#define r(off)	  (32*11 + off)(SP)
#define hsqr(off) (32*12 + off)(SP)
#define rsqr(off) (32*13 + off)(SP)
#define hcub(off) (32*14 + off)(SP)
#define rptr	  (32*15)(SP)
#define sel_save  (32*15 + 8)(SP)
#define zero_save (32*15 + 8 + 4)(SP)

// func p256PointAddAffineAsm(res, in1 *P256Point, in2 *p256AffinePoint, sign, sel, zero int)
TEXT ·p256PointAddAffineAsm(SB),0,$512-48
	// Move input to stack in order to free registers

  // CHECK-NEXT: %[[SIZE:.*]] = "tf.MlirLocalVarOp"() : () -> tensor<!tf_type.resource<tensor<1xi32>>>
  // CHECK-NEXT: %[[ZERO_SIZE:.*]] = "tf.Const"() <{value = dense<0> : tensor<1xi32>}> : () -> tensor<1xi32>
  // CHECK-NEXT: "tf.AssignVariableOp"(%[[SIZE]], %[[ZERO_SIZE]]) : (tensor<!tf_type.resource<tensor<1xi32>>>, tensor<1xi32>) -> ()

    *   Triangular solve performance improvements.
    *   Unify RNN cell interface between TF and Keras. Add new
        get_initial_state() to Keras and TF RNN cell, which will use to replace
        the existing zero_state() method.
    *   Update initialization of variables in Keras.
    *   Updates to "constrained_optimization" in tensorflow/contrib.
    *   boosted trees: adding pruning mode.