intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-22 10:17:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: pass outImplicitArgs to patchImplicitArgs function

Browse Source 
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
This commit is contained in:
Kamil Kopryk
2024-08-05 16:40:51 +00:00
committed by Compute-Runtime-Automation
parent fdf7b98502
commit 2a9bcdeb83
15 changed files with 28 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
level_zero/core/source/cmdlist/cmdlist_hw_skl_to_tgllp.inl
View File

@@ -203,6 +203,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
nullptr, // outWalkerPtr
nullptr, // cpuWalkerBuffer
nullptr, // cpuPayloadBuffer
nullptr, // outImplicitArgsPtr
&additionalCommands, // additionalCommands
commandListPreemptionMode, // preemptionMode
launchParams.requiredPartitionDim, // requiredPartitionDim

1
level_zero/core/source/cmdlist/cmdlist_hw_xehp_and_later.inl
View File

@@ -348,6 +348,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
nullptr, // outWalkerPtr
launchParams.cmdWalkerBuffer, // cpuWalkerBuffer
launchParams.hostPayloadBuffer, // cpuPayloadBuffer
nullptr, // outImplicitArgsPtr
&additionalCommands, // additionalCommands
kernelPreemptionMode, // preemptionMode
launchParams.requiredPartitionDim, // requiredPartitionDim

1
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_1.cpp
View File

@@ -202,6 +202,7 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenNotEnoughSpaceInCommandStreamWhenA
nullptr, // outWalkerPtr
nullptr, // cpuWalkerBuffer
nullptr, // cpuPayloadBuffer
nullptr, // outImplicitArgsPtr
nullptr, // additionalCommands
PreemptionMode::MidBatch, // preemptionMode
NEO::RequiredPartitionDim::none, // requiredPartitionDim

1
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@@ -749,6 +749,7 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenNotEnoughSpaceInCommandStreamWhenA
nullptr, // outWalkerPtr
nullptr, // cpuWalkerBuffer
nullptr, // cpuPayloadBuffer
nullptr, // outImplicitArgsPtr
nullptr, // additionalCommands
PreemptionMode::MidBatch, // preemptionMode
NEO::RequiredPartitionDim::none, // requiredPartitionDim

2
opencl/source/helpers/hardware_commands_helper_bdw_and_later.inl
View File

@@ -92,7 +92,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendCrossThreadData(
auto implicitArgsGpuVA = indirectHeap.getGraphicsAllocation()->getGpuAddress() + indirectHeap.getUsed();
auto ptrToPatchImplicitArgs = indirectHeap.getSpace(sizeForImplicitArgsProgramming);
ImplicitArgsHelper::patchImplicitArgs(ptrToPatchImplicitArgs, *pImplicitArgs, kernelDescriptor, {}, rootDeviceEnvironment);
ImplicitArgsHelper::patchImplicitArgs(ptrToPatchImplicitArgs, *pImplicitArgs, kernelDescriptor, {}, rootDeviceEnvironment, nullptr);
auto implicitArgsCrossThreadPtr = ptrOffset(reinterpret_cast<uint64_t *>(kernel.getCrossThreadData()), kernelDescriptor.payloadMappings.implicitArgs.implicitArgsBuffer);
*implicitArgsCrossThreadPtr = implicitArgsGpuVA;

2
opencl/source/helpers/hardware_commands_helper_xehp_and_later.inl
View File

@@ -95,7 +95,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendCrossThreadData(
auto ptrToPatchImplicitArgs = indirectHeap.getSpace(sizeForImplicitArgsProgramming);
ImplicitArgsHelper::patchImplicitArgs(ptrToPatchImplicitArgs, *pImplicitArgs, kernelDescriptor, std::make_pair(generationOfLocalIdsByRuntime, requiredWalkOrder), rootDeviceEnvironment);
ImplicitArgsHelper::patchImplicitArgs(ptrToPatchImplicitArgs, *pImplicitArgs, kernelDescriptor, std::make_pair(generationOfLocalIdsByRuntime, requiredWalkOrder), rootDeviceEnvironment, nullptr);
}
uint32_t sizeToCopy = sizeCrossThreadData;

2
shared/source/command_container/command_encoder.h
View File

@@ -58,6 +58,7 @@ struct EncodeDispatchKernelArgs {
void *outWalkerPtr = nullptr;
void *cpuWalkerBuffer = nullptr;
void *cpuPayloadBuffer = nullptr;
void *outImplicitArgsPtr = nullptr;
std::list<void *> *additionalCommands = nullptr;
PreemptionMode preemptionMode = PreemptionMode::Initial;
NEO::RequiredPartitionDim requiredPartitionDim = NEO::RequiredPartitionDim::none;
@@ -210,7 +211,6 @@ struct EncodeDispatchKernel {
static void patchScratchAddressInImplicitArgs(ImplicitArgs &implicitArgs, uint64_t scratchAddress, bool scratchPtrPatchingRequired);
static size_t getInlineDataOffset(EncodeDispatchKernelArgs &args);
static void *getImplicitArgsAddress(EncodeDispatchKernelArgs &args, const KernelDescriptor &kernelDescriptor);
static size_t getScratchPtrOffsetOfImplicitArgs();
template <typename WalkerType>

5
shared/source/command_container/command_encoder.inl
View File

@@ -906,11 +906,6 @@ size_t EncodeDispatchKernel<Family>::getDefaultDshAlignment() {
return Family::cacheLineSize;
}
template <typename GfxFamily>
void *EncodeDispatchKernel<GfxFamily>::getImplicitArgsAddress(EncodeDispatchKernelArgs &args, const KernelDescriptor &kernelDescriptor) {
return nullptr;
}
template <typename GfxFamily>
size_t EncodeDispatchKernel<GfxFamily>::getScratchPtrOffsetOfImplicitArgs() {
return 0;

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

@@ -198,7 +198,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
auto implicitArgsCrossThreadPtr = ptrOffset(const_cast<uint64_t *>(reinterpret_cast<const uint64_t *>(args.dispatchInterface->getCrossThreadData())), kernelDescriptor.payloadMappings.implicitArgs.implicitArgsBuffer);
*implicitArgsCrossThreadPtr = implicitArgsGpuVA;
ptr = NEO::ImplicitArgsHelper::patchImplicitArgs(ptr, *pImplicitArgs, kernelDescriptor, {}, rootDeviceEnvironment);
ptr = NEO::ImplicitArgsHelper::patchImplicitArgs(ptr, *pImplicitArgs, kernelDescriptor, {}, rootDeviceEnvironment, nullptr);
}
memcpy_s(ptr, sizeCrossThreadData,

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

@@ -267,7 +267,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
pImplicitArgs->localIdTablePtr = heap->getGraphicsAllocation()->getGpuAddress() + heap->getUsed() - iohRequiredSize;
EncodeDispatchKernel<Family>::patchScratchAddressInImplicitArgs<heaplessModeEnabled>(*pImplicitArgs, scratchAddressForImmediatePatching, args.immediateScratchAddressPatching);
ptr = NEO::ImplicitArgsHelper::patchImplicitArgs(ptr, *pImplicitArgs, kernelDescriptor, std::make_pair(localIdsGenerationByRuntime, requiredWorkgroupOrder), rootDeviceEnvironment);
ptr = NEO::ImplicitArgsHelper::patchImplicitArgs(ptr, *pImplicitArgs, kernelDescriptor, std::make_pair(localIdsGenerationByRuntime, requiredWorkgroupOrder), rootDeviceEnvironment, &args.outImplicitArgsPtr);
}
if (args.isIndirect) {

7
shared/source/kernel/implicit_args_helper.cpp
View File

@@ -66,7 +66,7 @@ uint32_t getSizeForImplicitArgsPatching(const ImplicitArgs *pImplicitArgs, const
}
}
void *patchImplicitArgs(void *ptrToPatch, const ImplicitArgs &implicitArgs, const KernelDescriptor &kernelDescriptor, std::optional<std::pair<bool, uint32_t>> hwGenerationOfLocalIdsParams, const RootDeviceEnvironment &rootDeviceEnvironment) {
void *patchImplicitArgs(void *ptrToPatch, const ImplicitArgs &implicitArgs, const KernelDescriptor &kernelDescriptor, std::optional<std::pair<bool, uint32_t>> hwGenerationOfLocalIdsParams, const RootDeviceEnvironment &rootDeviceEnvironment, void **outImplicitArgsAddress) {
auto localIdsGeneratedByHw = hwGenerationOfLocalIdsParams.has_value() ? hwGenerationOfLocalIdsParams.value().first : false;
auto totalSizeToProgram = getSizeForImplicitArgsPatching(&implicitArgs, kernelDescriptor, localIdsGeneratedByHw, rootDeviceEnvironment);
auto retVal = ptrOffset(ptrToPatch, totalSizeToProgram);
@@ -89,6 +89,11 @@ void *patchImplicitArgs(void *ptrToPatch, const ImplicitArgs &implicitArgs, cons
auto sizeForLocalIdsProgramming = totalSizeToProgram - ImplicitArgs::getSize();
ptrToPatch = ptrOffset(ptrToPatch, sizeForLocalIdsProgramming);
}
if (outImplicitArgsAddress) {
*outImplicitArgsAddress = ptrToPatch;
}
memcpy_s(ptrToPatch, ImplicitArgs::getSize(), &implicitArgs, ImplicitArgs::getSize());
return retVal;

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

@@ -25,6 +25,6 @@ namespace ImplicitArgsHelper {
std::array<uint8_t, 3> getDimensionOrderForLocalIds(const uint8_t *workgroupDimensionsOrder, std::optional<std::pair<bool /* localIdsGeneratedByRuntime */, uint32_t /* walkOrderForHwGenerationOfLocalIds */>> hwGenerationOfLocalIdsParams);
uint32_t getGrfSize(uint32_t simd);
uint32_t getSizeForImplicitArgsPatching(const ImplicitArgs *pImplicitArgs, const KernelDescriptor &kernelDescriptor, bool localIdsGeneratedByRuntime, const RootDeviceEnvironment &rootDeviceEnvironment);
void *patchImplicitArgs(void *ptrToPatch, const ImplicitArgs &implicitArgs, const KernelDescriptor &kernelDescriptor, std::optional<std::pair<bool /* localIdsGeneratedByRuntime */, uint32_t /* walkOrderForHwGenerationOfLocalIds */>> hwGenerationOfLocalIdsParams, const RootDeviceEnvironment &rootDeviceEnvironment);
void *patchImplicitArgs(void *ptrToPatch, const ImplicitArgs &implicitArgs, const KernelDescriptor &kernelDescriptor, std::optional<std::pair<bool /* localIdsGeneratedByRuntime */, uint32_t /* walkOrderForHwGenerationOfLocalIds */>> hwGenerationOfLocalIdsParams, const RootDeviceEnvironment &rootDeviceEnvironment, void **outImplicitArgs);
} // namespace ImplicitArgsHelper
} // namespace NEO

8
shared/test/unit_test/command_container/command_encoder_tests.cpp
View File

@@ -778,14 +778,6 @@ HWTEST_F(CommandEncoderTests, givenInterfaceDescriptorWhenEncodeEuSchedulingPoli
}
}
HWTEST_F(CommandEncoderTests, whenGetImplicitArgsAddressIsCalledThenNullptrIsReturned) {
KernelDescriptor kernelDescriptor{};
EncodeDispatchKernelArgs args{};
auto implicitArgsPtr = EncodeDispatchKernel<FamilyType>::getImplicitArgsAddress(args, kernelDescriptor);
EXPECT_EQ(nullptr, implicitArgsPtr);
}
HWTEST_F(CommandEncoderTests, whenGetScratchPtrOffsetOfImplicitArgsIsCalledThenZeroIsReturned) {
auto scratchOffset = EncodeDispatchKernel<FamilyType>::getScratchPtrOffsetOfImplicitArgs();

1
shared/test/unit_test/fixtures/command_container_fixture.cpp
View File

@@ -51,6 +51,7 @@ EncodeDispatchKernelArgs CommandEncodeStatesFixture::createDefaultDispatchKernel
nullptr, // outWalkerPtr
nullptr, // cpuWalkerBuffer
nullptr, // cpuPayloadBuffer
nullptr, // outImplicitArgsPtr
nullptr, // additionalCommands
PreemptionMode::Disabled, // preemptionMode
NEO::RequiredPartitionDim::none, // requiredPartitionDim

14
shared/test/unit_test/kernel/implicit_args_helper_tests.cpp
View File

@@ -102,6 +102,7 @@ TEST(ImplicitArgsHelperTest, givenImplicitArgsWithImplicitArgsBufferOffsetInPayl
TEST(ImplicitArgsHelperTest, givenImplicitArgsWithoutImplicitArgsBufferOffsetInPayloadMappingWhenPatchingImplicitArgsThenOnlyProperRegionIsPatched) {
ImplicitArgs implicitArgs{ImplicitArgs::getSize()};
void *outImplicitArgs = nullptr;
KernelDescriptor kernelDescriptor{};
kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[0] = 0;
kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[1] = 1;
@@ -119,6 +120,7 @@ TEST(ImplicitArgsHelperTest, givenImplicitArgsWithoutImplicitArgsBufferOffsetInP
auto totalWorkgroupSize = implicitArgs.localSizeX * implicitArgs.localSizeY * implicitArgs.localSizeZ;
auto localIdsPatchingSize = totalWorkgroupSize * 3 * sizeof(uint16_t);
auto localIdsOffset = alignUp(localIdsPatchingSize, MemoryConstants::cacheLineSize);
auto memoryToPatch = std::make_unique<uint8_t[]>(totalSizeForPatching);
@@ -126,10 +128,13 @@ TEST(ImplicitArgsHelperTest, givenImplicitArgsWithoutImplicitArgsBufferOffsetInP
memset(memoryToPatch.get(), pattern, totalSizeForPatching);
auto retVal = ImplicitArgsHelper::patchImplicitArgs(memoryToPatch.get(), implicitArgs, kernelDescriptor, {}, rootDeviceEnvironment);
auto retVal = ImplicitArgsHelper::patchImplicitArgs(memoryToPatch.get(), implicitArgs, kernelDescriptor, {}, rootDeviceEnvironment, &outImplicitArgs);
EXPECT_EQ(retVal, ptrOffset(memoryToPatch.get(), totalSizeForPatching));
void *expectedImplicitArgsPtr = ptrOffset(memoryToPatch.get(), localIdsOffset);
EXPECT_EQ(expectedImplicitArgsPtr, outImplicitArgs);
uint32_t offset = 0;
for (; offset < localIdsPatchingSize; offset++) {
@@ -147,7 +152,7 @@ TEST(ImplicitArgsHelperTest, givenImplicitArgsWithoutImplicitArgsBufferOffsetInP
TEST(ImplicitArgsHelperTest, givenImplicitArgsWithImplicitArgsBufferOffsetInPayloadMappingWhenPatchingImplicitArgsThenOnlyProperRegionIsPatched) {
ImplicitArgs implicitArgs{ImplicitArgs::getSize()};
void *outImplicitArgs = nullptr;
KernelDescriptor kernelDescriptor{};
kernelDescriptor.payloadMappings.implicitArgs.implicitArgsBuffer = 0x10;
EXPECT_TRUE(isValidOffset<>(kernelDescriptor.payloadMappings.implicitArgs.implicitArgsBuffer));
@@ -168,10 +173,13 @@ TEST(ImplicitArgsHelperTest, givenImplicitArgsWithImplicitArgsBufferOffsetInPayl
memset(memoryToPatch.get(), pattern, totalSizeForPatching);
auto retVal = ImplicitArgsHelper::patchImplicitArgs(memoryToPatch.get(), implicitArgs, kernelDescriptor, {}, rootDeviceEnvironment);
auto retVal = ImplicitArgsHelper::patchImplicitArgs(memoryToPatch.get(), implicitArgs, kernelDescriptor, {}, rootDeviceEnvironment, &outImplicitArgs);
EXPECT_EQ(retVal, ptrOffset(memoryToPatch.get(), totalSizeForPatching));
void *expectedImplicitArgsPtr = memoryToPatch.get();
EXPECT_EQ(expectedImplicitArgsPtr, outImplicitArgs);
uint32_t offset = 0;
for (; offset < ImplicitArgs::getSize(); offset++) {