// the 'go get' command, not an intermediate state, and is therefore
			// deterministic and therefore cacheable, and the constraints on the
			// selected version of each module can only narrow as we iterate.
			//
			// "all" is functionally very similar to a wildcard pattern. The set of
			// packages imported by the main module does not change, and the query

			// TODO not sure if this is best done in wide precision or narrow
			// Double-rounding might be an issue.
			// Note that the pre-SSA implementation does the entire calculation
			// in wide format, so wide is compatible.
			xreal = s.newValueOrSfCall2(divop, wt, xreal, denom)
			ximag = s.newValueOrSfCall2(divop, wt, ximag, denom)

			if pt != wt { // Narrow to store back

The internal quantization on `lhs` is a quantization to `Trhs`, dynamic range,
per-batch (per-axis along axis `dimension_numbers.input_batch_dimension`), asymmetric,
and not narrow range (the range is [Trhs_MIN, Trhs_MAX]).

`lhs` and `rhs` must be Tensors of same rank, and meet following shape conditions.
- lhs_feature % feature_group_count == 0
- lhs_feature % rhs_input_feature == 0

        of simply using the `dtype` & `shape`.
    *   [Improve support](https://github.com/tensorflow/tensorflow/issues/33862)
        for dynamically-sized TensorArray inside `tf.function`.
*   `tf.math`:
    *   Narrow down `argmin`/`argmax` contract to always return the smallest
        index for ties.
    *   `tf.math.reduce_variance` and `tf.math.reduce_std` return correct