[llvm] [LoongArch] Custom legalize vselect to vector_shuffle for most constant condition cases (PR #163489)

[Zhaoxin Yang via llvm-commits]

================
@@ -592,10 +592,57 @@ SDValue LoongArchTargetLowering::LowerOperation(SDValue Op,
     return lowerVECREDUCE(Op, DAG);
   case ISD::ConstantFP:
     return lowerConstantFP(Op, DAG);
+  case ISD::VSELECT:
+    return lowerVSELECT(Op, DAG);
   }
   return SDValue();
 }
 
+SDValue LoongArchTargetLowering::lowerVSELECT(SDValue Op,
+                                              SelectionDAG &DAG) const {
+  SDValue Cond = Op.getOperand(0);
+  SDValue LHS = Op.getOperand(1);
+  SDValue RHS = Op.getOperand(2);
+  MVT VT = Op.getSimpleValueType();
+  SDLoc DL(Op);
+
+  // Try to lower vselect to vector_shuffle. All cases with constant
+  // build_vector condition will be handled. NOTE: On 32-bit platform, vselect
+  // with v2i64/v4i64 type condition will never enter this because of the extra
+  // bitcast. It can be considered separatly later.
----------------
ylzsx wrote:

separatly -> separately.

https://github.com/llvm/llvm-project/pull/163489