvailableDevices, bool useGlobalAtomics, bool areMultipleSubDevicesInContext) {
const bool isL3Allowed = surfaceState->getMemoryObjectControlState() == gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER);
if (isL3Allowed) {
const bool isConstantSurface = allocation && allocation->getAllocationType() == GraphicsAllocation::AllocationType::CONSTANT_SURFACE;

4
shared/source/gen8/state_base_address_gen8.cpp
View File

@@ -29,7 +29,7 @@ void StateBaseAddressHelper<BDWFamily>::programStateBaseAddress(
bool isMultiOsContextCapable,
MemoryCompressionState memoryCompressionState,
bool useGlobalAtomics,
size_t numDevicesInContext) {
bool areMultipleSubDevicesInContext) {
*stateBaseAddress = BDWFamily::cmdInitStateBaseAddress;
@@ -77,7 +77,7 @@ void StateBaseAddressHelper<BDWFamily>::programStateBaseAddress(
stateBaseAddress->setStatelessDataPortAccessMemoryObjectControlState(statelessMocsIndex);
appendStateBaseAddressParameters(stateBaseAddress, ssh, setGeneralStateBaseAddress, indirectObjectHeapBaseAddress,
gmmHelper, isMultiOsContextCapable, memoryCompressionState, true, useGlobalAtomics, numDevicesInContext);
gmmHelper, isMultiOsContextCapable, memoryCompressionState, true, useGlobalAtomics, areMultipleSubDevicesInContext);
}
template struct StateBaseAddressHelper<BDWFamily>;
} // namespace NEO

4
shared/source/helpers/state_base_address.h
View File

@@ -38,7 +38,7 @@ struct StateBaseAddressHelper {
bool isMultiOsContextCapable,
MemoryCompressionState memoryCompressionState,
bool useGlobalAtomics,
size_t numDevicesInContext);
bool areMultipleSubDevicesInContext);
static void appendStateBaseAddressParameters(
STATE_BASE_ADDRESS *stateBaseAddress,
@@ -50,7 +50,7 @@ struct StateBaseAddressHelper {
MemoryCompressionState memoryCompressionState,
bool overrideBindlessSurfaceStateBase,
bool useGlobalAtomics,
size_t nnumDevicesInContext);
bool areMultipleSubDevicesInContext);
static void appendExtraCacheSettings(STATE_BASE_ADDRESS *stateBaseAddress, GmmHelper *gmmHelper);

4
shared/source/helpers/state_base_address_base.inl
View File

@@ -33,7 +33,7 @@ void StateBaseAddressHelper<GfxFamily>::programStateBaseAddress(
bool isMultiOsContextCapable,
MemoryCompressionState memoryCompressionState,
bool useGlobalAtomics,
size_t numDevicesInContext) {
bool areMultipleSubDevicesInContext) {
*stateBaseAddress = GfxFamily::cmdInitStateBaseAddress;
bool overrideBindlessSurfaceStateBase = true;
@@ -102,7 +102,7 @@ void StateBaseAddressHelper<GfxFamily>::programStateBaseAddress(
stateBaseAddress->setStatelessDataPortAccessMemoryObjectControlState(statelessMocsIndex);
appendStateBaseAddressParameters(stateBaseAddress, ssh, setGeneralStateBaseAddress, indirectObjectHeapBaseAddress, gmmHelper,
isMultiOsContextCapable, memoryCompressionState, overrideBindlessSurfaceStateBase, useGlobalAtomics, numDevicesInContext);
isMultiOsContextCapable, memoryCompressionState, overrideBindlessSurfaceStateBase, useGlobalAtomics, areMultipleSubDevicesInContext);
}
template <typename GfxFamily>

2
shared/source/helpers/state_base_address_bdw.inl
View File

@@ -20,7 +20,7 @@ void StateBaseAddressHelper<GfxFamily>::appendStateBaseAddressParameters(
MemoryCompressionState memoryCompressionState,
bool overrideBindlessSurfaceStateBase,
bool useGlobalAtomics,
size_t numDevicesInContext) {
bool areMultipleSubDevicesInContext) {
appendExtraCacheSettings(stateBaseAddress, gmmHelper);
}

2
shared/source/helpers/state_base_address_skl_plus.inl
View File

@@ -20,7 +20,7 @@ void StateBaseAddressHelper<GfxFamily>::appendStateBaseAddressParameters(
MemoryCompressionState memoryCompressionState,
bool overrideBindlessSurfaceStateBase,
bool useGlobalAtomics,
size_t numDevicesInContext) {
bool areMultipleSubDevicesInContext) {
if (overrideBindlessSurfaceStateBase && ssh) {
stateBaseAddress->setBindlessSurfaceStateBaseAddressModifyEnable(true);

2
shared/test/common/helpers/dispatch_flags_helper.h
View File

@@ -41,6 +41,6 @@ struct DispatchFlagsHelper {
false, //usePerDssBackedBuffer
false, //useSingleSubdevice
false, //useGlobalAtomics
1u); //numDevicesInContext
false); //areMultipleSubDevicesInContext
}
};