[PATCH] [X86] Teach how to custom lower double-to-half conversions under fast-math.

[Ahmed Bougacha]

Should this be in the target-independent legalizer instead, like FP16_TO_FP?  Only with UnsafeFPMath, of course (I'm a bit uncomfortable with that, but I don't find it shocking, and I see there's precedent.)

Otherwise, the change seems reasonable, thanks!

-Ahmed


================
Comment at: test/CodeGen/X86/fastmath-float-half-conversion.ll:17
@@ +16,3 @@
+; CHECK-NOT: callq  {{_+}}truncxfhf2
+; CHECK: vcvtps2ph $0, %xmm0, %xmm0
+; CHECK: ret
----------------
Not very important, but add a check for fp80->fp32, perhaps?

http://reviews.llvm.org/D7832

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