This seems to build on PR gh-16782, correct?

@mattip, yes this pull-request temporary merge #16782, so I can be able to test the new intrinsics.

All tests are successfully passed, I will move testing units of the new initrinics to #16782 so we can merge this pr.
https://travis-ci.org/github/numpy/numpy/builds/733983312
https://github.com/numpy/numpy/pull/17340/checks?check_run_id=1226297258

@mattip, These intrinsics already been used by #17587 and #16247 and proved a good efficiency almost similar to the replacement raw SIMD in case of AVX2 and AVX512F, provide massive improvements for non-contiguous memory access
in the case of SSE and VSX, on the other hand, NEON/ASIMD shows acceptable improvements but not that wow.

I hope we can merge this pull-request as soon as possible.

@seiko2plus I notice that you are still making commits here. Do you feel that there is more to do?

I was hoping to merge #16782 first, thinking that then we might be able to add some (maybe marked @slow) tests using that infrastructure here. Does that make sense?

@charris, no, the last change I made on this pr was 17 days ago,

 seiko2plus force-pushed the seiko2plus:npyv_partial_noncont_mem branch from bec733b to 1b8637d 17 days ago

the other messages due to build #16247 and #17587 on the top of this pr(reference commit).

@mattip,

I was hoping to merge #16782 first,

I totally agree with you without testing cases it would be chaos.

thinking that then we might be able to add some (maybe marked @slow) tests using that infrastructure here

there's no need for @slow #16782 is too fast in running time the current ratio 1 to 5 seconds depending on
the enabled SIMD extensions. The only issue is the binary size and maybe the building time.

Thanks Sayed.