elements are shifted positively (towards larger indices) along the dimension
specified by `axis[i]`. Negative shifts will roll the elements in the opposite
direction.}]>:$shift,
    Arg<TF_I32OrI64Tensor, [{Dimension must be 0-D or 1-D. `axis[i]` specifies the dimension that the shift
`shift[i]` should occur. If the same axis is referenced more than once, the

		// importing fmt or strconv
		// - try to use FixedZone, which would also require escaping the name
		// and would represent e.g. "America/Los_Angeles" daylight saving time
		// shifts inaccurately
		// - use the pointer format, which is no worse than you'd get with the
		// old fmt.Sprintf("%#v", t) format.
		//
		// Of these, Location(loc.name) is the least disruptive. This is an edge

      // offset = d<=0 ? max_diag_len - diag_len_d : 0
      cmp = rewriter.create<TF::LessEqualOp>(loc, d, b_zero);
    } else {
      // offset = 0
      cmp = b_false;
    }

    // This offset shifts the diagonals to the "left" or "right", depending
    // on alignment.
    Value offset = rewriter.create<SelectOp>(
        loc, b_zero.getType(), cmp,

    types. Examples of quantized types are TF_Qint8 or TF_Quint8.

    An example is TF_DequantizeOp, which converts a quantized type to a float.
    This op is rewritten to generic ops that perform the scale and shift
    and can operate on non-quantized types.

    Currently, TF_DequantizeOp is the only op with a lowering that falls
    in this category. When more lowerings are added (e.g. QuantizeV2Op),

	etype1 := s.concreteEtype(t)
	etype2 := s.concreteEtype(u)
	x, ok := shiftOpToSSA[opAndTwoTypes{op, etype1, etype2}]
	if !ok {
		s.Fatalf("unhandled shift op %v etype=%s/%s", op, etype1, etype2)
	}
	return x
}

func (s *state) uintptrConstant(v uint64) *ssa.Value {
	if s.config.PtrSize == 4 {

*   Update gRPC release to 0.14.
*   Eigen version upgrade.
*   Switch to eigen thread pool
*   `tf.nn.moments()` now accepts a `shift` argument. Shifting by a good
    estimate of the mean improves numerical stability. Also changes the behavior
    of the `shift` argument to `tf.nn.sufficient_statistics()`.
*   Performance improvements
*   Many bugfixes
*   Many documentation fixes