[llvm] InstCombine: Handle canonicalize in SimplifyDemandedFPClass (PR #173189)

[Matt Arsenault via llvm-commits]

================
@@ -112,8 +109,7 @@ define nofpclass(nan zero) float @ret_nofpclass_nan_zero__canonicalize_select_su
 define nofpclass(nan) float @ret_nofpclass_nan__canonicalize_only_psub_ieee() {
 ; CHECK-LABEL: define nofpclass(nan) float @ret_nofpclass_nan__canonicalize_only_psub_ieee() {
 ; CHECK-NEXT:    [[PSUB:%.*]] = call float @returns_psub()
-; CHECK-NEXT:    [[CANON:%.*]] = call float @llvm.canonicalize.f32(float [[PSUB]])
-; CHECK-NEXT:    ret float [[CANON]]
+; CHECK-NEXT:    ret float [[PSUB]]
----------------
arsenm wrote:

Because it's not supposed to be flushed with IEEE. It's any unknown denormal value 

https://github.com/llvm/llvm-project/pull/173189