[llvm] [X86][SSE] Improve ISD::FP_TO_*INT_SAT vector lowering (PR #199416)

[via llvm-commits]

================
@@ -1270,6 +1271,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::FP_TO_UINT,         MVT::v2i32, Custom);
     setOperationAction(ISD::STRICT_FP_TO_SINT,  MVT::v4i32, Custom);
     setOperationAction(ISD::STRICT_FP_TO_SINT,  MVT::v2i32, Custom);
+    if (!Subtarget.hasAVX10_2()) {
+      setOperationAction(ISD::FP_TO_SINT_SAT, MVT::v4i32, Custom);
+      setOperationAction(ISD::FP_TO_UINT_SAT, MVT::v4i32, Custom);
+    }
----------------
Panadulek wrote:

we set operation as legal in if  if (Subtarget.hasAVX10_2())  which is in 325 line.
if (!Subtarget.useSoftFloat() && Subtarget.hasAVX())  1492 line. 

So how i understand when code mark   operations ISD::FP_TO_UINT_SAT  and ISD::FP_TO_SINT_SAT as legal then code which is in if for hasAVX will make those operations as custom. So we will get regression for avx10_2. 

https://github.com/llvm/llvm-project/pull/199416