[llvm] [X86]Add NO_REVERSE attribute to X86 RMW instrs in memfold table (PR #67288)

Shengchen Kan via llvm-commits llvm-commits at lists.llvm.org
Mon Sep 25 19:02:26 PDT 2023


KanRobert wrote:

> X86 don't want to unfold RMW instrs to 1 load + 1 op + 1 store, because RMW could save code size. And from all the call position analysis, we could find we didn't unfold RMW in current code.

It's not only for code save. When the register pressure is very high, we can avoid clobbering one register by RMW.

https://github.com/llvm/llvm-project/pull/67288


More information about the llvm-commits mailing list