intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-01 04:23:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Remove isCleanLeftoverMemoryRequired() + refactor sampler support path

Browse Source 
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2022-09-30 13:20:48 +00:00
committed by Compute-Runtime-Automation
parent 6073702941
commit 52b63be026
11 changed files with 59 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -126,31 +126,33 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
PreemptionHelper::programInterfaceDescriptorDataPreemption<Family>(&idd, args.preemptionMode);
if constexpr (Family::supportsSampler) {
uint32_t samplerStateOffset = 0;
uint32_t samplerCount = 0;
uint32_t samplerCount = 0;
if (kernelDescriptor.payloadMappings.samplerTable.numSamplers > 0) {
auto heap = ApiSpecificConfig::getBindlessConfiguration() ? args.device->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : container.getIndirectHeap(HeapType::DYNAMIC_STATE);
UNRECOVERABLE_IF(!heap);
if (args.device->getDeviceInfo().imageSupport) {
if constexpr (Family::supportsSampler) {
uint32_t samplerStateOffset = 0;
samplerCount = kernelDescriptor.payloadMappings.samplerTable.numSamplers;
samplerStateOffset = EncodeStates<Family>::copySamplerState(
heap, kernelDescriptor.payloadMappings.samplerTable.tableOffset,
kernelDescriptor.payloadMappings.samplerTable.numSamplers, kernelDescriptor.payloadMappings.samplerTable.borderColor,
args.dispatchInterface->getDynamicStateHeapData(),
args.device->getBindlessHeapsHelper(), hwInfo);
if (ApiSpecificConfig::getBindlessConfiguration()) {
container.getResidencyContainer().push_back(args.device->getBindlessHeapsHelper()->getHeap(NEO::BindlessHeapsHelper::BindlesHeapType::GLOBAL_DSH)->getGraphicsAllocation());
if (kernelDescriptor.payloadMappings.samplerTable.numSamplers > 0) {
auto heap = ApiSpecificConfig::getBindlessConfiguration() ? args.device->getBindlessHeapsHelper()->getHeap(BindlessHeapsHelper::GLOBAL_DSH) : container.getIndirectHeap(HeapType::DYNAMIC_STATE);
UNRECOVERABLE_IF(!heap);
samplerCount = kernelDescriptor.payloadMappings.samplerTable.numSamplers;
samplerStateOffset = EncodeStates<Family>::copySamplerState(
heap, kernelDescriptor.payloadMappings.samplerTable.tableOffset,
kernelDescriptor.payloadMappings.samplerTable.numSamplers, kernelDescriptor.payloadMappings.samplerTable.borderColor,
args.dispatchInterface->getDynamicStateHeapData(),
args.device->getBindlessHeapsHelper(), hwInfo);
if (ApiSpecificConfig::getBindlessConfiguration()) {
container.getResidencyContainer().push_back(args.device->getBindlessHeapsHelper()->getHeap(NEO::BindlessHeapsHelper::BindlesHeapType::GLOBAL_DSH)->getGraphicsAllocation());
}
}
}
idd.setSamplerStatePointer(samplerStateOffset);
EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(idd, samplerCount, bindingTableStateCount);
} else {
EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(idd, 0u, bindingTableStateCount);
idd.setSamplerStatePointer(samplerStateOffset);
}
}
EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(idd, samplerCount, bindingTableStateCount);
uint64_t offsetThreadData = 0u;
const uint32_t inlineDataSize = sizeof(INLINE_DATA);
auto crossThreadData = args.dispatchInterface->getCrossThreadData();

3
shared/source/command_stream/command_stream_receiver_hw.h
View File

@@ -86,7 +86,8 @@ class CommandStreamReceiverHw : public CommandStreamReceiver {
const LinearStream *dsh,
const LinearStream *ioh,
const LinearStream *ssh,
uint64_t generalStateBase);
uint64_t generalStateBase,
bool imagesSupported);
void collectStateBaseAddresIohPatchInfo(uint64_t commandBufferAddress, uint64_t commandOffset, const LinearStream &ioh);

8
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@@ -444,7 +444,8 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
EncodeWA<GfxFamily>::encodeAdditionalPipelineSelect(commandStreamCSR, dispatchFlags.pipelineSelectArgs, false, hwInfo, isRcs());
if (DebugManager.flags.AddPatchInfoCommentsForAUBDump.get()) {
collectStateBaseAddresPatchInfo(commandStream.getGraphicsAllocation()->getGpuAddress(), stateBaseAddressCmdOffset, dsh, ioh, ssh, newGSHbase);
collectStateBaseAddresPatchInfo(commandStream.getGraphicsAllocation()->getGpuAddress(), stateBaseAddressCmdOffset, dsh, ioh, ssh, newGSHbase,
device.getDeviceInfo().imageSupport);
}
}
@@ -998,10 +999,11 @@ void CommandStreamReceiverHw<GfxFamily>::collectStateBaseAddresPatchInfo(
const LinearStream *dsh,
const LinearStream *ioh,
const LinearStream *ssh,
uint64_t generalStateBase) {
uint64_t generalStateBase,
bool imagesSupported) {
typedef typename GfxFamily::STATE_BASE_ADDRESS STATE_BASE_ADDRESS;
if constexpr (GfxFamily::supportsSampler) {
if (imagesSupported) {
PatchInfoData dynamicStatePatchInfo = {dsh->getGraphicsAllocation()->getGpuAddress(), 0u, PatchInfoAllocationType::DynamicStateHeap, baseAddress, commandOffset + STATE_BASE_ADDRESS::PATCH_CONSTANTS::DYNAMICSTATEBASEADDRESS_BYTEOFFSET, PatchInfoAllocationType::Default};
flatBatchBufferHelper->setPatchInfoData(dynamicStatePatchInfo);
}

24
shared/test/unit_test/encoders/test_encode_dispatch_kernel_xehp_and_later.cpp
View File

@@ -251,6 +251,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, CommandEncodeStatesTest, giveNumBindingTableZeroWhe
EXPECT_EQ(idd.getBindingTablePointer(), 0u);
}
struct SamplerSupportedMatcher {
template <PRODUCT_FAMILY productFamily>
static constexpr bool isMatched() {
@@ -265,6 +266,11 @@ HWTEST2_F(CommandEncodeStatesTest, giveNumSamplersOneWhenDispatchKernelThensampl
using SAMPLER_STATE = typename FamilyType::SAMPLER_STATE;
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
using WALKER_TYPE = typename FamilyType::WALKER_TYPE;
if (!pDevice->getDeviceInfo().imageSupport) {
GTEST_SKIP();
}
uint32_t numSamplers = 1;
SAMPLER_STATE samplerState;
memset(&samplerState, 2, sizeof(SAMPLER_STATE));
@@ -421,10 +427,12 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, CommandEncodeStatesTest, givenForceBtpPrefetchModeD
}
if constexpr (FamilyType::supportsSampler) {
if (EncodeSurfaceState<FamilyType>::doBindingTablePrefetch()) {
EXPECT_NE(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
} else {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
if (pDevice->getDeviceInfo().imageSupport) {
if (EncodeSurfaceState<FamilyType>::doBindingTablePrefetch()) {
EXPECT_NE(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
} else {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
}
}
}
}
@@ -447,7 +455,9 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, CommandEncodeStatesTest, givenForceBtpPrefetchModeD
EXPECT_EQ(0u, idd.getBindingTableEntryCount());
if constexpr (FamilyType::supportsSampler) {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
if (pDevice->getDeviceInfo().imageSupport) {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
}
}
}
@@ -469,7 +479,9 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, CommandEncodeStatesTest, givenForceBtpPrefetchModeD
EXPECT_NE(0u, idd.getBindingTableEntryCount());
if constexpr (FamilyType::supportsSampler) {
EXPECT_NE(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
if (pDevice->getDeviceInfo().imageSupport) {
EXPECT_NE(INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT_NO_SAMPLERS_USED, idd.getSamplerCount());
}
}
}
}

4
shared/test/unit_test/encoders/test_encode_states.cpp
View File

@@ -316,7 +316,7 @@ HWTEST2_F(CommandEncodeStatesTest, givenCommandContainerWithDirtyHeapsWhenSetSta
auto itorCmd = find<STATE_BASE_ADDRESS *>(commands.begin(), commands.end());
auto pCmd = genCmdCast<STATE_BASE_ADDRESS *>(*itorCmd);
if constexpr (FamilyType::supportsSampler) {
if (pDevice->getDeviceInfo().imageSupport) {
EXPECT_EQ(dsh->getHeapGpuBase(), pCmd->getDynamicStateBaseAddress());
} else {
EXPECT_EQ(dsh, nullptr);
@@ -356,7 +356,7 @@ HWTEST_F(CommandEncodeStatesTest, givenCommandContainerWhenSetStateBaseAddressCa
ASSERT_NE(itorCmd, commands.end());
auto cmd = genCmdCast<STATE_BASE_ADDRESS *>(*itorCmd);
if constexpr (FamilyType::supportsSampler) {
if (pDevice->getDeviceInfo().imageSupport) {
EXPECT_NE(dsh->getHeapGpuBase(), cmd->getDynamicStateBaseAddress());
} else {
EXPECT_EQ(dsh, nullptr);