*****************************************************************************/
#include "common.h"
+#include "encoder/me.h"
void x264_mb_predict_mv( x264_t *h, int i_list, int idx, int i_width, int16_t mvp[2] )
{
int i_refc = h->mb.cache.ref[i_list][i8 - 8 + i_width ];
int16_t *mv_c = h->mb.cache.mv[i_list][i8 - 8 + i_width];
- int i_count;
+ int i_count = 0;
- if( (idx&0x03) == 3 || ( i_width == 2 && (idx&0x3) == 2 )|| i_refc == -2 )
+ if( (idx&3) >= 2 + (i_width&1) || i_refc == -2 )
{
i_refc = h->mb.cache.ref[i_list][i8 - 8 - 1];
mv_c = h->mb.cache.mv[i_list][i8 - 8 - 1];
if( h->mb.i_partition == D_16x8 )
{
- if( idx == 0 && i_refb == i_ref )
+ if( idx == 0 )
{
- *(uint32_t*)mvp = *(uint32_t*)mv_b;
- return;
+ if( i_refb == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_b;
+ return;
+ }
}
- else if( idx != 0 && i_refa == i_ref )
+ else
{
- *(uint32_t*)mvp = *(uint32_t*)mv_a;
- return;
+ if( i_refa == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
+ return;
+ }
}
}
else if( h->mb.i_partition == D_8x16 )
{
- if( idx == 0 && i_refa == i_ref )
+ if( idx == 0 )
{
- *(uint32_t*)mvp = *(uint32_t*)mv_a;
- return;
+ if( i_refa == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
+ return;
+ }
}
- else if( idx != 0 && i_refc == i_ref )
+ else
{
- *(uint32_t*)mvp = *(uint32_t*)mv_c;
- return;
+ if( i_refc == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_c;
+ return;
+ }
}
}
- i_count = 0;
if( i_refa == i_ref ) i_count++;
if( i_refb == i_ref ) i_count++;
if( i_refc == i_ref ) i_count++;
if( i_count > 1 )
+ {
+median:
x264_median_mv( mvp, mv_a, mv_b, mv_c );
+ }
else if( i_count == 1 )
{
if( i_refa == i_ref )
else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
*(uint32_t*)mvp = *(uint32_t*)mv_a;
else
- x264_median_mv( mvp, mv_a, mv_b, mv_c );
+ goto median;
}
void x264_mb_predict_mv_16x16( x264_t *h, int i_list, int i_ref, int16_t mvp[2] )
int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
int16_t *mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
- int i_count;
+ int i_count = 0;
if( i_refc == -2 )
{
mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
}
- i_count = 0;
if( i_refa == i_ref ) i_count++;
if( i_refb == i_ref ) i_count++;
if( i_refc == i_ref ) i_count++;
if( i_count > 1 )
+ {
+median:
x264_median_mv( mvp, mv_a, mv_b, mv_c );
+ }
else if( i_count == 1 )
{
if( i_refa == i_ref )
else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
*(uint32_t*)mvp = *(uint32_t*)mv_a;
else
- x264_median_mv( mvp, mv_a, mv_b, mv_c );
+ goto median;
}
int16_t *mv_b = h->mb.cache.mv[0][X264_SCAN8_0 - 8];
if( i_refa == -2 || i_refb == -2 ||
- ( i_refa == 0 && *(uint32_t*)mv_a == 0 ) ||
- ( i_refb == 0 && *(uint32_t*)mv_b == 0 ) )
+ !( i_refa | *(uint32_t*)mv_a ) ||
+ !( i_refb | *(uint32_t*)mv_b ) )
{
*(uint32_t*)mv = 0;
}
{
int i_mb_4x4 = 16 * h->mb.i_mb_stride * h->mb.i_mb_y + 4 * h->mb.i_mb_x;
int i_mb_8x8 = 4 * h->mb.i_mb_stride * h->mb.i_mb_y + 2 * h->mb.i_mb_x;
- int i8, i4;
- int b8x8;
+ int i8;
const int type_col = h->fref1[0]->mb_type[ h->mb.i_mb_xy ];
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
return 1;
}
- b8x8 = h->sps->b_direct8x8_inference ||
- (type_col != P_8x8 && type_col != B_SKIP && type_col != B_DIRECT && type_col != B_8x8);
for( i8 = 0; i8 < 4; i8++ )
{
if( i_ref >= 0 )
{
const int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
-
+ const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
+ const int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
+ const int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
+ if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_col[1] > h->mb.mv_max_spel[1]) )
+ return 0;
x264_macroblock_cache_ref( h, 2*x8, 2*y8, 2, 2, 0, i_ref );
-
- if( b8x8 )
- {
- const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
- const int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
- const int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, pack16to32_mask(l0x, l0y) );
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
- }
- else
- {
- for( i4 = 0; i4 < 4; i4++ )
- {
- const int x4 = i4%2 + 2*x8;
- const int y4 = i4/2 + 2*y8;
- const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + x4 + y4 * h->mb.i_b4_stride ];
- const int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
- const int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, pack16to32_mask(l0x, l0y) );
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
- }
- }
+ x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, pack16to32_mask(l0x, l0y) );
+ x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
}
else
{
}
}
- if( h->param.i_threads > 1 )
- {
- int di = b8x8 ? 4 : 1;
- for( i4=0; i4<16; i4+=di )
- {
- if( h->mb.cache.mv[0][x264_scan8[i4]][1] > h->mb.mv_max_spel[1]
- || h->mb.cache.mv[1][x264_scan8[i4]][1] > h->mb.mv_max_spel[1] )
- {
-#if 0
- fprintf(stderr, "direct_temporal: (%d,%d) (%d,%d) > %d \n",
- h->mb.cache.mv[0][x264_scan8[i4]][0],
- h->mb.cache.mv[0][x264_scan8[i4]][1],
- h->mb.cache.mv[1][x264_scan8[i4]][0],
- h->mb.cache.mv[1][x264_scan8[i4]][1],
- h->mb.mv_max_spel[1]);
-#endif
- return 0;
- }
- }
- }
-
return 1;
}
static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
{
int ref[2];
- DECLARE_ALIGNED_8( int16_t mv[2][2] );
+ ALIGNED_8( int16_t mv[2][2] );
int i_list;
- int i8, i4;
- int b8x8;
+ int i8;
const int8_t *l1ref0 = &h->fref1[0]->ref[0][ h->mb.i_b8_xy ];
const int8_t *l1ref1 = &h->fref1[0]->ref[1][ h->mb.i_b8_xy ];
const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->fref1[0]->mv[0][ h->mb.i_b4_xy ];
if( IS_INTRA( type_col ) || (ref[0]&&ref[1]) )
return 1;
- b8x8 = h->sps->b_direct8x8_inference ||
- (type_col != P_8x8 && type_col != B_SKIP && type_col != B_DIRECT && type_col != B_8x8);
-
/* col_zero_flag */
for( i8=0; i8<4; i8++ )
{
const int x8 = i8%2;
const int y8 = i8/2;
const int o8 = x8 + y8 * h->mb.i_b8_stride;
- if( l1ref0[o8] == 0 || ( l1ref0[o8] < 0 && l1ref1[o8] == 0 ) )
+ const int o4 = 3*(x8 + y8 * h->mb.i_b4_stride);
+ if( l1ref0[o8] == 0 )
{
- const int16_t (*l1mv)[2] = (l1ref0[o8] == 0) ? l1mv0 : l1mv1;
- if( b8x8 )
+ if( abs( l1mv0[o4][0] ) <= 1 && abs( l1mv0[o4][1] ) <= 1 )
{
- const int16_t *mvcol = l1mv[3*x8 + 3*y8 * h->mb.i_b4_stride];
- if( abs( mvcol[0] ) <= 1 && abs( mvcol[1] ) <= 1 )
- {
- if( ref[0] == 0 )
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
- if( ref[1] == 0 )
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
- }
+ if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
+ if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
}
- else
+ }
+ else if( l1ref0[o8] < 0 && l1ref1[o8] == 0 )
+ {
+ if( abs( l1mv1[o4][0] ) <= 1 && abs( l1mv1[o4][1] ) <= 1 )
{
- for( i4=0; i4<4; i4++ )
- {
- const int x4 = i4%2 + 2*x8;
- const int y4 = i4/2 + 2*y8;
- const int16_t *mvcol = l1mv[x4 + y4 * h->mb.i_b4_stride];
- if( abs( mvcol[0] ) <= 1 && abs( mvcol[1] ) <= 1 )
- {
- if( ref[0] == 0 )
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, 0 );
- if( ref[1] == 0 )
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 1, 0 );
- }
- }
+ if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
+ if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
}
}
}
int mvy1 = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
int i_mode = x264_size2pixel[height][width];
int i_stride0 = 16, i_stride1 = 16;
- DECLARE_ALIGNED_16( uint8_t tmp0[16*16] );
- DECLARE_ALIGNED_16( uint8_t tmp1[16*16] );
+ ALIGNED_ARRAY_16( uint8_t, tmp0,[16*16] );
+ ALIGNED_ARRAY_16( uint8_t, tmp1,[16*16] );
uint8_t *src0, *src1;
src0 = h->mc.get_ref( tmp0, &i_stride0, h->mb.pic.p_fref[0][i_ref0], h->mb.pic.i_stride[0],
{
const int i8 = x264_scan8[0] + x + 8*y;
- /* FIXME: optimize based on current block size, not global settings? */
- if( h->sps->b_direct8x8_inference )
- {
- if( h->mb.cache.ref[0][i8] >= 0 )
- if( h->mb.cache.ref[1][i8] >= 0 )
- x264_mb_mc_01xywh( h, x, y, 2, 2 );
- else
- x264_mb_mc_0xywh( h, x, y, 2, 2 );
+ if( h->mb.cache.ref[0][i8] >= 0 )
+ if( h->mb.cache.ref[1][i8] >= 0 )
+ x264_mb_mc_01xywh( h, x, y, 2, 2 );
else
- x264_mb_mc_1xywh( h, x, y, 2, 2 );
- }
+ x264_mb_mc_0xywh( h, x, y, 2, 2 );
else
- {
- if( h->mb.cache.ref[0][i8] >= 0 )
- {
- if( h->mb.cache.ref[1][i8] >= 0 )
- {
- x264_mb_mc_01xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+1, 1, 1 );
- }
- else
- {
- x264_mb_mc_0xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_0xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_0xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_0xywh( h, x+1, y+1, 1, 1 );
- }
- }
- else
- {
- x264_mb_mc_1xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+1, 1, 1 );
- }
- }
+ x264_mb_mc_1xywh( h, x, y, 2, 2 );
}
void x264_mb_mc_8x8( x264_t *h, int i8 )
case D_L1_8x8:
x264_mb_mc_1xywh( h, x, y, 2, 2 );
break;
- case D_L1_8x4:
- x264_mb_mc_1xywh( h, x, y+0, 2, 1 );
- x264_mb_mc_1xywh( h, x, y+1, 2, 1 );
- break;
- case D_L1_4x8:
- x264_mb_mc_1xywh( h, x+0, y, 1, 2 );
- x264_mb_mc_1xywh( h, x+1, y, 1, 2 );
- break;
- case D_L1_4x4:
- x264_mb_mc_1xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+1, 1, 1 );
- break;
case D_BI_8x8:
x264_mb_mc_01xywh( h, x, y, 2, 2 );
break;
- case D_BI_8x4:
- x264_mb_mc_01xywh( h, x, y+0, 2, 1 );
- x264_mb_mc_01xywh( h, x, y+1, 2, 1 );
- break;
- case D_BI_4x8:
- x264_mb_mc_01xywh( h, x+0, y, 1, 2 );
- x264_mb_mc_01xywh( h, x+1, y, 1, 2 );
- break;
- case D_BI_4x4:
- x264_mb_mc_01xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+1, 1, 1 );
- break;
case D_DIRECT_8x8:
x264_mb_mc_direct8x8( h, x, y );
break;
}
else /* B_*x* */
{
- int b_list0[2];
- int b_list1[2];
+ const uint8_t *b_list0 = x264_mb_type_list_table[h->mb.i_type][0];
+ const uint8_t *b_list1 = x264_mb_type_list_table[h->mb.i_type][1];
- int i;
-
- /* init ref list utilisations */
- for( i = 0; i < 2; i++ )
- {
- b_list0[i] = x264_mb_type_list0_table[h->mb.i_type][i];
- b_list1[i] = x264_mb_type_list1_table[h->mb.i_type][i];
- }
if( h->mb.i_partition == D_16x16 )
{
if( b_list0[0] && b_list1[0] ) x264_mb_mc_01xywh( h, 0, 0, 4, 4 );
for( i=0; i<=h->param.b_interlaced; i++ )
for( j=0; j<3; j++ )
{
- CHECKED_MALLOC( h->mb.intra_border_backup[i][j], h->fdec->i_stride[j] );
+ /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
+ CHECKED_MALLOCZERO( h->mb.intra_border_backup[i][j], (h->sps->i_mb_width*16+32)>>!!j );
h->mb.intra_border_backup[i][j] += 8;
}
h->mb.i_neighbour4[15] =
h->mb.i_neighbour8[3] = MB_LEFT|MB_TOP|MB_TOPLEFT;
+ int buf_hpel = (h->param.i_width+48) * sizeof(int16_t);
+ int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
+ int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
+ int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
+ ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
+ int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
+ CHECKED_MALLOC( h->scratch_buffer, X264_MAX4( buf_hpel, buf_ssim, buf_tesa, buf_mbtree ) );
+
return 0;
fail: return -1;
}
x264_free( h->mb.skipbp );
x264_free( h->mb.cbp );
x264_free( h->mb.qp );
+ x264_free( h->scratch_buffer );
}
void x264_macroblock_slice_init( x264_t *h )
{
h->mc.copy[i?PIXEL_8x8:PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE,
h->mb.pic.p_fenc_plane[i], i_stride2, w );
memcpy( &h->mb.pic.p_fdec[i][-1-FDEC_STRIDE], intra_fdec-1, w*3/2+1 );
- if( h->mb.b_interlaced )
+ if( h->mb.b_interlaced || h->mb.b_reencode_mb )
{
const uint8_t *plane_fdec = &h->fdec->plane[i][i_pix_offset];
for( j = 0; j < w; j++ )
{
h->mb.i_mb_type_top =
i_top_type= h->mb.type[i_top_xy];
+ h->mb.cache.i_cbp_top = h->mb.cbp[i_top_xy];
h->mb.i_neighbour |= MB_TOP;
else
{
h->mb.i_mb_type_top = -1;
+ h->mb.cache.i_cbp_top = -1;
/* load intra4x4 */
h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] =
i_left_xy = i_mb_xy - 1;
h->mb.i_mb_type_left =
i_left_type = h->mb.type[i_left_xy];
+ h->mb.cache.i_cbp_left = h->mb.cbp[h->mb.i_mb_xy - 1];
h->mb.i_neighbour |= MB_LEFT;
else
{
h->mb.i_mb_type_left = -1;
+ 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.i_neighbour & MB_TOP);
}
- if( !h->mb.b_interlaced )
+ if( !h->mb.b_interlaced && !h->mb.b_reencode_mb )
{
copy_column8( h->mb.pic.p_fdec[0]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+ 4*FDEC_STRIDE );
copy_column8( h->mb.pic.p_fdec[0]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+12*FDEC_STRIDE );
h->mb.skipbp[i_mb_xy] = 0xf;
else if( i_mb_type == B_8x8 )
{
- int skipbp = 0;
- for( i = 0; i < 4; i++ )
- skipbp |= ( h->mb.i_sub_partition[i] == D_DIRECT_8x8 ) << i;
+ int skipbp = ( h->mb.i_sub_partition[0] == D_DIRECT_8x8 ) << 0;
+ skipbp |= ( h->mb.i_sub_partition[1] == D_DIRECT_8x8 ) << 1;
+ skipbp |= ( h->mb.i_sub_partition[2] == D_DIRECT_8x8 ) << 2;
+ skipbp |= ( h->mb.i_sub_partition[3] == D_DIRECT_8x8 ) << 3;
h->mb.skipbp[i_mb_xy] = skipbp;
}
else
projects / x264 / blobdiff