[llvm] [AMDGPU] Make fneg/fabs/copysign legal for bf16 (PR #91676)

[Matt Arsenault via llvm-commits]

================
@@ -2045,6 +2059,21 @@ def : GCNPat <
   (V_OR_B32_e64 (S_MOV_B32 (i32 0x00008000)), VGPR_32:$src) // Set sign bit
 >;
 
+def : GCNPat <
----------------
arsenm wrote:

These should be the same as the f16 case. Can you avoid repeating the patterns with a foreach vt : [f16, bf16] over the existing patterns 

https://github.com/llvm/llvm-project/pull/91676