const bool symmetric) {
  const float eps = 1e-7;
  ASSERT_THAT(*float_quant_params.min(), SizeIs(1));
  ASSERT_THAT(*float_quant_params.max(), SizeIs(1));
  float float_min = std::min(0.f, float_quant_params.min()->Get(0));
  float float_max = std::max(0.f, float_quant_params.max()->Get(0));
  if (symmetric) {
    // When the symmetric case, ConvertStatsToQDQs in PrepareQuantizePass

			u.tracef("depth %d >= %d", u.depth, unificationDepthLimit)
		}
		if panicAtUnificationDepthLimit {
			panic("unification reached recursion depth limit")
		}
		return false
	}

	// Unification is symmetric, so we can swap the operands.
	// Ensure that if we have at least one
	// - defined type, make sure one is in y
	// - type parameter recorded with u, make sure one is in x

			u.tracef("depth %d >= %d", u.depth, unificationDepthLimit)
		}
		if panicAtUnificationDepthLimit {
			panic("unification reached recursion depth limit")
		}
		return false
	}

	// Unification is symmetric, so we can swap the operands.
	// Ensure that if we have at least one
	// - defined type, make sure one is in y
	// - type parameter recorded with u, make sure one is in x

  }];

  let description = [{
The input is a tensor of shape `[..., M, M]` whose inner-most 2 dimensions
form square matrices.

The input has to be symmetric and positive definite. Only the lower-triangular
part of the input will be used for this operation. The upper-triangular part
will not be read.

The output is a tensor of the same shape as the input

        `TFLiteConverter.from_saved_model`.
    *   Added DEPTH_TO_SPACE support in Post training quantization.
    *   Added dynamic range quantization support for the BatchMatMul op.
        *   Both symmetric and asymmetric quantized input tensor are supported.
    *   Add `RFFT2D` as builtin op. (`RFFT2D` also supports `RFFTD`.) Currently
        only supports float32 input.
    *   Add 5D support to `SLICE` op.