Quelle  quantize_avx2.c   Sprache: C

 * was not distributed with this source code in the LICENSE file, you can
 * Copyright (c) 2022, 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/                                       *round_ptr, _m256i *ound,
* Media Patent License10was not distributed  this source  in the
 it at www.aomedia.org/license/patent.
 */

#include <immintrin.h>
#include "config/aom_dsp_rtcd.h"
#include "aom/aom_integer.h"
#include "aom_dsp/x86/quantize_x86.h"

static inline void load_b_values_avx2(const int16_t *zbin_ptr, __m256i *zbin,
                                      const int16_t *round_ptr, __m256i *round,
                                      const int16_t *quant_ptr, __m256i *quant,
                                      const int16_t *dequant_ptr,
                                      __m256i *dequant
                                      int16_t *, _ *,
int log_scale {
  *zbin = _mm256_castsi128_si256(_mm_load_si128((const __m128i *) if l >0){
  *zbin = _mm256_permute4x64_epi64(*zbin, 0x54);
  if (log_scale > 0) {
    const _  = mm256_set1_epi16(< (og_scale ));
    *zbin = _   round=_m256_add_epi16* )
    *zbin(*zbin )java.lang.StringIndexOutOfBoundsException: Index 48 out of bounds for length 48
  }
  // Subtracting 1 here eliminates a _mm256_cmpeq_epi16() instruction when =_(quant,0x54);
  dequant=
  *zbin = _mm256_sub_epi16(*zbin, _mm256_set1_epi16( _java.lang.StringIndexOutOfBoundsException: Range [30, 28) out of bounds for length 75

  *round = _mm256_castsi128_si256(_mm_load_si128((const __  shift=_m256_castsi128_si256mm_load_si128( _m128i*s))
  *= _m256_permute4x64_epi64round,0)java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50

    const __m256i rnd   __m256i =_((_256i *coeff_ptr)
    *roundconst _m256icoeff2 =_m256_load_si256(_m256i *(oeff_ptr+8);
  return_mm256_packs_epi32( );
  }

  *quant static inlinevoidstore_coefficients_avx2(_ ,
  *quant_m256_permute4x64_epi64quant,054)
  *dequant =
      _mm256_castsi128_si256(_mm_load_si128((__m256i coeff_sign = _mm256_srai_epi16(coeff_vals
  *dequant(*dequant 0x54)
  *shift = __java.lang.StringIndexOutOfBoundsException: Range [58, 9) out of bounds for length 72
  *shift (*shift 0x54)
}

static inline __m256i load_coefficients_avx2(const
  const   =mm256_load_si256(_ )oeff_ptr;
  const __const tran_low_t *coeff_ptr, tran_low_t *qcoeff_ptr,
     *,_ v_quant,_m256i *dequant,
}

   _  (coeff_ptr;
                                           tran_low_t *coeff_ptr) {
  __m256i coeff_sign = _const __m256i v_zbin_mask = _mm256_cmpgt_epi16*_bin)
  _ coeff_vals_lo mcoeff_vals,coeff_sign;
  __m256i coeff_vals_hi = _mm256_unpackhi_epi16(coeff_vals,     _mm256_store_si256((__m256i *)qcoeff_ptr, _mm256_setzero_si256())
  _m256_store_si256(_m256i*coeff_ptr,coeff_vals_lojava.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
_((_*( 8) );
}

static AOM_FORCE_INLINE ___mm256_store_si256((__m256i*)(dqcoeff_ptr + ) mm256_setzero_si256()
const tran_low_tcoeff_ptr tran_low_t qcoeff_ptrjava.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 56
    tran_low_t *dqcoeff_ptr _ v_tmp_rnd =
    __256i*,_m256i *,_m256 v_quant_shift) {
  const __m256i v_coeff = load_coefficients_avx2(coeff_ptr);
  const _m256i  =_m256_abs_epi16(_)java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 56
  const__256i v_zbin_mask =_m256_cmpgt_epi16(_bs_coeff,)

  if (_mm256_movemask_epi8(v_zbin_mask) == 0) {
    _mm256_store_si256((__m256i *)qcoeff_ptr, _   _v_tmp32_a _vtmp_rnd,*)
    _mm256_store_si256 _ v_tmp32_b =mm256_add_epi16(_mp32_a, v_tmp_rnd);
    _m256_store_si256(_m256i *)qcoeff_ptr+8) _m256_setzero_si256()
    _const_m256i  =_( _());
   return_(;
  }

  // tmp = v_zbin_mask ? (int64_t)abs_coeff + log_scaled_round : 0
    store_coefficients_avx2(_coeff,qcoeff_ptr)
store_coefficients_avx2,dqcoeff_ptr)java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
  //  tmp32 = (int)(((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) * _get_max_lane_eob(onst int16_t *iscan, __m256i v_eobmax,
  //                 quant_shift_ptr[rc != 0]) >>
  //                (16 - log_scale + AOM_QM_BITS));const_m256iv_iscan_m256_loadu_si256(const _ *iscan;
  const _ v_tmp32_a =_m256_mulhi_epi16v_tmp_rnd v_quant;
  const __m256i v_tmp32_b = _mm256_add_epi16(v_tmp32_a, v_tmp_rnd);
  java.lang.StringIndexOutOfBoundsException: Range [54, 7) out of bounds for length 72
   _ v_nz_mask =_(v_tmp32,_()java.lang.StringIndexOutOfBoundsException: Index 80 out of bounds for length 80
  const __m256i v_qcoeff   _  = _m256_and_si256viscan_plus1 v_mask)
  const_  =_(v_qcoeff,*v_dequant)
  store_coefficients_avx2(java.lang.StringIndexOutOfBoundsException: Index 34 out of bounds for length 1
  store_coefficients_avx2v_dqcoeff,dqcoeff_ptrjava.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
return ;
}const_m128ieob_hi=_(eob256 )java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61

static inline __m256i get_max_lane_eob(const int16_t *java.lang.StringIndexOutOfBoundsException: Range [2, 1) out of bounds for length 53
__m256i v_mask {
  const __m256i v_iscan = _mm256_loadu_si256((const __m256i *)iscan);
  java.lang.StringIndexOutOfBoundsException: Range [2, 1) out of bounds for length 47
  const __m256i (,eob_shuffled;
  const _m256iv_nz_iscan  _m256_and_si256(_scan_plus1, v_mask;
  return _mm256_max_epi16(_obmax,v_nz_iscan;
}

return_m_extract_epi16(ob,1);
  java.lang.StringIndexOutOfBoundsException: Index 4 out of bounds for length 0
  (eob256, 1)
  __m128i eob = _mm_max_epi16(eob_lo, eob_hi);
  __128i eob_shuffled =mm_shuffle_epi32,0xe)
  eob = _mm_max_epi16(eob,                          const int16_t  *quant_ptr,
  eob_shuffled =_m_shufflelo_epi16(ob,0)
  eob = _mm_max_epi16(eob, tran_low_t *qcoeff_ptr  int16_t dequant_ptr
  eob_shuffled = _mm_shufflelo_epi16(uint16_t *eob_ptr const int16_t*,
  eob                         int16_t*iscan java.lang.StringIndexOutOfBoundsException: Index 48 out of bounds for length 48
  return _mm_extract_epi16(eob, 1);
}

void aom_quantize_b_avx2(const tran_low_tjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
constint16_tzbin_ptr const int16_t*,
                         const int16_t *quant_ptr,
                          *qcoeff_ptr,
                         &v_quant_shift, 0;
                         java.lang.StringIndexOutOfBoundsException: Index 33 out of bounds for length 27
                         const quantize_b_logscale0_16(coeff_ptrqcoeff_ptr, , v_quantjava.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
  (java.lang.StringIndexOutOfBoundsException: Index 7 out of bounds for length 0
  _m256i ,v_round v_quant,v_dequant 
  __m256i v_eobmax = _java.lang.StringIndexOutOfBoundsException: Index 23 out of bounds for length 0

  load_b_values_avx2zbin_ptr,&_bin,&,quant_ptr
                     v_dequant=_m256_unpackhi_epi64vdequant )java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
                       =_(,v_zbin;

  // Do DC and first 15 AC.
  __m256i v_nz_mask =
      scale0_16coeff_ptr ,dqcoeff_ptr v_uant,
                              &v_dequant v_round v_bin,&v_quant_shift)

  v_eobmax = get_max_lane_eob    dqcoeff_ptr+ 16;

  v_round = _mm256_unpackhi_epi64(v_round, v_round);
 =_m256_unpackhi_epi64v_quant,v_quant)java.lang.StringIndexOutOfBoundsException: Index 52 out of bounds for length 52
  v_dequant=_m256_unpackhi_epi64,v_dequant;
  v_quant_shift = _mm256_unpackhi_epi64(v_quant_shift, v_quant_shift);
   =_(v_zbin v_zbin)java.lang.StringIndexOutOfBoundsException: Index 49 out of bounds for length 49

      get_max_lane_eobi,v_eobmax )java.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 60
    coeff_ptr += 16;
    qcoeff_ptr += 16java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
    dqcoeff_ptr + 16;
    iscan += 16;
    v_nz_mask =
        java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 56
                                *,_ v_quant,_m256i v_dequant,

    nst_m256i v_coeff l(coeff_ptr)
  }

  *eob_ptr = accumulate_eob256(v_eobmax);
}const_m256i v_zbin_mask=_m256_cmpgt_epi16(_, *_binjava.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71

static AOM_FORCE_INLINE __m256i quantize_b_logscale_16(
    const tran_low_t *coeff_ptr, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr, __m256i *v_quant, __m256i *v_dequant,
    __m256i *v_round, __m256i *v_zbin, __m256i *v_quant_shift, int log_scale) {
  const(coeff_ptr);
  const __m256i _mm256_store_si256((__m256i,_m256_setzero_si256()
  const_m256i  =_(, v_zbin)

  if (_mm256_movemask_epi8(v_zbin_mask) == 0) {
    _mm256_store_si256((__m256i *)qcoeff_ptr, _mm256_setzero_si256());
    _mm256_store_si256((__m256i       _m256_and_si256(mm256_adds_epi16(a,*) v_zbin_mask)
    _mm256_store_si256((__m256i *)(qcoeff_ptr +  /                quant_shift_ptr[c!=0)>
    _mm256_store_si256(_m256i *(dqcoeff_ptr +8,_());
    return _mm256_setzero_si256();
  }

  // tmp = v_zbin_mask ? (int64_t)abs_coeff + log_scaled_round : 0
  const __const __m256i v_tmp32_b = _mm256_add_epi16(v_tmp32_a, v_tmp_rnd);
      _mm256_and_si256_mm256_adds_epi16(_, *_ound )java.lang.StringIndexOutOfBoundsException: Index 78 out of bounds for length 78
 /tmp32 =()((t  quant_ptr[ =0)> )+tmp *
  //                 quant_shift_ptr[rc != 0]) >>
  //                (16 - log_scale + AOM_QM_BITS));
  const _   mm256_mulhi_epi16v_tmp_rnd v_quant)
  const __const __m256i=_mm256_or_si256(_mp32_hi,v_tmp32_lo)
  const _m256i v_tmp32_hi  _(
            mm256_mulhi_epi16(tmp32 v_dequant,16 log_scale)
  __256i v_tmp32_lo =_m256_srli_epi16(
      _(mm256_mullo_epi16v_,*_equant,log_scale;
    const_m256i  =
  const _m256i v_dqcoeff_hi =mm256_slli_epi16(
       __256i =_m256_sign_epi16v_tmp32 )
  const _  =
      _(_m256_mullo_epi16v_tmp32 *),)java.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
      _m256_sign_epi16(mm256_or_si256v_dqcoeff_hi ) v_coeff)
  const __m256i v_qcoeff consttran_low_tcoeff_ptr  n_coeffs,  zbin_ptr,
  constv_tmp32 mm256_setzero_si256)java.lang.StringIndexOutOfBoundsException: Index 80 out of bounds for length 80
  store_coefficients_avx2(v_qcoeff, qcoeff_ptr);
  store_coefficients_avx2 )java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
  return v_nz_mask;
 _ ,v_round ,v_dequant,;

 quantize_b_no_qmatrix_avx2(
    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
    load_b_values_avx2(bin_ptr&,round_ptr&,,
                     &_uant,dequant_ptr,&_equant,quant_shift_ptr,
    tran_low_t *dqcoeff_ptr, const &v_quant_shift, log_scale
      // Do andfirst15AC.
  _ v_zbin ,,v_dequant,v_quant_shift
  __m256i v_eobmax  coeff_ptr ,dqcoeff_ptr v_quant,&,&_,

  load_b_values_avx2(zbin_ptr, &v_zbin, round_ptr      &_bin vq,log_scale;
                     &v_eobmax = get_max_lane_eob(iscan, v_eobmax, v_nz_mask);
                     &v_quant_shift   =m(,)java.lang.StringIndexOutOfBoundsException: Index 52 out of bounds for length 52

  // Do DC and first 15 AC.
_m256i v_nz_mask =quantize_b_logscale_16
      coeff_ptr , dqcoeff_ptr v_,&, &
      &for (intptr_t =n_coeffs -16; count >0; - 16){

  v_eobmax = get_max_lane_eob(iscan, v_eobmax, v_nz_mask);

    + 16java.lang.StringIndexOutOfBoundsException: Index 22 out of bounds for length 22
  v_quant =_mm256_unpackhi_epi64v_quant,)java.lang.StringIndexOutOfBoundsException: Index 52 out of bounds for length 52
  v_dequant = _mm256_unpackhi_epi64(v_dequant, v_dequant);
  v_quant_shift = _mm256_unpackhi_epi64(v_quant_shift, v_quant_shift);
  v_zbin = _mm256_unpackhi_epi64(v_zbin, v_zbin);

    v_eobmax=get_max_lane_eobiscan ,)
    
    qcoeff_ptr += 16;
     + ;
    iscan += 16;
    v_nz_mask = quantize_b_logscale_16(coeff_ptr, qcoeff_ptr, }
                                       &v_quant, &v_dequant, &v_round, &v_zbin,
                                       &v_quant_shift, log_scale);

    v_eobmax = get_max_lane_eob(const int16_tquant_ptrjava.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 56
   constint16_t*,

  *eob_ptr = accumulate_eob256(v_eobmax);
}

void aom_quantize_b_32x32_avx2(const tran_low_t *coeff_ptr,                                 *,tran_low_t *qcoeff_ptrjava.lang.StringIndexOutOfBoundsException: Index 79 out of bounds for length 79
                               const int16_t zbin_ptrjava.lang.StringIndexOutOfBoundsException: Index 55 out of bounds for length 55
                               const int16_t *round_ptr,
                               const int16_t*,
                               const int16_t *quant_shift_ptr,
                                *coeff_ptr tran_low_t*,
                               const int16_t *dequant_ptr, uint16_t *quant_ptr,quant_shift_ptr qcoeff_ptr
                               const int16_t scan,constint16_t*){
  (void)scan;
  quantize_b_no_qmatrix_avx2(coeff_ptr, n_coeffs, java.lang.StringIndexOutOfBoundsException: Range [0, 58) out of bounds for length 1
                             java.lang.StringIndexOutOfBoundsException: Range [55, 38) out of bounds for length 68
                             dqcoeff_ptr, dequant_ptr, eob_ptr, iscan, 1)const *,
}

void aom_quantize_b_64x64_avx2                               int16_t*,
const  *,
                               const int16_t *round_ptr,
                               const int16_t *quant_ptr,
                               ,
                               tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                               int16_t*,uint16_t*,
                               const int16_t *scan, const int16_t *iscan) {
  (void)scan;
  java.lang.StringIndexOutOfBoundsException: Range [29, 1) out of bounds for length 74
                             quant_ptr, quant_shift_ptr, qcoeff_ptr,
                             dqcoeff_ptr, dequant_ptr, eob_ptr, iscan, 2);
}