// as inputs to TPU computations. If more than one dimension is sharded, then
// a tree of connected split ops are added before tf_device.parallel_execute op.
mlir::LogicalResult HandleTileShardedInputs(
    const mlir::Location& location, const xla::OpSharding& input_sharding,
    const mlir::Value& original_source, mlir::OpBuilder* builder,
    llvm::SmallVectorImpl<mlir::Value>* tiled_inputs) {