and `tf.keras.layers.LocallyConnected1D`. The new mode
        (`implementation=2`) performs forward pass as a single dense matrix
        multiplication, allowing dramatic speedups in certain scenarios (but
        worse performance in others - see docstring). The option also allows to
        use `padding=same`.
    *   Add documentation clarifying the differences between tf.fill and
        tf.constant.

2FB9          ; mapped                 ; 9999          # 3.0  KANGXI RADICAL FRAGRANT
2FBA          ; mapped                 ; 99AC          # 3.0  KANGXI RADICAL HORSE
2FBB          ; mapped                 ; 9AA8          # 3.0  KANGXI RADICAL BONE
2FBC          ; mapped                 ; 9AD8          # 3.0  KANGXI RADICAL TALL

186-5, Section 6.4.1, point 2 and Section 6.4.2, point 3. func hashToNat[P Point[P]](c *Curve[P], e *bigmod.Nat, hash []byte) { // ECDSA asks us to take the left-most log2(N) bits of hash, and use them as // an integer modulo N. This is the absolute worst of all worlds: we still // have to reduce, because the result might still overflow N, but to take // the left-most bits for P-521 we have to do a right shift. if size := c.N.Size(); len(hash) >= size { hash = hash[:size] if excess := len(hash)*8 -...

186-5, Section 6.4.1, point 2 and Section 6.4.2, point 3. func hashToNat[P Point[P]](c *Curve[P], e *bigmod.Nat, hash []byte) { // ECDSA asks us to take the left-most log2(N) bits of hash, and use them as // an integer modulo N. This is the absolute worst of all worlds: we still // have to reduce, because the result might still overflow N, but to take // the left-most bits for P-521 we have to do a right shift. if size := c.N.Size(); len(hash) >= size { hash = hash[:size] if excess := len(hash)*8 -...