[PATCH] D26090: [NVPTX] Compute 'rem' using the result of 'div', if possible.

[Justin Lebar via llvm-commits]

> It would be nice to have this as target-independent option

Sure.  I wonder how we should structure this -- a TLI option that we
use in DAGCombine?

> Only platforms with very slow hardware multiply want to use div and rem as separate instructions.

OOC do you know if we're doing this same optimization elsewhere,
and/or do we know of any specific targets that would immediately turn
on this optimization as a no-brainer?

On Sat, Oct 29, 2016 at 1:07 PM, Joerg Sonnenberger <joerg at netbsd.org> wrote:
> joerg added a comment.
>
> We've been talking about this on IRC. It would be nice to have this as target-independent option, since on many platforms a generic rem is much slower than a div followed by a multiply/subtract. Only platforms with very slow hardware multiply want to use div and rem as separate instructions.
>
>
> Repository:
>   rL LLVM
>
> https://reviews.llvm.org/D26090
>
>
>