intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-05 09:09:04 +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

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);