rg/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2004-hmv * Cost: 4M + 4S + 1*half + 5add + 2*2 + 1*3. * Source: 2004 Hankerson–Menezes–Vanstone, page 91. * A = 3(X₁-Z₁²)×(X₁+Z₁²) * B = 2Y₁ * Z₃ = B×Z₁ * C = B² * D = C×X₁ * X₃ = A²-2D * Y₃ = (D-X₃)×A-C²/2 * * Three-operand formula: * T1 = Z1² * T2 = X1-T1 * T1 = X1+T1 * T2 = T2*T1 * T2 = 3*T2 * Y3 = 2*Y1 * Z3 = Y3*Z1 * Y3 = Y3² * T3 = Y3*X1 * Y3 = Y3² * Y3 = half*Y3 * X3 = T2² * T1 = 2*T3 * X3 = X3-T1 * T1 = T3-X3 * T1 = T1*T2 * Y3 = T1-Y3 */...

*   Move skflow to `contrib/learn`
*   Better linear optimizer in `contrib/linear_optimizer`
*   Random forest implementation in `contrib/tensor_forest`
*   CTC loss and decoders in `contrib/ctc`
*   Basic support for `half` data type
*   Better support for loading user ops (see examples in `contrib/`)
*   Allow use of (non-blocking) Eigen threadpool with
    `TENSORFLOW_USE_EIGEN_THREADPOOL` define