[llvm] [X86][FP16][BF16] Improve vectorization of fcmp (PR #116153)

[Simon Pilgrim via llvm-commits]

================
@@ -23333,12 +23339,8 @@ static unsigned translateX86FSETCC(ISD::CondCode SetCCOpcode, SDValue &Op0,
 
 /// Break a VSETCC 256-bit integer VSETCC into two new 128 ones and then
 /// concatenate the result back.
-static SDValue splitIntVSETCC(EVT VT, SDValue LHS, SDValue RHS,
-                              ISD::CondCode Cond, SelectionDAG &DAG,
-                              const SDLoc &dl) {
-  assert(VT.isInteger() && VT == LHS.getValueType() &&
-         VT == RHS.getValueType() && "Unsupported VTs!");
----------------
RKSimon wrote:

Replace with this?
`assert(VT.isInteger() && LHS.getValueType() == RHS.getValueType() && "Unsupported VTs!");`

https://github.com/llvm/llvm-project/pull/116153