Improving avx/avx2 swizzles

[DiamonDinoia]

Hi,

This PR optimzes swizzles where possible
Fixed a bug in the float avx2 swizzle which did not allow duplicates
Adds more tests

[serge-sans-paille]

If you could reduce code duplication by syndicating patterns as suggested, that would be great!

[DiamonDinoia]

If this passes the tests it can be merged. I hope I did not break any build this time.

Should I add myself to the copyright?

[serge-sans-paille]

You can add yourself, yes. And rebase on master branch ;-)

[DiamonDinoia]

I rebased. I did not squash as I have some commits that I would like to keep for reference in the branch. Feel free to squash and merge in master.

[serge-sans-paille]

Hey @DiamonDinoia : I'll have a look at this next week, don't worry if you don't have any feedback since then 🙇

[DiamonDinoia]

I am not sure what the problem is with PowerPC. It should not touch any code I wrote and from the print the results seems correct.

[serge-sans-paille]

How strange: the failure shows batches that... look the same /o\

[DiamonDinoia]

I could not find an explanation for it.

[serge-sans-paille]

If you don't mind, I'll split your commit history in pieces and validate them step-by-step in order to understand what's happening - keeping your authorship information, of course.

[DiamonDinoia]

Sure, no problem, go ahead!

[serge-sans-paille]

perfect, can you just squash your commits into a nice history? Thanks o/

[serge-sans-paille]

Thanks for being patient and for the great patch set \o/

[serge-sans-paille]

See ##1155 for the sse2 part

[DiamonDinoia]

Thanks for being patient and for the great patch set \o/

My pleasure!