/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <emmintrin.h>

#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
#include "aom_ports/mem.h"

void aom_highbd_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t count,
                                const int16_t *zbin_ptr,
                                const int16_t *round_ptr,
                                const int16_t *quant_ptr,
                                const int16_t *quant_shift_ptr,
                                tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                                const int16_t *dequant_ptr, uint16_t *eob_ptr,
                                const int16_t *scan, const int16_t *iscan) {
  int i, j, non_zero_regs = (int)count / 4, eob_i = -1;
  __m128i zbins[2];
  __m128i nzbins[2];

  zbins[0] = _mm_set_epi32((int)zbin_ptr[1], (int)zbin_ptr[1], (int)zbin_ptr[1],
                           (int)zbin_ptr[0]);
  zbins[1] = _mm_set1_epi32((int)zbin_ptr[1]);

  nzbins[0] = _mm_setzero_si128();
  nzbins[1] = _mm_setzero_si128();
  nzbins[0] = _mm_sub_epi32(nzbins[0], zbins[0]);
  nzbins[1] = _mm_sub_epi32(nzbins[1], zbins[1]);

  (void)scan;

  memset(qcoeff_ptr, 0, count * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, count * sizeof(*dqcoeff_ptr));

  // Pre-scan pass
  for (i = ((int)count / 4) - 1; i >= 0; i--) {
    __m128i coeffs, cmp1, cmp2;
    int test;
    coeffs = _mm_load_si128((const __m128i *)(coeff_ptr + i * 4));
    cmp1 = _mm_cmplt_epi32(coeffs, zbins[i != 0]);
    cmp2 = _mm_cmpgt_epi32(coeffs, nzbins[i != 0]);
    cmp1 = _mm_and_si128(cmp1, cmp2);
    test = _mm_movemask_epi8(cmp1);
    if (test == 0xffff)
      non_zero_regs--;
    else
      break;
  }

  // Quantization pass:
  for (i = 0; i < non_zero_regs; i++) {
    __m128i coeffs, coeffs_sign, tmp1, tmp2;
    int test;
    int abs_coeff[4];
    int coeff_sign[4];

    coeffs = _mm_load_si128((const __m128i *)(coeff_ptr + i * 4));
    coeffs_sign = _mm_srai_epi32(coeffs, 31);
    coeffs = _mm_sub_epi32(_mm_xor_si128(coeffs, coeffs_sign), coeffs_sign);
    tmp1 = _mm_cmpgt_epi32(coeffs, zbins[i != 0]);
    tmp2 = _mm_cmpeq_epi32(coeffs, zbins[i != 0]);
    tmp1 = _mm_or_si128(tmp1, tmp2);
    test = _mm_movemask_epi8(tmp1);
    _mm_storeu_si128((__m128i *)abs_coeff, coeffs);
    _mm_storeu_si128((__m128i *)coeff_sign, coeffs_sign);

    for (j = 0; j < 4; j++) {
      if (test & (1 << (4 * j))) {
        int k = 4 * i + j;
        const int64_t tmp3 = abs_coeff[j] + round_ptr[k != 0];
        const int64_t tmp4 = ((tmp3 * quant_ptr[k != 0]) >> 16) + tmp3;
        const uint32_t abs_qcoeff =
            (uint32_t)((tmp4 * quant_shift_ptr[k != 0]) >> 16);
        qcoeff_ptr[k] =
            (int)(abs_qcoeff ^ (uint32_t)coeff_sign[j]) - coeff_sign[j];
        dqcoeff_ptr[k] = qcoeff_ptr[k] * dequant_ptr[k != 0];
        if (abs_qcoeff) eob_i = iscan[k] > eob_i ? iscan[k] : eob_i;
      }
    }
  }
  *eob_ptr = eob_i + 1;
}

void aom_highbd_quantize_b_32x32_sse2(
    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
    const int16_t *round_ptr, const int16_t *quant_ptr,
    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
    const int16_t *scan, const int16_t *iscan) {
  __m128i zbins[2];
  __m128i nzbins[2];
  int idx = 0;
  int idx_arr[1024];
  int i, eob = -1;
  const int zbin0_tmp = ROUND_POWER_OF_TWO(zbin_ptr[0], 1);
  const int zbin1_tmp = ROUND_POWER_OF_TWO(zbin_ptr[1], 1);
  (void)scan;
  zbins[0] = _mm_set_epi32(zbin1_tmp, zbin1_tmp, zbin1_tmp, zbin0_tmp);
  zbins[1] = _mm_set1_epi32(zbin1_tmp);

  nzbins[0] = _mm_setzero_si128();
  nzbins[1] = _mm_setzero_si128();
  nzbins[0] = _mm_sub_epi32(nzbins[0], zbins[0]);
  nzbins[1] = _mm_sub_epi32(nzbins[1], zbins[1]);

  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));

  // Pre-scan pass
  for (i = 0; i < n_coeffs / 4; i++) {
    __m128i coeffs, cmp1, cmp2;
    int test;
    coeffs = _mm_load_si128((const __m128i *)(coeff_ptr + i * 4));
    cmp1 = _mm_cmplt_epi32(coeffs, zbins[i != 0]);
    cmp2 = _mm_cmpgt_epi32(coeffs, nzbins[i != 0]);
    cmp1 = _mm_and_si128(cmp1, cmp2);
    test = _mm_movemask_epi8(cmp1);
    if (!(test & 0xf)) idx_arr[idx++] = i * 4;
    if (!(test & 0xf0)) idx_arr[idx++] = i * 4 + 1;
    if (!(test & 0xf00)) idx_arr[idx++] = i * 4 + 2;
    if (!(test & 0xf000)) idx_arr[idx++] = i * 4 + 3;
  }

  // Quantization pass: only process the coefficients selected in
  // pre-scan pass. Note: idx can be zero.
  for (i = 0; i < idx; i++) {
    const int rc = idx_arr[i];
    const int coeff = coeff_ptr[rc];
    const int coeff_sign = AOMSIGN(coeff);
    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
    const int64_t tmp1 = abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
    const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
    const uint32_t abs_qcoeff =
        (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 15);
    qcoeff_ptr[rc] = (int)(abs_qcoeff ^ (uint32_t)coeff_sign) - coeff_sign;
    dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
    if (abs_qcoeff) eob = iscan[idx_arr[i]] > eob ? iscan[idx_arr[i]] : eob;
  }
  *eob_ptr = eob + 1;
}

void aom_highbd_quantize_b_64x64_sse2(
    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
    const int16_t *round_ptr, const int16_t *quant_ptr,
    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
    const int16_t *scan, const int16_t *iscan) {
  __m128i zbins[2];
  __m128i nzbins[2];
  int idx = 0;
  int idx_arr[1024];
  int i, eob = -1;
  const int zbin0_tmp = ROUND_POWER_OF_TWO(zbin_ptr[0], 2);
  const int zbin1_tmp = ROUND_POWER_OF_TWO(zbin_ptr[1], 2);
  (void)scan;
  zbins[0] = _mm_set_epi32(zbin1_tmp, zbin1_tmp, zbin1_tmp, zbin0_tmp);
  zbins[1] = _mm_set1_epi32(zbin1_tmp);

  nzbins[0] = _mm_setzero_si128();
  nzbins[1] = _mm_setzero_si128();
  nzbins[0] = _mm_sub_epi32(nzbins[0], zbins[0]);
  nzbins[1] = _mm_sub_epi32(nzbins[1], zbins[1]);

  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));

  // Pre-scan pass
  for (i = 0; i < n_coeffs / 4; i++) {
    __m128i coeffs, cmp1, cmp2;
    int test;
    coeffs = _mm_load_si128((const __m128i *)(coeff_ptr + i * 4));
    cmp1 = _mm_cmplt_epi32(coeffs, zbins[i != 0]);
    cmp2 = _mm_cmpgt_epi32(coeffs, nzbins[i != 0]);
    cmp1 = _mm_and_si128(cmp1, cmp2);
    test = _mm_movemask_epi8(cmp1);
    if (!(test & 0xf)) idx_arr[idx++] = i * 4;
    if (!(test & 0xf0)) idx_arr[idx++] = i * 4 + 1;
    if (!(test & 0xf00)) idx_arr[idx++] = i * 4 + 2;
    if (!(test & 0xf000)) idx_arr[idx++] = i * 4 + 3;
  }

  // Quantization pass: only process the coefficients selected in
  // pre-scan pass. Note: idx can be zero.
  for (i = 0; i < idx; i++) {
    const int rc = idx_arr[i];
    const int coeff = coeff_ptr[rc];
    const int coeff_sign = AOMSIGN(coeff);
    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
    const int64_t tmp1 = abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 2);
    const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
    const uint32_t abs_qcoeff =
        (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 14);
    qcoeff_ptr[rc] = (int)(abs_qcoeff ^ (uint32_t)coeff_sign) - coeff_sign;
    dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 4;
    if (abs_qcoeff) eob = iscan[idx_arr[i]] > eob ? iscan[idx_arr[i]] : eob;
  }
  *eob_ptr = eob + 1;
}