SWDEV-189140 - Add P2P support in PAL path
- PAL requires P2P resource open on the usage device. Add the new interface to open the resource
- Add a hidden P2P device object creation into amd::Memory. It can be activated with OCL context that has a single device.
Affected files ...
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_p2p_amd.cpp#2 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#337 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#134 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palmemory.cpp#25 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#74 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.hpp#28 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#80 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#133 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#126 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#93 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.cpp#136 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.hpp#109 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#306 edit
SWDEV-187956 - [CQE OCL][NAVI 10][WIN][LNX] Assertion failure observed with Subtests of OpenMM on PAL/LC path
- Correct the number of SIMDs and SIMD width for Navi10.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldefs.hpp#51 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#24 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/runtime/OCLDeviceQueries.cpp#54 edit
SWDEV-145570 - Disable subwindow copy temporarily to see if TC random hangs go away
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#30 edit
SWDEV-132899 - [OCL] set lcWavefrontSize64_ on ROCm for all ASICs except gfx10+ so that we Link wavefrontsize64 correctly
ReviewBoardURL = http://ocltc.amd.com/reviews/r/17081/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#43 edit
SWDEV-168145 - Add ECC target feature to OpenCL runtime
- hard coded SRAM ECC target feature for now since ROCr disable sram-ecc reporting via ISA until HCC is fixed
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#123 edit
SWDEV-180407 - Observed failure while running OCL 2.0 conformance API : min_max_device_version
- revert CL1739455 to use OCL version 1.2 as default to avoid this issue for ROCm 2.2 release
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#117 edit
SWDEV-134107 - Add support for respecting target's xnack setting
- Enable the XNACK feature for all the APU system and remove the xnackEnabled_ field in AMDDeviceInfo struct
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#332 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#116 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#98 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#41 edit
SWDEV-178313 - Properly enable OpenCL 2.0 on ROCm/LC path for Vega10+.
OPENCL_VERSION_STR is 2.1, but we only enable 2.0 since we don't have compiler's support for 2.1.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#115 edit
SWDEV-178313 - Enable OpenCL 2.0 on ROCm/LC path for Vega10+
Doorbell self-ring doesn't work for Fiji, so we enable 2.0 only for Vega10+ for now.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#114 edit
SWDEV-127767 - Don't guess at the suffix for the device libraries
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/CMakeLists.txt#18 edit
SWDEV-172202 - Workaround the scheduler for systems don't support PCIe 3 atomics properly.
The idea is the scheduler uses a device side global as write_index, and only write the write_index back to the hsa queue when the last thread of the scheduler leaves.
This change along with the library side change have been tested on systems with or without proper PCIe 3 atomics support.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#29 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsched.hpp#2 edit
SWDEV-162389 - OpenCL Support for COMgr
- added the machineTargetLC_ values, which was introduced in CL1702548, for Carrizo and Hawaii
- requested by Joseph Greathouse for public users (https://github.com/RadeonOpenCompute/ROCm/issues/668)
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#22 edit
SWDEV-172202 - Back out changelist 1730757.
Failure in OCLDynamic tests in various TC Sanity tests.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#111 edit
SWDEV-79445 - OCL generic changes and code clean-up
- Keep the body of all methods in the Program interface
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#97 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#44 edit
SWDEV-79445 - OCL generic changes and code clean-up
- Move the constructor body of LightningProgram to the header
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#96 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#43 edit
SWDEV-79445 - Back out changelist 1722556
- More changes are necessary on ROCm backend to support a dynamic switch between HSAIL and LC
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#107 edit
SWDEV-145570 - Use Subwindow copy SDMA for D->H and H->D copies if possible or fall back to linebyline copies if unalinged pitch.
- Set correct flags for SVM finegrain buffer for ROC backend
ReviewBoardURL = http://ocltc.amd.com/reviews/r/16353/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#41 edit
SWDEV-79445 - OCL generic changes and code clean-up
- Allow ROCM build within the same workspace as PAL. Please note that ROCM defualt path in this case will be HSAIL.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#105 edit
SWDEV-169078 - Also copy private_segment_size/group_segment_size to runtime handle for COMgr support
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#75 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#46 edit
SWDEV-79445 - OCL generic changes and code clean-up
- Modify CMakeLists.txt to switch to different library interface to match the COMgr requirement
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/CMakeLists.txt#16 edit
SWDEV-170523 - [Sugon]OCL conformance Maths and integer-ops fails with commandqueue error
- increase the max command queue to 200
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#40 edit
SWDEV-169078 - Also copy private_segment_size/group_segment_size to runtime handle.
The library need these info to setup child kernel's AQL for device enqueue.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/devkernel.hpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#69 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#26 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#44 edit
SWDEV-164783 - [ROCm QA][G][QR] Wimpyfull ->Buffer->sub_buffers_read_write_dual_devices test is failing
- Problem looks to be due to a change made in rocmemory.cpp in allocMapTarget() function.
- The issue was that Vlad modified the host memory that was being returned by using 'mapMemory_' vs 'owner()' as a pointer to use 'getHostMem()'.
- Added the old logic back in while also maintaining Vlads logic, and the test is now passing.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#40 edit
foreman
foreman