Remove PGI compiler and add shared object versioning to nvcc #2784
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2784 +/- ##
==========================================
- Coverage 83.07% 78.20% -4.88%
==========================================
Files 248 248
Lines 50623 50644 +21
Branches 4460 4462 +2
==========================================
- Hits 42053 39604 -2449
- Misses 7808 10206 +2398
- Partials 762 834 +72
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
JDBetteridge
changed the title
JDBetteridge
force-pushed
the
JDBetteridge/tweak_compilers
branch
2 times, most recently
from
f2d5ce1 to
ecee115
Compare
JDBetteridge
force-pushed
the
JDBetteridge/tweak_compilers
branch
from
ecee115 to
e89e262
Compare
| except (FileNotFoundError, CalledProcessError): | ||
| pass | ||
| finally: | ||
| if out.returncode == 0: |
There was a problem hiding this comment.
Can probably use re to oneline it but that's fine.
There was a problem hiding this comment.
I opted for this to make it consistent with the BASH method used to extract the version used in the dockerfiles. There might be better ways of obtaining this information 😅
| start = out.stdout.find(',', start) | ||
| stop = out.stdout.find('\n', start) | ||
| with suppress(InvalidVersion): | ||
| cuda_version = parse(out.stdout[start:stop]) |
There was a problem hiding this comment.
IS this more robust than Version(....) ?
There was a problem hiding this comment.
AFAICT they are identical: https://packaging.pypa.io/en/stable/version.html#packaging.version.parse
parse just seemed to be more descriptive here.
| for soname in versioned | ||
| ]) | ||
| super().add_libraries(filterfalse(lambda s: s.startswith(':'), new)) | ||
|
|
There was a problem hiding this comment.
Would make CudaCompiler inherit from NvidiaCompiler to habe a single version/cxxstd
There was a problem hiding this comment.
Seems like a good idea, however:
nvcis derived from the old Portland compilernvccis only for compiling CUDA code and leverages a host compilergcc,clang, etc...
That is to say: nvcc isn't a type of nvc and I don't think inheritance is appropriate here, especially since the NvidiaCompiler differs significantly from the GCC compiler, which is unfortunately the base of the "generic" Compiler class . There are several issues with the inheritance pattern used in the compiler.py file and it would be nice to do a bit of a tidy up, like I previously did here. But that can wait for another PR, I think!
This is WIP, I need this for a separate piece of work