Media
/
compute-runtime
mirror of https://github.com/intel/compute-runtime.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
compute-runtime/unit_tests/command_queue/enqueue_copy_image_tests.cpp

358 lines
16 KiB
C++
Raw Blame History

/*
* Copyright (c) 2017, Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include <algorithm>
#include "runtime/gen_common/reg_configs.h"
#include "unit_tests/command_queue/enqueue_copy_image_fixture.h"
#include "unit_tests/gen_common/gen_commands_common_validation.h"
#include "unit_tests/mocks/mock_builtin_dispatch_info_builder.h"
#include "test.h"
using namespace OCLRT;
HWTEST_F(EnqueueCopyImageTest, gpgpuWalker) {
typedef typename FamilyType::GPGPU_WALKER GPGPU_WALKER;
enqueueCopyImage<FamilyType>();
auto *cmd = reinterpret_cast<GPGPU_WALKER *>(cmdWalker);
ASSERT_NE(nullptr, cmd);
// Verify GPGPU_WALKER parameters
EXPECT_NE(0u, cmd->getThreadGroupIdXDimension());
EXPECT_NE(0u, cmd->getThreadGroupIdYDimension());
EXPECT_NE(0u, cmd->getThreadGroupIdZDimension());
EXPECT_NE(0u, cmd->getRightExecutionMask());
EXPECT_NE(0u, cmd->getBottomExecutionMask());
EXPECT_EQ(GPGPU_WALKER::SIMD_SIZE_SIMD32, cmd->getSimdSize());
EXPECT_NE(0u, cmd->getIndirectDataLength());
EXPECT_FALSE(cmd->getIndirectParameterEnable());
// Compute the SIMD lane mask
size_t simd =
cmd->getSimdSize() == GPGPU_WALKER::SIMD_SIZE_SIMD32 ? 32 : cmd->getSimdSize() == GPGPU_WALKER::SIMD_SIZE_SIMD16 ? 16 : 8;
uint64_t simdMask = (1ull << simd) - 1;
// Mask off lanes based on the execution masks
auto laneMaskRight = cmd->getRightExecutionMask() & simdMask;
auto lanesPerThreadX = 0;
while (laneMaskRight) {
lanesPerThreadX += laneMaskRight & 1;
laneMaskRight >>= 1;
}
//auto numWorkItems = ( ( cmd->getThreadWidthCounterMaximum() - 1 ) * simd + lanesPerThreadX ) * cmd->getThreadGroupIdXDimension();
//EXPECT_EQ( expectedWorkItems, numWorkItems );
}
HWTEST_F(EnqueueCopyImageTest, alignsToCSR) {
//this test case assumes IOQ
auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
csr.taskCount = pCmdQ->taskCount + 100;
csr.taskLevel = pCmdQ->taskLevel + 50;
enqueueCopyImage<FamilyType>();
EXPECT_EQ(csr.peekTaskCount(), pCmdQ->taskCount);
EXPECT_EQ(csr.peekTaskLevel(), pCmdQ->taskLevel + 1);
}
HWTEST_F(EnqueueCopyImageTest, bumpsTaskLevel) {
auto taskLevelBefore = pCmdQ->taskLevel;
enqueueCopyImage<FamilyType>();
EXPECT_GT(pCmdQ->taskLevel, taskLevelBefore);
}
HWTEST_F(EnqueueCopyImageTest, addsCommands) {
auto usedCmdBufferBefore = pCS->getUsed();
enqueueCopyImage<FamilyType>();
EXPECT_NE(usedCmdBufferBefore, pCS->getUsed());
}
HWTEST_F(EnqueueCopyImageTest, addsIndirectData) {
auto dshBefore = pDSH->getUsed();
auto iohBefore = pIOH->getUsed();
auto sshBefore = pSSH->getUsed();
enqueueCopyImage<FamilyType>();
EXPECT_NE(dshBefore, pDSH->getUsed());
EXPECT_NE(iohBefore, pIOH->getUsed());
EXPECT_NE(sshBefore, pSSH->getUsed());
}
HWTEST_F(EnqueueCopyImageTest, loadRegisterImmediateL3CNTLREG) {
typedef typename FamilyType::MI_LOAD_REGISTER_IMM MI_LOAD_REGISTER_IMM;
enqueueCopyImage<FamilyType>();
// All state should be programmed before walker
auto itorCmd = findMmio<FamilyType>(cmdList.begin(), itorWalker, L3CNTLRegisterOffset<FamilyType>::registerOffset);
ASSERT_NE(itorWalker, itorCmd);
auto *cmd = genCmdCast<MI_LOAD_REGISTER_IMM *>(*itorCmd);
ASSERT_NE(nullptr, cmd);
auto RegisterOffset = L3CNTLRegisterOffset<FamilyType>::registerOffset;
EXPECT_EQ(RegisterOffset, cmd->getRegisterOffset());
auto l3Cntlreg = cmd->getDataDword();
auto numURBWays = (l3Cntlreg >> 1) & 0x7f;
auto L3ClientPool = (l3Cntlreg >> 25) & 0x7f;
EXPECT_NE(0u, numURBWays);
EXPECT_NE(0u, L3ClientPool);
}
HWTEST_F(EnqueueCopyImageTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyImage<FamilyType>();
validateStateBaseAddress<FamilyType>(this->pDevice->getCommandStreamReceiver().getMemoryManager()->getInternalHeapBaseAddress(),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}
HWTEST_F(EnqueueCopyImageTest, mediaInterfaceDescriptorLoad) {
typedef typename FamilyType::MEDIA_INTERFACE_DESCRIPTOR_LOAD MEDIA_INTERFACE_DESCRIPTOR_LOAD;
typedef typename FamilyType::INTERFACE_DESCRIPTOR_DATA INTERFACE_DESCRIPTOR_DATA;
enqueueCopyImage<FamilyType>();
// All state should be programmed before walker
auto cmd = reinterpret_cast<MEDIA_INTERFACE_DESCRIPTOR_LOAD *>(cmdMediaInterfaceDescriptorLoad);
ASSERT_NE(nullptr, cmd);
// Verify we have a valid length -- multiple of INTERFACE_DESCRIPTOR_DATAs
EXPECT_EQ(0u, cmd->getInterfaceDescriptorTotalLength() % sizeof(INTERFACE_DESCRIPTOR_DATA));
// Validate the start address
size_t alignmentStartAddress = 64 * sizeof(uint8_t);
EXPECT_EQ(0u, cmd->getInterfaceDescriptorDataStartAddress() % alignmentStartAddress);
// Validate the length
EXPECT_NE(0u, cmd->getInterfaceDescriptorTotalLength());
size_t alignmentTotalLength = 32 * sizeof(uint8_t);
EXPECT_EQ(0u, cmd->getInterfaceDescriptorTotalLength() % alignmentTotalLength);
// Generically validate this command
FamilyType::PARSE::template validateCommand<MEDIA_INTERFACE_DESCRIPTOR_LOAD *>(cmdList.begin(), itorMediaInterfaceDescriptorLoad);
}
HWTEST_F(EnqueueCopyImageTest, interfaceDescriptorData) {
typedef typename FamilyType::STATE_BASE_ADDRESS STATE_BASE_ADDRESS;
typedef typename FamilyType::INTERFACE_DESCRIPTOR_DATA INTERFACE_DESCRIPTOR_DATA;
enqueueCopyImage<FamilyType>();
// Extract the interfaceDescriptorData
auto cmdSBA = (STATE_BASE_ADDRESS *)cmdStateBaseAddress;
auto &interfaceDescriptorData = *(INTERFACE_DESCRIPTOR_DATA *)cmdInterfaceDescriptorData;
// Validate the kernel start pointer. Technically, a kernel can start at address 0 but let's force a value.
auto kernelStartPointer = ((uint64_t)interfaceDescriptorData.getKernelStartPointerHigh() << 32) + interfaceDescriptorData.getKernelStartPointer();
EXPECT_LE(kernelStartPointer, cmdSBA->getInstructionBufferSize() * MemoryConstants::pageSize);
size_t maxLocalSize = 256u;
auto localWorkSize = std::min(maxLocalSize,
Image2dDefaults::imageDesc.image_width * Image2dDefaults::imageDesc.image_height);
auto simd = 32u;
auto threadsPerThreadGroup = (localWorkSize + simd - 1) / simd;
EXPECT_EQ(threadsPerThreadGroup, interfaceDescriptorData.getNumberOfThreadsInGpgpuThreadGroup());
EXPECT_NE(0u, interfaceDescriptorData.getCrossThreadConstantDataReadLength());
EXPECT_NE(0u, interfaceDescriptorData.getConstantIndirectUrbEntryReadLength());
// We shouldn't have these pointers the same.
EXPECT_NE(kernelStartPointer, interfaceDescriptorData.getBindingTablePointer());
}
HWTEST_F(EnqueueCopyImageTest, surfaceState) {
typedef typename FamilyType::RENDER_SURFACE_STATE RENDER_SURFACE_STATE;
enqueueCopyImage<FamilyType>();
for (uint32_t i = 0; i < 2; ++i) {
const auto &surfaceState = getSurfaceState<FamilyType>(i);
const auto &imageDesc = dstImage->getImageDesc();
EXPECT_EQ(imageDesc.image_width, surfaceState.getWidth());
EXPECT_EQ(imageDesc.image_height, surfaceState.getHeight());
EXPECT_NE(0u, surfaceState.getSurfacePitch());
EXPECT_NE(0u, surfaceState.getSurfaceType());
auto surfaceFormat = surfaceState.getSurfaceFormat();
bool isRedescribedFormat =
surfaceFormat == RENDER_SURFACE_STATE::SURFACE_FORMAT_R32G32B32A32_UINT ||
surfaceFormat == RENDER_SURFACE_STATE::SURFACE_FORMAT_R32G32_UINT ||
surfaceFormat == RENDER_SURFACE_STATE::SURFACE_FORMAT_R32_UINT ||
surfaceFormat == RENDER_SURFACE_STATE::SURFACE_FORMAT_R16_UINT ||
surfaceFormat == RENDER_SURFACE_STATE::SURFACE_FORMAT_R8_UINT;
EXPECT_TRUE(isRedescribedFormat);
EXPECT_EQ(RENDER_SURFACE_STATE::SURFACE_HORIZONTAL_ALIGNMENT_HALIGN_4, surfaceState.getSurfaceHorizontalAlignment());
EXPECT_EQ(RENDER_SURFACE_STATE::SURFACE_VERTICAL_ALIGNMENT_VALIGN_4, surfaceState.getSurfaceVerticalAlignment());
}
const auto &srcSurfaceState = getSurfaceState<FamilyType>(0);
EXPECT_EQ(reinterpret_cast<uint64_t>(srcImage->getCpuAddress()), srcSurfaceState.getSurfaceBaseAddress());
const auto &dstSurfaceState = getSurfaceState<FamilyType>(1);
EXPECT_EQ(reinterpret_cast<uint64_t>(dstImage->getCpuAddress()), dstSurfaceState.getSurfaceBaseAddress());
}
HWTEST_F(EnqueueCopyImageTest, pipelineSelect) {
enqueueCopyImage<FamilyType>();
int numCommands = getNumberOfPipelineSelectsThatEnablePipelineSelect<FamilyType>();
EXPECT_EQ(1, numCommands);
}
HWTEST_F(EnqueueCopyImageTest, mediaVFEState) {
typedef typename FamilyType::MEDIA_VFE_STATE MEDIA_VFE_STATE;
enqueueCopyImage<FamilyType>();
auto *cmd = (MEDIA_VFE_STATE *)cmdMediaVfeState;
// Verify we have a valid length
EXPECT_EQ(pDevice->getHardwareInfo().pSysInfo->ThreadCount, cmd->getMaximumNumberOfThreads());
EXPECT_NE(0u, cmd->getNumberOfUrbEntries());
EXPECT_NE(0u, cmd->getUrbEntryAllocationSize());
// Generically validate this command
FamilyType::PARSE::template validateCommand<MEDIA_VFE_STATE *>(cmdList.begin(), itorMediaVfeState);
}
typedef EnqueueCopyImageMipMapTest MipMapCopyImageTest;
HWTEST_P(MipMapCopyImageTest, GivenImagesWithNonZeroMipLevelsWhenCopyImageIsCalledThenProperMipLevelsAreSet) {
cl_mem_object_type srcImageType, dstImageType;
std::tie(srcImageType, dstImageType) = GetParam();
auto &origBuilder = BuiltIns::getInstance().getBuiltinDispatchInfoBuilder(
EBuiltInOps::CopyImageToImage3d,
pCmdQ->getContext(),
pCmdQ->getDevice());
// substitute original builder with mock builder
auto oldBuilder = BuiltIns::getInstance().setBuiltinDispatchInfoBuilder(
EBuiltInOps::CopyImageToImage3d,
pCmdQ->getContext(),
pCmdQ->getDevice(),
std::unique_ptr<OCLRT::BuiltinDispatchInfoBuilder>(new MockBuiltinDispatchInfoBuilder(BuiltIns::getInstance(), &origBuilder)));
cl_int retVal = CL_SUCCESS;
cl_image_desc srcImageDesc = {};
uint32_t expectedSrcMipLevel = 3;
uint32_t expectedDstMipLevel = 4;
srcImageDesc.image_type = srcImageType;
srcImageDesc.num_mip_levels = 10;
srcImageDesc.image_width = 4;
srcImageDesc.image_height = 1;
srcImageDesc.image_depth = 1;
cl_image_desc dstImageDesc = srcImageDesc;
dstImageDesc.image_type = dstImageType;
size_t srcOrigin[] = {0, 0, 0, 0};
size_t dstOrigin[] = {0, 0, 0, 0};
size_t region[] = {srcImageDesc.image_width, 1, 1};
std::unique_ptr<Image> srcImage;
std::unique_ptr<Image> dstImage;
switch (srcImageType) {
case CL_MEM_OBJECT_IMAGE1D:
srcOrigin[1] = expectedSrcMipLevel;
srcImage = std::unique_ptr<Image>(ImageHelper<Image1dDefaults>::create(context, &srcImageDesc));
break;
case CL_MEM_OBJECT_IMAGE1D_ARRAY:
srcImageDesc.image_array_size = 2;
srcOrigin[2] = expectedSrcMipLevel;
srcImage = std::unique_ptr<Image>(ImageHelper<Image1dArrayDefaults>::create(context, &srcImageDesc));
break;
case CL_MEM_OBJECT_IMAGE2D:
srcOrigin[2] = expectedSrcMipLevel;
srcImage = std::unique_ptr<Image>(ImageHelper<Image2dDefaults>::create(context, &srcImageDesc));
break;
case CL_MEM_OBJECT_IMAGE2D_ARRAY:
srcImageDesc.image_array_size = 2;
srcOrigin[3] = expectedSrcMipLevel;
srcImage = std::unique_ptr<Image>(ImageHelper<Image2dArrayDefaults>::create(context, &srcImageDesc));
break;
case CL_MEM_OBJECT_IMAGE3D:
srcOrigin[3] = expectedSrcMipLevel;
srcImage = std::unique_ptr<Image>(ImageHelper<Image3dDefaults>::create(context, &srcImageDesc));
break;
}
EXPECT_NE(nullptr, srcImage.get());
switch (dstImageType) {
case CL_MEM_OBJECT_IMAGE1D:
dstOrigin[1] = expectedDstMipLevel;
dstImage = std::unique_ptr<Image>(ImageHelper<Image1dDefaults>::create(context, &dstImageDesc));
break;
case CL_MEM_OBJECT_IMAGE1D_ARRAY:
dstImageDesc.image_array_size = 2;
dstOrigin[2] = expectedDstMipLevel;
dstImage = std::unique_ptr<Image>(ImageHelper<Image1dArrayDefaults>::create(context, &dstImageDesc));
break;
case CL_MEM_OBJECT_IMAGE2D:
dstOrigin[2] = expectedDstMipLevel;
dstImage = std::unique_ptr<Image>(ImageHelper<Image2dDefaults>::create(context, &dstImageDesc));
break;
case CL_MEM_OBJECT_IMAGE2D_ARRAY:
dstImageDesc.image_array_size = 2;
dstOrigin[3] = expectedDstMipLevel;
dstImage = std::unique_ptr<Image>(ImageHelper<Image2dArrayDefaults>::create(context, &dstImageDesc));
break;
case CL_MEM_OBJECT_IMAGE3D:
dstOrigin[3] = expectedDstMipLevel;
dstImage = std::unique_ptr<Image>(ImageHelper<Image3dDefaults>::create(context, &dstImageDesc));
break;
}
EXPECT_NE(nullptr, dstImage.get());
retVal = pCmdQ->enqueueCopyImage(srcImage.get(),
dstImage.get(),
srcOrigin,
dstOrigin,
region,
0,
nullptr,
nullptr);
EXPECT_EQ(CL_SUCCESS, retVal);
auto &mockBuilder = static_cast<MockBuiltinDispatchInfoBuilder &>(BuiltIns::getInstance().getBuiltinDispatchInfoBuilder(EBuiltInOps::CopyImageToImage3d,
pCmdQ->getContext(),
pCmdQ->getDevice()));
auto params = mockBuilder.getBuiltinOpParams();
EXPECT_EQ(expectedSrcMipLevel, params->srcMipLevel);
EXPECT_EQ(expectedDstMipLevel, params->dstMipLevel);
// restore original builder and retrieve mock builder
auto newBuilder = BuiltIns::getInstance().setBuiltinDispatchInfoBuilder(
EBuiltInOps::CopyImageToImage3d,
pCmdQ->getContext(),
pCmdQ->getDevice(),
std::move(oldBuilder));
EXPECT_NE(nullptr, newBuilder);
}
uint32_t types[] = {CL_MEM_OBJECT_IMAGE1D, CL_MEM_OBJECT_IMAGE1D_ARRAY, CL_MEM_OBJECT_IMAGE2D, CL_MEM_OBJECT_IMAGE2D_ARRAY, CL_MEM_OBJECT_IMAGE3D};
INSTANTIATE_TEST_CASE_P(MipMapCopyImageTest_GivenImagesWithNonZeroMipLevelsWhenCopyImageIsCalledThenProperMipLevelsAreSet,
MipMapCopyImageTest,
::testing::Combine(
::testing::ValuesIn(types),
::testing::ValuesIn(types)));
Reference in New Issue View Git Blame Copy Permalink