[PATCH] D52134: [X86] Remove an fp->int->fp domain crossing in LowerUINT_TO_FP_i64.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon accepted this revision.
RKSimon added a comment.
This revision is now accepted and ready to land.

LGTM - since we're targetting older targets with pre-AVX code the domain penalty is more likely to be present.



================
Comment at: lib/Target/X86/X86ISelLowering.cpp:16740
   if (Subtarget.hasSSE3()) {
     // FIXME: The 'haddpd' instruction may be slower than 'movhlps + addsd'.
     Result = DAG.getNode(X86ISD::FHADD, dl, MVT::v2f64, Sub, Sub);
----------------
Update this to 'shuffle + addsd' ?


Repository:
  rL LLVM

https://reviews.llvm.org/D52134