refactor: remove obsolete cuda_kernels

[ManasviGoyal]

Removed the following cuda_kernels which had a corresponding cpu-kernels in v1 but not in v2:

1. awkward_ByteMaskedArray_getitem_carry.cu
2. awkward_ByteMaskedArray_mask.cu
3. awkward_Index_to_Index64
4. awkward_IndexedArray_getitem_nextcarry_outindex_mask
5. awkward_IndexedArray_mask
6. awkward_ListOffsetArray_reduce_global_startstop_64
7. awkward_NumpyArray_contiguous_init
8. awkward_NumpyArray_contiguous_next
9. awkward_NumpyArray_fill_tobool
10. awkward_NumpyArray_getitem_boolean_nonzero
11. awkward_NumpyArray_getitem_next_array
12. awkward_NumpyArray_getitem_next_array_advanced
13. awkward_NumpyArray_getitem_next_at
14. awkward_NumpyArray_getitem_next_range
15. awkward_NumpyArray_getitem_next_range_advanced
16. awkward_carry_arange
17. awkward_combinations
18. awkward_content_reduce_zeroparents_64
19. awkward_index_carry_nocheck
20. awkward_regularize_arrayslice
21. awkward_zero_mask

This one was in none of the versions:

22. awkward_index_carry

[codecov]

Codecov Report

Merging #2876 (46f7f54) into main (d117334) will not change coverage.
The diff coverage is n/a.

Additional details and impacted files

Files	Coverage Δ
src/awkward/_connect/cuda/__init__.py	0.00% <ø> (ø)

[ManasviGoyal]

@jpivarski @ianna
These are two more that had cpu-kernels in v1 and don't have one in v2

• awkward_reduce_argmax_bool_64
• awkward_reduce_argmin_bool_64

I see that they have a different if condition. But since their cpu-kernels were dropped, are they still needed?

[jpivarski]

This is good to do as a first step, so that you don't accidentally spend time on obsolete kennels. The number of removed kennels doesn't surprise me—going from v1 to v2, it became possible to use NumPy in places where we previously needed custom kernels. For the GPU, CuPy takes that role (assuming that CuPy is a good replacement for NumPy in those places, which it very likely is, and if not, a shim in its nplike would fix it).

I'd say you can merge when ready. You only need one review, not two, although if @ianna has anything to say about this, please do!

[ManasviGoyal]

This is good to do as a first step, so that you don't accidentally spend time on obsolete kennels. The number of removed kennels doesn't surprise me—going from v1 to v2, it became possible to use NumPy in places where we previously needed custom kernels. For the GPU, CuPy takes that role (assuming that CuPy is a good replacement for NumPy in those places, which it very likely is, and if not, a shim in its nplike would fix it).

@jpivarski These two should be removed too right?

• awkward_reduce_argmax_bool_64
• awkward_reduce_argmin_bool_64

[jpivarski]

If you don't see them in v2, then yes.

[ianna]

@ManasviGoyal - looks good to me! please, merge it when you are done with it. Thanks!