/*
  This file is part of ct.

  This program is free software: you can redistribute it and/or modify
  it under the terms of the GNU General Public License as published by
  the Free Software Foundation, either version 3 of the License, or
  (at your option) any later version.

  This program is distributed in the hope that it will be useful, but
  WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with ct. If not, see <https://www.gnu.org/licenses/>.
*/

#include "nn.h"
#include "tak.h"
#include "weights_5.h"
#include "weights_6.h"

// ===================================================================
// Helpers
// ===================================================================

#define RELU(x) ((x) = ((x) < 0) ? 0 : (x))

static inline void prepare_input(float *cur_board, tak_state_p stat);

// ===================================================================
// Implementation of small neural networks
// ===================================================================

float nn1986_evaluate_black_win(tak_state_p state) {
  static float cur_board[FIVE_INP_NUM];
  static float dense1[FIVE_DENSE_NUM];
  static float output[2];

  prepare_input(cur_board, state);
  /* ------------------ *
   *  First dense layer *
   * ------------------ */
  for (unsigned int d1 = 0; d1 < FIVE_DENSE_NUM; d1++) {
    dense1[d1] = five_dense_biases[d1];
    for (unsigned int fl = 0; fl < FIVE_INP_NUM; fl++) {
      dense1[d1] += cur_board[fl] * five_dense_weights[d1][fl];
    }
    RELU(dense1[d1]);
  }
  /* ------------- *
   *  Output layer *
   * ------------- */
  for (uint8_t k = 0; k < 2; k++) {
    output[k] = five_output_bias[k];
    for (unsigned int d2 = 0; d2 < FIVE_DENSE_NUM; d2++) {
      output[k] += dense1[d2] * five_output_weights[k][d2];
    }
    RELU(output[k]);
  }
  const float norm = output[0] + output[1];
  return (2 * output[0] / norm) - 1;
}

float nn2690_evaluate_black_win(tak_state_p state) {
  static float cur_board[SIX_INP_NUM];
  static float dense1[SIX_DENSE_NUM];
  static float output[2];

  prepare_input(cur_board, state);
  /* ------------------ *
   *  First dense layer *
   * ------------------ */
  for (unsigned int d1 = 0; d1 < SIX_DENSE_NUM; d1++) {
    dense1[d1] = six_dense_biases[d1];
    for (unsigned int fl = 0; fl < SIX_INP_NUM; fl++) {
      dense1[d1] += cur_board[fl] * six_dense_weights[d1][fl];
    }
    RELU(dense1[d1]);
  }
  /* ------------- *
   *  Output layer *
   * ------------- */
  for (uint8_t k = 0; k < 2; k++) {
    output[k] = six_output_bias[k];
    for (unsigned int d2 = 0; d2 < SIX_DENSE_NUM; d2++) {
      output[k] += dense1[d2] * six_output_weights[k][d2];
    }
    RELU(output[k]);
  }
  const float norm = output[0] + output[1];
  return (2 * output[0] / norm) - 1;
}

// ===================================================================
// Helper implementation
// ===================================================================

static inline void prepare_input(float *cur_board, tak_state_p state) {
  /* --------------- *
   *  Populate input *
   * --------------- */
  for (unsigned int y = 0; y < state->board_size; y++) {
    for (unsigned int x = 0; x < state->board_size; x++) {
      const unsigned int loc = x + y * state->board_size;
      const unsigned int count = COUNT_AT(state, loc);
      float lookup = 0;
      if (count > 0) {
        if (STONE_AT(state, loc) == STONE_STANDING) {
          lookup = (state->colours[loc] & 1) ? +0.25 : -0.25;
        } else if (STONE_AT(state, loc) == STONE_CAPSTONE) {
          lookup = (state->colours[loc] & 1) ? +1.00 : -1.00;
        } else {
          lookup = (state->colours[loc] & 1) ? +0.50 : -0.50;
        }
      }
      cur_board[3 + loc] = lookup;
    }
  }
  // Add input of flat counts and ply parity
  cur_board[0] = (state->ply & 1) ? 1 : -1;
  cur_board[1] = (float)(state->white_count & 127) / 21.0;
  cur_board[2] = (float)(state->black_count & 127) / 21.0;
}