Add intel+cuda toolchain#1976
Conversation
| """Compiler toolchain with Intel compilers (icc/ifort), Intel MPI, | ||
| Intel Math Kernel Library (MKL), Intel FFTW wrappers and CUDA""" | ||
|
|
||
| NAME = 'intelc' |
There was a problem hiding this comment.
@akesandgren hmm, won't this raise eyebrows w.r.t. naming? would intelcuda be better?
There was a problem hiding this comment.
Yeah, i though of that too, but the other cuda enabled toolchains are also c, not cuda.
But, i'll leave it up to you as for the actual naming here...
| comp_versions = {ec['name']: self.det_full_version(ec)} | ||
| if ec['name'] == 'ifort': | ||
| # 'icc' key should be provided since it's the only one used in the template | ||
| comp_versions.update({'icc': self.det_full_version(ec)}) |
There was a problem hiding this comment.
@akesandgren there's a lot of copy-pasting going on here, not a big fan of that
Also, I'd like to see this handled through a separate PR, preferably combined with #1753 which we should push through as well.
There's still an open question holding back that PR, but I feel that shouldn't be held back much longer...
There was a problem hiding this comment.
Yeah, i was a bit lazy writing that the first time around. Easy coding makes for fewer bugs :-)
Not quite sure what #1753 is trying to do though.
I can break this part out and make the rest of this a dependency on it if you really want that.
|
Waiting for PR #1982 after which this one will be changed to depend on that instead. |
intel+cuda toolchain." Depend on PR# 1982 instead.
| 'Clang,GCC': ('Clang-GCC', '%(Clang)s-%(GCC)s'), | ||
| 'CUDA,GCC': ('GCC-CUDA', '%(GCC)s-%(CUDA)s'), | ||
| 'CUDA,icc,ifort': ('intel-CUDA', '%(icc)s-%(CUDA)s'), | ||
| 'CUDA,intel': ('intel-CUDA', '%(intel)s-%(CUDA)s'), |
There was a problem hiding this comment.
@akesandgren is this last entry also required? I would expect that only the CUDA,icc,ifort entry is required for the iccifortcuda toolchain?
You are not combining intel with CUDA anywhere, only icc+ifort with CUDA?
sync with develop & fix conflict
add additional test case for use of iccifortcuda in HMNS
fix broken test due to adding more test easyconfigs
|
Tested using easybuilders/easybuild-easyconfigs#3750, all good to go, thanks @akesandgren! |
This is the framework part of a new intel+cuda toolchain.
The intention is that it should work the same way the gcc+cuda toolchain works.