[PATCH] D69989: Assume ieee behavior without denormal-fp-math attribute

[Matt Arsenault via Phabricator via llvm-commits]

arsenm marked an inline comment as done.
arsenm added inline comments.


================
Comment at: llvm/test/CodeGen/X86/sqrt-fastmath-mir.ll:115
 
-attributes #0 = { "unsafe-fp-math"="true" "reciprocal-estimates"="sqrt:2" }
-attributes #1 = { nounwind readnone }
+attributes #0 = { "unsafe-fp-math"="true" "reciprocal-estimates"="sqrt:2" "denormal-fp-math"="ieee,ieee" }
+attributes #1 = { "unsafe-fp-math"="true" "reciprocal-estimates"="sqrt:2" "denormal-fp-math"="ieee,preserve-sign" }
----------------
spatel wrote:
> Double-check my understanding: we have the "ieee,ieee" attribute here, but we would never expect to see that in practice?
Yes, ieee,ieee is the default, but explicitly setting it is also valid. In the AMDGPU specific patches, I switched the default based on calling convention for graphics shaders (but I like this less over time, and may remove this detail)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69989/new/

https://reviews.llvm.org/D69989