[PATCH] D51347: AMDGPU: Force shrinking of add/sub even if the carry is used

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In https://reviews.llvm.org/D51347#1216048, @arsenm wrote:

> In https://reviews.llvm.org/D51347#1216006, @rampitec wrote:
>
> > I am not sure that a copy + e32 instruction is better than a single e64 instruction. In fact I think it is worse.
>
>
> In practice the copy is always eliminated since its usually paired with a carry in operation. The total cycle count is the same with reduced code size


Usually the add is shrunk, so it’s more code size neutral


https://reviews.llvm.org/D51347