[Fix] missing upstream fixes #82
Conversation
|
@willend I don't have an easy way to test that I've implemented your changes correctly, especially McStasMcXtrace/McCode#1761 Could you possibly use this branch with your testing infrastructure to check if things look right? |
|
@g5t thanks, I have installed locally via the .whl which came in very handy! Current test incoming is for CPU MPI initially (should I try with --openacc / multicore also?) |
|
It's the OPENACC/MULTICORE defines that I'm least confident about, so a test that exercises them would be welcome 😁 |
|
Since I only had mccode-antlr tooling on the CPU-only box, testing with GPU/OPENACC required a little more handheld configuration.... But now an 8-way GPU run is also incoming. |
(Sync with upstream top-level cogen defines that control/influence the use of __builtin_isnan() and friends.)
|
@g5t just added a commit to PR home-branch that adds a little define-hierarchy at the top of the cogen header: (The With this, most of the differences in compilation success between "CPU-antlr" and "GPU-antlr" seen in https://new-nightly.mcstas.org/2024-11-18_output/2024-11-18_output.html (col 2-3 vs 4) should go away... |
There was a problem hiding this comment.
As mentioned on the PR itself, I just a commit to PR home-branch that adds a little define-hierarchy at the top of the cogen header:
#ifndef WIN32
# ifndef OPENACC
# define _GNU_SOURCE
# endif
# define _POSIX_C_SOURCE 200809L
#endif
(The _GNU_SOURCE needs to be there to ensure correct use of __builtin_isnan on (also very old libc) gcc-environments like conda, for the Nvidia nvc (openacc) compiler, that define should not be there, and naturally cl.exe understands none of it all ;-) )
With this, most of the differences in compilation success between "CPU-antlr" and "GPU-antlr" seen in https://new-nightly.mcstas.org/2024-11-18_output/2024-11-18_output.html (col 2-3 vs 4) should go away...
No description provided.