*   Add delegate support for `QUANTIZED_16BIT_LSTM`.
    *   Converts hardswish subgraphs into atomic ops.
*   Add support for defaulting the value of `cycle_length` argument of
    `tf.data.Dataset.interleave` to the number of schedulable CPU cores.
*   `parallel_for`: Add converter for `MatrixDiag`.
*   Add `narrow_range` attribute to `QuantizeAndDequantizeV2` and V3.
*   Added new op: `tf.strings.unsorted_segment_join`.

0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2, } func blockGeneric(dig *Digest, p []byte) { var w [64]uint32 h0, h1, h2, h3, h4, h5, h6, h7 := dig.h[0], dig.h[1], dig.h[2], dig.h[3], dig.h[4], dig.h[5], dig.h[6], dig.h[7] for len(p) >= chunk { // Can interlace the computation of w with the // rounds below if needed for speed. for i := 0; i < 16; i++ { j := i * 4 w[i] = uint32(p[j])<<24 | uint32(p[j+1])<<16 | uint32(p[j+2])<<8 | uint32(p[j+3]) } for i := 16; i < 64; i++ { v1 := w[i-2] t1 := (bits.RotateLeft32(v1,...