// double for the jacobian to capture the higher precision
  // between X_T and Y_T.
  TestResize<double, float, double>(RESIZE_BILINEAR);
}

TEST_F(ImageGradTest, TestBicubic) {
  TestResizedShape(RESIZE_BICUBIC);
  TestResize<float, float, float>(RESIZE_BICUBIC);
  // Note that Y_T is always float for this op. We choose
  // double for the jacobian to capture the higher precision
  // between X_T and Y_T.

	} else {
		s.scanNum()
	}
	return true
}

// parsePrecision scans for a precision. It returns false if there's a bad index expression.
func (s *formatState) parsePrecision() bool {
	// If there's a period, there may be a precision.
	if s.nbytes < len(s.format) && s.format[s.nbytes] == '.' {
		s.flags = append(s.flags, '.') // Treat precision as a flag.
		s.nbytes++
		if !s.parseIndex() {
			return false
		}

					(len(matches2) == 1 && matches2[0][1] != fmt.Sprint(length) && ok) {
					alterColumn = true
				}
			}
		}

		// check precision
		if precision, _, ok := columnType.DecimalSize(); ok && int64(field.Precision) != precision {
			if regexp.MustCompile(fmt.Sprintf("[^0-9]%d[^0-9]", field.Precision)).MatchString(m.DataTypeOf(field)) {
				alterColumn = true
			}
		}
	}

	// check nullable

      lhs_batching_dimensions = [0],
      lhs_contracting_dimensions = [1, 2],
      rhs_batching_dimensions = [0],
      rhs_contracting_dimensions = [1, 3]
    >,
    precision_config = [#mhlo<precision DEFAULT>, #mhlo<precision DEFAULT>]
  } : (tensor<3x2x6x5x1xf32>, tensor<3x2x4x6xf32>) -> tensor<3x5x1x4xf32>
  func.return %0 : tensor<3x5x1x4xf32>

// CHECK-LABEL:   convert_dot_general

	struct { A int; B float }	// change of type for B
	struct { }			// no field names in common
	struct { C, D int }		// no field names in common

Integers are transmitted two ways: arbitrary precision signed integers or
arbitrary precision unsigned integers. There is no int8, int16 etc.
discrimination in the gob format; there are only signed and unsigned integers. As
described below, the transmitter sends the value in a variable-length encoding;

//   - asInteger: returns a representation of the current value as an int64 if
//     possible or results in an error if conversion would result in overflow
//	   or loss of precision.
//
//   - asApproximateFloat: returns a float64 representation of the quantity which may
//     lose precision. If the value of the quantity is outside the range of a float64
//     +Inf/-Inf will be returned.
//
//     <Quantity>.isInteger() <bool>

    // CHECK-NOT: tf.CustomAggregator
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<?x100352xf32>, tensor<100352x10xf32>) -> tensor<?x10xf32>
    return %0 : tensor<?x10xf32>
  }
}

// -----

    %cst = stablehlo.constant dense<0.000000e+00> : tensor<10x1x3xf32>
    %0 = stablehlo.dot_general %arg0, %arg1, batching_dims = [0] x [0], contracting_dims = [2] x [1], precision = [DEFAULT, DEFAULT] {mhlo.frontend_attributes = {grad_x = "false", grad_y = "false"}} : (tensor<10x1x1024xf32>, tensor<10x1024x3xf32>) -> tensor<10x1x3xf32>
    %1 = stablehlo.maximum %0, %cst : tensor<10x1x3xf32>

     * Computes the quantile value of the given dataset.
     *
     * @param dataset the dataset to do the calculation on, which must be non-empty, which will be
     *     cast to doubles (with any associated lost of precision), and which will not be mutated by
     *     this call (it is copied instead)
     * @return the quantile value
     */
    public double compute(Collection<? extends Number> dataset) {

        return new DependencyResolvers(variantResolver, componentResolver);
    }

    /**
     * Adapts a {@link ComponentDependencyResolver} to a {@link VariantDependencyResolver}
     * by returning component-precision coordinates.
     */
    private static class VariantResolverAdapter implements VariantDependencyResolver {

        private final ComponentDependencyResolver delegate;