intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 14:02:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(ocl): Use GPU address of SVM alloc. when using builtins

Browse Source 
Use GPU address of SVM allocation instead of CPU one. It fixes a bug
visible on platforms, where potentially GPU VA does not have to match
CPU VA.

Related-To: NEO-7970
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit is contained in:
Kacper Nowak
2023-05-19 00:43:16 +00:00
committed by Compute-Runtime-Automation
parent 3c35ca5f20
commit d05fee8087
2 changed files with 89 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/source/built_ins/builtins_dispatch_builder.cpp
View File

@@ -340,7 +340,7 @@ class BuiltInOp<EBuiltInOps::FillBuffer> : public BuiltinDispatchInfoBuilder {
// Set-up srcMemObj with pattern
auto graphicsAllocation = operationParams.srcMemObj->getMultiGraphicsAllocation().getDefaultGraphicsAllocation();
kernelSplit1DBuilder.setArgSvm(2, operationParams.srcMemObj->getSize(), graphicsAllocation->getUnderlyingBuffer(), graphicsAllocation, CL_MEM_READ_ONLY);
kernelSplit1DBuilder.setArgSvm(2, operationParams.srcMemObj->getSize(), reinterpret_cast<void *>(graphicsAllocation->getGpuAddress()), graphicsAllocation, CL_MEM_READ_ONLY);
// Set-up patternSizeInEls
kernelSplit1DBuilder.setArg(SplitDispatch::RegionCoordX::Left, 3, static_cast<OffsetType>(operationParams.srcMemObj->getSize()));

102
opencl/test/unit_test/command_queue/enqueue_svm_mem_fill_tests.cpp
View File

@@ -5,8 +5,10 @@
*
*/
#include "shared/source/memory_manager/internal_allocation_storage.h"
#include "shared/source/memory_manager/unified_memory_manager.h"
#include "shared/test/common/mocks/mock_builtins.h"
#include "shared/test/common/mocks/mock_graphics_allocation.h"
#include "shared/test/common/test_macros/test.h"
#include "opencl/source/built_ins/builtins_dispatch_builder.h"
@@ -19,20 +21,14 @@
using namespace NEO;
struct EnqueueSvmMemFillTest : public ClDeviceFixture,
public CommandQueueHwFixture,
public ::testing::TestWithParam<size_t> {
struct BaseEnqueueSvmMemFillFixture : public ClDeviceFixture,
public CommandQueueHwFixture {
typedef CommandQueueHwFixture CommandQueueFixture;
EnqueueSvmMemFillTest() {
}
void SetUp() override {
void setUp() {
ClDeviceFixture::setUp();
CommandQueueFixture::setUp(pClDevice, 0);
REQUIRE_SVM_OR_SKIP(pDevice);
patternSize = (size_t)GetParam();
ASSERT_TRUE((0 < patternSize) && (patternSize <= 128));
SVMAllocsManager::SvmAllocationProperties svmProperties;
svmProperties.coherent = true;
svmPtr = context->getSVMAllocsManager()->createSVMAlloc(256, svmProperties, context->getRootDeviceIndices(), context->getDeviceBitfields());
@@ -43,19 +39,97 @@ struct EnqueueSvmMemFillTest : public ClDeviceFixture,
ASSERT_NE(nullptr, svmAlloc);
}
void TearDown() override {
void tearDown() {
if (svmPtr) {
context->getSVMAllocsManager()->freeSVMAlloc(svmPtr);
}
CommandQueueFixture::tearDown();
ClDeviceFixture::tearDown();
}
const uint64_t pattern[16] = {0x0011223344556677, 0x8899AABBCCDDEEFF, 0xFFEEDDCCBBAA9988, 0x7766554433221100,
0xFFEEDDCCBBAA9988, 0x7766554433221100, 0x0011223344556677, 0x8899AABBCCDDEEFF};
size_t patternSize = 0;
void *svmPtr = nullptr;
GraphicsAllocation *svmAlloc = nullptr;
const uint64_t pattern[16] = {0x0011223344556677, 0x8899AABBCCDDEEFF, 0xFFEEDDCCBBAA9988, 0x7766554433221100,
0xFFEEDDCCBBAA9988, 0x7766554433221100, 0x0011223344556677, 0x8899AABBCCDDEEFF};
size_t patternSize = 0x10;
};
using BaseEnqueueSvmMemFillTest = Test<BaseEnqueueSvmMemFillFixture>;
HWTEST_F(BaseEnqueueSvmMemFillTest, givenEnqueueSVMMemFillWhenUsingFillBufferBuilderThenUseGpuAddressOfPatternSVMAllocation) {
struct MockFillBufferBuilder : MockBuiltinDispatchInfoBuilder {
MockFillBufferBuilder(BuiltIns &kernelLib, ClDevice &clDevice, BuiltinDispatchInfoBuilder *origBuilder, const void *pattern, size_t patternSize)
: MockBuiltinDispatchInfoBuilder(kernelLib, clDevice, origBuilder),
pattern(pattern), patternSize(patternSize) {
}
const void *pattern;
size_t patternSize;
};
auto builtIns = new MockBuiltins();
pCmdQ->getDevice().getExecutionEnvironment()->rootDeviceEnvironments[pCmdQ->getDevice().getRootDeviceIndex()]->builtins.reset(builtIns);
auto &origBuilder = BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(
EBuiltInOps::FillBuffer,
pCmdQ->getClDevice());
ASSERT_NE(nullptr, &origBuilder);
// note that we need to store the returned value, as it is an unique pointer storing original builder, which will be later invoked
auto oldBuilder = pClExecutionEnvironment->setBuiltinDispatchInfoBuilder(
rootDeviceIndex,
EBuiltInOps::FillBuffer,
std::unique_ptr<NEO::BuiltinDispatchInfoBuilder>(new MockFillBufferBuilder(*builtIns, pCmdQ->getClDevice(), &origBuilder, pattern, patternSize)));
size_t patternSize = 0x10u;
auto patternAllocation = static_cast<MockGraphicsAllocation *>(context->getMemoryManager()->allocateGraphicsMemoryWithProperties({pCmdQ->getDevice().getRootDeviceIndex(), 2 * patternSize, AllocationType::FILL_PATTERN, pCmdQ->getDevice().getDeviceBitfield()}));
// offset cpuPtr so cpuPtr != gpuAddress in order to ensure that setArgSVM will be called using gpu address of the pattern allocation
auto origCpuPtr = patternAllocation->cpuPtr;
patternAllocation->cpuPtr = ptrOffset(patternAllocation->cpuPtr, patternSize);
ASSERT_NE((uint64_t)patternAllocation->cpuPtr, patternAllocation->getGpuAddress());
auto internalAllocStorage = pCmdQ->getGpgpuCommandStreamReceiver().getInternalAllocationStorage();
internalAllocStorage->storeAllocation(std::unique_ptr<GraphicsAllocation>(patternAllocation), REUSABLE_ALLOCATION);
auto retVal = pCmdQ->enqueueSVMMemFill(
svmPtr,
pattern,
patternSize,
256,
0,
nullptr,
nullptr);
EXPECT_EQ(CL_SUCCESS, retVal);
// return mock builder
auto &newBuilder = BuiltInDispatchBuilderOp::getBuiltinDispatchInfoBuilder(
EBuiltInOps::FillBuffer,
pCmdQ->getClDevice());
ASSERT_NE(nullptr, &origBuilder);
auto &mockBuilder = static_cast<MockFillBufferBuilder &>(newBuilder);
auto kernel = mockBuilder.getMultiDispatchInfo()->begin()->getKernel();
auto patternArgIndex = 2;
const auto &patternArg = kernel->getKernelArguments().at(patternArgIndex);
EXPECT_EQ(patternAllocation->getGpuAddress(), reinterpret_cast<uint64_t>(patternArg.value));
patternAllocation->cpuPtr = origCpuPtr;
}
struct EnqueueSvmMemFillTest : public BaseEnqueueSvmMemFillFixture,
public ::testing::TestWithParam<size_t> {
typedef CommandQueueHwFixture CommandQueueFixture;
void SetUp() override {
BaseEnqueueSvmMemFillFixture::setUp();
REQUIRE_SVM_OR_SKIP(pDevice);
patternSize = (size_t)GetParam();
ASSERT_TRUE((0 < patternSize) && (patternSize <= 128));
}
void TearDown() override {
BaseEnqueueSvmMemFillFixture::tearDown();
}
};
HWTEST_P(EnqueueSvmMemFillTest, givenEnqueueSVMMemFillWhenUsingFillBufferBuilderThenItIsConfiguredWithBuitinOpParamsAndProducesDispatchInfo) {