[mlir][Vector] Fold consecutive bitcast.

[jacquesguan]

Summary

This patch supports to fold consecutive bitcast into one bitcast.

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

 }
 
 OpFoldResult BitCastOp::fold(ArrayRef<Attribute> operands) {
   // Nop cast.
   if (getSource().getType() == getResult().getType())
     return getSource();
 
   // Canceling bitcasts.
-  if (auto otherOp = getSource().getDefiningOp<BitCastOp>())
+  if (auto otherOp = getSource().getDefiningOp<BitCastOp>()) {
     if (getResult().getType() == otherOp.getSource().getType())
       return otherOp.getSource();
 
+    setOperand(otherOp.getSource());
+    return getResult();
+  }
+
   Attribute sourceConstant = operands.front();
   if (!sourceConstant)
     return {};
 
   Type srcElemType = getSourceVectorType().getElementType();
   Type dstElemType = getResultVectorType().getElementType();
 
   if (auto floatPack = sourceConstant.dyn_cast<DenseFPElementsAttr>()) {

mlir/test/Dialect/Vector/canonicalize.mlir

 // -----
 
 // CHECK-LABEL: func @dont_reduce_one_element_vector
 //       CHECK: vector.reduction
 func.func @dont_reduce_one_element_vector(%a : vector<4xf32>) -> f32 {
   %s = vector.reduction <add>, %a : vector<4xf32> into f32
   return %s : f32
 }
+
+// -----
+
+// CHECK-LABEL: func @bitcast(
+//  CHECK-SAME:               %[[ARG:.*]]: vector<4x8xf32>) -> vector<4x16xi16> {
+//       CHECK: vector.bitcast %[[ARG:.*]] : vector<4x8xf32> to vector<4x16xi16>
+func.func @bitcast(%a: vector<4x8xf32>) -> vector<4x16xi16> {
+  %0 = vector.bitcast %a : vector<4x8xf32> to vector<4x8xi32>
+  %1 = vector.bitcast %0 : vector<4x8xi32> to vector<4x16xi16>
+  return %1 : vector<4x16xi16>
+}