Closed matthiasdiener closed 3 years ago
@matthiasdiener Could you please revert a1b97a6?
@matthiasdiener Could you please revert a1b97a6?
sure, done
get rid of most of meshmode e9156d5
:+1: Thanks!
I think this is ready for another review @inducer @kaushikcfd . I had to work around some test failures, maybe you have ideas for better fixes.
add pytato to dependencies
I think we should follow pyopencl
and not make pytato a dep.
add pytato to dependencies
I think we should follow
pyopencl
and not make pytato a dep.
Packages using arraycontext now need pytato to run the tests; I'm not sure how to better handle this.
This has merge conflicts. Could you take a look?
This has merge conflicts. Could you take a look?
We'll have to implement ravel now for pytato I guess, but could you please take a look at the rest?
We'll have to implement ravel now for pytato I guess, but could you please take a look at the rest?
That's actually pretty complicated. (@kaushikcfd, correct? How is pytato's reshape
support?) I would recommend overriding the norm instead. The ravel
test can be xfail
ed.
We'll have to implement ravel now for pytato I guess, but could you please take a look at the rest?
That's actually pretty complicated. (@kaushikcfd, correct? How is pytato's
reshape
support?) I would recommend overriding the norm instead. Theravel
test can bexfail
ed.
I implemented a version of ravel() (that passes the test), but not sure if it's correct.
Edit:
If it is correct, we could move _rec_ravel
to pytato itself probably.
Supporting ravel fully (including the order "K" and "A") requires some thought, essentially it asks the question: "What is the order/strides of a pytato array expression's dims?". order = "C" for non-parametric shapes is fully supported till now. order="F" is not difficult to support, but order="K" and "A" are ill-defined for current pytato trunk.
Supporting ravel fully (including the order "K" and "A") requires some thought, essentially it asks the question: "What is the order/strides of a pytato array expression's dims?".
https://github.com/inducer/pytato/pull/92. IOW: IMO, that question is not supposed to have a well-defined answer.
I think this is officially ready for review.
Unsubscribing... @-mention or request review once it's ready for a look or needs attention.
This is ready for review I think.
This is ready for review I think.
I still need to take care of one comment. I will update here, once that's done.
This is ready to go in from end too!
Ready for review @inducer.
The followup PR with meshmode's version of transform_loopy_program is here: https://github.com/inducer/meshmode/pull/216
For anyone wondering about the CI failures: https://github.com/inducer/meshmode/pull/238 should fix those.
Thanks all for seeing this through. Looks great! In it goes.
Originally at https://github.com/inducer/meshmode/pull/116