/*asymmetric_quantize_inputs=*/mlir::BoolAttr());
    rewriter.replaceOp(bmm_op, {fc_op.getResult(0)});

    return success();
  };
};

// Converts batch_matmul operation with a ones tensor to a reduce_sum.
struct ConvertBatchMatMulOpToReduceSum
    : public OpRewritePattern<TFL::BatchMatMulOp> {
  using OpRewritePattern<TFL::BatchMatMulOp>::OpRewritePattern;
  LogicalResult matchAndRewrite(TFL::BatchMatMulOp bmm_op,

                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    /* Given upstream grad U and a BiasAdd: A + bias, the gradients are:
     *
     *    dA = U
     *    dbias = reduceSum(U, dims = channel_dim)
     */

    AbstractTensorHandle* upstream_grad = grad_outputs[0];
    DCHECK(upstream_grad);

    // Recover data format from forward pass for gradient.
    std::string data_format;