[llvm] [AMDGPU] Fix FMA Combine (PR #119217)

[Matt Arsenault via llvm-commits]

================
@@ -14696,7 +14696,7 @@ SDValue SITargetLowering::performFMACombine(SDNode *N,
   EVT VT = N->getValueType(0);
   SDLoc SL(N);
 
-  if (!Subtarget->hasDot7Insts() || VT != MVT::f32)
+  if (!Subtarget->hasDot10Insts() || VT != MVT::f32)
----------------
arsenm wrote:

I still have no idea what these numbers mean 

https://github.com/llvm/llvm-project/pull/119217