Quellcodebibliothek Statistik Leitseite products/Sources/formale Sprachen/C/Firefox/third_party/aom/av1/encoder/x86/   (Browser von der Mozilla Stiftung Version 136.0.1©)  Datei vom 10.2.2025 mit Größe 11 kB image not shown  

Impressum av1_quantize_sse2.c   Sprache: C

 
/*
 * 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 <xmmintrin.h>

#include "config/av1_rtcd.h"

#include "aom/aom_integer.h"
#include "aom_dsp/x86/quantize_x86.h"

static inline void read_coeff(const tran_low_t *coeff, intptr_t offset,
                              __m128i *c0, __m128i *c1) {
  const tran_low_t *addr = coeff + offset;
  if (sizeof(tran_low_t) == 4) {
    const __m128i x0 = _mm_load_si128((const __m128i *)addr);
    const __m128i x1 = _mm_load_si128((const __m128i *)addr + 1);
    const __m128i x2 = _mm_load_si128((const __m128i *)addr + 2);
    const __m128i x3 = _mm_load_si128((const __m128i *)addr + 3);
    *c0 = _mm_packs_epi32(x0, x1);
    *c1 = _mm_packs_epi32(x2, x3);
  } else {
    *c0 = _mm_load_si128((const __m128i *)addr);
    *c1 = _mm_load_si128((const __m128i *)addr + 1);
  }
}

static inline void write_qcoeff(const __m128i *qc0, const __m128i *qc1,
                                tran_low_t *qcoeff, intptr_t offset) {
  tran_low_t *addr = qcoeff + offset;
  if (sizeof(tran_low_t) == 4) {
    const __m128i zero = _mm_setzero_si128();
    __m128i sign_bits = _mm_cmplt_epi16(*qc0, zero);
    __m128i y0 = _mm_unpacklo_epi16(*qc0, sign_bits);
    __m128i y1 = _mm_unpackhi_epi16(*qc0, sign_bits);
    _mm_store_si128((__m128i *)addr, y0);
    _mm_store_si128((__m128i *)addr + 1, y1);

    sign_bits = _mm_cmplt_epi16(*qc1, zero);
    y0 = _mm_unpacklo_epi16(*qc1, sign_bits);
    y1 = _mm_unpackhi_epi16(*qc1, sign_bits);
    _mm_store_si128((__m128i *)addr + 2, y0);
    _mm_store_si128((__m128i *)addr + 3, y1);
  } else {
    _mm_store_si128((__m128i *)addr, *qc0);
    _mm_store_si128((__m128i *)addr + 1, *qc1);
  }
}

static inline void write_zero(tran_low_t *qcoeff, intptr_t offset) {
  const __m128i zero = _mm_setzero_si128();
  tran_low_t *addr = qcoeff + offset;
  if (sizeof(tran_low_t) == 4) {
    _mm_store_si128((__m128i *)addr, zero);
    _mm_store_si128((__m128i *)addr + 1, zero);
    _mm_store_si128((__m128i *)addr + 2, zero);
    _mm_store_si128((__m128i *)addr + 3, zero);
  } else {
    _mm_store_si128((__m128i *)addr, zero);
    _mm_store_si128((__m128i *)addr + 1, zero);
  }
}

static inline void quantize(const int16_t *iscan_ptr,
                            const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                            tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                            const __m128i *round0, const __m128i *round1,
                            const __m128i *quant0, const __m128i *quant1,
                            const __m128i *dequant0, const __m128i *dequant1,
                            const __m128i *thr0, const __m128i *thr1,
                            __m128i *eob) {
  __m128i coeff0, coeff1;
  // Do DC and first 15 AC
  read_coeff(coeff_ptr, n_coeffs, &coeff0, &coeff1);

  // Poor man's sign extract
  const __m128i coeff0_sign = _mm_srai_epi16(coeff0, 15);
  const __m128i coeff1_sign = _mm_srai_epi16(coeff1, 15);
  __m128i qcoeff0 = _mm_xor_si128(coeff0, coeff0_sign);
  __m128i qcoeff1 = _mm_xor_si128(coeff1, coeff1_sign);
  qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
  qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
  const __m128i mask0 = _mm_or_si128(_mm_cmpgt_epi16(qcoeff0, *thr0),
                                     _mm_cmpeq_epi16(qcoeff0, *thr0));
  const __m128i mask1 = _mm_or_si128(_mm_cmpgt_epi16(qcoeff1, *thr1),
                                     _mm_cmpeq_epi16(qcoeff1, *thr1));
  const int nzflag = _mm_movemask_epi8(mask0) | _mm_movemask_epi8(mask1);

  if (nzflag) {
    qcoeff0 = _mm_adds_epi16(qcoeff0, *round0);
    qcoeff1 = _mm_adds_epi16(qcoeff1, *round1);
    const __m128i qtmp0 = _mm_mulhi_epi16(qcoeff0, *quant0);
    const __m128i qtmp1 = _mm_mulhi_epi16(qcoeff1, *quant1);

    // Reinsert signs
    qcoeff0 = _mm_xor_si128(qtmp0, coeff0_sign);
    qcoeff1 = _mm_xor_si128(qtmp1, coeff1_sign);
    qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
    qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);

    write_qcoeff(&qcoeff0, &qcoeff1, qcoeff_ptr, n_coeffs);

    coeff0 = _mm_mullo_epi16(qcoeff0, *dequant0);
    coeff1 = _mm_mullo_epi16(qcoeff1, *dequant1);

    write_qcoeff(&coeff0, &coeff1, dqcoeff_ptr, n_coeffs);

    const __m128i zero = _mm_setzero_si128();
    // Scan for eob
    const __m128i zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
    const __m128i zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
    const __m128i nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
    const __m128i nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
    const __m128i iscan0 =
        _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
    const __m128i iscan1 =
        _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
    // Add one to convert from indices to counts
    const __m128i iscan0_nz = _mm_sub_epi16(iscan0, nzero_coeff0);
    const __m128i iscan1_nz = _mm_sub_epi16(iscan1, nzero_coeff1);
    const __m128i eob0 = _mm_and_si128(iscan0_nz, nzero_coeff0);
    const __m128i eob1 = _mm_and_si128(iscan1_nz, nzero_coeff1);
    const __m128i eob2 = _mm_max_epi16(eob0, eob1);
    *eob = _mm_max_epi16(*eob, eob2);
  } else {
    write_zero(qcoeff_ptr, n_coeffs);
    write_zero(dqcoeff_ptr, n_coeffs);
  }
}

void av1_quantize_fp_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_ptr, const int16_t *iscan_ptr) {
  (void)scan_ptr;
  (void)zbin_ptr;
  (void)quant_shift_ptr;

  coeff_ptr += n_coeffs;
  iscan_ptr += n_coeffs;
  qcoeff_ptr += n_coeffs;
  dqcoeff_ptr += n_coeffs;
  n_coeffs = -n_coeffs;

  const __m128i round0 = _mm_load_si128((const __m128i *)round_ptr);
  const __m128i round1 = _mm_unpackhi_epi64(round0, round0);
  const __m128i quant0 = _mm_load_si128((const __m128i *)quant_ptr);
  const __m128i quant1 = _mm_unpackhi_epi64(quant0, quant0);
  const __m128i dequant0 = _mm_load_si128((const __m128i *)dequant_ptr);
  const __m128i dequant1 = _mm_unpackhi_epi64(dequant0, dequant0);
  const __m128i thr0 = _mm_srai_epi16(dequant0, 1);
  const __m128i thr1 = _mm_srai_epi16(dequant1, 1);
  __m128i eob = _mm_setzero_si128();

  quantize(iscan_ptr, coeff_ptr, n_coeffs, qcoeff_ptr, dqcoeff_ptr, &round0,
           &round1, &quant0, &quant1, &dequant0, &dequant1, &thr0, &thr1, &eob);

  n_coeffs += 8 * 2;

  // AC only loop
  while (n_coeffs < 0) {
    quantize(iscan_ptr, coeff_ptr, n_coeffs, qcoeff_ptr, dqcoeff_ptr, &round1,
             &round1, &quant1, &quant1, &dequant1, &dequant1, &thr1, &thr1,
             &eob);
    n_coeffs += 8 * 2;
  }

  // Accumulate EOB
  {
    __m128i eob_shuffled;
    eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
    eob = _mm_max_epi16(eob, eob_shuffled);
    eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
    eob = _mm_max_epi16(eob, eob_shuffled);
    eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
    eob = _mm_max_epi16(eob, eob_shuffled);
    *eob_ptr = _mm_extract_epi16(eob, 1);
  }
}

static inline void quantize_lp(const int16_t *iscan_ptr,
                               const int16_t *coeff_ptr, intptr_t n_coeffs,
                               int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
                               const __m128i *round0, const __m128i *round1,
                               const __m128i *quant0, const __m128i *quant1,
                               const __m128i *dequant0, const __m128i *dequant1,
                               __m128i *eob) {
  const int16_t *read = coeff_ptr + n_coeffs;
  __m128i coeff0 = _mm_load_si128((const __m128i *)read);
  __m128i coeff1 = _mm_load_si128((const __m128i *)read + 1);

  // Poor man's sign extract
  const __m128i coeff0_sign = _mm_srai_epi16(coeff0, 15);
  const __m128i coeff1_sign = _mm_srai_epi16(coeff1, 15);
  __m128i qcoeff0 = _mm_xor_si128(coeff0, coeff0_sign);
  __m128i qcoeff1 = _mm_xor_si128(coeff1, coeff1_sign);
  qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
  qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);

  qcoeff0 = _mm_adds_epi16(qcoeff0, *round0);
  qcoeff1 = _mm_adds_epi16(qcoeff1, *round1);
  const __m128i qtmp0 = _mm_mulhi_epi16(qcoeff0, *quant0);
  const __m128i qtmp1 = _mm_mulhi_epi16(qcoeff1, *quant1);

  // Reinsert signs
  qcoeff0 = _mm_xor_si128(qtmp0, coeff0_sign);
  qcoeff1 = _mm_xor_si128(qtmp1, coeff1_sign);
  qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
  qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);

  int16_t *addr = qcoeff_ptr + n_coeffs;
  _mm_store_si128((__m128i *)addr, qcoeff0);
  _mm_store_si128((__m128i *)addr + 1, qcoeff1);

  coeff0 = _mm_mullo_epi16(qcoeff0, *dequant0);
  coeff1 = _mm_mullo_epi16(qcoeff1, *dequant1);

  addr = dqcoeff_ptr + n_coeffs;
  _mm_store_si128((__m128i *)addr, coeff0);
  _mm_store_si128((__m128i *)addr + 1, coeff1);

  const __m128i zero = _mm_setzero_si128();
  // Scan for eob
  const __m128i zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
  const __m128i zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
  const __m128i nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
  const __m128i nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);

  const __m128i iscan0 =
      _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
  const __m128i iscan1 =
      _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);

  // Add one to convert from indices to counts
  const __m128i iscan0_nz = _mm_sub_epi16(iscan0, nzero_coeff0);
  const __m128i iscan1_nz = _mm_sub_epi16(iscan1, nzero_coeff1);
  const __m128i eob0 = _mm_and_si128(iscan0_nz, nzero_coeff0);
  const __m128i eob1 = _mm_and_si128(iscan1_nz, nzero_coeff1);
  const __m128i eob2 = _mm_max_epi16(eob0, eob1);
  *eob = _mm_max_epi16(*eob, eob2);
}

void av1_quantize_lp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
                          const int16_t *round_ptr, const int16_t *quant_ptr,
                          int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
                          const int16_t *dequant_ptr, uint16_t *eob_ptr,
                          const int16_t *scan, const int16_t *iscan) {
  (void)scan;
  coeff_ptr += n_coeffs;
  iscan += n_coeffs;
  qcoeff_ptr += n_coeffs;
  dqcoeff_ptr += n_coeffs;
  n_coeffs = -n_coeffs;

  // Setup global values
  const __m128i round0 = _mm_load_si128((const __m128i *)round_ptr);
  const __m128i round1 = _mm_unpackhi_epi64(round0, round0);
  const __m128i quant0 = _mm_load_si128((const __m128i *)quant_ptr);
  const __m128i quant1 = _mm_unpackhi_epi64(quant0, quant0);
  const __m128i dequant0 = _mm_load_si128((const __m128i *)dequant_ptr);
  const __m128i dequant1 = _mm_unpackhi_epi64(dequant0, dequant0);
  __m128i eob = _mm_setzero_si128();

  // DC and first 15 AC
  quantize_lp(iscan, coeff_ptr, n_coeffs, qcoeff_ptr, dqcoeff_ptr, &round0,
              &round1, &quant0, &quant1, &dequant0, &dequant1, &eob);
  n_coeffs += 8 * 2;

  // AC only loop
  while (n_coeffs < 0) {
    quantize_lp(iscan, coeff_ptr, n_coeffs, qcoeff_ptr, dqcoeff_ptr, &round1,
                &round1, &quant1, &quant1, &dequant1, &dequant1, &eob);
    n_coeffs += 8 * 2;
  }

  // Accumulate EOB
  *eob_ptr = accumulate_eob(eob);
}

Messung V0.5
C=94 H=99 G=96

¤ Die Informationen auf dieser Webseite wurden nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit, noch Qualität der bereit gestellten Informationen zugesichert.0.12Bemerkung:  (vorverarbeitet)  ¤

*Bot Zugriff






Wurzel

Suchen

Beweissystem der NASA

Beweissystem Isabelle

NIST Cobol Testsuite

Cephes Mathematical Library

Wiener Entwicklungsmethode

Haftungshinweis

Die Informationen auf dieser Webseite wurden nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit, noch Qualität der bereit gestellten Informationen zugesichert.

Bemerkung:

Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.