[PATCH] [X86][SSE] Vector double -> float conversion memory folding (cvtpd2ps)

[Quentin Colombet]

Hi Simon,

LGTM.

Regarding this problem:

> it reloads the whole ymm and then references the lower xmm as the src for the conversion


Could you elaborate on the variants that are causing problems? IIRC, all inputs are xmm, not ymm.

Thanks,
-Quentin


REPOSITORY
  rL LLVM

http://reviews.llvm.org/D6663

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/