-
Notifications
You must be signed in to change notification settings - Fork 66
Use enable_language(HIP) #702
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
host-configs/llnl/toss_4_x86_64_ib_cray/[email protected][email protected]_hip.cmake
Show resolved
Hide resolved
Co-authored-by: Chris White <[email protected]>
adayton1
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks, David!
|
|
||
| # AMDGPU_TARGETS should be defined in the hip-config.cmake that gets "included" via find_package(hip) | ||
| # This file is also what hardcodes the --offload-arch flags we're removing here | ||
| if(DEFINED AMDGPU_TARGETS) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It seems like there's no longer a point for having AMDGPU_TARGETS.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Or is it used internally by hip's CMake setup?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't see it in CMake's source, maybe it is in the FindRocm source. I would have expected this to only be used as a backwards compatiblity and maybe just set the CMake variable.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I could do backwards compatibility but I think removal might be best since we are going to be having breaking BLT changes soon
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
i am okay with that as well.
…o feature/enable_language_hip
host-configs/llnl/toss_4_x86_64_ib_cray/[email protected]_hip.cmake
Outdated
Show resolved
Hide resolved
| endif() | ||
|
|
||
| blt_import_library(NAME blt_hip | ||
| COMPILE_FLAGS ${_blt_hip_compile_flags} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I just noticed this: _blt_hip_compile_flags is still set above but it is now unused. Is the fix for --rocm-path with crayftn no longer needed? @davidbeckingsale @white238
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@LLNL/blt my question still stands. We've noticed --rocm-path has disappeared from our compile lines.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If something is being compiled as LANGUAGE HIP then why would you need the rocm-path to be manually added (or for it to be guarded from CrayFTN)?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not sure, but something about the BLT update now causes some run time failures for us. We're experimenting to try to narrow it down. Adding --rocm-path back in doesn't fix it for us, but if it's really not needed, then all the code to set _blt_hip_compile_flags should be removed.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks like it is not --rocm-path that was the issue. We had a dependency that was ignoring the BLT we were telling it to use and instead using its own internal BLT. We switched from using AMDGPU_TARGETS to CMAKE_HIP_ARCHITECTURES when we migrated to BLT v0.7.0, but the old BLT was still relying on AMDGPU_TARGETS. So the code to set _blt_hip_compile_flags should probably be removed.
Inspired by #596 , but doesn't support hipcc