[PATCH] D71945: [X86] Potential improvement for v2i32->v2f64 uint_to_fp

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: llvm/test/CodeGen/X86/vec_int_to_fp.ll:1038
 ; AVX2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
----------------
craig.topper wrote:
> RKSimon wrote:
> > Would AVX1/AVX2 benefit for the v4i32-v4f64 case?
> What are our options for zext v4i32->v4i64 on avx1. We don’t get the instruction until avx2
We'd probably end up with a PMOVZX(xmm) for the lower v2i32, a PUNPCKH(xmm, zero) for the upper v2i32 followed by a VINSERTF128 and the VORPD(ymm) - and that removes a PBLENDW, PSRLD, 2*CVTDQ2PD and MULPD (+replace VADDPD with VSUBPD) - so that should be an improvement.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D71945/new/

https://reviews.llvm.org/D71945