[llvm] [SDAG] Heed enableAggressiveFMAFusion when folding fma(a,b,c*d)+e (PR #94209)

[Matt Arsenault via llvm-commits]

================
@@ -39,7 +39,7 @@ define amdgpu_ps float @_amdgpu_ps_main() #0 {
 ; GFX10-NEXT:    v_sub_f32_e32 v8, s0, v1
 ; GFX10-NEXT:    v_fma_f32 v7, -s2, v6, s6
 ; GFX10-NEXT:    v_fma_f32 v5, v6, v5, 1.0
-; GFX10-NEXT:    v_mad_f32 v10, s2, v6, v2
+; GFX10-NEXT:    v_fma_f32 v10, s2, v6, v2
----------------
arsenm wrote:

Some of these test diffs don't make sense to me. Here we traded mad for fma (which probably shouldn't have happened), and below the opposite (traded fma for mad) 

https://github.com/llvm/llvm-project/pull/94209