[PATCH] D141980: [X86] Transform slow LEA_B_I_D/LEA_SLOWBASE_I -> LEA_IS_D/LEA_IS iff base == index

Tue Jan 17 18:57:47 PST 2023

craig.topper added a comment.

In D141980#4060945 <https://reviews.llvm.org/D141980#4060945>, @pengfei wrote:

> scale > 1 also has bad performation on some microarch, see Intel SOM.

I didn't check the SOM. But I think I remember that. One of the conditions for slow LEA is existence of a SIB byte in the encoding if I remember correctly. That's why RBP and R13 <https://reviews.llvm.org/source/pstl/> are always slow. They are always encoded using SIB.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D141980/new/

https://reviews.llvm.org/D141980