SWDEV-112016 - [ROCm CQE][OCLonLC][QR][G] Seg fault observed with "sub_buffers_read_write" of Buffers, due to CL#1364923
- Take the host pointer for CPU access in bufferfill.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#8 edit
[ROCm/clr commit: 027f8d9108]
SWDEV-94644 - Make sure we are processing metadata note entry with supported n_type. Update build log and fail for not supported metadata n_type. Use constants defined in AMDGPUPTNote.h
This change is needed for https://reviews.llvm.org/D29115
This change is required for CL 1366203
ReviewBoardURL: http://ocltc.amd.com/reviews/r/12223/
Testing: lightning conformance tests locally
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#31 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#54 edit
[ROCm/clr commit: 4ad12d089c]
SWDEV-110504 - OpenCL on ROCm: MAX_WORK_GROUP_SIZE is 256 but max work items per dimension is 1024
- Report Orca stack capabilities. OCL runtime currently can't report 1024 due to possible regressions in the apps.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#32 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#12 edit
[ROCm/clr commit: 0673572c79]
SWDEV-108384 - Force map memory offset to 0
Copy what we do in ORCA and PAL backend: When mapping an image, default the offset to 0 instead of the map offset for the staging buffer.
ReviewBoardURL = http://ocltc.amd.com/reviews/r/12010/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#25 edit
[ROCm/clr commit: 320d5de3a1]
SWDEV-108384 - Send down row pitch and slice pitch
This fixes test case #8 and 9.
ReviewBoardURL = http://ocltc.amd.com/reviews/r/12009/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#24 edit
[ROCm/clr commit: d027173ee5]
SWDEV-107442, SWDEV-108405 - Need to release GPU memory fence before releasing resources
When we call submitKernelInternal, we need to release GPU memory fence before release resources used in the kernel otherwise we can page fault.
ReviewBoardURL = http://ocltc.amd.com/reviews/r/11955/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#5 edit
[ROCm/clr commit: 824a34cf54]
SWDEV-107568 - [ROCm CQE][OCL][CZ] Basic 2.0 conformance test giving Segmentation fault (core dumped) at "progvar_prog_scope_uninit"
- Detect if writable program scope variables are present in the program, and if so, insert barrier each dispatch of a kernel from this program.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#11 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#48 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#16 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#21 edit
[ROCm/clr commit: b705f74166]
SWDEV-102966 - Dump code object disassembly in OpenCL rocm device.
Invoke DumpExecutableAsText from driver library.
Update build to depend on some more LLVM libraries.
LLVM changes are included, but will come through amd-common.
Driver changes will come through ROCm-OpenCL-driver.
Testing: Run some SDK samples/test_basic with AMD_OCL_BUILD_OPTIONS_APPEND=-save-temps
Reviewed by: Laurent Morichetti, German Andryeyev.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#47 edit
[ROCm/clr commit: c7d2b188de]
SWDEV-107558 - Wrong path for map/unmap images
If mapMemory is allocated, we shouldn't use the read/writeImage path since they lock/unlock the host pointer which in this case, is mapped to GPU memory already.
ReviewBoardURL = http://ocltc.amd.com/reviews/r/11891/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#20 edit
[ROCm/clr commit: 45bf0e3ba6]
SWDEV-108024 - [ROCm CQE] Printf broken on ROCm , LC and HSAIL path
A zero-lenth printfBuffer is not an error. Even if a printfInfo is present, it does not mean that the kernel is going to call printf. Return true if offsetSize is 0.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprintf.cpp#5 edit
[ROCm/clr commit: f07e2533df]
SWDEV-105136 - Use the "execution" view rather than the "linking" view to find the metadata and size of the program scope variables.In the "execution" view, the section header table is optional, so we should iterate through the segments to add up the size of PT_LOAD segments with read but not execute flags. We will also find the metadata in the PT_NOTE segment.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#24 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#45 edit
[ROCm/clr commit: 66c5d710bc]
SWDEV-102510 - Need a way to control cl_khr/cl_amd extension macros
- Use -cl-ext option to enable OpenCL extensions
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palcompiler.cpp#11 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#44 edit
[ROCm/clr commit: 90d4cf8e78]
SWDEV-103793 - [ROCm CQE][OCLonLC][ISV] QCDGPU test fails
- The runtime should not report cl_amd_fp64, cl_khr_fp64 should be used instead.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#9 edit
[ROCm/clr commit: 30a6048658]
SWDEV-95488 - Limiting max cap Context on ROCm, it is limited to 100
- Re-order the member variables to match the initializer list.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#8 edit
[ROCm/clr commit: 1f5d19b75b]
SWDEV-94611 - [OCL-LC-ROCm] Add "HSAIL" or "LC" to the driver version string to tell which stack is used.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#24 edit
[ROCm/clr commit: 248840bd54]
SWDEV-102937, SWDEV-103130 - Set agent acquire fence all the time in dispatch
Misunderstanding of fence acquire and release: The fence release doesn't mean another dispatch will wait. So we need to set the acquire in the dispatch kernel even if we insert a barrier packet to try to sync.
We could try to fix this by including the memory tracking to all the image/buffer operations provided by OpenCL (ReadImage, WriteImage, etc.) but that's a much bigger change and not what it's in place right now at all.
ReviewBoardURL = http://ocltc.amd.com/reviews/r/11508/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#15 edit
[ROCm/clr commit: da9d770820]
SWDEV-104875 - ROCm/HSA: use the finalizer from source tree
Right now ROCm/HSA uses finalizer from the HSA RT installed.
This finalizer version has outdated stale SC sources.
At the same time source tree has fresh finalizer sources matching ORCA.
The offline tool amdhsafin is built from that sources.
This change switches from HSA RT finalizer to the in tree finalizer.
Testing: precheckin
Reviewed by Laurent Morichetti and Evgeny Mankov
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.cpp#20 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#39 edit
[ROCm/clr commit: 9385a3778f]
foreman