[PATCH] D130151: [X86][FP16] Do not split FP64->FP16 to FP64->FP32->FP16
Phoebe Wang via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 21 05:56:52 PDT 2022
pengfei added inline comments.
================
Comment at: llvm/test/CodeGen/X86/vector-half-conversions.ll:727
; AVX2-NEXT: vcvtpd2ps %ymm0, %xmm0
; AVX2-NEXT: vcvtps2ph $0, %xmm0, %xmm0
; AVX2-NEXT: vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
----------------
pengfei wrote:
> LuoYuanke wrote:
> > It seems transform from double to float and then float to half. The same for AVX1.
> Good catch! Will investigate.
On non-AVX512 case, the v8f64->v8f16 was firstly split to v4f64->v4f16. Then v4f16 will be widden to v8f16 by another path which I missed to change :(
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D130151/new/
https://reviews.llvm.org/D130151
More information about the llvm-commits
mailing list