[PATCH] D62024: [InstCombine] fold shuffles of insert_subvectors

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

In D62024#1506977 <https://reviews.llvm.org/D62024#1506977>, @efriedma wrote:

> > This transform relies on expanding the shuffle of narrowed operands back to
>
> I think that qualifies as its own category; you're basically saying that certain shuffles should be equivalent to certain other shuffles, which is sort of orthogonal to whether any particular shuffle is legal.


I'm struggling to find words to describe this without opening the door completely to arbitrary shuffle combining. Any suggestions are welcome. :)

The motivating transform that led me here is the last test diff (ends up as just a single insert_subvector), so I could restrict the transform further, but it seemed like the more general fold would be beneficial and reversible if necessary.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62024/new/

https://reviews.llvm.org/D62024