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

[Craig Topper via Phabricator via llvm-commits]

craig.topper marked an inline comment as done.
craig.topper added inline comments.


================
Comment at: llvm/test/CodeGen/X86/vec_int_to_fp.ll:734
 ; VEX-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
 ; VEX-NEXT:    retq
 ;
----------------
RKSimon wrote:
> Would AVX1/AVX2 benefit here?
This test case is weird. It explicitly use a v4i32->v4f64 and then extracts it to v2f64. The sse tests changed because we had to split the v4f64 during type legalizaiton and then half the split became dead. With AVX we don't split it and then push the extract through later after vector op legalization.


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

https://reviews.llvm.org/D71945