intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 22:12:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

Pass root device index to Buffer::setArgStateful

Browse Source 
Related-To: NEO-4672
Change-Id: Ic846eac488809bd7d9534bb7378d9398acd36451
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2020-06-25 18:17:44 +02:00
committed by sys_ocldev
parent 28aa29bcbd
commit c96fd30e4c
10 changed files with 65 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/source/gtpin/gtpin_callbacks.cpp
View File

@@ -139,7 +139,7 @@ void gtpinNotifyKernelSubmit(cl_kernel kernel, void *pCmdQueue) {
void *pSurfaceState = gtpinHelper.getSurfaceState(pKernel, gtpinBTI);
cl_mem buffer = (cl_mem)resource;
auto pBuffer = castToObjectOrAbort<Buffer>(buffer);
pBuffer->setArgStateful(pSurfaceState, false, false, false, false);
pBuffer->setArgStateful(pSurfaceState, false, false, false, false, device.getRootDeviceIndex());
}
}

2
opencl/source/kernel/kernel.cpp
View File

@@ -1316,7 +1316,7 @@ cl_int Kernel::setArgBuffer(uint32_t argIndex,
if (requiresSshForBuffers()) {
auto surfaceState = ptrOffset(getSurfaceStateHeap(), kernelArgInfo.offsetHeap);
buffer->setArgStateful(surfaceState, forceNonAuxMode, disableL3, isAuxTranslationKernel, kernelArgInfo.isReadOnly);
buffer->setArgStateful(surfaceState, forceNonAuxMode, disableL3, isAuxTranslationKernel, kernelArgInfo.isReadOnly, getDevice().getRootDeviceIndex());
}
kernelArguments[argIndex].isStatelessUncacheable = kernelArgInfo.pureStatefulBufferAccess ? false : buffer->isMemObjUncacheable();

19
opencl/source/mem_obj/buffer.cpp
View File

@@ -626,16 +626,21 @@ Buffer *Buffer::createBufferHwFromDevice(const Device *device,
MemoryProperties memoryProperties = MemoryPropertiesHelper::createMemoryProperties(flags, flagsIntel, 0, device);
auto pBuffer = funcCreate(nullptr, memoryProperties, flags, flagsIntel, size, memoryStorage, hostPtr, gfxAllocation,
zeroCopy, isHostPtrSVM, isImageRedescribed);
if (!gfxAllocation) {
auto multiGraphicsAllocation = MultiGraphicsAllocation(device->getRootDeviceIndex());
std::swap(pBuffer->multiGraphicsAllocation, multiGraphicsAllocation);
}
pBuffer->offset = offset;
pBuffer->executionEnvironment = device->getExecutionEnvironment();
pBuffer->rootDeviceEnvironment = pBuffer->executionEnvironment->rootDeviceEnvironments[device->getRootDeviceIndex()].get();
return pBuffer;
}
uint32_t Buffer::getMocsValue(bool disableL3Cache, bool isReadOnlyArgument) const {
uint32_t Buffer::getMocsValue(bool disableL3Cache, bool isReadOnlyArgument, uint32_t rootDeviceIndex) const {
uint64_t bufferAddress = 0;
size_t bufferSize = 0;
auto graphicsAllocation = multiGraphicsAllocation.getDefaultGraphicsAllocation();
auto graphicsAllocation = multiGraphicsAllocation.getGraphicsAllocation(rootDeviceIndex);
if (graphicsAllocation) {
bufferAddress = graphicsAllocation->getGpuAddress();
bufferSize = graphicsAllocation->getUnderlyingBufferSize();
@@ -657,8 +662,8 @@ uint32_t Buffer::getMocsValue(bool disableL3Cache, bool isReadOnlyArgument) cons
}
}
uint32_t Buffer::getSurfaceSize(bool alignSizeForAuxTranslation) const {
auto bufferAddress = getBufferAddress();
uint32_t Buffer::getSurfaceSize(bool alignSizeForAuxTranslation, uint32_t rootDeviceIndex) const {
auto bufferAddress = getBufferAddress(rootDeviceIndex);
auto bufferAddressAligned = alignDown(bufferAddress, 4);
auto bufferOffset = ptrDiff(bufferAddress, bufferAddressAligned);
@@ -666,9 +671,9 @@ uint32_t Buffer::getSurfaceSize(bool alignSizeForAuxTranslation) const {
return surfaceSize;
}
uint64_t Buffer::getBufferAddress() const {
auto graphicsAllocation = multiGraphicsAllocation.getDefaultGraphicsAllocation();
uint64_t Buffer::getBufferAddress(uint32_t rootDeviceIndex) const {
// The graphics allocation for Host Ptr surface will be created in makeResident call and GPU address is expected to be the same as CPU address
auto graphicsAllocation = multiGraphicsAllocation.getGraphicsAllocation(rootDeviceIndex);
auto bufferAddress = (graphicsAllocation != nullptr) ? graphicsAllocation->getGpuAddress() : castToUint64(getHostPtr());
bufferAddress += this->offset;
return bufferAddress;
@@ -695,7 +700,7 @@ void Buffer::setSurfaceState(const Device *device,
cl_mem_flags flags,
cl_mem_flags_intel flagsIntel) {
auto buffer = Buffer::createBufferHwFromDevice(device, flags, flagsIntel, svmSize, svmPtr, svmPtr, gfxAlloc, offset, true, false, false);
buffer->setArgStateful(surfaceState, false, false, false, false);
buffer->setArgStateful(surfaceState, false, false, false, false, device->getRootDeviceIndex());
buffer->graphicsAllocation = nullptr;
delete buffer;
}

10
opencl/source/mem_obj/buffer.h
View File

@@ -135,7 +135,7 @@ class Buffer : public MemObj {
bool isSubBuffer();
bool isValidSubBufferOffset(size_t offset);
uint64_t setArgStateless(void *memory, uint32_t patchSize, uint32_t rootDeviceIndex, bool set32BitAddressing);
virtual void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) = 0;
virtual void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) = 0;
bool bufferRectPitchSet(const size_t *bufferOrigin,
const size_t *region,
size_t &bufferRowPitch,
@@ -151,9 +151,9 @@ class Buffer : public MemObj {
bool isReadWriteOnCpuAllowed(uint32_t rootDeviceIndex);
bool isReadWriteOnCpuPreferred(void *ptr, size_t size, const Device &device);
uint32_t getMocsValue(bool disableL3Cache, bool isReadOnlyArgument) const;
uint32_t getSurfaceSize(bool alignSizeForAuxTranslation) const;
uint64_t getBufferAddress() const;
uint32_t getMocsValue(bool disableL3Cache, bool isReadOnlyArgument, uint32_t rootDeviceIndex) const;
uint32_t getSurfaceSize(bool alignSizeForAuxTranslation, uint32_t rootDeviceIndex) const;
uint64_t getBufferAddress(uint32_t rootDeviceIndex) const;
bool isCompressed(uint32_t rootDeviceIndex) const;
@@ -204,7 +204,7 @@ class BufferHw : public Buffer {
: Buffer(context, memoryProperties, flags, flagsIntel, size, memoryStorage, hostPtr, gfxAllocation,
zeroCopy, isHostPtrSVM, isObjectRedescribed) {}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnlyArgument) override;
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnlyArgument, uint32_t rootDeviceIndex) override;
void appendBufferState(void *memory, Context *context, GraphicsAllocation *gfxAllocation, bool isReadOnlyArgument);
void appendSurfaceStateExt(void *memory);

8
opencl/source/mem_obj/buffer_base.inl
View File

@@ -32,9 +32,11 @@ union SURFACE_STATE_BUFFER_LENGTH {
};
template <typename GfxFamily>
void BufferHw<GfxFamily>::setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnlyArgument) {
EncodeSurfaceState<GfxFamily>::encodeBuffer(memory, getBufferAddress(), getSurfaceSize(alignSizeForAuxTranslation), getMocsValue(disableL3, isReadOnlyArgument), true);
EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(multiGraphicsAllocation.getDefaultGraphicsAllocation(), rootDeviceEnvironment->getGmmHelper(), memory, forceNonAuxMode, isReadOnlyArgument);
void BufferHw<GfxFamily>::setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnlyArgument, uint32_t rootDeviceIndex) {
auto graphicsAllocation = multiGraphicsAllocation.getGraphicsAllocation(rootDeviceIndex);
EncodeSurfaceState<GfxFamily>::encodeBuffer(memory, getBufferAddress(rootDeviceIndex), getSurfaceSize(alignSizeForAuxTranslation, rootDeviceIndex), getMocsValue(disableL3, isReadOnlyArgument, rootDeviceIndex), true);
EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(multiGraphicsAllocation.getGraphicsAllocation(rootDeviceIndex),
executionEnvironment->rootDeviceEnvironments[rootDeviceIndex]->getGmmHelper(), memory, forceNonAuxMode, isReadOnlyArgument);
appendBufferState(memory, context, graphicsAllocation, isReadOnlyArgument);
appendSurfaceStateExt(memory);

4
opencl/test/unit_test/helpers/base_object_tests.cpp
View File

@@ -79,7 +79,7 @@ class MockObject : public MockObjectBase<BaseType> {};
template <>
class MockObject<Buffer> : public MockObjectBase<Buffer> {
public:
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) override {}
};
template <>
@@ -293,7 +293,7 @@ class MockBuffer : public MockBufferStorage, public Buffer {
CL_MEM_USE_HOST_PTR, 0, sizeof(data), &data, &data, &mockGfxAllocation, true, false, false) {
}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) override {
}
};

74
opencl/test/unit_test/mem_obj/buffer_tests.cpp
View File

@@ -2656,6 +2656,7 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferSetSurfaceThatAddressIsForcedTo32bitW
{
DebugManager.flags.Force32bitAddressing.set(true);
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
auto size = MemoryConstants::pageSize;
auto ptr = (void *)alignedMalloc(size * 2, MemoryConstants::pageSize);
auto retVal = CL_SUCCESS;
@@ -2668,15 +2669,15 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferSetSurfaceThatAddressIsForcedTo32bitW
retVal);
EXPECT_EQ(CL_SUCCESS, retVal);
EXPECT_TRUE(is64bit ? buffer->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->is32BitAllocation() : true);
EXPECT_TRUE(is64bit ? buffer->getGraphicsAllocation(rootDeviceIndex)->is32BitAllocation() : true);
using RENDER_SURFACE_STATE = typename FamilyType::RENDER_SURFACE_STATE;
RENDER_SURFACE_STATE surfaceState = {};
buffer->setArgStateful(&surfaceState, false, false, false, false);
buffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
auto surfBaseAddress = surfaceState.getSurfaceBaseAddress();
auto bufferAddress = buffer->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getGpuAddress();
auto bufferAddress = buffer->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress();
EXPECT_EQ(bufferAddress, surfBaseAddress);
delete buffer;
@@ -2687,6 +2688,7 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferSetSurfaceThatAddressIsForcedTo32bitW
HWTEST_F(BufferSetSurfaceTests, givenBufferWithOffsetWhenSetArgStatefulIsCalledThenSurfaceBaseAddressIsProperlyOffseted) {
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
auto size = MemoryConstants::pageSize;
auto ptr = (void *)alignedMalloc(size * 2, MemoryConstants::pageSize);
auto retVal = CL_SUCCESS;
@@ -2708,10 +2710,10 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferWithOffsetWhenSetArgStatefulIsCalledT
using RENDER_SURFACE_STATE = typename FamilyType::RENDER_SURFACE_STATE;
RENDER_SURFACE_STATE surfaceState = {};
subBuffer->setArgStateful(&surfaceState, false, false, false, false);
subBuffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
auto surfBaseAddress = surfaceState.getSurfaceBaseAddress();
auto bufferAddress = buffer->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->getGpuAddress();
auto bufferAddress = buffer->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress();
EXPECT_EQ(bufferAddress + region.origin, surfBaseAddress);
subBuffer->release();
@@ -2723,6 +2725,7 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferWithOffsetWhenSetArgStatefulIsCalledT
HWTEST_F(BufferSetSurfaceTests, givenBufferWhenSetArgStatefulWithL3ChacheDisabledIsCalledThenL3CacheShouldBeOffAndSizeIsAlignedTo512) {
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
auto size = 128;
auto retVal = CL_SUCCESS;
@@ -2737,7 +2740,7 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferWhenSetArgStatefulWithL3ChacheDisable
using RENDER_SURFACE_STATE = typename FamilyType::RENDER_SURFACE_STATE;
RENDER_SURFACE_STATE surfaceState = {};
buffer->setArgStateful(&surfaceState, false, true, true, false);
buffer->setArgStateful(&surfaceState, false, true, true, false, rootDeviceIndex);
auto mocs = surfaceState.getMemoryObjectControlState();
auto gmmHelper = device->getGmmHelper();
@@ -2748,6 +2751,7 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferWhenSetArgStatefulWithL3ChacheDisable
HWTEST_F(BufferSetSurfaceTests, givenBufferThatIsMisalignedButIsAReadOnlyArgumentWhenSurfaceStateIsSetThenL3IsOn) {
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
auto size = 128;
auto retVal = CL_SUCCESS;
@@ -2762,9 +2766,9 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferThatIsMisalignedButIsAReadOnlyArgumen
using RENDER_SURFACE_STATE = typename FamilyType::RENDER_SURFACE_STATE;
RENDER_SURFACE_STATE surfaceState = {};
buffer->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->setSize(127);
buffer->getGraphicsAllocation(rootDeviceIndex)->setSize(127);
buffer->setArgStateful(&surfaceState, false, false, false, true);
buffer->setArgStateful(&surfaceState, false, false, false, true, rootDeviceIndex);
auto mocs = surfaceState.getMemoryObjectControlState();
auto gmmHelper = device->getGmmHelper();
@@ -2775,6 +2779,7 @@ HWTEST_F(BufferSetSurfaceTests, givenBufferThatIsMisalignedButIsAReadOnlyArgumen
HWTEST_F(BufferSetSurfaceTests, givenAlignedCacheableReadOnlyBufferThenChoseOclBufferPolicy) {
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
const auto size = MemoryConstants::pageSize;
const auto ptr = (void *)alignedMalloc(size * 2, MemoryConstants::pageSize);
const auto flags = CL_MEM_USE_HOST_PTR | CL_MEM_READ_ONLY;
@@ -2789,7 +2794,7 @@ HWTEST_F(BufferSetSurfaceTests, givenAlignedCacheableReadOnlyBufferThenChoseOclB
EXPECT_EQ(CL_SUCCESS, retVal);
typename FamilyType::RENDER_SURFACE_STATE surfaceState = {};
buffer->setArgStateful(&surfaceState, false, false, false, false);
buffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
const auto expectedMocs = device->getGmmHelper()->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER);
const auto actualMocs = surfaceState.getMemoryObjectControlState();
@@ -2800,6 +2805,7 @@ HWTEST_F(BufferSetSurfaceTests, givenAlignedCacheableReadOnlyBufferThenChoseOclB
HWTEST_F(BufferSetSurfaceTests, givenAlignedCacheableNonReadOnlyBufferThenChooseOclBufferPolicy) {
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
const auto size = MemoryConstants::pageSize;
const auto ptr = (void *)alignedMalloc(size * 2, MemoryConstants::pageSize);
const auto flags = CL_MEM_USE_HOST_PTR;
@@ -2814,7 +2820,7 @@ HWTEST_F(BufferSetSurfaceTests, givenAlignedCacheableNonReadOnlyBufferThenChoose
EXPECT_EQ(CL_SUCCESS, retVal);
typename FamilyType::RENDER_SURFACE_STATE surfaceState = {};
buffer->setArgStateful(&surfaceState, false, false, false, false);
buffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
const auto expectedMocs = device->getGmmHelper()->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER);
const auto actualMocs = surfaceState.getMemoryObjectControlState();
@@ -2829,23 +2835,24 @@ HWTEST_F(BufferSetSurfaceTests, givenRenderCompressedGmmResourceWhenSurfaceState
RENDER_SURFACE_STATE surfaceState = {};
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
auto retVal = CL_SUCCESS;
std::unique_ptr<Buffer> buffer(Buffer::create(&context, CL_MEM_READ_WRITE, 1, nullptr, retVal));
auto graphicsAllocation = buffer->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex());
auto graphicsAllocation = buffer->getGraphicsAllocation(rootDeviceIndex);
graphicsAllocation->setAllocationType(GraphicsAllocation::AllocationType::BUFFER_COMPRESSED);
auto gmm = new Gmm(context.getDevice(0)->getGmmClientContext(), nullptr, 1, false);
graphicsAllocation->setDefaultGmm(gmm);
gmm->isRenderCompressed = true;
buffer->setArgStateful(&surfaceState, false, false, false, false);
buffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
EXPECT_EQ(0u, surfaceState.getAuxiliarySurfaceBaseAddress());
EXPECT_TRUE(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E == surfaceState.getAuxiliarySurfaceMode());
EXPECT_TRUE(RENDER_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT == surfaceState.getCoherencyType());
graphicsAllocation->setAllocationType(GraphicsAllocation::AllocationType::BUFFER);
buffer->setArgStateful(&surfaceState, false, false, false, false);
buffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
EXPECT_TRUE(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE == surfaceState.getAuxiliarySurfaceMode());
}
@@ -2855,14 +2862,15 @@ HWTEST_F(BufferSetSurfaceTests, givenNonRenderCompressedGmmResourceWhenSurfaceSt
RENDER_SURFACE_STATE surfaceState = {};
MockContext context;
auto rootDeviceIndex = context.getDevice(0)->getRootDeviceIndex();
auto retVal = CL_SUCCESS;
std::unique_ptr<Buffer> buffer(Buffer::create(&context, CL_MEM_READ_WRITE, 1, nullptr, retVal));
auto gmm = new Gmm(context.getDevice(0)->getGmmClientContext(), nullptr, 1, false);
buffer->getGraphicsAllocation(context.getDevice(0)->getRootDeviceIndex())->setDefaultGmm(gmm);
buffer->getGraphicsAllocation(rootDeviceIndex)->setDefaultGmm(gmm);
gmm->isRenderCompressed = false;
buffer->setArgStateful(&surfaceState, false, false, false, false);
buffer->setArgStateful(&surfaceState, false, false, false, false, rootDeviceIndex);
EXPECT_EQ(0u, surfaceState.getAuxiliarySurfaceBaseAddress());
EXPECT_TRUE(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE == surfaceState.getAuxiliarySurfaceMode());
@@ -3094,7 +3102,7 @@ TEST_F(MultiRootDeviceBufferTest, givenBufferWhenGetSurfaceSizeCalledWithoutAlig
uint32_t size = 0x131;
std::unique_ptr<Buffer> buffer(Buffer::create(context.get(), flags, size, nullptr, retVal));
auto surfaceSize = buffer->getSurfaceSize(false);
auto surfaceSize = buffer->getSurfaceSize(false, expectedRootDeviceIndex);
EXPECT_EQ(surfaceSize, alignUp(size, 4));
}
@@ -3104,35 +3112,21 @@ TEST_F(MultiRootDeviceBufferTest, givenBufferWhenGetSurfaceSizeCalledWithAlignSi
uint32_t size = 0x131;
std::unique_ptr<Buffer> buffer(Buffer::create(context.get(), flags, size, nullptr, retVal));
auto surfaceSize = buffer->getSurfaceSize(true);
auto surfaceSize = buffer->getSurfaceSize(true, expectedRootDeviceIndex);
EXPECT_EQ(surfaceSize, alignUp(size, 512));
}
TEST_F(MultiRootDeviceBufferTest, givenHostPtrBufferWhenGetBufferAddressCalledThenHostPtrReturned) {
class MockBuffer : public Buffer {
public:
using Buffer::multiGraphicsAllocation;
MockBuffer(void *hostPtr) {
this->hostPtr = hostPtr;
}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {
}
};
void *hostPtr = reinterpret_cast<void *>(0x3000);
std::unique_ptr<MockBuffer> buffer(new MockBuffer(hostPtr));
auto address = buffer->getBufferAddress();
ASSERT_EQ(hostPtr, reinterpret_cast<void *>(address));
}
TEST_F(MultiRootDeviceBufferTest, givenBufferWithoutMultiGAWhenGetBufferAddressCalledThenCorrectAddressReturned) {
TEST_F(MultiRootDeviceBufferTest, givenNullptrGraphicsAllocationForRootDeviceIndexWhenGettingBufferAddressThenHostPtrReturned) {
cl_int retVal = 0;
cl_mem_flags flags = CL_MEM_READ_WRITE;
cl_mem_flags flags = CL_MEM_READ_WRITE | CL_MEM_USE_HOST_PTR;
std::unique_ptr<Buffer> buffer(Buffer::create(context.get(), flags, MemoryConstants::pageSize, nullptr, retVal));
char *hostPtr[MemoryConstants::pageSize]{};
std::unique_ptr<Buffer> buffer(Buffer::create(context.get(), flags, MemoryConstants::pageSize, hostPtr, retVal));
auto address = buffer->getBufferAddress();
auto address = buffer->getBufferAddress(expectedRootDeviceIndex);
auto graphicsAllocation = buffer->getGraphicsAllocation(expectedRootDeviceIndex);
ASSERT_EQ(graphicsAllocation->getGpuAddress(), address);
EXPECT_EQ(graphicsAllocation->getGpuAddress(), address);
address = buffer->getBufferAddress(0);
EXPECT_EQ(reinterpret_cast<uint64_t>(buffer->getHostPtr()), address);
}

6
opencl/test/unit_test/mocks/mock_buffer.h
View File

@@ -48,7 +48,7 @@ class MockBuffer : public MockBufferStorage, public Buffer {
this->graphicsAllocation = &this->mockGfxAllocation;
}
}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) override {
Buffer::setSurfaceState(device.get(), memory, getSize(), getCpuAddress(), 0, (externalAlloc != nullptr) ? externalAlloc : &mockGfxAllocation, 0, 0);
}
GraphicsAllocation *externalAlloc = nullptr;
@@ -61,7 +61,7 @@ class AlignedBuffer : public MockBufferStorage, public Buffer {
}
AlignedBuffer(GraphicsAllocation *gfxAllocation) : MockBufferStorage(), Buffer(nullptr, MemoryPropertiesHelper::createMemoryProperties(CL_MEM_USE_HOST_PTR, 0, 0, nullptr), CL_MEM_USE_HOST_PTR, 0, sizeof(data) / 2, alignUp(&data, 64), alignUp(&data, 64), gfxAllocation, true, false, false) {
}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) override {
Buffer::setSurfaceState(device.get(), memory, getSize(), getCpuAddress(), 0, &mockGfxAllocation, 0, 0);
}
};
@@ -73,7 +73,7 @@ class UnalignedBuffer : public MockBufferStorage, public Buffer {
}
UnalignedBuffer(GraphicsAllocation *gfxAllocation) : MockBufferStorage(true), Buffer(nullptr, MemoryPropertiesHelper::createMemoryProperties(CL_MEM_USE_HOST_PTR, 0, 0, nullptr), CL_MEM_USE_HOST_PTR, 0, sizeof(data) / 2, alignUp(&data, 4), alignUp(&data, 4), gfxAllocation, false, false, false) {
}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) override {
Buffer::setSurfaceState(device.get(), memory, getSize(), getCpuAddress(), 0, &mockGfxAllocation, 0, 0);
}
};

2
opencl/test/unit_test/os_interface/linux/drm_command_stream_tests.cpp
View File

@@ -1447,7 +1447,7 @@ class DrmMockBuffer : public Buffer {
gfxAllocation(alloc) {
}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly) override {
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnly, uint32_t rootDeviceIndex) override {
}
protected:

2
shared/source/memory_manager/multi_graphics_allocation.h
View File

@@ -11,7 +11,7 @@
namespace NEO {
class MultiGraphicsAllocation : NonCopyableOrMovableClass {
class MultiGraphicsAllocation : NonCopyableClass {
public:
MultiGraphicsAllocation(uint32_t maxRootDeviceIndex);