*
     * @param outputType The type to cast the input to
     * @param object The object to be cast (must not be {@code null})
     * @param <O> The type to be cast to
     * @param <I> The type of the object to be vast
     * @return The input object, cast to the output type
     */
    public static <O, I> O cast(Class<O> outputType, I object) {
        try {

     *
     * ```kt
     * if (this is String) {
     *   this.substring() // 'this' receiver is explicit, so no implicit smart-cast here.
     *
     *   smartcast() // 'this' receiver is implicit, therefore there is implicit smart-cast involved.
     * }
     * ```
     */
    public fun KtExpression.getImplicitReceiverSmartCast(): Collection<KaImplicitReceiverSmartCast> =

				},
			},
		},
	}
	for _, tc := range cases {
		for _, op := range tc.opts {
			for _, typ := range tc.costs {
				t.Run(typ.comparableLiteral+op, func(t *testing.T) {
					e := typ.comparableLiteral + op
					testCost(t, e, typ.expectedEstimatedCost, typ.expectedRuntimeCost)
				})
			}
		}
	}
}

func TestIndexOfCost(t *testing.T) {

      >();
}

OpFoldResult StorageCastOp::fold(FoldAdaptor) {
  // Matches x -> [scast -> scast] -> y, replacing the second scast with the
  // value of x if the casts invert each other.
  auto srcScastOp = getArg().getDefiningOp<StorageCastOp>();
  if (!srcScastOp || srcScastOp.getArg().getType() != getType())
    return OpFoldResult();
  return srcScastOp.getArg();
}

      >();
}

OpFoldResult StorageCastOp::fold(FoldAdaptor) {
  // Matches x -> [scast -> scast] -> y, replacing the second scast with the
  // value of x if the casts invert each other.
  auto srcScastOp = getArg().getDefiningOp<StorageCastOp>();
  if (!srcScastOp || srcScastOp.getArg().getType() != getType())
    return OpFoldResult();
  return srcScastOp.getArg();
}

//===----------------------------------------------------------------------===//
// Op quantization pass-through patterns.
//===----------------------------------------------------------------------===//
// Casts result type of $1 to a quantized type by using the quantization
// parameters from the type in $0.
class UpdateShapeWithAxis<int i> : NativeCodeCall<

  pm.addNestedPass<mlir::func::FuncOp>(
      mlir::quant::CreateReplaceCastHacksWithTFXLAOpsPass());
  pm.addNestedPass<mlir::func::FuncOp>(mlir::createCSEPass());
  // Use optimize pass to remove double casts that are inserted when inlining
  // functions.
  pm.addNestedPass<mlir::func::FuncOp>(mlir::quant::CreateOptimizePass());
}

}  // namespace quantization

  pass_manager->addPass(
      mlir::tf_saved_model::CreateOptimizeGlobalTensorsPass());

  if (pass_config.shape_inference) {
    // Add a shape inference pass to optimize away the unnecessary casts.
    pass_manager->addPass(mlir::TF::CreateTFShapeInferencePass());
  }

  // Legalize while early to allow further constant folding.
  // TODO(jpienaar): This may not actually matter as we do canonicalization

#include "tensorflow/compiler/tf2xla/tf2xla_defs.h"

namespace mlir {
namespace TF {

// TODO(b/229028654) Use definitions from tf2xla_defs.h directly. We currently
// don't do this to avoid explicit casts (implicit conversion from
// `absl::string_view` to `llvm::StringRef` is not supported until C++17).

// Whether soft placement is allowed. If true, the marked node is eligible for
// outside compilation.

            IsCompatibleTypeWithTFLCastOp(type)) {
          auto cast = b.create<CastOp>(yield_op->getLoc(), type, value);
          args.push_back(cast);
        } else {
          auto cast = b.create<TF::CastOp>(yield_op->getLoc(), type, value);
          args.push_back(cast);
        }
      }
    }
    args.append(new_args.begin(), new_args.end());
  } else {