[llvm] [AMDGPU] Fix pattern selecting fmul to v_fma_mix_f32 (PR #180210)

[Jay Foad via llvm-commits]

================
@@ -1853,12 +1853,14 @@ define float @v_mad_mix_f32_f16lo_f16lo_f32_denormals(half %src0, half %src1, fl
   ret float %result
 }
 
+; TODO: Constant 0x8000 should fold into v_fma_mix_f32 on GFX10+.
----------------
jayfoad wrote:

That's much nicer, thanks!!

https://github.com/llvm/llvm-project/pull/180210