Branch: refs/heads/main
Revision: 717e225
Author: stonea
Link: Fix failing GPU tests. by stonea · Pull Request #19848 · chapel-lang/chapel · GitHub
Log Message:
Merge pull request #19848 from stonea/fix_failing_gpu_tests
Fix failing GPU tests.
There are a few changes:
Add PREDIFF symbolic link to multiGpu dir
The outermost PREDIFF script is written generically enough to handle this. Anyway I've updated the good files to not explicitly do this and am updated the multi/ directory to follow the precedent of having a symbolically linked PREDIFF to the more generic script.
Update multiGPU .good file to recognize multiple printouts of arrays
I'm not sure why this has changed (maybe due to changes in the locale model?).
But it seems like the new output is correct (we're supposed to reprint the arrays for each GPU, right?).
Update fnWithForall to use new locale model
Instead of doing getChild(1) do gpus[0].
Update memory allocation in diags
See this diff, from:
< 0 (cpu): diags.chpl:12: allocate 16B of array elements at 0xPREDIFFED
to:
0 (cpu): diags.chpl:12: allocate 128B of array elements at 0xPREDIFFED
I'm guessing this reflects an internal change in the array module and we should just update the .good file but I want a second opinion.
[Reviewed by @e-kayrakli]
Modified Files:
M test/gpu/native/diags.good
M test/gpu/native/kernelFnCalls/fnWithForall.chpl
M test/gpu/native/multiGPU/multiGPU.good
M test/gpu/native/multiGPU/worksharing.good
M test/gpu/native/multiGPU/worksharingBasic.good
Compare: https://github.com/chapel-lang/chapel/compare/567a17651997...717e22529f4a