Test failure on i386 and armhf platforms about compyle HOT 17 CLOSED

@avalentino -- I suspected as much, that cython will pass and opencl would not. At least I know where the issue is potentially and can work on a fix that works for both. Thanks for your patience.

from compyle.

@avalentino -- this is now merged in master. Thanks for your patience!

from compyle.

Is there any progress on this?
I have just tested the new release 0.7 and the problem seems to be still there.

from compyle.

Sorry about that, is this both on i386 and armhf? I suspect this is just a poor mapping between the types that we have -- we have perhaps assumed that the world is now 64 bit. Is there a simple way for us to test this ourselves? I can try to spend some time on this later this week but cannot test it, would you be willing to help test this if I push to a public branch for instance?

from compyle.

Thanks @prabhuramachandran for the quick reply.
Sure I can help to test and I can also try to help with the fix if you provide some guidance.
I can test quite easily on i386, Unfortunately testing on armhf is more complex for me, I have to wait for the standard debian CI.

from compyle.

@avalentino -- sorry my response times can be quite terrible sometimes. Thank you for your kind offer! I will try to push a fix to a branch and let you know this week.

from compyle.

@avalentino -- can you please try the fix-32bit branch I have just pushed and see if that fixes the problem? Many thanks! Also there a way to easily test this on travis-ci to prevent issues going forward?

from compyle.

Hi @prabhuramachandran unfortunately the test still fails on i386:

https://salsa.debian.org/science-team/compyle/-/jobs/1047959

from compyle.

Thanks @avalentino, I have pushed another small change. Would it be possible to test the latest version of the branch? Thanks!

from compyle.

Dear @prabhuramachandran the test still fails: https://salsa.debian.org/science-team/compyle/-/jobs/1049211
This time I have also enabled verbose test output and do not caprure output streams.
Please note that the problem seems t affect only the cython backend:

compyle/tests/test_array.py::test_align_multiple[cython] FAILED
compyle/tests/test_array.py::test_align_multiple[opencl] 1 warning generated.
2 warnings generated.
PASSED
compyle/tests/test_array.py::test_align_multiple[cuda] SKIPPED

from compyle.

@avalentino -- Thanks for testing. I've made another fix which removes some unnecessary code. Let's hope this works.

from compyle.

Hi, @prabhuramachandran it seems you are on the right way.
The test still fails unfortunately [1].
The good news is that now the cython backend works, while the pyopencl is broken (in the previous run it was the contrary)

compyle/tests/test_array.py::test_align_multiple[cython] PASSED
compyle/tests/test_array.py::test_align_multiple[opencl] FAILED
compyle/tests/test_array.py::test_align_multiple[cuda] SKIPPED

Is there a way to change the dtype_to_ctype behavior depending on the calling backend?

[1] https://salsa.debian.org/science-team/compyle/-/jobs/1054489

from compyle.

Hi @prabhuramachandran, just a reminder.
Now that we have found the origin of the issue it would be nice to have this fixed for Debian 11.

from compyle.

@avalentino -- yes sorry, I am chasing down another urgent bug and should get to the issue later this week.

from compyle.

@avalentino -- I have attempted a more robust fix and force pushed to the fix-32bit branch. Please let me know if this works. Thanks!

from compyle.

@prabhuramachandran I can confirm that the new fix works correctly. thanks a lot.

from compyle.

@avalentino -- thank you very much for testing!

from compyle.