Andy Porter

An update, I've patched: ``` --- components/science/source/algorithm/solver/sci_r_solver_field_vector_mod.x90(revision 52054) +++ components/science/source/algorithm/solver/sci_r_solver_field_vector_mod.x90(working copy) @@ -12,7 +12,7 @@ use constants_mod, only : i_def, r_def, r_solver use r_solver_field_mod, only : r_solver_field_type use field_mod, only...

Adi has confirmed that it's a PSyclone failure, not a compile-time failure so this is not #2601.

``` Generation Error: generator: specified PSyclone transformation module 'psct-omp_gpu_trans' raised the following exception during execution... { File "/home/dn-sad1/rds/hpc-work/ngarch/nemo/cfgs/BENCH_ST/BLD_SCT_PSYCLONE/psct-omp_gpu_trans.py", line 123, in trans omp_loop_trans.apply(loop, options={"force": True}) File "/home/dn-sad1/.local/lib/python3.11/site-packages/psyclone/psyir/transformations/omp_loop_trans.py", line 265, in...

I think (from yesterday) that Adi is just using the standard OMP offload script from the examples directory.

Yes, it's #2027. I have a fix I'm working on in that Issue.

Also, NG-Arch is working with ORCA2-ICE (less PISCES) and that is showing more issues that we see with BENCH.

Thanks for this one Sergi. I noticed today that `ACCRegionDirective` has `signatures` which should also be replaced somehow with `reference_accesses`: https://github.com/stfc/PSyclone/blob/433d31bdceb199754e2a260716e749ec7a4222f2/src/psyclone/psyir/nodes/acc_directives.py#L97-L128

#2278 could also be considered a part of this.

Just a note for the future: there's no need to open an Issue 'towards' another Issue :-) Just add to the existing Issue. It is then the PR that is...

I think collapse has been around quite a long time (at least 6 or 7 years). We support it in OpenACC so should be able to generalise that.