analyzing input pipeline performance with TF Profiler.

*   [`tf.distribute.TPUStrategy`](https://www.tensorflow.org/api_docs/python/tf/distribute/TPUStrategy)
    is now a stable API and no longer considered experimental for TensorFlow.
    (earlier `tf.distribute.experimental.TPUStrategy`).

*   [TF Profiler](https://www.tensorflow.org/guide/profiler) introduces two new
    tools: a memory profiler to visualize your model’s memory usage over time

{ panic("ecdsa: reseed interval exceeded") } tlen := 0 for tlen < len(out) { // V = HMAC_K(V) // T = T || V d.hK.Reset() d.hK.Write(d.V) d.V = d.hK.Sum(d.V[:0]) tlen += copy(out[tlen:], d.V) } // Note that if this function shows up on ECDSA-level profiles, this can be // optimized in the common case by deferring the rest to the next Generate // call, which will never come in nearly all cases. // HMAC_DRBG_Update, per Section 10.1.2.2, without provided_data. // K = HMAC (K, V || 0x00) d.hK.Reset()...