[clang] [llvm] [IR] Allow fast math flags on fptrunc and fpext (PR #115894)

[Matt Arsenault via cfe-commits]

================
@@ -42,6 +42,14 @@ entry:
   %f = fneg float %x
 ; CHECK:  %f_vec = fneg <3 x float> %vec
   %f_vec = fneg <3 x float> %vec
+; CHECK: %g = fpext float %x to double
----------------
arsenm wrote:

Needs bitcode compatibility test 

https://github.com/llvm/llvm-project/pull/115894