public int inDegree(N node) {
      return delegate().outDegree(node); // transpose
    }

    @Override
    public int outDegree(N node) {
      return delegate().inDegree(node); // transpose
    }

    @Override
    public boolean hasEdgeConnecting(N nodeU, N nodeV) {
      return delegate().hasEdgeConnecting(nodeV, nodeU); // transpose
    }

    @Override

      input.getLoc(), input, rewriter.getDenseI64ArrayAttr(permutation));
}

// Defers the transpose of the left-hand side (LHS) to the right-hand side and
// the result of a binary operation. In detail, this rewrites the
// `op(transpose(%rhs), %lhs)` to `transpose(op(%rhs, transpose(%lhs)))`. The
// LHS transpose permutation must be a NCHW->NHWC permutation.
template <typename OpT>

  // CHECK-DAG: %[[cst_0:.*]] = arith.constant dense<[0, 2, 1]> : tensor<3xi32>
  // CHECK: %[[v0:.*]] = "tf.Transpose"(%arg0, %[[cst]]) : (tensor<2x5x7xf32>, tensor<3xi32>) -> tensor<5x7x2xf32>
  // CHECK: %[[v1:.*]] = "tf.Transpose"(%arg1, %[[cst_0]]) : (tensor<5x3x2xf32>, tensor<3xi32>) -> tensor<5x2x3xf32>

      // out only isn't supported
      return std::nullopt;
    }
  }
  return dnums;
}

// Function to replace a unary einsum op, that can undergo simple transpose, to
// an explicit transpose op.
LogicalResult rewriteToReduceSumAndTranspose(TF::EinsumOp op,
                                             EinsumDimensionNumbers dnums,

// CHECK-SAME:                            %[[VAL_0:.*]]: tensor<1x3x6x6xf32>) -> tensor<*xf32> {
// CHECK:           %[[VAL_1:.*]] = arith.constant dense<[0, 2, 3, 1]> : tensor<4xi32>
// CHECK:           %[[VAL_2:.*]] = "tfl.transpose"(%[[VAL_0]], %[[VAL_1]]) : (tensor<1x3x6x6xf32>, tensor<4xi32>) -> tensor<1x6x6x3xf32>
// CHECK:           %[[VAL_3:.*]] = arith.constant dense<0> : tensor<4x2xi32>

   * instance, {@code rowKeySet().iterator()} supports {@code remove()} but {@code
   * columnKeySet().iterator()} doesn't. With a transposed {@link HashBasedTable}, it's the other
   * way around.
   */
  public static <R extends @Nullable Object, C extends @Nullable Object, V extends @Nullable Object>
      Table<C, R, V> transpose(Table<R, C, V> table) {
    return (table instanceof TransposeTable)
        ? ((TransposeTable<R, C, V>) table).original

func.func @main(%arg0: tensor<5x7xf32>) -> tensor<5x7xf32> {
  func.return %arg0: tensor<5x7xf32>
// CHECK-LABEL: main
// CHECK: return %arg0 : tensor<5x7xf32>
}

// - transpose
//
func.func @transpose_2d(%arg0: tensor<2x3xf32>) -> tensor<3x2xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[1, 0]> : tensor<2xi64>}> : (tensor<2x3xf32>) -> tensor<3x2xf32>
  func.return %0 : tensor<3x2xf32>

// CHECK-LABEL:   transpose_2d

    reserve_spaces[1] = op.input(4);  // pop_var
    if (data_format == "NCHW") {
      x = Transpose(scope, x, {0, 2, 3, 1});
      grad_y = Transpose(scope, grad_y, {0, 2, 3, 1});
    } else if (data_format == "NCDHW") {
      x = Transpose(scope, x, {0, 2, 3, 4, 1});
      grad_y = Transpose(scope, grad_y, {0, 2, 3, 4, 1});
    }

    StringPiece target_data_format;

   * instance, {@code rowKeySet().iterator()} supports {@code remove()} but {@code
   * columnKeySet().iterator()} doesn't. With a transposed {@link HashBasedTable}, it's the other
   * way around.
   */
  public static <R extends @Nullable Object, C extends @Nullable Object, V extends @Nullable Object>
      Table<C, R, V> transpose(Table<R, C, V> table) {
    return (table instanceof TransposeTable)
        ? ((TransposeTable<R, C, V>) table).original

  auto reshaped = rewriter.create<mhlo::DynamicReshapeOp>(
      loc, result_type, matmul.getResult(), result_shape);
  return reshaped.getResult();
}

// Converts mhlo.dot_general to tfl.BatchMatMul. Reshape and Transpose ops will
// be inserted when necessary. See ConvertDotGeneralOp for additional notes.
Value ConvertDotOp(PatternRewriter& rewriter, Operation* old_op) {
  auto dot_op = cast<mhlo::DotOp>(old_op);