Co_broadcast_cptr#2
Conversation
1ade4f5 to
1a27941
Compare
1a27941 to
37aa64a
Compare
3c116f1 to
ad1daf7
Compare
|
@certik while you're fixing LFortran bugs impacting Caffeine, here's another branch for an orthogonal forthcoming Caffeine feature that also fails (at runtime) with LFortran 0.62 and latest. It's getting a SEGV inside broadcast_derived_type() in test/prif_co_broadcast_test.F90 Here is a crash stack from LFortran 0.62: |
|
I can confirm that the previous main (328531f62f3d5c50b575003f936d0853712d8c1b) indeed failed, for me it fails with: The latest LFortran main (e9c464d2521a3bb5225c47e7752d91b65c2fdec6) compiles and runs all tests for this. So I think this is now fixed thanks to lfortran/lfortran#11181. |
@certik The failing CI run I linked above used |
|
Ok, we'll have to figure out how to extract the MRE for this. I made this lfortran/lfortran#11191. |
|
@certik: Failed again just now with lfortran:latest: |
6858cf0 to
a0ee388
Compare
|
The original code where the LFortran defect arose is archived in this branch: https://github.com/bonachea/caffeine/tree/refs/heads/co_broadcast_cptr-lfortran1 This pull request will advance by deploying a workaround that skips the failing test on LFortran, at least for now. |
882af9b to
434ba42
Compare
This particular part of the test code is unreachable in older versions of GFortran.
lfortran/lfortran#11191 Skip the failing test when using LFortran
Add `sequence` to derived types used in PRIF's contiguous communication calls, to ensure a flat linear storage layout for use in communicating the raw storage sequence.
434ba42 to
61121bc
Compare
|
Superseded by BerkeleyLab#319 |
Implements https://github.com/BerkeleyLab/prif/issues/156