common/bitstream.c

   1 /*****************************************************************************
   2  * bitstream.c: bitstream writing
   3  *****************************************************************************
   4  * Copyright (C) 2003-2017 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Fiona Glaser <fiona@x264.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  22  *
  23  * This program is also available under a commercial proprietary license.
  24  * For more information, contact us at licensing@x264.com.
  25  *****************************************************************************/
  26
  27 #include "common.h"
  28
  29 static uint8_t *x264_nal_escape_c( uint8_t *dst, uint8_t *src, uint8_t *end )
  30 {
  31     if( src < end ) *dst++ = *src++;
  32     if( src < end ) *dst++ = *src++;
  33     while( src < end )
  34     {
  35         if( src[0] <= 0x03 && !dst[-2] && !dst[-1] )
  36             *dst++ = 0x03;
  37         *dst++ = *src++;
  38     }
  39     return dst;
  40 }
  41
  42 uint8_t *x264_nal_escape_mmx2( uint8_t *dst, uint8_t *src, uint8_t *end );
  43 uint8_t *x264_nal_escape_sse2( uint8_t *dst, uint8_t *src, uint8_t *end );
  44 uint8_t *x264_nal_escape_avx2( uint8_t *dst, uint8_t *src, uint8_t *end );
  45 void x264_cabac_block_residual_rd_internal_sse2       ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  46 void x264_cabac_block_residual_rd_internal_lzcnt      ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  47 void x264_cabac_block_residual_rd_internal_ssse3      ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  48 void x264_cabac_block_residual_rd_internal_ssse3_lzcnt( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  49 void x264_cabac_block_residual_rd_internal_avx512     ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  50 void x264_cabac_block_residual_8x8_rd_internal_sse2       ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  51 void x264_cabac_block_residual_8x8_rd_internal_lzcnt      ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  52 void x264_cabac_block_residual_8x8_rd_internal_ssse3      ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  53 void x264_cabac_block_residual_8x8_rd_internal_ssse3_lzcnt( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  54 void x264_cabac_block_residual_8x8_rd_internal_avx512     ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  55 void x264_cabac_block_residual_internal_sse2  ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  56 void x264_cabac_block_residual_internal_lzcnt ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  57 void x264_cabac_block_residual_internal_avx2  ( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  58 void x264_cabac_block_residual_internal_avx512( dctcoef *l, int b_interlaced, intptr_t ctx_block_cat, x264_cabac_t *cb );
  59
  60 uint8_t *x264_nal_escape_neon( uint8_t *dst, uint8_t *src, uint8_t *end );
  61
  62 /****************************************************************************
  63  * x264_nal_encode:
  64  ****************************************************************************/
  65 void x264_nal_encode( x264_t *h, uint8_t *dst, x264_nal_t *nal )
  66 {
  67     uint8_t *src = nal->p_payload;
  68     uint8_t *end = nal->p_payload + nal->i_payload;
  69     uint8_t *orig_dst = dst;
  70
  71     if( h->param.b_annexb )
  72     {
  73         if( nal->b_long_startcode )
  74             *dst++ = 0x00;
  75         *dst++ = 0x00;
  76         *dst++ = 0x00;
  77         *dst++ = 0x01;
  78     }
  79     else /* save room for size later */
  80         dst += 4;
  81
  82     /* nal header */
  83     *dst++ = ( 0x00 << 7 ) | ( nal->i_ref_idc << 5 ) | nal->i_type;
  84
  85     dst = h->bsf.nal_escape( dst, src, end );
  86     int size = dst - orig_dst;
  87
  88     /* Apply AVC-Intra padding */
  89     if( h->param.i_avcintra_class )
  90     {
  91         int padding = nal->i_payload + nal->i_padding + NALU_OVERHEAD - size;
  92         if( padding > 0 )
  93         {
  94             memset( dst, 0, padding );
  95             size += padding;
  96         }
  97         nal->i_padding = X264_MAX( padding, 0 );
  98     }
  99
 100     /* Write the size header for mp4/etc */
 101     if( !h->param.b_annexb )
 102     {
 103         /* Size doesn't include the size of the header we're writing now. */
 104         int chunk_size = size - 4;
 105         orig_dst[0] = chunk_size >> 24;
 106         orig_dst[1] = chunk_size >> 16;
 107         orig_dst[2] = chunk_size >> 8;
 108         orig_dst[3] = chunk_size >> 0;
 109     }
 110
 111     nal->i_payload = size;
 112     nal->p_payload = orig_dst;
 113     x264_emms();
 114 }
 115
 116 void x264_bitstream_init( int cpu, x264_bitstream_function_t *pf )
 117 {
 118     memset( pf, 0, sizeof(*pf) );
 119
 120     pf->nal_escape = x264_nal_escape_c;
 121 #if HAVE_MMX
 122 #if ARCH_X86_64 && !defined( __MACH__ )
 123     pf->cabac_block_residual_internal = x264_cabac_block_residual_internal_sse2;
 124     pf->cabac_block_residual_rd_internal = x264_cabac_block_residual_rd_internal_sse2;
 125     pf->cabac_block_residual_8x8_rd_internal = x264_cabac_block_residual_8x8_rd_internal_sse2;
 126 #endif
 127
 128     if( cpu&X264_CPU_MMX2 )
 129         pf->nal_escape = x264_nal_escape_mmx2;
 130     if( cpu&X264_CPU_SSE2 )
 131     {
 132         if( cpu&X264_CPU_SSE2_IS_FAST )
 133             pf->nal_escape = x264_nal_escape_sse2;
 134     }
 135 #if ARCH_X86_64 && !defined( __MACH__ )
 136     if( cpu&X264_CPU_LZCNT )
 137     {
 138         pf->cabac_block_residual_internal = x264_cabac_block_residual_internal_lzcnt;
 139         pf->cabac_block_residual_rd_internal = x264_cabac_block_residual_rd_internal_lzcnt;
 140         pf->cabac_block_residual_8x8_rd_internal = x264_cabac_block_residual_8x8_rd_internal_lzcnt;
 141     }
 142
 143     if( cpu&X264_CPU_SSSE3 )
 144     {
 145         pf->cabac_block_residual_rd_internal = x264_cabac_block_residual_rd_internal_ssse3;
 146         pf->cabac_block_residual_8x8_rd_internal = x264_cabac_block_residual_8x8_rd_internal_ssse3;
 147         if( cpu&X264_CPU_LZCNT )
 148         {
 149             pf->cabac_block_residual_rd_internal = x264_cabac_block_residual_rd_internal_ssse3_lzcnt;
 150             pf->cabac_block_residual_8x8_rd_internal = x264_cabac_block_residual_8x8_rd_internal_ssse3_lzcnt;
 151         }
 152     }
 153
 154     if( cpu&X264_CPU_AVX2 )
 155     {
 156         pf->nal_escape = x264_nal_escape_avx2;
 157         pf->cabac_block_residual_internal = x264_cabac_block_residual_internal_avx2;
 158     }
 159
 160     if( cpu&X264_CPU_AVX512 )
 161     {
 162         pf->cabac_block_residual_internal = x264_cabac_block_residual_internal_avx512;
 163         pf->cabac_block_residual_rd_internal = x264_cabac_block_residual_rd_internal_avx512;
 164         pf->cabac_block_residual_8x8_rd_internal = x264_cabac_block_residual_8x8_rd_internal_avx512;
 165     }
 166 #endif
 167 #endif
 168 #if HAVE_ARMV6
 169     if( cpu&X264_CPU_NEON )
 170         pf->nal_escape = x264_nal_escape_neon;
 171 #endif
 172 #if ARCH_AARCH64
 173     if( cpu&X264_CPU_NEON )
 174         pf->nal_escape = x264_nal_escape_neon;
 175 #endif
 176 }