[llvm] [DRAFT] Added handling for AVX512DQ+AVX512VL+AVX512F to lowerFPToIntToFP (PR #160582)

[Simon Pilgrim via llvm-commits]

================
@@ -19897,21 +19897,50 @@ static SDValue lowerFPToIntToFP(SDValue CastToFP, const SDLoc &DL,
   // See if we have 128-bit vector cast instructions for this type of cast.
   // We need cvttps2dq/cvttpd2dq and cvtdq2ps/cvtdq2pd.
   if (!Subtarget.hasSSE2() || (VT != MVT::f32 && VT != MVT::f64) ||
-      IntVT != MVT::i32)
+      (IntVT != MVT::i32 && IntVT != MVT::i64))
----------------
RKSimon wrote:

You'd be better off moving the hasDQI logic in here to simplify below:
```
!(IntVT == MVT::i32 || (IntVT == MVT::i64 && Subtarget.hasDQI())))
```

https://github.com/llvm/llvm-project/pull/160582