/*****************************************************************************
- * macroblock.c: h264 encoder library
+ * macroblock.c: macroblock common functions
*****************************************************************************
- * Copyright (C) 2003-2008 x264 project
+ * Copyright (C) 2003-2011 x264 project
*
* Authors: Fiona Glaser <fiona@x264.com>
* Laurent Aimar <fenrir@via.ecp.fr>
* Loren Merritt <lorenm@u.washington.edu>
+ * Henrik Gramner <hengar-6@student.ltu.se>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
+ *
+ * This program is also available under a commercial proprietary license.
+ * For more information, contact us at licensing@x264.com.
*****************************************************************************/
#include "common.h"
#include "encoder/me.h"
+#define MC_LUMA(list,p) \
+ h->mc.mc_luma( &h->mb.pic.p_fdec[p][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE, \
+ &h->mb.pic.p_fref[list][i_ref][p*4], h->mb.pic.i_stride[p], \
+ mvx, mvy, 4*width, 4*height, \
+ list ? x264_weight_none : &h->sh.weight[i_ref][p] );
+
static NOINLINE void x264_mb_mc_0xywh( x264_t *h, int x, int y, int width, int height )
{
int i8 = x264_scan8[0]+x+8*y;
int mvx = x264_clip3( h->mb.cache.mv[0][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] ) + 4*4*x;
int mvy = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
- h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
- h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0],
- mvx, mvy, 4*width, 4*height, &h->sh.weight[i_ref][0] );
-
- // chroma is offset if MCing from a field of opposite parity
- if( h->mb.b_interlaced & i_ref )
- mvy += (h->mb.i_mb_y & 1)*4 - 2;
-
- h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x],
- &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- h->mb.pic.p_fref[0][i_ref][4], h->mb.pic.i_stride[1],
- mvx, mvy, 2*width, 2*height );
-
- if( h->sh.weight[i_ref][1].weightfn )
- h->sh.weight[i_ref][1].weightfn[width>>1]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- &h->sh.weight[i_ref][1], height*2 );
- if( h->sh.weight[i_ref][2].weightfn )
- h->sh.weight[i_ref][2].weightfn[width>>1]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- &h->sh.weight[i_ref][2],height*2 );
+ MC_LUMA( 0, 0 );
+ if( CHROMA444 )
+ {
+ MC_LUMA( 0, 1 );
+ MC_LUMA( 0, 2 );
+ }
+ else
+ {
+ int v_shift = h->mb.chroma_v_shift;
+ // Chroma in 4:2:0 is offset if MCing from a field of opposite parity
+ if( v_shift & MB_INTERLACED & i_ref )
+ mvy += (h->mb.i_mb_y & 1)*4 - 2;
+
+ int offset = (4*FDEC_STRIDE>>v_shift)*y + 2*x;
+ height = 4*height >> v_shift;
+
+ h->mc.mc_chroma( &h->mb.pic.p_fdec[1][offset],
+ &h->mb.pic.p_fdec[2][offset], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][i_ref][4], h->mb.pic.i_stride[1],
+ mvx, 2*mvy>>v_shift, 2*width, height );
+
+ if( h->sh.weight[i_ref][1].weightfn )
+ h->sh.weight[i_ref][1].weightfn[width>>1]( &h->mb.pic.p_fdec[1][offset], FDEC_STRIDE,
+ &h->mb.pic.p_fdec[1][offset], FDEC_STRIDE,
+ &h->sh.weight[i_ref][1], height );
+ if( h->sh.weight[i_ref][2].weightfn )
+ h->sh.weight[i_ref][2].weightfn[width>>1]( &h->mb.pic.p_fdec[2][offset], FDEC_STRIDE,
+ &h->mb.pic.p_fdec[2][offset], FDEC_STRIDE,
+ &h->sh.weight[i_ref][2], height );
+ }
}
static NOINLINE void x264_mb_mc_1xywh( x264_t *h, int x, int y, int width, int height )
{
int mvx = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] ) + 4*4*x;
int mvy = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
- h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
- h->mb.pic.p_fref[1][i_ref], h->mb.pic.i_stride[0],
- mvx, mvy, 4*width, 4*height, weight_none );
+ MC_LUMA( 1, 0 );
- if( h->mb.b_interlaced & i_ref )
- mvy += (h->mb.i_mb_y & 1)*4 - 2;
-
- h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x],
- &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- h->mb.pic.p_fref[1][i_ref][4], h->mb.pic.i_stride[1],
- mvx, mvy, 2*width, 2*height );
+ if( CHROMA444 )
+ {
+ MC_LUMA( 1, 1 );
+ MC_LUMA( 1, 2 );
+ }
+ else
+ {
+ int v_shift = h->mb.chroma_v_shift;
+ if( v_shift & MB_INTERLACED & i_ref )
+ mvy += (h->mb.i_mb_y & 1)*4 - 2;
+
+ int offset = (4*FDEC_STRIDE>>v_shift)*y + 2*x;
+ h->mc.mc_chroma( &h->mb.pic.p_fdec[1][offset],
+ &h->mb.pic.p_fdec[2][offset], FDEC_STRIDE,
+ h->mb.pic.p_fref[1][i_ref][4], h->mb.pic.i_stride[1],
+ mvx, 2*mvy>>v_shift, 2*width, 4*height>>v_shift );
+ }
}
+#define MC_LUMA_BI(p) \
+ src0 = h->mc.get_ref( tmp0, &i_stride0, &h->mb.pic.p_fref[0][i_ref0][p*4], h->mb.pic.i_stride[p], \
+ mvx0, mvy0, 4*width, 4*height, x264_weight_none ); \
+ src1 = h->mc.get_ref( tmp1, &i_stride1, &h->mb.pic.p_fref[1][i_ref1][p*4], h->mb.pic.i_stride[p], \
+ mvx1, mvy1, 4*width, 4*height, x264_weight_none ); \
+ h->mc.avg[i_mode]( &h->mb.pic.p_fdec[p][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE, \
+ src0, i_stride0, src1, i_stride1, weight );
+
static NOINLINE void x264_mb_mc_01xywh( x264_t *h, int x, int y, int width, int height )
{
int i8 = x264_scan8[0]+x+8*y;
ALIGNED_ARRAY_16( pixel, tmp1,[16*16] );
pixel *src0, *src1;
- src0 = h->mc.get_ref( tmp0, &i_stride0, h->mb.pic.p_fref[0][i_ref0], h->mb.pic.i_stride[0],
- mvx0, mvy0, 4*width, 4*height, weight_none );
- src1 = h->mc.get_ref( tmp1, &i_stride1, h->mb.pic.p_fref[1][i_ref1], h->mb.pic.i_stride[0],
- mvx1, mvy1, 4*width, 4*height, weight_none );
- h->mc.avg[i_mode]( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
- src0, i_stride0, src1, i_stride1, weight );
+ MC_LUMA_BI( 0 );
- if( h->mb.b_interlaced & i_ref0 )
- mvy0 += (h->mb.i_mb_y & 1)*4 - 2;
- if( h->mb.b_interlaced & i_ref1 )
- mvy1 += (h->mb.i_mb_y & 1)*4 - 2;
-
- h->mc.mc_chroma( tmp0, tmp0+8, 16, h->mb.pic.p_fref[0][i_ref0][4], h->mb.pic.i_stride[1],
- mvx0, mvy0, 2*width, 2*height );
- h->mc.mc_chroma( tmp1, tmp1+8, 16, h->mb.pic.p_fref[1][i_ref1][4], h->mb.pic.i_stride[1],
- mvx1, mvy1, 2*width, 2*height );
- h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp0, 16, tmp1, 16, weight );
- h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp0+8, 16, tmp1+8, 16, weight );
+ if( CHROMA444 )
+ {
+ MC_LUMA_BI( 1 );
+ MC_LUMA_BI( 2 );
+ }
+ else
+ {
+ int v_shift = h->mb.chroma_v_shift;
+ if( v_shift & MB_INTERLACED & i_ref0 )
+ mvy0 += (h->mb.i_mb_y & 1)*4 - 2;
+ if( v_shift & MB_INTERLACED & i_ref1 )
+ mvy1 += (h->mb.i_mb_y & 1)*4 - 2;
+
+ h->mc.mc_chroma( tmp0, tmp0+8, 16, h->mb.pic.p_fref[0][i_ref0][4], h->mb.pic.i_stride[1],
+ mvx0, 2*mvy0>>v_shift, 2*width, 4*height>>v_shift );
+ h->mc.mc_chroma( tmp1, tmp1+8, 16, h->mb.pic.p_fref[1][i_ref1][4], h->mb.pic.i_stride[1],
+ mvx1, 2*mvy1>>v_shift, 2*width, 4*height>>v_shift );
+
+ int chromapix = h->luma2chroma_pixel[i_mode];
+ int offset = (4*FDEC_STRIDE>>v_shift)*y + 2*x;
+ h->mc.avg[chromapix]( &h->mb.pic.p_fdec[1][offset], FDEC_STRIDE, tmp0, 16, tmp1, 16, weight );
+ h->mc.avg[chromapix]( &h->mb.pic.p_fdec[2][offset], FDEC_STRIDE, tmp0+8, 16, tmp1+8, 16, weight );
+ }
}
+#undef MC_LUMA
+#undef MC_LUMA_BI
+
void x264_mb_mc_8x8( x264_t *h, int i8 )
{
int x = 2*(i8&1);
h->mb.i_b8_stride = h->mb.i_mb_width * 2;
h->mb.i_b4_stride = h->mb.i_mb_width * 4;
- h->mb.b_interlaced = h->param.b_interlaced;
+ h->mb.b_interlaced = PARAM_INTERLACED;
CHECKED_MALLOC( h->mb.qp, i_mb_count * sizeof(int8_t) );
CHECKED_MALLOC( h->mb.cbp, i_mb_count * sizeof(int16_t) );
- CHECKED_MALLOC( h->mb.skipbp, i_mb_count * sizeof(int8_t) );
CHECKED_MALLOC( h->mb.mb_transform_size, i_mb_count * sizeof(int8_t) );
CHECKED_MALLOC( h->mb.slice_table, i_mb_count * sizeof(uint16_t) );
memset( h->mb.slice_table, -1, i_mb_count * sizeof(uint16_t) );
CHECKED_MALLOC( h->mb.intra4x4_pred_mode, i_mb_count * 8 * sizeof(int8_t) );
/* all coeffs */
- CHECKED_MALLOC( h->mb.non_zero_count, i_mb_count * 24 * sizeof(uint8_t) );
+ CHECKED_MALLOC( h->mb.non_zero_count, i_mb_count * 48 * sizeof(uint8_t) );
if( h->param.b_cabac )
{
+ CHECKED_MALLOC( h->mb.skipbp, i_mb_count * sizeof(int8_t) );
CHECKED_MALLOC( h->mb.chroma_pred_mode, i_mb_count * sizeof(int8_t) );
CHECKED_MALLOC( h->mb.mvd[0], i_mb_count * sizeof( **h->mb.mvd ) );
- CHECKED_MALLOC( h->mb.mvd[1], i_mb_count * sizeof( **h->mb.mvd ) );
+ if( h->param.i_bframe )
+ CHECKED_MALLOC( h->mb.mvd[1], i_mb_count * sizeof( **h->mb.mvd ) );
}
for( int i = 0; i < 2; i++ )
{
- int i_refs = X264_MIN(16, (i ? 1 + !!h->param.i_bframe_pyramid : h->param.i_frame_reference) ) << h->param.b_interlaced;
+ int i_refs = X264_MIN(X264_REF_MAX, (i ? 1 + !!h->param.i_bframe_pyramid : h->param.i_frame_reference) ) << PARAM_INTERLACED;
if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
- i_refs = X264_MIN(16, i_refs + 2); //smart weights add two duplicate frames
- else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
- i_refs = X264_MIN(16, i_refs + 1); //blind weights add one duplicate frame
+ i_refs = X264_MIN(X264_REF_MAX, i_refs + 1 + (BIT_DEPTH == 8)); //smart weights add two duplicate frames, one in >8-bit
for( int j = !i; j < i_refs; j++ )
{
if( h->param.analyse.i_weighted_pred )
{
- int i_padv = PADV << h->param.b_interlaced;
- int align = h->param.cpu&X264_CPU_CACHELINE_64 ? 64 : h->param.cpu&X264_CPU_CACHELINE_32 ? 32 : 16;
- int i_stride, luma_plane_size = 0;
+ int i_padv = PADV << PARAM_INTERLACED;
+ int luma_plane_size = 0;
int numweightbuf;
if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE )
if( !h->param.i_sync_lookahead || h == h->thread[h->param.i_threads] )
{
// Fake analysis only works on lowres
- i_stride = ALIGN( h->mb.i_mb_width*8 + 2*PADH, align );
- luma_plane_size = i_stride * (h->mb.i_mb_height*8+2*i_padv);
+ luma_plane_size = h->fdec->i_stride_lowres * (h->mb.i_mb_height*8+2*i_padv);
// Only need 1 buffer for analysis
numweightbuf = 1;
}
}
else
{
- i_stride = ALIGN( h->mb.i_mb_width*16 + 2*PADH, align );
- luma_plane_size = i_stride * (h->mb.i_mb_height*16+2*i_padv);
+ /* Both ref and fenc is stored for 4:2:0 and 4:2:2 which means that 4:2:0 and 4:4:4
+ * needs the same amount of space and 4:2:2 needs twice that much */
+ luma_plane_size = h->fdec->i_stride[0] * (h->mb.i_mb_height*(16<<(CHROMA_FORMAT==CHROMA_422))+2*i_padv);
if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
- //SMART can weight one ref and one offset -1
- numweightbuf = 2;
+ //smart can weight one ref and one offset -1 in 8-bit
+ numweightbuf = 1 + (BIT_DEPTH == 8);
else
- //blind only has one weighted copy (offset -1)
+ //simple only has one weighted ref
numweightbuf = 1;
}
void x264_macroblock_cache_free( x264_t *h )
{
for( int i = 0; i < 2; i++ )
- for( int j = !i; j < 32; j++ )
+ for( int j = !i; j < X264_REF_MAX*2; j++ )
if( h->mb.mvr[i][j] )
x264_free( h->mb.mvr[i][j]-1 );
- for( int i = 0; i < 16; i++ )
+ for( int i = 0; i < X264_REF_MAX; i++ )
x264_free( h->mb.p_weight_buf[i] );
if( h->param.b_cabac )
{
+ x264_free( h->mb.skipbp );
x264_free( h->mb.chroma_pred_mode );
x264_free( h->mb.mvd[0] );
x264_free( h->mb.mvd[1] );
x264_free( h->mb.intra4x4_pred_mode );
x264_free( h->mb.non_zero_count );
x264_free( h->mb.mb_transform_size );
- x264_free( h->mb.skipbp );
x264_free( h->mb.cbp );
x264_free( h->mb.qp );
}
int x264_macroblock_thread_allocate( x264_t *h, int b_lookahead )
{
if( !b_lookahead )
- for( int i = 0; i <= h->param.b_interlaced; i++ )
- {
- for( int j = 0; j < 2; j++ )
+ {
+ for( int i = 0; i <= 4*PARAM_INTERLACED; i++ )
+ for( int j = 0; j < (CHROMA444 ? 3 : 2); j++ )
{
- /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
- CHECKED_MALLOCZERO( h->intra_border_backup[i][j], (h->sps->i_mb_width*16+32) * sizeof(pixel) );
+ CHECKED_MALLOC( h->intra_border_backup[i][j], (h->sps->i_mb_width*16+32) * sizeof(pixel) );
h->intra_border_backup[i][j] += 16;
+ if( !PARAM_INTERLACED )
+ h->intra_border_backup[1][j] = h->intra_border_backup[i][j];
}
+ for( int i = 0; i <= PARAM_INTERLACED; i++ )
+ {
CHECKED_MALLOC( h->deblock_strength[i], sizeof(**h->deblock_strength) * h->mb.i_mb_width );
+ h->deblock_strength[1] = h->deblock_strength[i];
}
+ }
/* Allocate scratch buffer */
int scratch_size = 0;
if( !b_lookahead )
{
- int buf_hpel = (h->thread[0]->fdec->i_width[0]+48) * sizeof(dctcoef);
+ int buf_hpel = (h->thread[0]->fdec->i_width[0]+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+24) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
scratch_size = X264_MAX3( buf_hpel, buf_ssim, buf_tesa );
}
- int buf_mbtree = h->param.rc.b_mb_tree * ((h->mb.i_mb_width+3)&~3) * sizeof(int);
+ int buf_mbtree = h->param.rc.b_mb_tree * ((h->mb.i_mb_width+7)&~7) * sizeof(int);
scratch_size = X264_MAX( scratch_size, buf_mbtree );
- CHECKED_MALLOC( h->scratch_buffer, scratch_size );
+ if( scratch_size )
+ CHECKED_MALLOC( h->scratch_buffer, scratch_size );
+ else
+ h->scratch_buffer = NULL;
return 0;
fail:
void x264_macroblock_thread_free( x264_t *h, int b_lookahead )
{
if( !b_lookahead )
- for( int i = 0; i <= h->param.b_interlaced; i++ )
- {
+ {
+ for( int i = 0; i <= PARAM_INTERLACED; i++ )
x264_free( h->deblock_strength[i] );
- for( int j = 0; j < 2; j++ )
+ for( int i = 0; i <= 4*PARAM_INTERLACED; i++ )
+ for( int j = 0; j < (CHROMA444 ? 3 : 2); j++ )
x264_free( h->intra_border_backup[i][j] - 16 );
- }
+ }
x264_free( h->scratch_buffer );
}
h->mb.ref[1] = h->fdec->ref[1];
h->mb.type = h->fdec->mb_type;
h->mb.partition = h->fdec->mb_partition;
+ h->mb.field = h->fdec->field;
- h->fdec->i_ref[0] = h->i_ref0;
- h->fdec->i_ref[1] = h->i_ref1;
- for( int i = 0; i < h->i_ref0; i++ )
- h->fdec->ref_poc[0][i] = h->fref0[i]->i_poc;
+ h->fdec->i_ref[0] = h->i_ref[0];
+ h->fdec->i_ref[1] = h->i_ref[1];
+ for( int i = 0; i < h->i_ref[0]; i++ )
+ h->fdec->ref_poc[0][i] = h->fref[0][i]->i_poc;
if( h->sh.i_type == SLICE_TYPE_B )
{
- for( int i = 0; i < h->i_ref1; i++ )
- h->fdec->ref_poc[1][i] = h->fref1[i]->i_poc;
+ for( int i = 0; i < h->i_ref[1]; i++ )
+ h->fdec->ref_poc[1][i] = h->fref[1][i]->i_poc;
map_col_to_list0(-1) = -1;
map_col_to_list0(-2) = -2;
- for( int i = 0; i < h->fref1[0]->i_ref[0]; i++ )
+ for( int i = 0; i < h->fref[1][0]->i_ref[0]; i++ )
{
- int poc = h->fref1[0]->ref_poc[0][i];
+ int poc = h->fref[1][0]->ref_poc[0][i];
map_col_to_list0(i) = -2;
- for( int j = 0; j < h->i_ref0; j++ )
- if( h->fref0[j]->i_poc == poc )
+ for( int j = 0; j < h->i_ref[0]; j++ )
+ if( h->fref[0][j]->i_poc == poc )
{
map_col_to_list0(i) = j;
break;
{
memset( h->mb.cache.skip, 0, sizeof( h->mb.cache.skip ) );
- if( h->sh.i_disable_deblocking_filter_idc != 1 && h->param.analyse.i_weighted_pred )
+ if( h->sh.i_disable_deblocking_filter_idc != 1 && h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
{
deblock_ref_table(-2) = -2;
deblock_ref_table(-1) = -1;
- for( int i = 0; i < h->i_ref0 << h->sh.b_mbaff; i++ )
+ for( int i = 0; i < h->i_ref[0] << SLICE_MBAFF; i++ )
{
/* Mask off high bits to avoid frame num collisions with -1/-2.
* In current x264 frame num values don't cover a range of more
* than 32, so 6 bits is enough for uniqueness. */
- if( !h->mb.b_interlaced )
- deblock_ref_table(i) = h->fref0[i]->i_frame_num&63;
+ if( !MB_INTERLACED )
+ deblock_ref_table(i) = h->fref[0][i]->i_frame_num&63;
else
- deblock_ref_table(i) = ((h->fref0[i>>1]->i_frame_num&63)<<1) + (i&1);
+ deblock_ref_table(i) = ((h->fref[0][i>>1]->i_frame_num&63)<<1) + (i&1);
}
}
}
/* init with not available (for top right idx=7,15) */
memset( h->mb.cache.ref, -2, sizeof( h->mb.cache.ref ) );
- if( h->i_ref0 > 0 )
- for( int field = 0; field <= h->sh.b_mbaff; field++ )
+ if( h->i_ref[0] > 0 )
+ for( int field = 0; field <= SLICE_MBAFF; field++ )
{
- int curpoc = h->fdec->i_poc + field*h->sh.i_delta_poc_bottom;
- int refpoc = h->fref0[0]->i_poc;
- if( h->sh.b_mbaff && field )
- refpoc += h->sh.i_delta_poc_bottom;
+ int curpoc = h->fdec->i_poc + h->fdec->i_delta_poc[field];
+ int refpoc = h->fref[0][0]->i_poc + h->fref[0][0]->i_delta_poc[field];
int delta = curpoc - refpoc;
h->fdec->inv_ref_poc[field] = (256 + delta/2) / delta;
h->mb.i_subpel_refine = h->param.analyse.i_subpel_refine;
if( h->sh.i_type == SLICE_TYPE_B && (h->mb.i_subpel_refine == 6 || h->mb.i_subpel_refine == 8) )
h->mb.i_subpel_refine--;
- h->mb.b_chroma_me = h->param.analyse.b_chroma_me && h->sh.i_type == SLICE_TYPE_P
- && h->mb.i_subpel_refine >= 5;
+ h->mb.b_chroma_me = h->param.analyse.b_chroma_me &&
+ ((h->sh.i_type == SLICE_TYPE_P && h->mb.i_subpel_refine >= 5) ||
+ (h->sh.i_type == SLICE_TYPE_B && h->mb.i_subpel_refine >= 9));
h->mb.b_dct_decimate = h->sh.i_type == SLICE_TYPE_B ||
(h->param.analyse.b_dct_decimate && h->sh.i_type != SLICE_TYPE_I);
-
-
- /* fdec: fenc:
- * yyyyyyy
- * yYYYY YYYY
- * yYYYY YYYY
- * yYYYY YYYY
- * yYYYY YYYY
- * uuu vvv UUVV
- * uUU vVV UUVV
- * uUU vVV
+ h->mb.i_mb_prev_xy = -1;
+
+ /* 4:2:0 4:2:2 4:4:4
+ * fdec fenc fdec fenc fdec fenc
+ * y y y y y y y Y Y Y Y y y y y y y y Y Y Y Y y y y y y y y Y Y Y Y
+ * y Y Y Y Y Y Y Y Y y Y Y Y Y Y Y Y Y y Y Y Y Y Y Y Y Y
+ * y Y Y Y Y Y Y Y Y y Y Y Y Y Y Y Y Y y Y Y Y Y Y Y Y Y
+ * y Y Y Y Y Y Y Y Y y Y Y Y Y Y Y Y Y y Y Y Y Y Y Y Y Y
+ * y Y Y Y Y U U V V y Y Y Y Y U U V V y Y Y Y Y U U U U
+ * u u u v v v U U V V u u u v v v U U V V u u u u u u u U U U U
+ * u U U v V V u U U v V V U U V V u U U U U U U U U
+ * u U U v V V u U U v V V U U V V u U U U U U U U U
+ * u U U v V V u U U U U V V V V
+ * u U U v V V u U U U U V V V V
+ * v v v v v v v V V V V
+ * v V V V V V V V V
+ * v V V V V
+ * v V V V V
+ * v V V V V
*/
h->mb.pic.p_fenc[0] = h->mb.pic.fenc_buf;
- h->mb.pic.p_fenc[1] = h->mb.pic.fenc_buf + 16*FENC_STRIDE;
- h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 16*FENC_STRIDE + 8;
h->mb.pic.p_fdec[0] = h->mb.pic.fdec_buf + 2*FDEC_STRIDE;
+ h->mb.pic.p_fenc[1] = h->mb.pic.fenc_buf + 16*FENC_STRIDE;
h->mb.pic.p_fdec[1] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE;
- h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE + 16;
+ if( CHROMA444 )
+ {
+ h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 32*FENC_STRIDE;
+ h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 36*FDEC_STRIDE;
+ }
+ else
+ {
+ h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 16*FENC_STRIDE + 8;
+ h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE + 16;
+ }
}
void x264_prefetch_fenc( x264_t *h, x264_frame_t *fenc, int i_mb_x, int i_mb_y )
int stride_y = fenc->i_stride[0];
int stride_uv = fenc->i_stride[1];
int off_y = 16 * i_mb_x + 16 * i_mb_y * stride_y;
- int off_uv = 16 * i_mb_x + 8 * i_mb_y * stride_uv;
+ int off_uv = 16 * i_mb_x + (16 * i_mb_y * stride_uv >> h->mb.chroma_v_shift);
h->mc.prefetch_fenc( fenc->plane[0]+off_y, stride_y,
fenc->plane[1]+off_uv, stride_uv, i_mb_x );
}
dst[i*FDEC_STRIDE] = src[i*FDEC_STRIDE];
}
-static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x, int mb_y, int i, int b_interlaced )
+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 w = (i ? 8 : 16);
+ int mb_interlaced = b_mbaff && MB_INTERLACED;
+ int height = b_chroma ? 16 >> h->mb.chroma_v_shift : 16;
int i_stride = h->fdec->i_stride[i];
- int i_stride2 = i_stride << b_interlaced;
- int i_pix_offset = b_interlaced
- ? 16 * mb_x + w * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
- : 16 * mb_x + w * mb_y * i_stride;
+ int i_stride2 = i_stride << mb_interlaced;
+ 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;
pixel *plane_fdec = &h->fdec->plane[i][i_pix_offset];
- pixel *intra_fdec = &h->intra_border_backup[mb_y & h->sh.b_mbaff][i][mb_x*16];
+ int fdec_idx = b_mbaff ? (mb_interlaced ? (3 + (mb_y&1)) : (mb_y&1) ? 2 : 4) : 0;
+ pixel *intra_fdec = &h->intra_border_backup[fdec_idx][i][mb_x*16];
int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
- x264_frame_t **fref[2] = { h->fref0, h->fref1 };
- if( b_interlaced )
+ /* 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;
h->mb.pic.p_fenc_plane[i] = &h->fenc->plane[i][i_pix_offset];
- if( i )
+ if( b_chroma )
{
- h->mc.load_deinterleave_8x8x2_fenc( h->mb.pic.p_fenc[1], h->mb.pic.p_fenc_plane[1], i_stride2 );
+ 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) );
+ if( b_mbaff )
+ {
+ 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
{
- h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE, h->mb.pic.p_fenc_plane[0], i_stride2, 16 );
- memcpy( h->mb.pic.p_fdec[0]-FDEC_STRIDE, intra_fdec, 24*sizeof(pixel) );
+ h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE, h->mb.pic.p_fenc_plane[i], i_stride2, 16 );
+ memcpy( h->mb.pic.p_fdec[i]-FDEC_STRIDE, intra_fdec, 24*sizeof(pixel) );
+ if( b_mbaff )
+ h->mb.pic.p_fdec[i][-FDEC_STRIDE-1] = intra_fdec[-1];
}
- if( b_interlaced )
+ if( b_mbaff )
{
- for( int j = 0; j < w; j++ )
- if( i )
+ 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[0][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
+ 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++ )
{
- h->mb.pic.p_fref[0][j][i?4:0] = &fref[0][j >> b_interlaced]->plane[i][ref_pix_offset[j&1]];
- if( !i )
+ // 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][k] = &fref[0][j >> b_interlaced]->filtered[k][ref_pix_offset[j&1]];
- if( h->sh.weight[j][0].weightfn )
- h->mb.pic.p_fref_w[j] = &h->fenc->weighted[j >> b_interlaced][ref_pix_offset[j&1]];
- else
- h->mb.pic.p_fref_w[j] = h->mb.pic.p_fref[0][j][0];
+ 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++ )
{
- h->mb.pic.p_fref[1][j][i?4:0] = &fref[1][j >> b_interlaced]->plane[i][ref_pix_offset[j&1]];
- if( !i )
+ 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][k] = &fref[1][j >> b_interlaced]->filtered[k][ref_pix_offset[j&1]];
+ h->mb.pic.p_fref[1][j][i*4+k] = filtered_src[k] + ref_pix_offset[j&1];
}
}
-static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, int mb_y )
+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}}
+};
+
+static void ALWAYS_INLINE x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, int mb_y, int b_interlaced )
{
- int top = (mb_y - (1 << h->mb.b_interlaced)) * h->mb.i_mb_stride + mb_x;
+ const int mb_interlaced = b_interlaced && MB_INTERLACED;
+ int top_y = mb_y - (1 << mb_interlaced);
+ int top = top_y * h->mb.i_mb_stride + mb_x;
h->mb.i_mb_x = mb_x;
h->mb.i_mb_y = mb_y;
h->mb.i_mb_xy = mb_y * h->mb.i_mb_stride + mb_x;
h->mb.i_b8_xy = 2*(mb_y * h->mb.i_b8_stride + mb_x);
h->mb.i_b4_xy = 4*(mb_y * h->mb.i_b4_stride + mb_x);
+ h->mb.left_b8[0] =
+ h->mb.left_b8[1] = -1;
+ h->mb.left_b4[0] =
+ h->mb.left_b4[1] = -1;
h->mb.i_neighbour = 0;
h->mb.i_neighbour_intra = 0;
h->mb.i_neighbour_frame = 0;
h->mb.i_mb_top_xy = -1;
- h->mb.i_mb_left_xy = -1;
+ h->mb.i_mb_top_y = -1;
+ h->mb.i_mb_left_xy[0] = h->mb.i_mb_left_xy[1] = -1;
h->mb.i_mb_topleft_xy = -1;
h->mb.i_mb_topright_xy = -1;
h->mb.i_mb_type_top = -1;
- h->mb.i_mb_type_left = -1;
+ h->mb.i_mb_type_left[0] = h->mb.i_mb_type_left[1] = -1;
h->mb.i_mb_type_topleft = -1;
h->mb.i_mb_type_topright = -1;
+ h->mb.left_index_table = &left_indices[3];
+ h->mb.topleft_partition = 0;
+
+ int topleft_y = top_y;
+ int topright_y = top_y;
+ int left[2];
+
+ left[0] = left[1] = h->mb.i_mb_xy - 1;
+ h->mb.left_b8[0] = h->mb.left_b8[1] = h->mb.i_b8_xy - 2;
+ h->mb.left_b4[0] = h->mb.left_b4[1] = h->mb.i_b4_xy - 4;
+
+ if( b_interlaced )
+ {
+ h->mb.i_mb_top_mbpair_xy = h->mb.i_mb_xy - 2*h->mb.i_mb_stride;
+ h->mb.i_mb_topleft_y = -1;
+ h->mb.i_mb_topright_y = -1;
+
+ if( mb_y&1 )
+ {
+ if( mb_x && mb_interlaced != h->mb.field[h->mb.i_mb_xy-1] )
+ {
+ left[0] = left[1] = h->mb.i_mb_xy - 1 - h->mb.i_mb_stride;
+ h->mb.left_b8[0] = h->mb.left_b8[1] = h->mb.i_b8_xy - 2 - 2*h->mb.i_b8_stride;
+ h->mb.left_b4[0] = h->mb.left_b4[1] = h->mb.i_b4_xy - 4 - 4*h->mb.i_b4_stride;
+
+ if( mb_interlaced )
+ {
+ h->mb.left_index_table = &left_indices[2];
+ left[1] += h->mb.i_mb_stride;
+ h->mb.left_b8[1] += 2*h->mb.i_b8_stride;
+ h->mb.left_b4[1] += 4*h->mb.i_b4_stride;
+ }
+ else
+ {
+ h->mb.left_index_table = &left_indices[1];
+ topleft_y++;
+ h->mb.topleft_partition = 1;
+ }
+ }
+ if( !mb_interlaced )
+ topright_y = -1;
+ }
+ else
+ {
+ if( mb_interlaced && top >= 0 )
+ {
+ if( !h->mb.field[top] )
+ {
+ top += h->mb.i_mb_stride;
+ top_y++;
+ }
+ if( mb_x )
+ topleft_y += !h->mb.field[h->mb.i_mb_stride*topleft_y + mb_x - 1];
+ if( mb_x < h->mb.i_mb_width-1 )
+ topright_y += !h->mb.field[h->mb.i_mb_stride*topright_y + mb_x + 1];
+ }
+ if( mb_x && mb_interlaced != h->mb.field[h->mb.i_mb_xy-1] )
+ {
+ if( mb_interlaced )
+ {
+ h->mb.left_index_table = &left_indices[2];
+ left[1] += h->mb.i_mb_stride;
+ h->mb.left_b8[1] += 2*h->mb.i_b8_stride;
+ h->mb.left_b4[1] += 4*h->mb.i_b4_stride;
+ }
+ else
+ h->mb.left_index_table = &left_indices[0];
+ }
+ }
+ }
if( mb_x > 0 )
{
h->mb.i_neighbour_frame |= MB_LEFT;
- h->mb.i_mb_left_xy = h->mb.i_mb_xy - 1;
- h->mb.i_mb_type_left = h->mb.type[h->mb.i_mb_left_xy];
- if( h->mb.i_mb_xy > h->sh.i_first_mb )
+ h->mb.i_mb_left_xy[0] = left[0];
+ h->mb.i_mb_left_xy[1] = left[1];
+ h->mb.i_mb_type_left[0] = h->mb.type[h->mb.i_mb_left_xy[0]];
+ h->mb.i_mb_type_left[1] = h->mb.type[h->mb.i_mb_left_xy[1]];
+ if( h->mb.slice_table[left[0]] == h->sh.i_first_mb )
{
h->mb.i_neighbour |= MB_LEFT;
- if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_left ) )
+ // FIXME: We don't currently support constrained intra + mbaff.
+ if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_left[0] ) )
h->mb.i_neighbour_intra |= MB_LEFT;
}
}
/* We can't predict from the previous threadslice since it hasn't been encoded yet. */
- if( (h->i_threadslice_start >> h->mb.b_interlaced) != (mb_y >> h->mb.b_interlaced) )
+ if( (h->i_threadslice_start >> mb_interlaced) != (mb_y >> mb_interlaced) )
{
if( top >= 0 )
{
h->mb.i_neighbour_frame |= MB_TOP;
h->mb.i_mb_top_xy = top;
+ h->mb.i_mb_top_y = top_y;
h->mb.i_mb_type_top = h->mb.type[h->mb.i_mb_top_xy];
- if( top >= h->sh.i_first_mb )
+ if( h->mb.slice_table[top] == h->sh.i_first_mb )
{
h->mb.i_neighbour |= MB_TOP;
}
}
- if( mb_x > 0 && top - 1 >= 0 )
+ if( mb_x > 0 && topleft_y >= 0 )
{
h->mb.i_neighbour_frame |= MB_TOPLEFT;
- h->mb.i_mb_topleft_xy = top - 1;
+ h->mb.i_mb_topleft_xy = h->mb.i_mb_stride*topleft_y + mb_x - 1;
+ h->mb.i_mb_topleft_y = topleft_y;
h->mb.i_mb_type_topleft = h->mb.type[h->mb.i_mb_topleft_xy];
- if( top - 1 >= h->sh.i_first_mb )
+ if( h->mb.slice_table[h->mb.i_mb_topleft_xy] == h->sh.i_first_mb )
{
h->mb.i_neighbour |= MB_TOPLEFT;
}
}
- if( mb_x < h->mb.i_mb_width - 1 && top + 1 >= 0 )
+ if( mb_x < h->mb.i_mb_width - 1 && topright_y >= 0 )
{
h->mb.i_neighbour_frame |= MB_TOPRIGHT;
- h->mb.i_mb_topright_xy = top + 1;
+ h->mb.i_mb_topright_xy = h->mb.i_mb_stride*topright_y + mb_x + 1;
+ h->mb.i_mb_topright_y = topright_y;
h->mb.i_mb_type_topright = h->mb.type[h->mb.i_mb_topright_xy];
- if( top + 1 >= h->sh.i_first_mb )
+ if( h->mb.slice_table[h->mb.i_mb_topright_xy] == h->sh.i_first_mb )
{
h->mb.i_neighbour |= MB_TOPRIGHT;
}
}
-void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
+#define LTOP 0
+#if HAVE_INTERLACED
+# define LBOT 1
+#else
+# define LBOT 0
+#endif
+
+static void ALWAYS_INLINE x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y, int b_mbaff )
{
- x264_macroblock_cache_load_neighbours( h, mb_x, mb_y );
+ x264_macroblock_cache_load_neighbours( h, mb_x, mb_y, b_mbaff );
- int left = h->mb.i_mb_left_xy;
+ int *left = h->mb.i_mb_left_xy;
int top = h->mb.i_mb_top_xy;
- int top_y = mb_y - (1 << h->mb.b_interlaced);
+ int top_y = h->mb.i_mb_top_y;
int s8x8 = h->mb.i_b8_stride;
int s4x4 = h->mb.i_b4_stride;
int top_8x8 = (2*top_y+1) * s8x8 + 2*mb_x;
/* 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)[24] = h->mb.non_zero_count;
+ uint8_t (*nnz)[48] = h->mb.non_zero_count;
int16_t *cbp = h->mb.cbp;
+ const x264_left_table_t *left_index_table = h->mb.left_index_table;
+
/* load cache */
if( h->mb.i_neighbour & MB_TOP )
{
CP32( &h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8], &i4x4[top][0] );
/* load non_zero_count */
- CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &nnz[top][12] );
- /* shift because x264_scan8[16] is misaligned */
- M32( &h->mb.cache.non_zero_count[x264_scan8[16+0] - 9] ) = M16( &nnz[top][18] ) << 8;
- M32( &h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] ) = M16( &nnz[top][22] ) << 8;
+ CP32( &h->mb.cache.non_zero_count[x264_scan8[ 0] - 8], &nnz[top][12] );
+ CP32( &h->mb.cache.non_zero_count[x264_scan8[16] - 8], &nnz[top][16-4 + (16>>h->mb.chroma_v_shift)] );
+ CP32( &h->mb.cache.non_zero_count[x264_scan8[32] - 8], &nnz[top][32-4 + (16>>h->mb.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
{
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+0] - 9] ) = 0x80808080U;
- M32( &h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] ) = 0x80808080U;
+ 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 )
{
- h->mb.cache.i_cbp_left = cbp[left];
+ int ltop = left[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
+ h->mb.cache.i_cbp_left = cbp[ltop];
/* load intra4x4 */
- h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = i4x4[left][4];
- h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = i4x4[left][5];
- h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = i4x4[left][6];
- h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = i4x4[left][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 */
- h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][3];
- h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][7];
- h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][11];
- h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][15];
-
- h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = nnz[left][16+1];
- h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = nnz[left][16+3];
-
- h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = nnz[left][16+4+1];
- h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = nnz[left][16+4+3];
+ 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]];
- /* Finish the prefetching */
- for( int l = 0; l < lists; l++ )
+ if( CHROMA_FORMAT >= CHROMA_422 )
{
- 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] );
+ int offset = (4>>h->mb.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
{
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[ 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[ 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[16+4+0] - 1] =
- h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = 0x80;
+ 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] )
+ ( (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( h->sh.b_mbaff )
+ if( b_mbaff )
{
- h->mb.pic.i_fref[0] = h->i_ref0 << h->mb.b_interlaced;
- h->mb.pic.i_fref[1] = h->i_ref1 << h->mb.b_interlaced;
- h->mb.cache.i_neighbour_interlaced =
- !!(h->mb.i_neighbour & MB_LEFT)
- + !!(h->mb.i_neighbour & MB_TOP);
+ 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( !h->mb.b_interlaced )
+ if( !b_mbaff )
{
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_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 );
- x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 0, 0 );
- x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 1, 0 );
+ 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
+ {
+ 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, 1 );
- x264_macroblock_load_pic_pointers( h, mb_x, mb_y, 1, 1 );
+ 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 i = 0; i < h->mb.pic.i_fref[0]; i++ )
- h->mb.pic.p_integral[0][i] = &h->fref0[i]->integral[offset];
- for( int i = 0; i < h->mb.pic.i_fref[1]; i++ )
- h->mb.pic.p_integral[1][i] = &h->fref1[i]->integral[offset];
+ 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 );
int i8 = x264_scan8[0] - 1 - 1*8;
if( h->mb.i_neighbour & MB_TOPLEFT )
{
- h->mb.cache.ref[l][i8] = ref[top_8x8 - 1];
- CP32( h->mb.cache.mv[l][i8], mv[top_4x4 - 1] );
+ 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
{
i8 = x264_scan8[0] + 4 - 1*8;
if( h->mb.i_neighbour & MB_TOPRIGHT )
{
- h->mb.cache.ref[l][i8] = ref[top_8x8 + 2];
- CP32( h->mb.cache.mv[l][i8], mv[top_4x4 + 4] );
+ 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 )
{
- 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] );
+ 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
{
}
}
+ /* 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];
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 )
{
- CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left][4] );
- CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left][5] );
- CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left][6] );
- CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left][3] );
+ 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
- for( int i = 0; i < 4; i++ )
- M16( h->mb.cache.mvd[l][x264_scan8[0]-1+i*8] ) = 0;
+ 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[h->mb.b_interlaced&(mb_y&1)];
- h->mb.dist_scale_factor = h->mb.dist_scale_factor_buf[h->mb.b_interlaced&(mb_y&1)];
+ 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 );
- skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left] : 0;
- h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
- h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
+ 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;
| ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOP|MB_TOPLEFT : 0);
}
-void x264_macroblock_cache_load_neighbours_deblock( x264_t *h, int mb_x, int mb_y )
+void x264_macroblock_cache_load_progressive( x264_t *h, int mb_x, int mb_y )
{
- int deblock_on_slice_edges = h->sh.i_disable_deblocking_filter_idc != 2;
+ x264_macroblock_cache_load( h, mb_x, mb_y, 0 );
+}
- h->mb.i_neighbour = 0;
- h->mb.i_mb_xy = mb_y * h->mb.i_mb_stride + mb_x;
+void x264_macroblock_cache_load_interlaced( x264_t *h, int mb_x, int mb_y )
+{
+ x264_macroblock_cache_load( h, mb_x, mb_y, 1 );
+}
- if( mb_x > 0 )
+static void x264_macroblock_deblock_strength_mbaff( x264_t *h, uint8_t (*bs)[8][4] )
+{
+ if( (h->mb.i_neighbour & MB_LEFT) && h->mb.field[h->mb.i_mb_left_xy[0]] != MB_INTERLACED )
{
- h->mb.i_mb_left_xy = h->mb.i_mb_xy - 1;
- if( deblock_on_slice_edges || h->mb.slice_table[h->mb.i_mb_left_xy] == h->mb.slice_table[h->mb.i_mb_xy] )
- h->mb.i_neighbour |= MB_LEFT;
+ static const uint8_t offset[2][2][8] =
+ { { { 0, 0, 0, 0, 1, 1, 1, 1 },
+ { 2, 2, 2, 2, 3, 3, 3, 3 }, },
+ { { 0, 1, 2, 3, 0, 1, 2, 3 },
+ { 0, 1, 2, 3, 0, 1, 2, 3 }, }
+ };
+ ALIGNED_ARRAY_8( uint8_t, tmpbs, [8] );
+
+ const uint8_t *off = offset[MB_INTERLACED][h->mb.i_mb_y&1];
+ uint8_t (*nnz)[48] = h->mb.non_zero_count;
+
+ for( int i = 0; i < 8; i++ )
+ {
+ int left = h->mb.i_mb_left_xy[MB_INTERLACED ? i>>2 : i&1];
+ int nnz_this = h->mb.cache.non_zero_count[x264_scan8[0]+8*(i>>1)];
+ int nnz_left = nnz[left][3 + 4*off[i]];
+ if( !h->param.b_cabac && h->pps->b_transform_8x8_mode )
+ {
+ int j = off[i]&~1;
+ if( h->mb.mb_transform_size[left] )
+ nnz_left = !!(M16( &nnz[left][2+4*j] ) | M16( &nnz[left][2+4*(1+j)] ));
+ }
+ tmpbs[i] = (nnz_left || nnz_this) ? 2 : 1;
+ }
+
+ if( MB_INTERLACED )
+ {
+ CP32( bs[0][0], &tmpbs[0] );
+ CP32( bs[0][4], &tmpbs[4] );
+ }
+ else
+ {
+ for( int i = 0; i < 4; i++ ) bs[0][0][i] = tmpbs[2*i];
+ for( int i = 0; i < 4; i++ ) bs[0][4][i] = tmpbs[1+2*i];
+ }
}
- if( mb_y > h->mb.b_interlaced )
+ if( (h->mb.i_neighbour & MB_TOP) && MB_INTERLACED != h->mb.field[h->mb.i_mb_top_xy] )
{
- h->mb.i_mb_top_xy = h->mb.i_mb_xy - (h->mb.i_mb_stride << h->mb.b_interlaced);
- if( deblock_on_slice_edges || h->mb.slice_table[h->mb.i_mb_top_xy] == h->mb.slice_table[h->mb.i_mb_xy] )
- h->mb.i_neighbour |= MB_TOP;
+ if( !(h->mb.i_mb_y&1) && !MB_INTERLACED )
+ {
+ /* Need to filter both fields (even for frame macroblocks).
+ * Filter top two rows using the top macroblock of the above
+ * pair and then the bottom one. */
+ int mbn_xy = h->mb.i_mb_xy - 2 * h->mb.i_mb_stride;
+ uint8_t *nnz_cur = &h->mb.cache.non_zero_count[x264_scan8[0]];
+
+ for( int j = 0; j < 2; j++, mbn_xy += h->mb.i_mb_stride )
+ {
+ uint8_t (*nnz)[48] = h->mb.non_zero_count;
+
+ ALIGNED_4( uint8_t nnz_top[4] );
+ CP32( nnz_top, &nnz[mbn_xy][3*4] );
+
+ if( !h->param.b_cabac && h->pps->b_transform_8x8_mode && h->mb.mb_transform_size[mbn_xy] )
+ {
+ int nnz_top0 = M16( &nnz[mbn_xy][8] ) | M16( &nnz[mbn_xy][12] );
+ int nnz_top1 = M16( &nnz[mbn_xy][10] ) | M16( &nnz[mbn_xy][14] );
+ nnz_top[0] = nnz_top[1] = nnz_top0 ? 0x0101 : 0;
+ nnz_top[2] = nnz_top[3] = nnz_top1 ? 0x0101 : 0;
+ }
+
+ for( int i = 0; i < 4; i++ )
+ bs[1][4*j][i] = (nnz_cur[i] || nnz_top[i]) ? 2 : 1;
+ }
+ }
+ else
+ for( int i = 0; i < 4; i++ )
+ bs[1][0][i] = X264_MAX( bs[1][0][i], 1 );
}
}
-void x264_macroblock_cache_load_deblock( x264_t *h )
+void x264_macroblock_deblock_strength( x264_t *h )
{
- if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
+ uint8_t (*bs)[8][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
+ if( IS_INTRA( h->mb.i_type ) )
+ {
+ memset( bs[0][1], 3, 3*4*sizeof(uint8_t) );
+ memset( bs[1][1], 3, 3*4*sizeof(uint8_t) );
return;
+ }
+
+ /* Early termination: in this case, nnz guarantees all edges use strength 2.*/
+ if( h->mb.b_transform_8x8 && !CHROMA444 )
+ {
+ int cbp_mask = 0xf >> h->mb.chroma_v_shift;
+ if( (h->mb.i_cbp_luma&cbp_mask) == cbp_mask )
+ {
+ M32( bs[0][0] ) = 0x02020202;
+ M32( bs[0][2] ) = 0x02020202;
+ M32( bs[0][4] ) = 0x02020202;
+ memset( bs[1][0], 2, 5*4*sizeof(uint8_t) ); /* [1][1] and [1][3] has to be set for 4:2:2 */
+ return;
+ }
+ }
+
+ int neighbour_changed = 0;
+ if( h->sh.i_disable_deblocking_filter_idc != 2 )
+ {
+ neighbour_changed = h->mb.i_neighbour_frame&~h->mb.i_neighbour;
+ h->mb.i_neighbour = h->mb.i_neighbour_frame;
+ }
+
+ /* MBAFF deblock uses different left neighbors from encoding */
+ if( SLICE_MBAFF && (h->mb.i_neighbour & MB_LEFT) && (h->mb.field[h->mb.i_mb_xy - 1] != MB_INTERLACED) )
+ {
+ h->mb.i_mb_left_xy[1] =
+ h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1;
+ if( h->mb.i_mb_y&1 )
+ h->mb.i_mb_left_xy[0] -= h->mb.i_mb_stride;
+ else
+ h->mb.i_mb_left_xy[1] += h->mb.i_mb_stride;
+ }
/* If we have multiple slices and we're deblocking on slice edges, we
* have to reload neighbour data. */
- if( h->sh.i_first_mb && h->sh.i_disable_deblocking_filter_idc != 2 )
+ if( neighbour_changed )
{
- int old_neighbour = h->mb.i_neighbour;
- int mb_x = h->mb.i_mb_x;
- int mb_y = h->mb.i_mb_y;
- x264_macroblock_cache_load_neighbours_deblock( h, mb_x, mb_y );
- int new_neighbour = h->mb.i_neighbour;
- h->mb.i_neighbour &= ~old_neighbour;
- if( h->mb.i_neighbour )
- {
- int top_y = mb_y - (1 << h->mb.b_interlaced);
- int top_8x8 = (2*top_y+1) * h->mb.i_b8_stride + 2*mb_x;
- int top_4x4 = (4*top_y+3) * h->mb.i_b4_stride + 4*mb_x;
- int s8x8 = h->mb.i_b8_stride;
- int s4x4 = h->mb.i_b4_stride;
+ int top_y = h->mb.i_mb_top_y;
+ int top_8x8 = (2*top_y+1) * h->mb.i_b8_stride + 2*h->mb.i_mb_x;
+ int top_4x4 = (4*top_y+3) * h->mb.i_b4_stride + 4*h->mb.i_mb_x;
+ int s8x8 = h->mb.i_b8_stride;
+ int s4x4 = h->mb.i_b4_stride;
- uint8_t (*nnz)[24] = h->mb.non_zero_count;
+ uint8_t (*nnz)[48] = h->mb.non_zero_count;
+ const x264_left_table_t *left_index_table = SLICE_MBAFF ? h->mb.left_index_table : &left_indices[3];
- if( h->mb.i_neighbour & MB_TOP )
- CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &nnz[h->mb.i_mb_top_xy][12] );
+ if( neighbour_changed & MB_TOP )
+ CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &nnz[h->mb.i_mb_top_xy][12] );
- if( h->mb.i_neighbour & MB_LEFT )
+ if( neighbour_changed & MB_LEFT )
+ {
+ int *left = h->mb.i_mb_left_xy;
+ h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left[0]][left_index_table->nnz[0]];
+ h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left[0]][left_index_table->nnz[1]];
+ h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left[1]][left_index_table->nnz[2]];
+ h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left[1]][left_index_table->nnz[3]];
+ }
+
+ for( int l = 0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
+ {
+ int16_t (*mv)[2] = h->mb.mv[l];
+ int8_t *ref = h->mb.ref[l];
+
+ int i8 = x264_scan8[0] - 8;
+ if( neighbour_changed & MB_TOP )
{
- int left = h->mb.i_mb_left_xy;
- h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][3];
- h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][7];
- h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][11];
- h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][15];
+ 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] );
}
- for( int l = 0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
+ i8 = x264_scan8[0] - 1;
+ if( neighbour_changed & MB_LEFT )
{
- int16_t (*mv)[2] = h->mb.mv[l];
- int8_t *ref = h->mb.ref[l];
-
- int 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] );
- }
-
- i8 = x264_scan8[0] - 1;
- if( h->mb.i_neighbour & MB_LEFT )
- {
- int ir = h->mb.i_b8_xy - 1;
- 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] );
- }
+ h->mb.cache.ref[l][i8+0*8] =
+ h->mb.cache.ref[l][i8+1*8] = ref[h->mb.left_b8[0] + 1 + s8x8*left_index_table->ref[0]];
+ h->mb.cache.ref[l][i8+2*8] =
+ h->mb.cache.ref[l][i8+3*8] = ref[h->mb.left_b8[1] + 1 + s8x8*left_index_table->ref[2]];
+
+ CP32( h->mb.cache.mv[l][i8+0*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table->mv[0]] );
+ CP32( h->mb.cache.mv[l][i8+1*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table->mv[1]] );
+ CP32( h->mb.cache.mv[l][i8+2*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table->mv[2]] );
+ CP32( h->mb.cache.mv[l][i8+3*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table->mv[3]] );
}
}
- h->mb.i_neighbour = new_neighbour;
}
- if( h->param.analyse.i_weighted_pred && h->sh.i_type == SLICE_TYPE_P )
+ if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->sh.i_type == SLICE_TYPE_P )
{
/* Handle reference frame duplicates */
int i8 = x264_scan8[0] - 8;
/* Munge NNZ for cavlc + 8x8dct */
if( !h->param.b_cabac && h->pps->b_transform_8x8_mode )
{
- uint8_t (*nnz)[24] = h->mb.non_zero_count;
+ uint8_t (*nnz)[48] = h->mb.non_zero_count;
int top = h->mb.i_mb_top_xy;
- int left = h->mb.i_mb_left_xy;
+ int *left = h->mb.i_mb_left_xy;
if( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] )
{
M16( &h->mb.cache.non_zero_count[i8+2] ) = nnz_top1 ? 0x0101 : 0;
}
- if( (h->mb.i_neighbour & MB_LEFT) && h->mb.mb_transform_size[left] )
+ if( h->mb.i_neighbour & MB_LEFT )
{
int i8 = x264_scan8[0] - 1;
- int nnz_left0 = M16( &nnz[left][2] ) | M16( &nnz[left][6] );
- int nnz_left1 = M16( &nnz[left][10] ) | M16( &nnz[left][14] );
- h->mb.cache.non_zero_count[i8+8*0] = !!nnz_left0;
- h->mb.cache.non_zero_count[i8+8*1] = !!nnz_left0;
- h->mb.cache.non_zero_count[i8+8*2] = !!nnz_left1;
- h->mb.cache.non_zero_count[i8+8*3] = !!nnz_left1;
+ if( h->mb.mb_transform_size[left[0]] )
+ {
+ int nnz_left0 = M16( &nnz[left[0]][2] ) | M16( &nnz[left[0]][6] );
+ h->mb.cache.non_zero_count[i8+8*0] = !!nnz_left0;
+ h->mb.cache.non_zero_count[i8+8*1] = !!nnz_left0;
+ }
+ if( h->mb.mb_transform_size[left[1]] )
+ {
+ int nnz_left1 = M16( &nnz[left[1]][10] ) | M16( &nnz[left[1]][14] );
+ h->mb.cache.non_zero_count[i8+8*2] = !!nnz_left1;
+ h->mb.cache.non_zero_count[i8+8*3] = !!nnz_left1;
+ }
}
- if( h->mb.mb_transform_size[h->mb.i_mb_xy] )
+ if( h->mb.b_transform_8x8 )
{
int nnz0 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 0]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[ 2]] );
int nnz1 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 4]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[ 6]] );
M32( &h->mb.cache.non_zero_count[x264_scan8[0]+8*3] ) = nnzbot;
}
}
+
+ h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
+ bs, 4 >> MB_INTERLACED, h->sh.i_type == SLICE_TYPE_B );
+
+ if( SLICE_MBAFF )
+ x264_macroblock_deblock_strength_mbaff( h, bs );
}
-static void ALWAYS_INLINE twiddle_topleft_pixel( pixel *dst, pixel *src, int b_interlaced )
+static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb_y, int i, int b_chroma, int b_mbaff )
{
- // We update intra_border_backup in-place, so the topleft neighbor will no longer
- // exist there when load_pic_pointers wants it. Move it within p_fdec instead.
- if( b_interlaced )
- {
- dst[0] = dst[-1];
- dst[-1] = src[0];
- }
+ int height = b_chroma ? 16>>h->mb.chroma_v_shift : 16;
+ int i_stride = h->fdec->i_stride[i];
+ int i_stride2 = i_stride << (b_mbaff && MB_INTERLACED);
+ int i_pix_offset = (b_mbaff && MB_INTERLACED)
+ ? 16 * mb_x + height * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
+ : 16 * mb_x + height * mb_y * i_stride;
+ if( b_chroma )
+ h->mc.store_interleave_chroma( &h->fdec->plane[1][i_pix_offset], i_stride2, h->mb.pic.p_fdec[1], h->mb.pic.p_fdec[2], height );
else
- dst[0] = src[0];
+ h->mc.copy[PIXEL_16x16]( &h->fdec->plane[i][i_pix_offset], i_stride2, h->mb.pic.p_fdec[i], FDEC_STRIDE, 16 );
}
-static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb_y, int i, int b_interlaced )
+static void ALWAYS_INLINE x264_macroblock_backup_intra( x264_t *h, int mb_x, int mb_y, int b_mbaff )
{
- int w = i ? 8 : 16;
- int i_stride = h->fdec->i_stride[i];
- int i_stride2 = i_stride << b_interlaced;
- int i_pix_offset = b_interlaced
- ? 16 * mb_x + w * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
- : 16 * mb_x + w * mb_y * i_stride;
- pixel *intra_fdec = &h->intra_border_backup[mb_y & h->sh.b_mbaff][i][mb_x*16];
- if( i )
+ /* In MBAFF we store the last two rows in intra_border_backup[0] and [1].
+ * For progressive mbs this is the bottom two rows, and for interlaced the
+ * bottom row of each field. We also store samples needed for the next
+ * mbpair in intra_border_backup[2]. */
+ int backup_dst = !b_mbaff ? 0 : (mb_y&1) ? 1 : MB_INTERLACED ? 0 : 2;
+ memcpy( &h->intra_border_backup[backup_dst][0][mb_x*16 ], h->mb.pic.p_fdec[0]+FDEC_STRIDE*15, 16*sizeof(pixel) );
+ if( CHROMA444 )
{
- h->mc.store_interleave_8x8x2( &h->fdec->plane[1][i_pix_offset], i_stride2, h->mb.pic.p_fdec[1], h->mb.pic.p_fdec[2] );
- memcpy( intra_fdec, h->mb.pic.p_fdec[1]+FDEC_STRIDE*7, 8*sizeof(pixel) );
- memcpy( intra_fdec+8, h->mb.pic.p_fdec[2]+FDEC_STRIDE*7, 8*sizeof(pixel) );
- twiddle_topleft_pixel( h->mb.pic.p_fdec[1]-FDEC_STRIDE-1, h->mb.pic.p_fdec[1]-FDEC_STRIDE+7, b_interlaced );
- twiddle_topleft_pixel( h->mb.pic.p_fdec[2]-FDEC_STRIDE-1, h->mb.pic.p_fdec[2]-FDEC_STRIDE+7, b_interlaced );
+ memcpy( &h->intra_border_backup[backup_dst][1][mb_x*16 ], h->mb.pic.p_fdec[1]+FDEC_STRIDE*15, 16*sizeof(pixel) );
+ memcpy( &h->intra_border_backup[backup_dst][2][mb_x*16 ], h->mb.pic.p_fdec[2]+FDEC_STRIDE*15, 16*sizeof(pixel) );
}
else
{
- h->mc.copy[PIXEL_16x16]( &h->fdec->plane[0][i_pix_offset], i_stride2, h->mb.pic.p_fdec[0], FDEC_STRIDE, 16 );
- memcpy( intra_fdec, h->mb.pic.p_fdec[0]+FDEC_STRIDE*15, 16*sizeof(pixel) );
- twiddle_topleft_pixel( h->mb.pic.p_fdec[0]-FDEC_STRIDE-1, h->mb.pic.p_fdec[0]-FDEC_STRIDE+15, b_interlaced );
+ int backup_src = (15>>h->mb.chroma_v_shift) * FDEC_STRIDE;
+ memcpy( &h->intra_border_backup[backup_dst][1][mb_x*16 ], h->mb.pic.p_fdec[1]+backup_src, 8*sizeof(pixel) );
+ memcpy( &h->intra_border_backup[backup_dst][1][mb_x*16+8], h->mb.pic.p_fdec[2]+backup_src, 8*sizeof(pixel) );
+ }
+ if( b_mbaff )
+ {
+ if( mb_y&1 )
+ {
+ int backup_src = (MB_INTERLACED ? 7 : 14) * FDEC_STRIDE;
+ backup_dst = MB_INTERLACED ? 2 : 0;
+ memcpy( &h->intra_border_backup[backup_dst][0][mb_x*16 ], h->mb.pic.p_fdec[0]+backup_src, 16*sizeof(pixel) );
+ if( CHROMA444 )
+ {
+ memcpy( &h->intra_border_backup[backup_dst][1][mb_x*16 ], h->mb.pic.p_fdec[1]+backup_src, 16*sizeof(pixel) );
+ memcpy( &h->intra_border_backup[backup_dst][2][mb_x*16 ], h->mb.pic.p_fdec[2]+backup_src, 16*sizeof(pixel) );
+ }
+ else
+ {
+ if( CHROMA_FORMAT == CHROMA_420 )
+ backup_src = (MB_INTERLACED ? 3 : 6) * FDEC_STRIDE;
+ memcpy( &h->intra_border_backup[backup_dst][1][mb_x*16 ], h->mb.pic.p_fdec[1]+backup_src, 8*sizeof(pixel) );
+ memcpy( &h->intra_border_backup[backup_dst][1][mb_x*16+8], h->mb.pic.p_fdec[2]+backup_src, 8*sizeof(pixel) );
+ }
+ }
+ }
+ else
+ {
+ /* In progressive we update intra_border_backup in-place, so the topleft neighbor will
+ * no longer exist there when load_pic_pointers wants it. Move it within p_fdec instead. */
+ h->mb.pic.p_fdec[0][-FDEC_STRIDE-1] = h->mb.pic.p_fdec[0][-FDEC_STRIDE+15];
+ h->mb.pic.p_fdec[1][-FDEC_STRIDE-1] = h->mb.pic.p_fdec[1][-FDEC_STRIDE+(15>>h->mb.chroma_h_shift)];
+ h->mb.pic.p_fdec[2][-FDEC_STRIDE-1] = h->mb.pic.p_fdec[2][-FDEC_STRIDE+(15>>h->mb.chroma_h_shift)];
}
}
int8_t *i4x4 = h->mb.intra4x4_pred_mode[i_mb_xy];
uint8_t *nnz = h->mb.non_zero_count[i_mb_xy];
- if( h->mb.b_interlaced )
+ if( SLICE_MBAFF )
{
- x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 0, 1 );
- x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 1 );
+ x264_macroblock_backup_intra( h, h->mb.i_mb_x, h->mb.i_mb_y, 1 );
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 0, 0, 1 );
+ if( CHROMA444 )
+ {
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 0, 1 );
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 2, 0, 1 );
+ }
+ else
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 1, 1 );
}
else
{
- x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 0, 0 );
- x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 0 );
+ x264_macroblock_backup_intra( h, h->mb.i_mb_x, h->mb.i_mb_y, 0 );
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 0, 0, 0 );
+ if( CHROMA444 )
+ {
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 0, 0 );
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 2, 0, 0 );
+ }
+ else
+ x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 1, 0 );
}
x264_prefetch_fenc( h, h->fdec, h->mb.i_mb_x, h->mb.i_mb_y );
{
h->mb.qp[i_mb_xy] = 0;
h->mb.i_last_dqp = 0;
- h->mb.i_cbp_chroma = 2;
+ h->mb.i_cbp_chroma = CHROMA444 ? 0 : 2;
h->mb.i_cbp_luma = 0xf;
- h->mb.cbp[i_mb_xy] = 0x72f; /* all set */
+ h->mb.cbp[i_mb_xy] = (h->mb.i_cbp_chroma << 4) | h->mb.i_cbp_luma | 0x700;
h->mb.b_transform_8x8 = 0;
- for( int i = 0; i < 24; i++ )
+ for( int i = 0; i < 48; i++ )
h->mb.cache.non_zero_count[x264_scan8[i]] = h->param.b_cabac ? 1 : 16;
}
else
}
/* save non zero count */
- CP32( &nnz[0*4], &h->mb.cache.non_zero_count[x264_scan8[0]+0*8] );
- CP32( &nnz[1*4], &h->mb.cache.non_zero_count[x264_scan8[0]+1*8] );
- CP32( &nnz[2*4], &h->mb.cache.non_zero_count[x264_scan8[0]+2*8] );
- CP32( &nnz[3*4], &h->mb.cache.non_zero_count[x264_scan8[0]+3*8] );
- M16( &nnz[16+0*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+0*2]-1] ) >> 8;
- M16( &nnz[16+1*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+1*2]-1] ) >> 8;
- M16( &nnz[16+2*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+2*2]-1] ) >> 8;
- M16( &nnz[16+3*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+3*2]-1] ) >> 8;
+ 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]] );
+ 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]] );
+ 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]] );
+ if( CHROMA_FORMAT >= CHROMA_422 )
+ {
+ CP32( &nnz[16+2*4], &h->mb.cache.non_zero_count[x264_scan8[16+ 8]] );
+ CP32( &nnz[16+3*4], &h->mb.cache.non_zero_count[x264_scan8[16+10]] );
+ CP32( &nnz[32+2*4], &h->mb.cache.non_zero_count[x264_scan8[32+ 8]] );
+ CP32( &nnz[32+3*4], &h->mb.cache.non_zero_count[x264_scan8[32+10]] );
+ }
if( h->mb.i_cbp_luma == 0 && h->mb.i_type != I_8x8 )
h->mb.b_transform_8x8 = 0;
uint8_t (*mvd0)[2] = h->mb.mvd[0][i_mb_xy];
uint8_t (*mvd1)[2] = h->mb.mvd[1][i_mb_xy];
if( IS_INTRA(i_mb_type) && i_mb_type != I_PCM )
- h->mb.chroma_pred_mode[i_mb_xy] = x264_mb_pred_mode8x8c_fix[ h->mb.i_chroma_pred_mode ];
+ h->mb.chroma_pred_mode[i_mb_xy] = x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode];
else
h->mb.chroma_pred_mode[i_mb_xy] = I_PRED_CHROMA_DC;
void x264_macroblock_bipred_init( x264_t *h )
{
- for( int field = 0; field <= h->sh.b_mbaff; field++ )
- for( int i_ref0 = 0; i_ref0 < (h->i_ref0<<h->sh.b_mbaff); i_ref0++ )
- {
- int poc0 = h->fref0[i_ref0>>h->sh.b_mbaff]->i_poc;
- if( h->sh.b_mbaff && field^(i_ref0&1) )
- poc0 += h->sh.i_delta_poc_bottom;
- for( int i_ref1 = 0; i_ref1 < (h->i_ref1<<h->sh.b_mbaff); i_ref1++ )
+ for( int mbfield = 0; mbfield <= SLICE_MBAFF; mbfield++ )
+ for( int field = 0; field <= SLICE_MBAFF; field++ )
+ for( int i_ref0 = 0; i_ref0 < (h->i_ref[0]<<mbfield); i_ref0++ )
{
- int dist_scale_factor;
- int poc1 = h->fref1[i_ref1>>h->sh.b_mbaff]->i_poc;
- if( h->sh.b_mbaff && field^(i_ref1&1) )
- poc1 += h->sh.i_delta_poc_bottom;
- int cur_poc = h->fdec->i_poc + field*h->sh.i_delta_poc_bottom;
- int td = x264_clip3( poc1 - poc0, -128, 127 );
- if( td == 0 /* || pic0 is a long-term ref */ )
- dist_scale_factor = 256;
- else
+ x264_frame_t *l0 = h->fref[0][i_ref0>>mbfield];
+ int poc0 = l0->i_poc + mbfield*l0->i_delta_poc[field^(i_ref0&1)];
+ for( int i_ref1 = 0; i_ref1 < (h->i_ref[1]<<mbfield); i_ref1++ )
{
- int tb = x264_clip3( cur_poc - poc0, -128, 127 );
- int tx = (16384 + (abs(td) >> 1)) / td;
- dist_scale_factor = x264_clip3( (tb * tx + 32) >> 6, -1024, 1023 );
+ int dist_scale_factor;
+ x264_frame_t *l1 = h->fref[1][i_ref1>>mbfield];
+ int cur_poc = h->fdec->i_poc + mbfield*h->fdec->i_delta_poc[field];
+ int poc1 = l1->i_poc + mbfield*l1->i_delta_poc[field^(i_ref1&1)];
+ int td = x264_clip3( poc1 - poc0, -128, 127 );
+ if( td == 0 /* || pic0 is a long-term ref */ )
+ dist_scale_factor = 256;
+ else
+ {
+ int tb = x264_clip3( cur_poc - poc0, -128, 127 );
+ int tx = (16384 + (abs(td) >> 1)) / td;
+ dist_scale_factor = x264_clip3( (tb * tx + 32) >> 6, -1024, 1023 );
+ }
+
+ h->mb.dist_scale_factor_buf[mbfield][field][i_ref0][i_ref1] = dist_scale_factor;
+
+ dist_scale_factor >>= 2;
+ if( h->param.analyse.b_weighted_bipred
+ && dist_scale_factor >= -64
+ && dist_scale_factor <= 128 )
+ {
+ h->mb.bipred_weight_buf[mbfield][field][i_ref0][i_ref1] = 64 - dist_scale_factor;
+ // ssse3 implementation of biweight doesn't support the extrema.
+ // if we ever generate them, we'll have to drop that optimization.
+ assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
+ }
+ else
+ h->mb.bipred_weight_buf[mbfield][field][i_ref0][i_ref1] = 32;
}
-
- h->mb.dist_scale_factor_buf[field][i_ref0][i_ref1] = dist_scale_factor;
-
- dist_scale_factor >>= 2;
- if( h->param.analyse.b_weighted_bipred
- && dist_scale_factor >= -64
- && dist_scale_factor <= 128 )
- {
- h->mb.bipred_weight_buf[field][i_ref0][i_ref1] = 64 - dist_scale_factor;
- // ssse3 implementation of biweight doesn't support the extrema.
- // if we ever generate them, we'll have to drop that optimization.
- assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
- }
- else
- h->mb.bipred_weight_buf[field][i_ref0][i_ref1] = 32;
}
- }
}