[PATCH] D50705: AMDGPU: Address todo for handling 1/(2 pi)

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: lib/Target/AMDGPU/AMDGPUISelLowering.cpp:3457
+
+  return APF.bitwiseIsEqual(KF16) ||
+         APF.bitwiseIsEqual(KF32) ||
----------------
arsenm wrote:
> rampitec wrote:
> > You need to check bitwidth here. 0x3118 is a valid float 1.7612E-41. I assume same is true for double.
> .bitwiseIsEqual already checks the FP semantic type. This isn't a floating point compare
The BitsToFloat shouldn't be used for half though


https://reviews.llvm.org/D50705