x264 - x264_macroblock_cache_load
来源:互联网 发布:iptv网络高清机顶盒 编辑:程序博客网 时间:2024/05/20 01:11
1280x720 => w = 80 16x16mbs
=> h = 45 16x16mbs
total mbs = 3600
Now analysing mb 1, mb_x = 1, mb_y = 0
static void ALWAYS_INLINE x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y, int b_mbaff )
{
// load neighbours relations
x264_macroblock_cache_load_neighbours( h, mb_x, mb_y, b_mbaff );
// left mb coordinate of current mb,
// take account into MBAFF, it splitted as two parts,
// one component denote top of MBAFF,
// another component denote bottom of MBAFF
int *left = h->mb.i_mb_left_xy;
// top mb xy coordinate of current mb in 16x16 mb unit.
int top = h->mb.i_mb_top_xy;
// top mb y coordinate of current mb in 16x16 mb unit.
int top_y = h->mb.i_mb_top_y;
// h->mb.i_b8_stride = 160 in 8x8mb unit
int s8x8 = h->mb.i_b8_stride;
// h->mb.i_b4_stride = 320 in 4x4mb unit
int s4x4 = h->mb.i_b4_stride;
// top_8x8 mb of current mb in 8x8mb unit
int top_8x8 = (2*top_y+1) * s8x8 + 2*mb_x;
// top_4x4 mb of current mb in 4x4mb unit
int top_4x4 = (4*top_y+3) * s4x4 + 4*mb_x;
// used for loading ref/mv/mvd,
int lists = (1 << h->sh.i_type) & 3;
/* GCC pessimizes direct loads from heap-allocated arrays due to aliasing. */
/* By only dereferencing them once, we avoid this issue. */
int8_t (*i4x4)[8] = h->mb.intra4x4_pred_mode;
uint8_t (*nnz)[48] = h->mb.non_zero_count;
int16_t *cbp = h->mb.cbp;
// typedef struct x264_left_table_t
// {
// uint8_t intra[4];
// uint8_t nnz[4];
// uint8_t nnz_chroma[4];
// uint8_t mv[4];
// uint8_t ref[4];
// } x264_left_table_t;
// in x264_macroblock_cache_load_neighbours
// h->mb.left_index_table = &left_indices[3];
// ==> left_index_table = &left_indices[3]
// static const x264_left_table_t left_indices[4] =
// {
// /* Current is progressive */
// {{ 4, 4, 5, 5}, { 3, 3, 7, 7}, {16+1, 16+1, 32+1, 32+1}, {0, 0, 1, 1}, {0, 0, 0, 0}},
// {{ 6, 6, 3, 3}, {11, 11, 15, 15}, {16+5, 16+5, 32+5, 32+5}, {2, 2, 3, 3}, {1, 1, 1, 1}},
// /* Current is interlaced */
// {{ 4, 6, 4, 6}, { 3, 11, 3, 11}, {16+1, 16+1, 32+1, 32+1}, {0, 2, 0, 2}, {0, 1, 0, 1}},
// /* Both same */
// {{ 4, 5, 6, 3}, { 3, 7, 11, 15}, {16+1, 16+5, 32+1, 32+5}, {0, 1, 2, 3}, {0, 0, 1, 1}}
// };
// {intra = { 4, 5, 6, 3}, nnz = { 3, 7, 11, 15},
// nnz_chroma = {16+1, 16+5, 32+1, 32+5},
// mv = {0, 1, 2, 3}, ref = {0, 0, 1, 1}}
const x264_left_table_t *left_index_table = h->mb.left_index_table;
h->mb.cache.deblock_strength = h->deblock_strength[mb_y&1][h->param.b_sliced_threads?h->mb.i_mb_xy:mb_x];
/* load cache */
if( h->mb.i_neighbour & MB_TOP )
{ // if mb_top valid
// get cbp of top mb
h->mb.cache.i_cbp_top = cbp[top];
/* load intra4x4 */
// copy i4x4[top] to h->mb.cache.intra4x4_pred_mode[top]
CP32( &h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8], &i4x4[top][0] );
/* load non_zero_count */
// copy top Y
CP32( &h->mb.cache.non_zero_count[x264_scan8[ 0] - 8], &nnz[top][12] );
// copy top U
CP32( &h->mb.cache.non_zero_count[x264_scan8[16] - 8], &nnz[top][16-4 + (16>>CHROMA_V_SHIFT)] );
// copy top V
CP32( &h->mb.cache.non_zero_count[x264_scan8[32] - 8], &nnz[top][32-4 + (16>>CHROMA_V_SHIFT)] );
/* Finish the prefetching */
for( int l = 0; l < lists; l++ )
{
x264_prefetch( &h->mb.mv[l][top_4x4-1] );
/* Top right being not in the same cacheline as top left will happen
* once every 4 MBs, so one extra prefetch is worthwhile */
x264_prefetch( &h->mb.mv[l][top_4x4+4] );
x264_prefetch( &h->mb.ref[l][top_8x8-1] );
x264_prefetch( &h->mb.mvd[l][top] );
}
}
else
{
h->mb.cache.i_cbp_top = -1;
// mb top invalid
// init mb.cache.intra4x4_pre_mode[top] = 0xFFFFFFFFU, i.e, -1
// init top Y, U, V = 0x80808080U
/* load intra4x4 */
M32( &h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] ) = 0xFFFFFFFFU;
/* load non_zero_count */
M32( &h->mb.cache.non_zero_count[x264_scan8[ 0] - 8] ) = 0x80808080U;
M32( &h->mb.cache.non_zero_count[x264_scan8[16] - 8] ) = 0x80808080U;
M32( &h->mb.cache.non_zero_count[x264_scan8[32] - 8] ) = 0x80808080U;
}
if( h->mb.i_neighbour & MB_LEFT )
{
// get lt coordinate of current mb in 16x16 mb unit
int ltop = left[LTOP];
// if b_mbaff, lbot = left[LBOT] = left[1]
// else lbot = ltop
int lbot = b_mbaff ? left[LBOT] : ltop;
if( b_mbaff )
{ // ???
const int16_t top_luma = (cbp[ltop] >> (left_index_table->mv[0]&(~1))) & 2;
const int16_t bot_luma = (cbp[lbot] >> (left_index_table->mv[2]&(~1))) & 2;
h->mb.cache.i_cbp_left = (cbp[ltop] & 0xfff0) | (bot_luma<<2) | top_luma;
}
else // copy cbp[ltop] to mb.cache.i_cbp_left
h->mb.cache.i_cbp_left = cbp[ltop];
/* load intra4x4 */
// for non mbaff, ltop = left
// load i4x4[ltop] to mb.cache.intra4x4_pred_mode[left]
// 这里需要对照 x264_macroblock_cache_save 看
// 在 x264_macroblock_cache_save
// if( i_mb_type == I_4x4 )
// { // save bottom side predict mode of current mb [0, 1, 2, 3]
// CP32( &i4x4[0], &h->mb.cache.intra4x4_pred_mode[x264_scan8[10]] );
// // save right side predict mode of current mb [4, 5, 6,]
// M32( &i4x4[4] ) = pack8to32( h->mb.cache.intra4x4_pred_mode[x264_scan8[5] ],
// h->mb.cache.intra4x4_pred_mode[x264_scan8[7] ],
// h->mb.cache.intra4x4_pred_mode[x264_scan8[13] ], 0);
// }
// else if( !h->param.b_constrained_intra || IS_INTRA(i_mb_type) )
// M64( i4x4 ) = I_PRED_4x4_DC * 0x0101010101010101ULL;
// Now load right side of previous mb as left predict of current mb
// left_index_table->intra[0] = 4, [1] = 5, [2] = 6, [3] = 3
h->mb.cache.intra4x4_pred_mode[x264_scan8[ 0] - 1] = i4x4[ltop][left_index_table->intra[0]];
h->mb.cache.intra4x4_pred_mode[x264_scan8[ 2] - 1] = i4x4[ltop][left_index_table->intra[1]];
h->mb.cache.intra4x4_pred_mode[x264_scan8[ 8] - 1] = i4x4[lbot][left_index_table->intra[2]];
h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = i4x4[lbot][left_index_table->intra[3]];
/* load non_zero_count */
// for non mbaff, ltop = left
// load nnz[ltop] to mb.cache.non_zero_count
// left_index_table->nnz[0] = 3, [1] = 7, [2] = 11, [3] = 15
// see x264_macroblock_cache_save
// copy Ynnz of previous mb as left Y of current mb, 4 mbs once.
// CP32( &nnz[ 0+0*4], &h->mb.cache.non_zero_count[x264_scan8[ 0]] );
// CP32( &nnz[ 0+1*4], &h->mb.cache.non_zero_count[x264_scan8[ 2]] );
// CP32( &nnz[ 0+2*4], &h->mb.cache.non_zero_count[x264_scan8[ 8]] );
// CP32( &nnz[ 0+3*4], &h->mb.cache.non_zero_count[x264_scan8[10]] );
// copy Unnz of previous mb as left U of current mb, 4 mbs once,
// only 2 mbs valid
// CP32( &nnz[16+0*4], &h->mb.cache.non_zero_count[x264_scan8[16+0]] );
// CP32( &nnz[16+1*4], &h->mb.cache.non_zero_count[x264_scan8[16+2]] );
// copy Vnnz of previous mb as left V of current mb, 4 mbs once,
// only 2 mbs valid
// CP32( &nnz[32+0*4], &h->mb.cache.non_zero_count[x264_scan8[32+0]] );
// CP32( &nnz[32+1*4], &h->mb.cache.non_zero_count[x264_scan8[32+2]] );
// left_index_table->nnz[0] = 3, nnz[1] = 7, nnz[2] = 11, nnz[3] = 15
// exactly is most right 4 4x4 mbs of prevois 16x16 mb
// Therefore, this code is load right side 4x4 mb of prevoius 16x16 mb
// as left of current mb
h->mb.cache.non_zero_count[x264_scan8[ 0] - 1] = nnz[ltop][left_index_table->nnz[0]];
h->mb.cache.non_zero_count[x264_scan8[ 2] - 1] = nnz[ltop][left_index_table->nnz[1]];
h->mb.cache.non_zero_count[x264_scan8[ 8] - 1] = nnz[lbot][left_index_table->nnz[2]];
h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[lbot][left_index_table->nnz[3]];
if( CHROMA_FORMAT >= CHROMA_422 )
{
int offset = (4>>CHROMA_H_SHIFT) - 4;
h->mb.cache.non_zero_count[x264_scan8[16+ 0] - 1] = nnz[ltop][left_index_table->nnz[0]+16+offset];
h->mb.cache.non_zero_count[x264_scan8[16+ 2] - 1] = nnz[ltop][left_index_table->nnz[1]+16+offset];
h->mb.cache.non_zero_count[x264_scan8[16+ 8] - 1] = nnz[lbot][left_index_table->nnz[2]+16+offset];
h->mb.cache.non_zero_count[x264_scan8[16+10] - 1] = nnz[lbot][left_index_table->nnz[3]+16+offset];
h->mb.cache.non_zero_count[x264_scan8[32+ 0] - 1] = nnz[ltop][left_index_table->nnz[0]+32+offset];
h->mb.cache.non_zero_count[x264_scan8[32+ 2] - 1] = nnz[ltop][left_index_table->nnz[1]+32+offset];
h->mb.cache.non_zero_count[x264_scan8[32+ 8] - 1] = nnz[lbot][left_index_table->nnz[2]+32+offset];
h->mb.cache.non_zero_count[x264_scan8[32+10] - 1] = nnz[lbot][left_index_table->nnz[3]+32+offset];
}
else
{
h->mb.cache.non_zero_count[x264_scan8[16+ 0] - 1] = nnz[ltop][left_index_table->nnz_chroma[0]];
h->mb.cache.non_zero_count[x264_scan8[16+ 2] - 1] = nnz[lbot][left_index_table->nnz_chroma[1]];
h->mb.cache.non_zero_count[x264_scan8[32+ 0] - 1] = nnz[ltop][left_index_table->nnz_chroma[2]];
h->mb.cache.non_zero_count[x264_scan8[32+ 2] - 1] = nnz[lbot][left_index_table->nnz_chroma[3]];
}
}
else
{ // mb_left invalid, init relative variables
h->mb.cache.i_cbp_left = -1;
h->mb.cache.intra4x4_pred_mode[x264_scan8[ 0] - 1] =
h->mb.cache.intra4x4_pred_mode[x264_scan8[ 2] - 1] =
h->mb.cache.intra4x4_pred_mode[x264_scan8[ 8] - 1] =
h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = -1;
/* load non_zero_count */
h->mb.cache.non_zero_count[x264_scan8[ 0] - 1] =
h->mb.cache.non_zero_count[x264_scan8[ 2] - 1] =
h->mb.cache.non_zero_count[x264_scan8[ 8] - 1] =
h->mb.cache.non_zero_count[x264_scan8[10] - 1] =
h->mb.cache.non_zero_count[x264_scan8[16+ 0] - 1] =
h->mb.cache.non_zero_count[x264_scan8[16+ 2] - 1] =
h->mb.cache.non_zero_count[x264_scan8[32+ 0] - 1] =
h->mb.cache.non_zero_count[x264_scan8[32+ 2] - 1] = 0x80;
if( CHROMA_FORMAT >= CHROMA_422 )
{
h->mb.cache.non_zero_count[x264_scan8[16+ 8] - 1] =
h->mb.cache.non_zero_count[x264_scan8[16+10] - 1] =
h->mb.cache.non_zero_count[x264_scan8[32+ 8] - 1] =
h->mb.cache.non_zero_count[x264_scan8[32+10] - 1] = 0x80;
}
}
if( h->pps->b_transform_8x8_mode )
{
h->mb.cache.i_neighbour_transform_size =
( (h->mb.i_neighbour & MB_LEFT) && h->mb.mb_transform_size[left[0]] )
+ ( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] );
}
if( b_mbaff )
{
h->mb.pic.i_fref[0] = h->i_ref[0] << MB_INTERLACED;
h->mb.pic.i_fref[1] = h->i_ref[1] << MB_INTERLACED;
}
if( !b_mbaff )
{ // copy col 15 of p_fdec[0](16x16 mb) to col 31 of p_fdec[0]
// 我猜测, 它是将以前重构的像素块 (16x16 mb) 最后一列拷贝
// 当前宏块的左边, 准备作为参考列。
x264_copy_column8( h->mb.pic.p_fdec[0]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+ 4*FDEC_STRIDE );
x264_copy_column8( h->mb.pic.p_fdec[0]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+12*FDEC_STRIDE );
//
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 0, 0, 0 );
if( CHROMA444 )
{
x264_copy_column8( h->mb.pic.p_fdec[1]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[1]+15+ 4*FDEC_STRIDE );
x264_copy_column8( h->mb.pic.p_fdec[1]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[1]+15+12*FDEC_STRIDE );
x264_copy_column8( h->mb.pic.p_fdec[2]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[2]+15+ 4*FDEC_STRIDE );
x264_copy_column8( h->mb.pic.p_fdec[2]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[2]+15+12*FDEC_STRIDE );
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 1, 0, 0 );
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 2, 0, 0 );
}
else
{
// 拷贝 uv 最后一列(col 7)到 (col 31) 作为当前宏块的left
x264_copy_column8( h->mb.pic.p_fdec[1]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[1]+ 7+ 4*FDEC_STRIDE );
x264_copy_column8( h->mb.pic.p_fdec[2]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[2]+ 7+ 4*FDEC_STRIDE );
if( CHROMA_FORMAT == CHROMA_422 )
{
x264_copy_column8( h->mb.pic.p_fdec[1]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[1]+ 7+12*FDEC_STRIDE );
x264_copy_column8( h->mb.pic.p_fdec[2]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[2]+ 7+12*FDEC_STRIDE );
}
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 1, 1, 0 );
}
}
else
{
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 0, 0, 1 );
if( CHROMA444 )
{
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 1, 0, 1 );
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 2, 0, 1 );
}
else
x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 1, 1, 1 );
}
if( h->fdec->integral )
{
int offset = 16 * (mb_x + mb_y * h->fdec->i_stride[0]);
for( int list = 0; list < 2; list++ )
for( int i = 0; i < h->mb.pic.i_fref[list]; i++ )
h->mb.pic.p_integral[list][i] = &h->fref[list][i]->integral[offset];
}
x264_prefetch_fenc( h, h->fenc, mb_x, mb_y );
/* load ref/mv/mvd */
for( int l = 0; l < lists; l++ )
{
int16_t (*mv)[2] = h->mb.mv[l];
int8_t *ref = h->mb.ref[l];
int i8 = x264_scan8[0] - 1 - 1*8;
if( h->mb.i_neighbour & MB_TOPLEFT )
{
int ir = b_mbaff ? 2*(s8x8*h->mb.i_mb_topleft_y + mb_x-1)+1+s8x8 : top_8x8 - 1;
int iv = b_mbaff ? 4*(s4x4*h->mb.i_mb_topleft_y + mb_x-1)+3+3*s4x4 : top_4x4 - 1;
if( b_mbaff && h->mb.topleft_partition )
{
/* Take motion vector from the middle of macroblock instead of
* the bottom right as usual. */
iv -= 2*s4x4;
ir -= s8x8;
}
h->mb.cache.ref[l][i8] = ref[ir];
CP32( h->mb.cache.mv[l][i8], mv[iv] );
}
else
{
h->mb.cache.ref[l][i8] = -2;
M32( h->mb.cache.mv[l][i8] ) = 0;
}
i8 = x264_scan8[0] - 8;
if( h->mb.i_neighbour & MB_TOP )
{
h->mb.cache.ref[l][i8+0] =
h->mb.cache.ref[l][i8+1] = ref[top_8x8 + 0];
h->mb.cache.ref[l][i8+2] =
h->mb.cache.ref[l][i8+3] = ref[top_8x8 + 1];
CP128( h->mb.cache.mv[l][i8], mv[top_4x4] );
}
else
{
M128( h->mb.cache.mv[l][i8] ) = M128_ZERO;
M32( &h->mb.cache.ref[l][i8] ) = (uint8_t)(-2) * 0x01010101U;
}
i8 = x264_scan8[0] + 4 - 1*8;
if( h->mb.i_neighbour & MB_TOPRIGHT )
{
int ir = b_mbaff ? 2*(s8x8*h->mb.i_mb_topright_y + (mb_x+1))+s8x8 : top_8x8 + 2;
int iv = b_mbaff ? 4*(s4x4*h->mb.i_mb_topright_y + (mb_x+1))+3*s4x4 : top_4x4 + 4;
h->mb.cache.ref[l][i8] = ref[ir];
CP32( h->mb.cache.mv[l][i8], mv[iv] );
}
else
h->mb.cache.ref[l][i8] = -2;
i8 = x264_scan8[0] - 1;
if( h->mb.i_neighbour & MB_LEFT )
{
if( b_mbaff )
{
h->mb.cache.ref[l][i8+0*8] = ref[h->mb.left_b8[LTOP] + 1 + s8x8*left_index_table->ref[0]];
h->mb.cache.ref[l][i8+1*8] = ref[h->mb.left_b8[LTOP] + 1 + s8x8*left_index_table->ref[1]];
h->mb.cache.ref[l][i8+2*8] = ref[h->mb.left_b8[LBOT] + 1 + s8x8*left_index_table->ref[2]];
h->mb.cache.ref[l][i8+3*8] = ref[h->mb.left_b8[LBOT] + 1 + s8x8*left_index_table->ref[3]];
CP32( h->mb.cache.mv[l][i8+0*8], mv[h->mb.left_b4[LTOP] + 3 + s4x4*left_index_table->mv[0]] );
CP32( h->mb.cache.mv[l][i8+1*8], mv[h->mb.left_b4[LTOP] + 3 + s4x4*left_index_table->mv[1]] );
CP32( h->mb.cache.mv[l][i8+2*8], mv[h->mb.left_b4[LBOT] + 3 + s4x4*left_index_table->mv[2]] );
CP32( h->mb.cache.mv[l][i8+3*8], mv[h->mb.left_b4[LBOT] + 3 + s4x4*left_index_table->mv[3]] );
}
else
{
const int ir = h->mb.i_b8_xy - 1;
const int iv = h->mb.i_b4_xy - 1;
h->mb.cache.ref[l][i8+0*8] =
h->mb.cache.ref[l][i8+1*8] = ref[ir + 0*s8x8];
h->mb.cache.ref[l][i8+2*8] =
h->mb.cache.ref[l][i8+3*8] = ref[ir + 1*s8x8];
CP32( h->mb.cache.mv[l][i8+0*8], mv[iv + 0*s4x4] );
CP32( h->mb.cache.mv[l][i8+1*8], mv[iv + 1*s4x4] );
CP32( h->mb.cache.mv[l][i8+2*8], mv[iv + 2*s4x4] );
CP32( h->mb.cache.mv[l][i8+3*8], mv[iv + 3*s4x4] );
}
}
else
{
for( int i = 0; i < 4; i++ )
{
h->mb.cache.ref[l][i8+i*8] = -2;
M32( h->mb.cache.mv[l][i8+i*8] ) = 0;
}
}
/* Extra logic for top right mv in mbaff.
* . . . d . . a .
* . . . e . . . .
* . . . f b . c .
* . . . . . . . .
*
* If the top right of the 4x4 partitions labeled a, b and c in the
* above diagram do not exist, but the entries d, e and f exist (in
* the macroblock to the left) then use those instead.
*/
if( b_mbaff && (h->mb.i_neighbour & MB_LEFT) )
{
if( MB_INTERLACED && !h->mb.field[h->mb.i_mb_xy-1] )
{
h->mb.cache.topright_ref[l][0] = ref[h->mb.left_b8[0] + 1 + s8x8*0];
h->mb.cache.topright_ref[l][1] = ref[h->mb.left_b8[0] + 1 + s8x8*1];
h->mb.cache.topright_ref[l][2] = ref[h->mb.left_b8[1] + 1 + s8x8*0];
CP32( h->mb.cache.topright_mv[l][0], mv[h->mb.left_b4[0] + 3 + s4x4*(left_index_table->mv[0]+1)] );
CP32( h->mb.cache.topright_mv[l][1], mv[h->mb.left_b4[0] + 3 + s4x4*(left_index_table->mv[1]+1)] );
CP32( h->mb.cache.topright_mv[l][2], mv[h->mb.left_b4[1] + 3 + s4x4*(left_index_table->mv[2]+1)] );
}
else if( !MB_INTERLACED && h->mb.field[h->mb.i_mb_xy-1] )
{
// Looking at the bottom field so always take the bottom macroblock of the pair.
h->mb.cache.topright_ref[l][0] = ref[h->mb.left_b8[0] + 1 + s8x8*2 + s8x8*left_index_table->ref[0]];
h->mb.cache.topright_ref[l][1] = ref[h->mb.left_b8[0] + 1 + s8x8*2 + s8x8*left_index_table->ref[0]];
h->mb.cache.topright_ref[l][2] = ref[h->mb.left_b8[0] + 1 + s8x8*2 + s8x8*left_index_table->ref[2]];
CP32( h->mb.cache.topright_mv[l][0], mv[h->mb.left_b4[0] + 3 + s4x4*4 + s4x4*left_index_table->mv[0]] );
CP32( h->mb.cache.topright_mv[l][1], mv[h->mb.left_b4[0] + 3 + s4x4*4 + s4x4*left_index_table->mv[1]] );
CP32( h->mb.cache.topright_mv[l][2], mv[h->mb.left_b4[0] + 3 + s4x4*4 + s4x4*left_index_table->mv[2]] );
}
}
if( h->param.b_cabac )
{
uint8_t (*mvd)[8][2] = h->mb.mvd[l];
if( h->mb.i_neighbour & MB_TOP )
CP64( h->mb.cache.mvd[l][x264_scan8[0] - 8], mvd[top][0] );
else
M64( h->mb.cache.mvd[l][x264_scan8[0] - 8] ) = 0;
if( h->mb.i_neighbour & MB_LEFT && (!b_mbaff || h->mb.cache.ref[l][x264_scan8[0]-1] >= 0) )
{
CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left[LTOP]][left_index_table->intra[0]] );
CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left[LTOP]][left_index_table->intra[1]] );
}
else
{
M16( h->mb.cache.mvd[l][x264_scan8[0]-1+0*8] ) = 0;
M16( h->mb.cache.mvd[l][x264_scan8[0]-1+1*8] ) = 0;
}
if( h->mb.i_neighbour & MB_LEFT && (!b_mbaff || h->mb.cache.ref[l][x264_scan8[0]-1+2*8] >=0) )
{
CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left[LBOT]][left_index_table->intra[2]] );
CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left[LBOT]][left_index_table->intra[3]] );
}
else
{
M16( h->mb.cache.mvd[l][x264_scan8[0]-1+2*8] ) = 0;
M16( h->mb.cache.mvd[l][x264_scan8[0]-1+3*8] ) = 0;
}
}
/* If motion vectors are cached from frame macroblocks but this
* macroblock is a field macroblock then the motion vector must be
* halved. Similarly, motion vectors from field macroblocks are doubled. */
if( b_mbaff )
{
#define MAP_MVS\
if( FIELD_DIFFERENT(h->mb.i_mb_topleft_xy) )\
MAP_F2F(mv, ref, x264_scan8[0] - 1 - 1*8)\
if( FIELD_DIFFERENT(top) )\
{\
MAP_F2F(mv, ref, x264_scan8[0] + 0 - 1*8)\
MAP_F2F(mv, ref, x264_scan8[0] + 1 - 1*8)\
MAP_F2F(mv, ref, x264_scan8[0] + 2 - 1*8)\
MAP_F2F(mv, ref, x264_scan8[0] + 3 - 1*8)\
}\
if( FIELD_DIFFERENT(h->mb.i_mb_topright_xy) )\
MAP_F2F(mv, ref, x264_scan8[0] + 4 - 1*8)\
if( FIELD_DIFFERENT(left[0]) )\
{\
MAP_F2F(mv, ref, x264_scan8[0] - 1 + 0*8)\
MAP_F2F(mv, ref, x264_scan8[0] - 1 + 1*8)\
MAP_F2F(mv, ref, x264_scan8[0] - 1 + 2*8)\
MAP_F2F(mv, ref, x264_scan8[0] - 1 + 3*8)\
MAP_F2F(topright_mv, topright_ref, 0)\
MAP_F2F(topright_mv, topright_ref, 1)\
MAP_F2F(topright_mv, topright_ref, 2)\
}
if( MB_INTERLACED )
{
#define FIELD_DIFFERENT(macroblock) (macroblock >= 0 && !h->mb.field[macroblock])
#define MAP_F2F(varmv, varref, index)\
if( h->mb.cache.varref[l][index] >= 0 )\
{\
h->mb.cache.varref[l][index] <<= 1;\
h->mb.cache.varmv[l][index][1] /= 2;\
h->mb.cache.mvd[l][index][1] >>= 1;\
}
MAP_MVS
#undef MAP_F2F
#undef FIELD_DIFFERENT
}
else
{
#define FIELD_DIFFERENT(macroblock) (macroblock >= 0 && h->mb.field[macroblock])
#define MAP_F2F(varmv, varref, index)\
if( h->mb.cache.varref[l][index] >= 0 )\
{\
h->mb.cache.varref[l][index] >>= 1;\
h->mb.cache.varmv[l][index][1] <<= 1;\
h->mb.cache.mvd[l][index][1] <<= 1;\
}
MAP_MVS
#undef MAP_F2F
#undef FIELD_DIFFERENT
}
}
}
if( b_mbaff && mb_x == 0 && !(mb_y&1) && mb_y > 0 )
h->mb.field_decoding_flag = h->mb.field[h->mb.i_mb_xy - h->mb.i_mb_stride];
/* Check whether skip here would cause decoder to predict interlace mode incorrectly.
* FIXME: It might be better to change the interlace type rather than forcing a skip to be non-skip. */
h->mb.b_allow_skip = 1;
if( b_mbaff )
{
if( MB_INTERLACED != h->mb.field_decoding_flag &&
h->mb.i_mb_prev_xy >= 0 && IS_SKIP(h->mb.type[h->mb.i_mb_prev_xy]) )
h->mb.b_allow_skip = 0;
if( (mb_y&1) && IS_SKIP(h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride]) )
{
if( h->mb.i_neighbour & MB_LEFT )
{
if( h->mb.field[h->mb.i_mb_xy - 1] != MB_INTERLACED )
h->mb.b_allow_skip = 0;
}
else if( h->mb.i_neighbour & MB_TOP )
{
if( h->mb.field[h->mb.i_mb_top_xy] != MB_INTERLACED )
h->mb.b_allow_skip = 0;
}
else // Frame mb pair is predicted
{
if( MB_INTERLACED )
h->mb.b_allow_skip = 0;
}
}
}
if( h->param.b_cabac )
{
if( b_mbaff )
{
int left_xy, top_xy;
/* Neighbours here are calculated based on field_decoding_flag */
int mb_xy = mb_x + (mb_y&~1)*h->mb.i_mb_stride;
left_xy = mb_xy - 1;
if( (mb_y&1) && mb_x > 0 && h->mb.field_decoding_flag == h->mb.field[left_xy] )
left_xy += h->mb.i_mb_stride;
if( h->mb.field_decoding_flag )
{
top_xy = mb_xy - h->mb.i_mb_stride;
if( !(mb_y&1) && top_xy >= 0 && h->mb.slice_table[top_xy] == h->sh.i_first_mb && h->mb.field[top_xy] )
top_xy -= h->mb.i_mb_stride;
}
else
top_xy = mb_x + (mb_y-1)*h->mb.i_mb_stride;
h->mb.cache.i_neighbour_skip = (mb_x > 0 && h->mb.slice_table[left_xy] == h->sh.i_first_mb && !IS_SKIP( h->mb.type[left_xy] ))
+ (top_xy >= 0 && h->mb.slice_table[top_xy] == h->sh.i_first_mb && !IS_SKIP( h->mb.type[top_xy] ));
}
else
{
h->mb.cache.i_neighbour_skip = ((h->mb.i_neighbour & MB_LEFT) && !IS_SKIP( h->mb.i_mb_type_left[0] ))
+ ((h->mb.i_neighbour & MB_TOP) && !IS_SKIP( h->mb.i_mb_type_top ));
}
}
/* load skip */
if( h->sh.i_type == SLICE_TYPE_B )
{
h->mb.bipred_weight = h->mb.bipred_weight_buf[MB_INTERLACED][MB_INTERLACED&(mb_y&1)];
h->mb.dist_scale_factor = h->mb.dist_scale_factor_buf[MB_INTERLACED][MB_INTERLACED&(mb_y&1)];
if( h->param.b_cabac )
{
uint8_t skipbp;
x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
if( b_mbaff )
{
skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left[LTOP]] : 0;
h->mb.cache.skip[x264_scan8[0] - 1] = (skipbp >> (1+(left_index_table->mv[0]&~1))) & 1;
skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left[LBOT]] : 0;
h->mb.cache.skip[x264_scan8[8] - 1] = (skipbp >> (1+(left_index_table->mv[2]&~1))) & 1;
}
else
{
skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left[0]] : 0;
h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
}
skipbp = (h->mb.i_neighbour & MB_TOP) ? h->mb.skipbp[top] : 0;
h->mb.cache.skip[x264_scan8[0] - 8] = skipbp & 0x4;
h->mb.cache.skip[x264_scan8[4] - 8] = skipbp & 0x8;
}
}
if( h->sh.i_type == SLICE_TYPE_P )
x264_mb_predict_mv_pskip( h, h->mb.cache.pskip_mv );
// 根据当前16x16宏块的mb.i_neighbour_intra关系
// 确定其被分割为4x4, 或8x8宏块后宏块的关系
h->mb.i_neighbour4[0] =
h->mb.i_neighbour8[0] = (h->mb.i_neighbour_intra & (MB_TOP|MB_LEFT|MB_TOPLEFT))
| ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOPRIGHT : 0);
h->mb.i_neighbour4[4] =
h->mb.i_neighbour4[1] = MB_LEFT | ((h->mb.i_neighbour_intra & MB_TOP) ? (MB_TOP|MB_TOPLEFT|MB_TOPRIGHT) : 0);
h->mb.i_neighbour4[2] =
h->mb.i_neighbour4[8] =
h->mb.i_neighbour4[10] =
h->mb.i_neighbour8[2] = MB_TOP|MB_TOPRIGHT | ((h->mb.i_neighbour_intra & MB_LEFT) ? (MB_LEFT|MB_TOPLEFT) : 0);
h->mb.i_neighbour4[5] =
h->mb.i_neighbour8[1] = MB_LEFT | (h->mb.i_neighbour_intra & MB_TOPRIGHT)
| ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOP|MB_TOPLEFT : 0);
}
static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x, int mb_y, int i, int b_chroma, int b_mbaff )
{
int mb_interlaced = b_mbaff && MB_INTERLACED;
int height = b_chroma ? 16 >> CHROMA_V_SHIFT : 16;
int i_stride = h->fdec->i_stride[i];
int i_stride2 = i_stride << mb_interlaced;
// 对于当前的调试情况, i_pix_offset = 16
int i_pix_offset = mb_interlaced
? 16 * mb_x + height * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
: 16 * mb_x + height * mb_y * i_stride;
// 指向平面i的当前宏块对应的像素偏移位置
pixel *plane_fdec = &h->fdec->plane[i][i_pix_offset];
// 对于当前测试, fdec_idx = 1, 表示序号为 1 的 mb
int fdec_idx = b_mbaff ? (mb_interlaced ? (3 + (mb_y&1)) : (mb_y&1) ? 2 : 4) : !(mb_y&1);
// ???
pixel *intra_fdec = &h->intra_border_backup[fdec_idx][i][mb_x*16];
int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
/* ref_pix_offset[0] references the current field and [1] the opposite field. */
if( mb_interlaced )
ref_pix_offset[1] += (1-2*(mb_y&1)) * i_stride;
h->mb.pic.i_stride[i] = i_stride2;
// 将编码平面指针指向编码平面的当前宏块的像素位
h->mb.pic.p_fenc_plane[i] = &h->fenc->plane[i][i_pix_offset];
if( b_chroma )
{
h->mc.load_deinterleave_chroma_fenc( h->mb.pic.p_fenc[1], h->mb.pic.p_fenc_plane[1], i_stride2, height );
memcpy( h->mb.pic.p_fdec[1]-FDEC_STRIDE, intra_fdec, 8*sizeof(pixel) );
memcpy( h->mb.pic.p_fdec[2]-FDEC_STRIDE, intra_fdec+8, 8*sizeof(pixel) );
h->mb.pic.p_fdec[1][-FDEC_STRIDE-1] = intra_fdec[-1-8];
h->mb.pic.p_fdec[2][-FDEC_STRIDE-1] = intra_fdec[-1];
}
else
{
// 拷贝16x16的像素块到编码平面指针
h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE, h->mb.pic.p_fenc_plane[i], i_stride2, 16 );
// copy top and topright
memcpy( h->mb.pic.p_fdec[i]-FDEC_STRIDE, intra_fdec, 24*sizeof(pixel) );
// copy topleft
h->mb.pic.p_fdec[i][-FDEC_STRIDE-1] = intra_fdec[-1];
}
if( b_mbaff || h->mb.b_reencode_mb )
{
for( int j = 0; j < height; j++ )
if( b_chroma )
{
h->mb.pic.p_fdec[1][-1+j*FDEC_STRIDE] = plane_fdec[-2+j*i_stride2];
h->mb.pic.p_fdec[2][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
}
else
h->mb.pic.p_fdec[i][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
}
pixel *plane_src, **filtered_src;
for( int j = 0; j < h->mb.pic.i_fref[0]; j++ )
{
// Interpolate between pixels in same field.
if( mb_interlaced )
{
plane_src = h->fref[0][j>>1]->plane_fld[i];
filtered_src = h->fref[0][j>>1]->filtered_fld[i];
}
else
{
plane_src = h->fref[0][j]->plane[i];
filtered_src = h->fref[0][j]->filtered[i];
}
h->mb.pic.p_fref[0][j][i*4] = plane_src + ref_pix_offset[j&1];
if( !b_chroma )
{
for( int k = 1; k < 4; k++ )
h->mb.pic.p_fref[0][j][i*4+k] = filtered_src[k] + ref_pix_offset[j&1];
if( !i )
{
if( h->sh.weight[j][0].weightfn )
h->mb.pic.p_fref_w[j] = &h->fenc->weighted[j >> mb_interlaced][ref_pix_offset[j&1]];
else
h->mb.pic.p_fref_w[j] = h->mb.pic.p_fref[0][j][0];
}
}
}
if( h->sh.i_type == SLICE_TYPE_B )
for( int j = 0; j < h->mb.pic.i_fref[1]; j++ )
{
if( mb_interlaced )
{
plane_src = h->fref[1][j>>1]->plane_fld[i];
filtered_src = h->fref[1][j>>1]->filtered_fld[i];
}
else
{
plane_src = h->fref[1][j]->plane[i];
filtered_src = h->fref[1][j]->filtered[i];
}
h->mb.pic.p_fref[1][j][i*4] = plane_src + ref_pix_offset[j&1];
if( !b_chroma )
for( int k = 1; k < 4; k++ )
h->mb.pic.p_fref[1][j][i*4+k] = filtered_src[k] + ref_pix_offset[j&1];
}
}
- x264 - x264_macroblock_cache_load
- x264 - x264_macroblock_cache_load
- X264 中的x264_macroblock_cache_load 的问题
- X264 中的x264_macroblock_cache_load 的问题2
- x264_macroblock_cache_load()
- x264_macroblock_cache_load分析
- x264_macroblock_cache_load分析
- x264_macroblock_cache_load()函数分析
- x264_macroblock_cache_load()函数分析
- x264_macroblock_cache_load()函数分析
- x264
- x264 - x264.mb.i_neighbour8
- X264 笔记
- 编译x264
- x264编译
- x264 git
- x264优化
- X264设定
- 苏宁将牵手国外豪门俱乐部巴萨?
- qt中修改系统或用户环境变量
- 用Gradle 构建你的android程序
- 关于Mysql中的三种日期类型
- [算法导论] 动态规划与贪心算法的区别与联系
- x264 - x264_macroblock_cache_load
- ubuntu系统下 eclipse快捷键设置,同时觉得eclipse菜单不显示的问题
- 仿豆丁百度文库网页版阅读器完整解决方案,linux windows
- Tapestry使用示例及注意点总结
- StringBuffer
- 数据库常规top 10
- Xcode 打包 ipa 包
- IOS5 ARC unsafe_unretained等说明
- 简单的下载文件程序