SWDEV-104441 - [SSG] OpenCL has not implemented the asynchronous transfer
- Use lock protection for multiple maps of persistent memory
- Don't mark persistent as host mem
- Implement file write for invisible memory
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.cpp#129 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#236 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#408 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#13 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#77 edit
[ROCm/clr commit: 9d739c200c]
SWDEV-76870 - Add write feature to AMD extension (Merging //depot/rel/r6/15.30.1023/stream/opencl/... to //depot/stg/opencl/drivers/opencl/...)
Add clEnqueueReadBufferToFileAMD function to AMD extension (added declaration in cl_context.cpp)
1. Added clEnqueueReadBufferToFileAMD_fn function declaration to cl_ext.h
2. Added LiquidFlashFile::writeBlock method who implements transfer from GPU to SSD. In order to avoid code duplication LiquidFlashFile::readBlock and LiquidFlashFile::writeBlock is called from new method LiquidFlashFile::transferBlock (changes in cl_lqdflash_amd.cpp)
3. clEnqueueWriteBufferFromFileAMD and clEnqueueWriteBufferFromFileAMD call internal function EnqueueTransferBufferFromFileAMD who makes the same preparations as clEnqueueWriteBufferFromFileAMD in the prvious release
4. WriteBufferFromFileCommand class is renamed to TransferBufferFromFileCommand and new class makes the same preparation except assigning transfer direction (read or write)
Affected files ...
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_context.cpp#48 integrate
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_lqdflash_amd.cpp#13 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_lqdflash_amd.h#4 integrate
... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/headers/opencl2.0/CL/cl_ext.h#24 integrate
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#267 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#396 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.hpp#138 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#74 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.hpp#81 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.hpp#96 edit
[ROCm/clr commit: 1cb5415538]
ECR #304775 - Fix for a memory "leak" in one GEHC sample
- The app builds a dependency graph when the new command waits for the previous one. Our runtime couldn't release wait commands in this situation, because release is done in the command destructor. Release the events from the wait list when the command is done.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#71 edit
[ROCm/clr commit: e0b2a44b3f]
ECR #304775 - Wave limiter: Fix bug in adaptation.
Dumped waves/simd value is incorrect.
Should exit adptation only after the changed waves/simd value is applied.
Added wave limiter manager to handle situation that one kernel is enqueued to more than one queues. Create wave limiter for each virtual device.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#245 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#283 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#109 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#360 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuwavelimiter.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuwavelimiter.hpp#3 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#70 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.hpp#76 edit
[ROCm/clr commit: 51efa976bf]
ECR #304775 - Some code clean-up
- Remove amd::Context reference from the event object. Runtime can use queue to get the context for the limited number of use cases
- User/GL events will keep the context as a member
Affected files ...
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_event.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#69 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.hpp#75 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.cpp#120 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.hpp#91 edit
[ROCm/clr commit: 458cca25ce]
ECR #304775 - Back out changelist 1117184
- There is a hang on CPU in conversion test
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#67 edit
[ROCm/clr commit: 31ca701a3e]
foreman