[llvm] [AMDGPU] Fold redundant inf/nan checks into frexp instructions (PR #214936)

[Matt Arsenault via llvm-commits]

================
@@ -18829,6 +18829,131 @@ SDValue SITargetLowering::performClampCombine(SDNode *N,
   return getCanonicalConstantFP(DCI.DAG, SDLoc(N), N->getValueType(0), F);
 }
 
+/// Check if a value is a positive or negative infinity constant.
+static bool isInfinityFPConstant(SDValue V) {
----------------
arsenm wrote:

Use sd_match for this. This also won't handle vectors directly 

https://github.com/llvm/llvm-project/pull/214936