Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improve buffer surface state programming

Browse Source 
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2021-01-29 15:40:17 +00:00
committed by Compute-Runtime-Automation
parent 3245e86485
commit 580fdd757c
16 changed files with 60 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
opencl/source/mem_obj/image.inl
View File

@ -89,7 +89,7 @@ void ImageHw<GfxFamily>::setImageArg(void *memory, bool setAsMediaBlockImage, ui
if (imageDesc.num_samples > 1) {
setAuxParamsForMultisamples(surfaceState);
} else if (gmm && gmm->isRenderCompressed) {
EncodeSurfaceState<GfxFamily>::setAuxParamsForCCS(surfaceState, gmm);
EncodeSurfaceState<GfxFamily>::setImageAuxParamsForCCS(surfaceState, gmm);
}
appendSurfaceStateDepthParams(surfaceState, gmm);
EncodeSurfaceState<GfxFamily>::appendImageCompressionParams(surfaceState, graphicsAllocation, gmmHelper, isImageFromBuffer());
@ -111,7 +111,7 @@ void ImageHw<GfxFamily>::setAuxParamsForMultisamples(RENDER_SURFACE_STATE *surfa
EncodeSurfaceState<GfxFamily>::setClearColorParams(surfaceState, mcsGmm);
setUnifiedAuxBaseAddress<GfxFamily>(surfaceState, mcsGmm);
} else if (mcsGmm->unifiedAuxTranslationCapable()) {
EncodeSurfaceState<GfxFamily>::setAuxParamsForCCS(surfaceState, mcsGmm);
EncodeSurfaceState<GfxFamily>::setImageAuxParamsForCCS(surfaceState, mcsGmm);
} else {
surfaceState->setAuxiliarySurfaceMode((typename RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE)1);
surfaceState->setAuxiliarySurfacePitch(mcsSurfaceInfo.pitch);

24
opencl/test/unit_test/built_ins/built_in_tests.cpp
View File

@ -41,6 +41,7 @@
#include "compiler_options.h"
#include "gtest/gtest.h"
#include "os_inc.h"
#include "test_traits_common.h"
#include <string>
@ -130,6 +131,13 @@ struct AuxBuiltInTests : BuiltInTests, public ::testing::WithParamInterface<Kern
KernelObjForAuxTranslation::Type kernelObjType;
};
struct AuxBuiltinsMatcher {
template <PRODUCT_FAMILY productFamily>
static constexpr bool isMatched() {
return TestTraits<NEO::ToGfxCoreFamily<productFamily>::get()>::auxBuiltinsSupported;
}
};
INSTANTIATE_TEST_CASE_P(,
AuxBuiltInTests,
testing::ValuesIn({KernelObjForAuxTranslation::Type::MEM_OBJ, KernelObjForAuxTranslation::Type::GFX_ALLOC}));
@ -235,7 +243,7 @@ TEST_F(BuiltInTests, GivenCopyBufferToBufferWhenDispatchInfoIsCreatedThenParamsA
delete dstPtr;
}
HWTEST_P(AuxBuiltInTests, givenInputBufferWhenBuildingNonAuxDispatchInfoForAuxTranslationThenPickAndSetupCorrectKernels) {
HWTEST2_P(AuxBuiltInTests, givenInputBufferWhenBuildingNonAuxDispatchInfoForAuxTranslationThenPickAndSetupCorrectKernels, AuxBuiltinsMatcher) {
BuiltinDispatchInfoBuilder &baseBuilder = BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(EBuiltInOps::AuxTranslation, *pClDevice);
auto &builder = static_cast<BuiltInOp<EBuiltInOps::AuxTranslation> &>(baseBuilder);
@ -301,7 +309,7 @@ HWTEST_P(AuxBuiltInTests, givenInputBufferWhenBuildingNonAuxDispatchInfoForAuxTr
EXPECT_NE(builtinKernels[1], builtinKernels[2]);
}
HWTEST_P(AuxBuiltInTests, givenInputBufferWhenBuildingAuxDispatchInfoForAuxTranslationThenPickAndSetupCorrectKernels) {
HWTEST2_P(AuxBuiltInTests, givenInputBufferWhenBuildingAuxDispatchInfoForAuxTranslationThenPickAndSetupCorrectKernels, AuxBuiltinsMatcher) {
BuiltinDispatchInfoBuilder &baseBuilder = BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(EBuiltInOps::AuxTranslation, *pClDevice);
auto &builder = static_cast<BuiltInOp<EBuiltInOps::AuxTranslation> &>(baseBuilder);
@ -367,7 +375,7 @@ HWTEST_P(AuxBuiltInTests, givenInputBufferWhenBuildingAuxDispatchInfoForAuxTrans
EXPECT_NE(builtinKernels[1], builtinKernels[2]);
}
HWTEST_P(AuxBuiltInTests, givenInputBufferWhenBuildingAuxTranslationDispatchThenPickDifferentKernelsDependingOnRequest) {
HWTEST2_P(AuxBuiltInTests, givenInputBufferWhenBuildingAuxTranslationDispatchThenPickDifferentKernelsDependingOnRequest, AuxBuiltinsMatcher) {
BuiltinDispatchInfoBuilder &baseBuilder = BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(EBuiltInOps::AuxTranslation, *pClDevice);
auto &builder = static_cast<BuiltInOp<EBuiltInOps::AuxTranslation> &>(baseBuilder);
@ -405,7 +413,7 @@ HWTEST_P(AuxBuiltInTests, givenInputBufferWhenBuildingAuxTranslationDispatchThen
EXPECT_NE(builtinKernels[2], builtinKernels[5]);
}
HWTEST_P(AuxBuiltInTests, givenInvalidAuxTranslationDirectionWhenBuildingDispatchInfosThenAbort) {
HWTEST2_P(AuxBuiltInTests, givenInvalidAuxTranslationDirectionWhenBuildingDispatchInfosThenAbort, AuxBuiltinsMatcher) {
BuiltinDispatchInfoBuilder &baseBuilder = BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(EBuiltInOps::AuxTranslation, *pClDevice);
auto &builder = static_cast<BuiltInOp<EBuiltInOps::AuxTranslation> &>(baseBuilder);
@ -441,7 +449,7 @@ TEST_F(BuiltInTests, whenAuxBuiltInIsConstructedThenResizeKernelInstancedTo5) {
EXPECT_EQ(5u, mockAuxBuiltInOp.convertToNonAuxKernel.size());
}
HWTEST_P(AuxBuiltInTests, givenMoreKernelObjectsForAuxTranslationThanKernelInstancesWhenDispatchingThenResize) {
HWTEST2_P(AuxBuiltInTests, givenMoreKernelObjectsForAuxTranslationThanKernelInstancesWhenDispatchingThenResize, AuxBuiltinsMatcher) {
MockAuxBuilInOp mockAuxBuiltInOp(*pBuiltIns, *pClDevice);
EXPECT_EQ(5u, mockAuxBuiltInOp.convertToAuxKernel.size());
EXPECT_EQ(5u, mockAuxBuiltInOp.convertToNonAuxKernel.size());
@ -482,7 +490,7 @@ TEST_F(BuiltInTests, givenkAuxBuiltInWhenResizeIsCalledThenCloneAllNewInstancesF
}
}
HWTEST_P(AuxBuiltInTests, givenKernelWithAuxTranslationRequiredWhenEnqueueCalledThenLockOnBuiltin) {
HWTEST2_P(AuxBuiltInTests, givenKernelWithAuxTranslationRequiredWhenEnqueueCalledThenLockOnBuiltin, AuxBuiltinsMatcher) {
BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(EBuiltInOps::AuxTranslation, *pClDevice);
auto mockAuxBuiltInOp = new MockAuxBuilInOp(*pBuiltIns, *pClDevice);
pBuiltIns->BuiltinOpsBuilders[static_cast<uint32_t>(EBuiltInOps::AuxTranslation)].first.reset(mockAuxBuiltInOp);
@ -585,7 +593,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, AuxBuiltInTests, givenAuxTranslationKernelWhenSettin
}
}
HWTEST_P(AuxBuiltInTests, givenAuxToNonAuxTranslationWhenSettingSurfaceStateThenSetValidAuxMode) {
HWTEST2_P(AuxBuiltInTests, givenAuxToNonAuxTranslationWhenSettingSurfaceStateThenSetValidAuxMode, AuxBuiltinsMatcher) {
if (this->pDevice->areSharedSystemAllocationsAllowed()) {
GTEST_SKIP();
}
@ -643,7 +651,7 @@ HWTEST_P(AuxBuiltInTests, givenAuxToNonAuxTranslationWhenSettingSurfaceStateThen
}
}
HWTEST_P(AuxBuiltInTests, givenNonAuxToAuxTranslationWhenSettingSurfaceStateThenSetValidAuxMode) {
HWTEST2_P(AuxBuiltInTests, givenNonAuxToAuxTranslationWhenSettingSurfaceStateThenSetValidAuxMode, AuxBuiltinsMatcher) {
if (this->pDevice->areSharedSystemAllocationsAllowed()) {
GTEST_SKIP();
}

4
opencl/test/unit_test/gen12lp/image_tests_gen12lp.inl
View File

@ -120,7 +120,7 @@ GEN12LPTEST_F(gen12LpImageTests, givenRenderCompressionThenSurfaceStateParamsAre
auto surfaceState = FamilyType::cmdInitRenderSurfaceState;
auto imageHw = static_cast<ImageHw<FamilyType> *>(image.get());
imageHw->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getDefaultGmm()->gmmResourceInfo->getResourceFlags()->Info.RenderCompressed = true;
EncodeSurfaceState<FamilyType>::setAuxParamsForCCS(&surfaceState, imageHw->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getDefaultGmm());
EncodeSurfaceState<FamilyType>::setImageAuxParamsForCCS(&surfaceState, imageHw->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getDefaultGmm());
EXPECT_FALSE(surfaceState.getMemoryCompressionEnable());
EXPECT_EQ(surfaceState.getAuxiliarySurfaceMode(), RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E);
@ -136,7 +136,7 @@ GEN12LPTEST_F(gen12LpImageTests, givenMediaCompressionThenSurfaceStateParamsAreS
auto imageHw = static_cast<ImageHw<FamilyType> *>(image.get());
imageHw->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getDefaultGmm()->gmmResourceInfo->getResourceFlags()->Info.MediaCompressed = true;
surfaceState.setAuxiliarySurfaceMode(RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E);
EncodeSurfaceState<FamilyType>::setAuxParamsForCCS(&surfaceState, imageHw->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getDefaultGmm());
EncodeSurfaceState<FamilyType>::setImageAuxParamsForCCS(&surfaceState, imageHw->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getDefaultGmm());
EXPECT_TRUE(surfaceState.getMemoryCompressionEnable());
EXPECT_EQ(surfaceState.getAuxiliarySurfaceMode(), RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE);

2
opencl/test/unit_test/helpers/hw_helper_tests.cpp
View File

@ -455,7 +455,7 @@ HWTEST_F(HwHelperTest, givenCreatedSurfaceStateBufferWhenGmmAndAllocationCompres
SURFACE_TYPE type = RENDER_SURFACE_STATE::SURFACE_TYPE_SURFTYPE_BUFFER;
helper.setRenderSurfaceStateForBuffer(rootDeviceEnvironment, stateBuffer, size, addr, 0, pitch, &allocation, false, type, false, false);
EXPECT_EQ(RENDER_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT, state->getCoherencyType());
EXPECT_EQ(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E, state->getAuxiliarySurfaceMode());
EXPECT_TRUE(EncodeSurfaceState<FamilyType>::isAuxModeEnabled(state, allocation.getDefaultGmm()));
delete allocation.getDefaultGmm();
alignedFree(stateBuffer);

4
opencl/test/unit_test/mem_obj/buffer_set_arg_tests.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2017-2020 Intel Corporation
* Copyright (C) 2017-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@ -199,7 +199,7 @@ HWTEST_F(BufferSetArgTest, givenNonPureStatefulArgWhenRenderCompressedBufferIsSe
pKernelInfo->kernelArgInfo.at(0).pureStatefulBufferAccess = true;
ret = pKernel->setArgBuffer(0, sizeof(cl_mem), &clMem);
EXPECT_EQ(CL_SUCCESS, ret);
EXPECT_TRUE(RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E == surfaceState->getAuxiliarySurfaceMode());
EXPECT_TRUE(EncodeSurfaceState<FamilyType>::isAuxModeEnabled(surfaceState, graphicsAllocation->getDefaultGmm()));
}
TEST_F(BufferSetArgTest, Given32BitAddressingWhenSettingArgStatelessThenGpuAddressIsSetCorrectly) {

2
opencl/test/unit_test/mem_obj/buffer_tests.cpp
View File

@ -1553,7 +1553,7 @@ HWTEST_F(BufferSetSurfaceTests, givenRenderCompressedGmmResourceWhenSurfaceState
buffer->setArgStateful(&surfaceState, false, false, false, false, context.getDevice(0)->getDevice());
EXPECT_EQ(0u, surfaceState.getAuxiliarySurfaceBaseAddress());
EXPECT_TRUE(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E == surfaceState.getAuxiliarySurfaceMode());
EXPECT_TRUE(EncodeSurfaceState<FamilyType>::isAuxModeEnabled(&surfaceState, gmm));
EXPECT_TRUE(RENDER_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT == surfaceState.getCoherencyType());
}

12
opencl/test/unit_test/mem_obj/image_set_arg_tests.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2017-2020 Intel Corporation
* Copyright (C) 2017-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@ -690,13 +690,15 @@ HWTEST_F(ImageSetArgTest, givenMcsAllocationWhenSetArgIsCalledWithUnifiedAuxCapa
mockMcsGmmResInfo->setUnifiedAuxTranslationCapable();
EXPECT_TRUE(mcsAlloc->getDefaultGmm()->unifiedAuxTranslationCapable());
mcsAlloc->getDefaultGmm()->isRenderCompressed = true;
retVal = clSetKernelArg(pKernel, 0, sizeof(memObj), &memObj);
ASSERT_EQ(CL_SUCCESS, retVal);
auto surfaceState = reinterpret_cast<const RENDER_SURFACE_STATE *>(ptrOffset(pKernel->getSurfaceStateHeap(rootDeviceIndex),
pKernelInfo->kernelArgInfo[0].offsetHeap));
auto surfaceState = reinterpret_cast<RENDER_SURFACE_STATE *>(ptrOffset(pKernel->getSurfaceStateHeap(rootDeviceIndex),
pKernelInfo->kernelArgInfo[0].offsetHeap));
EXPECT_TRUE(surfaceState->getAuxiliarySurfaceMode() == AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E);
EXPECT_TRUE(EncodeSurfaceState<FamilyType>::isAuxModeEnabled(surfaceState, mcsAlloc->getDefaultGmm()));
EXPECT_EQ(1u, surfaceState->getAuxiliarySurfacePitch());
EXPECT_EQ(0u, surfaceState->getAuxiliarySurfaceQpitch());
}
@ -885,7 +887,7 @@ HWTEST_F(ImageSetArgTest, givenRenderCompressedResourceWhenSettingImgArgThenSetC
srcImage->setImageArg(&surfaceState, false, 0, pClDevice->getRootDeviceIndex());
EXPECT_TRUE(surfaceState.getAuxiliarySurfaceMode() == AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E);
EXPECT_TRUE(EncodeSurfaceState<FamilyType>::isAuxModeEnabled(&surfaceState, srcAllocation->getDefaultGmm()));
EXPECT_EQ(1u, surfaceState.getAuxiliarySurfacePitch());
EXPECT_EQ(0u, surfaceState.getAuxiliarySurfaceQpitch());
}