intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-30 09:58:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: introduce staging reads from image

Browse Source 
Related-To: NEO-12968

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
This commit is contained in:
Szymon Morek
2024-12-05 13:56:30 +00:00
committed by Compute-Runtime-Automation
parent f2725f217e
commit 6c4eb322b1
16 changed files with 702 additions and 241 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
opencl/source/command_queue/command_queue.h
View File

@@ -45,6 +45,7 @@ struct BuiltinOpParams;
struct CsrSelectionArgs;
struct MultiDispatchInfo;
struct TimestampPacketDependencies;
struct StagingTransferStatus;
enum class QueuePriority {
low,
@@ -147,6 +148,10 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
size_t rowPitch, size_t slicePitch, void *ptr, GraphicsAllocation *mapAllocation,
cl_uint numEventsInWaitList, const cl_event *eventWaitList, cl_event *event) = 0;
virtual cl_int enqueueReadImageImpl(Image *srcImage, cl_bool blockingRead, const size_t *origin, const size_t *region,
size_t rowPitch, size_t slicePitch, void *ptr, GraphicsAllocation *mapAllocation,
cl_uint numEventsInWaitList, const cl_event *eventWaitList, cl_event *event, CommandStreamReceiver &csr) = 0;
virtual cl_int enqueueWriteBuffer(Buffer *buffer, cl_bool blockingWrite, size_t offset, size_t cb,
const void *ptr, GraphicsAllocation *mapAllocation, cl_uint numEventsInWaitList,
const cl_event *eventWaitList, cl_event *event) = 0;
@@ -396,8 +401,11 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
cl_int enqueueStagingBufferMemcpy(cl_bool blockingCopy, void *dstPtr, const void *srcPtr, size_t size, cl_event *event);
cl_int enqueueStagingWriteImage(Image *dstImage, cl_bool blockingCopy, const size_t *globalOrigin, const size_t *globalRegion,
size_t inputRowPitch, size_t inputSlicePitch, const void *ptr, cl_event *event);
cl_int enqueueStagingReadImage(Image *dstImage, cl_bool blockingCopy, const size_t *globalOrigin, const size_t *globalRegion,
size_t inputRowPitch, size_t inputSlicePitch, const void *ptr, cl_event *event);
bool isValidForStagingBufferCopy(Device &device, void *dstPtr, const void *srcPtr, size_t size, bool hasDependencies);
bool isValidForStagingWriteImage(Image *image, const void *ptr, bool hasDependencies);
bool isValidForStagingTransferImage(Image *image, const void *ptr, bool hasDependencies);
protected:
void *enqueueReadMemObjForMap(TransferProperties &transferProperties, EventsRequest &eventsRequest, cl_int &errcodeRet);
@@ -441,7 +449,8 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
void unregisterGpgpuAndBcsCsrClients();
cl_int postStagingTransferSync(cl_event *event, const Event &profilingEvent, bool isSingleTransfer, bool isBlocking);
cl_int postStagingTransferSync(const StagingTransferStatus &status, cl_event *event, const cl_event profilingEvent, bool isSingleTransfer, bool isBlocking, cl_command_type commandType);
cl_event *assignEventForStaging(cl_event *userEvent, cl_event *profilingEvent, bool isFirstTransfer, bool isLastTransfer) const;
Context *context = nullptr;
ClDevice *device = nullptr;