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This finalizer version has outdated stale SC sources.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]
SWDEV-103420 - [ROCm CQE][OCLonLC][Fiji] 9 of 47 tests failed in oclperf.so component in Fiji
Enable image support for the Lightning path.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#20 edit
[ROCm/clr commit: 59612c30f9]
SWDEV-103418 - [ROCm CQE][OCLonLC][Fiji] OCLCreateBuffer results in Segmentation fault
Fix KernelBlitManager::createProgram to report an error if one or more blit kernel could not be created. Re-enable the image blit kernels in the blit program.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/blitcl.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#4 edit
[ROCm/clr commit: fd73443cb0]
SWDEV-94610 - Remove the padding at the end of the kernargs (It was for the hidden arguments, but now, LC reports the correct size). Set the LLVM triple to amdgcn-amd-amdhsa-opencl when building the built-in library.
Affected files ...
... //depot/stg/opencl/drivers/opencl/opencldefs#186 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#20 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#37 edit
[ROCm/clr commit: 0fb0fb1d8b]
SWDEV-102540 - Fill the device info_.boardName_ field from the info returned by ROCR's HSA_AMD_AGENT_INFO_PRODUCT_NAME query.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#19 edit
[ROCm/clr commit: 0dd6aae926]
SWDEV-94610 - Add gfx700 to the list of suported targets in HSAILProgram::linkImpl_LC. When dumping the source (-save-temps), print the options actually sent to clang as well as the options passed to OpenCL.
Affected files ...
... //depot/stg/opencl/drivers/opencl/library/build/Makefile.library#56 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/amdgpu_metadata.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#18 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#36 edit
[ROCm/clr commit: 383e97425b]
SWDEV-96479 - changing tab to space as Victor pointed out
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#18 edit
[ROCm/clr commit: c2b57553dc]
SWDEV-96479 - bug fix, exposing atomics capabilities based on GFX IP. For Hawaii the atomics is not supported hence disabled.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#17 edit
[ROCm/clr commit: d793001561]
SWDEV-94610 - Target features are only needed in the CL->IR stage. The attributes remain on the function, so they should not be set again in the IR->ISA stage.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#16 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#34 edit
[ROCm/clr commit: a1009a5d11]
SWDEV-94610 - Don't use the -cl-denorms-are-zero, but instead set the fp32/fp64 denorms with the target features +fp32-denormals and +fp64-denormals. fp64-denormals is always set, fp32-denormals in only set if device >= gfx900 and -cl-denorms-are-zero is not set.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#33 edit
[ROCm/clr commit: 7239172265]
SWDEV-94610 - Revert to the HSA 1.0 API for now.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#16 edit
[ROCm/clr commit: fe6ef85918]
SWDEV-94611 - [OCL-LC-ROCm] Use GFX IP for device name. Set the name to "gfx[M][m][s]" (M:major,m:minor,stepping). Removed the device name strings from the DeviceInfo table. Keep the machineTarget_ field until the compiler is changed to accept gfxip strings.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#15 edit
[ROCm/clr commit: c828fa8498]
SWDEV-94644 - Run prepare-builtins from the modules build directory, instead of right before generating the include files. Renamed the files to match the opensource build names (except for the .amdgcn suffix). Automatically generate a single include file for all libraries.
Affected files ...
... //depot/stg/opencl/drivers/opencl/library/build/Makefile.library#54 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/irif/build/Makefile.irif#7 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/ockl/build/Makefile.ockl#8 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/oclc/build/Makefile.oclc#10 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/ocml/build/Makefile.ocml#8 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/opencl/build/Makefile.opencl#10 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#30 edit
[ROCm/clr commit: 8bb15b463b]
SWDEV-96354 - Wrong usage of hsaImageData_ and deviceMemory_.
Use hsaImageData_ as the original pointer before alignment and only for that purpose. The deviceMemory_ is where the data is located. No one ever needs to use hsaImageData_ really. This is only an issue with tiled images
ReviewBoardURL = http://ocltc.amd.com/reviews/r/11331/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.hpp#4 edit
[ROCm/clr commit: 51948f577c]
SWDEV-94610 - Make sure each kernarg segment sits on a different cache line (align the kernargs on cache lines at minimum). Minor misc cleanups.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#13 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#8 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#13 edit
[ROCm/clr commit: 3a61b24dd5]
SWDEV-94610 - Fill the compileSize_ and compileSizeHint_ info from the LC metadata.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#13 edit
[ROCm/clr commit: 7d5471671b]
SWDEV-101853 - roc::Kernel cleanups:
- Remove unused classes & member functions/variables.
- Flatten vector arguments for the HSAIL path to remove the need for numElem_.
- Consolidate initArguments in a single loop for the HSAIL path.
- Use the Kernel::Argument to fill the OCL descriptor as much as possible.
- Set the access qualifier for both buffers and images.
- Fix the indentation and coding conventions.
- Add new ROC_ARG_TYPE type for hidden arguments
- Add an index_ field the roc::Kernel::Argument to record the OCL signature index for this argument, or -1 for hidden arguments
- Handle the hidden arguments as any other argument at dispatch (now included in the hsailArgList_)
- roc::Kernel::hsailArgAt(int) now returns the kernel argument for the given position in the OCL signature, not the position the the hsailArgList_.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#12 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#12 edit
[ROCm/clr commit: 1c37625aa2]
SWDEV-101169 - Compile the PCH file from <stdin> instead of a file reference. This removes the requirement to have the original file present when using the PCH file.
Affected files ...
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/headers/build/Makefile.headers#9 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/irif/build/Makefile.irif#6 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/ockl/build/Makefile.ockl#7 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/oclc/build/Makefile.oclc#9 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/ocml/build/Makefile.ocml#7 edit
... //depot/stg/opencl/drivers/opencl/make/amdgcn.git/opencl/build/Makefile.opencl#9 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#14 edit
[ROCm/clr commit: 7937fd1812]
SWDEV-94610 - The spec says that the value returned for HSA_EXECUTABLE_SYMBOL_INFO_NAME_LENGTH does not include the NUL terminator. We should add one before using the string.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#25 edit
[ROCm/clr commit: 557d2bfddf]
SWDEV-94610 - Restore the amdgpu_metadata.[ch]pp files. We need to share these files between different projects, and should avoid branching them. Ideally, they would be part of a metadata utility library.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/amdgpu_metadata.cpp#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/amdgpu_metadata.hpp#1 branch
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#9 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmetadata.cpp#3 delete
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmetadata.hpp#4 delete
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#11 edit
[ROCm/clr commit: ee978692f5]
SWDEV-94610 - Use the metadata to set the correct size for pointer arguments. Pointers to different address spaces may be of different sizes.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#11 edit
[ROCm/clr commit: 741816f873]
SWDEV-94610 - Fix the argName length issue. The string returned by the ROCR is already NUL-terminated.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#22 edit
[ROCm/clr commit: 52e3652f92]
SWDEV-94610 - Fix the API::get_kernel_arg_info conformance test failure. The runtime metadata needs to return references from Name() and TypeName() instead of temporary strings. Name().c_str() should be valid until the program is destroyed.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmetadata.cpp#2 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmetadata.hpp#3 edit
[ROCm/clr commit: 29a3f5d56a]
foreman