[PATCH] D91515: [X86] Use GF2P8AFFINEQB to implement vector bitreverse.

Mon Nov 16 01:43:09 PST 2020

RKSimon added a comment.

Cheers for looking at this!

Please can you add cost model coverage as well?

FYI I raised a bug about using GFNI for counting leading/trailing bits here https://bugs.llvm.org/show_bug.cgi?id=48050

The InstLatX64 guys found a load of other potential uses for GFNI at https://github.com/InstLatx64/InstLatX64_Demo/blob/master/GFNI_Demo.h

================
Comment at: llvm/test/CodeGen/X86/vector-bitreverse.ll:82
+; GFNISSE-NEXT:    # kill: def $al killed $al killed $eax
+; GFNISSE-NEXT:    retq
+;
----------------
We found it useful for XOP to handle scalar cases on the vector unit - would it be useful for GFNI as well?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D91515/new/

https://reviews.llvm.org/D91515