[PATCH] D32085: AMDGPU: Do not lower fast unsafe div for safe, f32, with fp32 denormals

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:3384-3385
 
+  if (!Unsafe && VT == MVT::f32 && Subtarget->hasFP32Denormals())
+    return SDValue();
+
----------------
How is this different from the check right below in the constant case? Is it just the multiply by reciprocal case for non-constant below that is the problem?


https://reviews.llvm.org/D32085