Commit 60455fff authored by Fiona Glaser's avatar Fiona Glaser

Rework subme system, add RD refinement in B-frames

The new system is as follows: subme6 is RD in I/P frames, subme7 is RD in all frames, subme8 is RD refinement in I/P frames, and subme9 is RD refinement in all frames.
subme6 == old subme6, subme7 == old subme6+brdo, subme8 == old subme7+brdo, subme9 == no equivalent
--b-rdo has, accordingly, been removed.  --bime has also been removed, and instead enabled automatically at subme >= 5.
RD refinement in B-frames (subme9) includes both qpel-RD and an RD version of bime.
parent 9b10152f
......@@ -112,10 +112,10 @@ SRC2 = $(SRCS) $(SRCCLI)
OPT0 = --crf 30 -b1 -m1 -r1 --me dia --no-cabac --pre-scenecut --direct temporal --no-ssim --no-psnr
OPT1 = --crf 16 -b2 -m3 -r3 --me hex -8 --direct spatial --no-dct-decimate
OPT2 = --crf 26 -b2 -m5 -r2 --me hex -8 -w --cqm jvt --nr 100
OPT3 = --crf 18 -b3 -m7 -r5 --me umh -8 -t1 -A all --mixed-refs --b-rdo -w --b-pyramid --direct auto --bime --no-fast-pskip
OPT4 = --crf 22 -b3 -m6 -r4 --me esa -8 -t2 -A all --mixed-refs --b-rdo --bime
OPT5 = --frames 50 --crf 24 -b3 -m7 -r3 --me tesa -8 -t1 --mixed-refs --b-rdo --bime
OPT6 = --frames 50 -q0 -m6 -r2 --me hex -Aall
OPT3 = --crf 18 -b3 -m9 -r5 --me umh -8 -t1 -A all --mixed-refs -w --b-pyramid --direct auto --no-fast-pskip
OPT4 = --crf 22 -b3 -m7 -r4 --me esa -8 -t2 -A all --mixed-refs
OPT5 = --frames 50 --crf 24 -b3 -m9 -r3 --me tesa -8 -t1 --mixed-refs
OPT6 = --frames 50 -q0 -m9 -r2 --me hex -Aall
OPT7 = --frames 50 -q0 -m2 -r1 --me hex --no-cabac
ifeq (,$(VIDS))
......
......@@ -487,12 +487,8 @@ int x264_param_parse( x264_param_t *p, const char *name, const char *value )
p->analyse.f_psy_trellis = 0;
}
}
OPT("bime")
p->analyse.b_bidir_me = atobool(value);
OPT("chroma-me")
p->analyse.b_chroma_me = atobool(value);
OPT2("b-rdo", "brdo")
p->analyse.b_bframe_rdo = atobool(value);
OPT("mixed-refs")
p->analyse.b_mixed_references = atobool(value);
OPT("trellis")
......@@ -842,7 +838,6 @@ char *x264_param2string( x264_param_t *p, int b_res )
s += sprintf( s, " me=%s", x264_motion_est_names[ p->analyse.i_me_method ] );
s += sprintf( s, " subme=%d", p->analyse.i_subpel_refine );
s += sprintf( s, " psy_rd=%.1f:%.1f", p->analyse.f_psy_rd, p->analyse.f_psy_trellis );
s += sprintf( s, " brdo=%d", p->analyse.b_bframe_rdo );
s += sprintf( s, " mixed_ref=%d", p->analyse.b_mixed_references );
s += sprintf( s, " me_range=%d", p->analyse.i_me_range );
s += sprintf( s, " chroma_me=%d", p->analyse.b_chroma_me );
......@@ -859,10 +854,9 @@ char *x264_param2string( x264_param_t *p, int b_res )
s += sprintf( s, " bframes=%d", p->i_bframe );
if( p->i_bframe )
{
s += sprintf( s, " b_pyramid=%d b_adapt=%d b_bias=%d direct=%d wpredb=%d bime=%d",
s += sprintf( s, " b_pyramid=%d b_adapt=%d b_bias=%d direct=%d wpredb=%d",
p->b_bframe_pyramid, p->i_bframe_adaptive, p->i_bframe_bias,
p->analyse.i_direct_mv_pred, p->analyse.b_weighted_bipred,
p->analyse.b_bidir_me );
p->analyse.i_direct_mv_pred, p->analyse.b_weighted_bipred );
}
s += sprintf( s, " keyint=%d keyint_min=%d scenecut=%d%s",
......
......@@ -76,7 +76,7 @@ typedef struct
int i_lambda2;
int i_qp;
int16_t *p_cost_mv;
int b_mbrd;
int i_mbrd;
/* I: Intra part */
......@@ -203,19 +203,19 @@ static void x264_mb_analyse_load_costs( x264_t *h, x264_mb_analysis_t *a )
static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int i_qp )
{
int i = h->param.analyse.i_subpel_refine - (h->sh.i_type == SLICE_TYPE_B);
/* mbrd == 1 -> RD mode decision */
/* mbrd == 2 -> RD refinement */
a->i_mbrd = (i>=6) + (i>=8);
/* conduct the analysis using this lamda and QP */
a->i_qp = h->mb.i_qp = i_qp;
h->mb.i_chroma_qp = h->chroma_qp_table[i_qp];
a->i_lambda = x264_lambda_tab[i_qp];
a->i_lambda2 = x264_lambda2_tab[i_qp];
a->b_mbrd = h->param.analyse.i_subpel_refine >= 6 &&
( h->sh.i_type != SLICE_TYPE_B || h->param.analyse.b_bframe_rdo );
h->mb.i_me_method = h->param.analyse.i_me_method;
h->mb.i_subpel_refine = h->param.analyse.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_trellis = h->param.analyse.i_trellis > 1 && a->b_mbrd;
h->mb.b_trellis = h->param.analyse.i_trellis > 1 && a->i_mbrd;
h->mb.b_transform_8x8 = 0;
h->mb.b_noise_reduction = 0;
......@@ -226,12 +226,12 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int i_qp )
a->i_satd_i8x8chroma = COST_MAX;
/* non-RD PCM decision is inaccurate (as is psy-rd), so don't do it */
a->i_satd_pcm = !h->mb.i_psy_rd && a->b_mbrd ? ((uint64_t)X264_PCM_COST*a->i_lambda2 + 128) >> 8 : COST_MAX;
a->i_satd_pcm = !h->mb.i_psy_rd && a->i_mbrd ? ((uint64_t)X264_PCM_COST*a->i_lambda2 + 128) >> 8 : COST_MAX;
a->b_fast_intra = 0;
h->mb.i_skip_intra =
h->mb.b_lossless ? 0 :
a->b_mbrd ? 2 :
a->i_mbrd ? 2 :
!h->param.analyse.i_trellis && !h->param.analyse.i_noise_reduction;
/* II: Inter part P/B frame */
......@@ -650,7 +650,7 @@ static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_
{
DECLARE_ALIGNED_16( uint8_t edge[33] );
x264_pixel_cmp_t sa8d = (h->pixf.mbcmp[0] == h->pixf.satd[0]) ? h->pixf.sa8d[PIXEL_8x8] : h->pixf.mbcmp[PIXEL_8x8];
int i_satd_thresh = a->b_mbrd ? COST_MAX : X264_MIN( i_satd_inter, a->i_satd_i16x16 );
int i_satd_thresh = a->i_mbrd ? COST_MAX : X264_MIN( i_satd_inter, a->i_satd_i16x16 );
int i_cost = 0;
b_merged_satd = h->pixf.intra_sa8d_x3_8x8 && h->pixf.mbcmp[0] == h->pixf.satd[0];
......@@ -728,7 +728,7 @@ static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_
a->i_satd_i8x8 = COST_MAX;
i_cost = i_cost * 4/(idx+1);
}
if( X264_MIN(i_cost, a->i_satd_i16x16) > i_satd_inter*(5+a->b_mbrd)/4 )
if( X264_MIN(i_cost, a->i_satd_i16x16) > i_satd_inter*(5+!!a->i_mbrd)/4 )
return;
}
......@@ -738,7 +738,7 @@ static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_
int i_cost;
int i_satd_thresh = X264_MIN3( i_satd_inter, a->i_satd_i16x16, a->i_satd_i8x8 );
b_merged_satd = h->pixf.intra_satd_x3_4x4 && h->pixf.mbcmp[0] == h->pixf.satd[0];
if( a->b_mbrd )
if( a->i_mbrd )
i_satd_thresh = i_satd_thresh * (10-a->b_fast_intra)/8;
i_cost = a->i_lambda * 24; /* from JVT (SATD0) */
......@@ -1094,7 +1094,7 @@ static void x264_mb_analyse_inter_p16x16( x264_t *h, x264_mb_analysis_t *a )
assert( a->l0.me16x16.mv[1] <= h->mb.mv_max_spel[1] || h->param.i_threads == 1 );
h->mb.i_type = P_L0;
if( a->b_mbrd )
if( a->i_mbrd )
{
x264_mb_cache_fenc_satd( h );
if( a->l0.me16x16.i_ref == 0 && *(uint32_t*)a->l0.me16x16.mv == *(uint32_t*)h->mb.cache.pskip_mv )
......@@ -2029,31 +2029,34 @@ static void x264_mb_analyse_b_rd( x264_t *h, x264_mb_analysis_t *a, int i_satd_i
}
}
static void refine_bidir( x264_t *h, x264_mb_analysis_t *a )
static void x264_refine_bidir( x264_t *h, x264_mb_analysis_t *a )
{
const int i_biweight = h->mb.bipred_weight[a->l0.i_ref][a->l1.i_ref];
int i;
if( IS_INTRA(h->mb.i_type) )
return;
switch( h->mb.i_partition )
{
case D_16x16:
if( h->mb.i_type == B_BI_BI )
x264_me_refine_bidir( h, &a->l0.me16x16, &a->l1.me16x16, i_biweight );
x264_me_refine_bidir_satd( h, &a->l0.me16x16, &a->l1.me16x16, i_biweight );
break;
case D_16x8:
for( i=0; i<2; i++ )
if( a->i_mb_partition16x8[i] == D_BI_8x8 )
x264_me_refine_bidir( h, &a->l0.me16x8[i], &a->l1.me16x8[i], i_biweight );
x264_me_refine_bidir_satd( h, &a->l0.me16x8[i], &a->l1.me16x8[i], i_biweight );
break;
case D_8x16:
for( i=0; i<2; i++ )
if( a->i_mb_partition8x16[i] == D_BI_8x8 )
x264_me_refine_bidir( h, &a->l0.me8x16[i], &a->l1.me8x16[i], i_biweight );
x264_me_refine_bidir_satd( h, &a->l0.me8x16[i], &a->l1.me8x16[i], i_biweight );
break;
case D_8x8:
for( i=0; i<4; i++ )
if( h->mb.i_sub_partition[i] == D_BI_8x8 )
x264_me_refine_bidir( h, &a->l0.me8x8[i], &a->l1.me8x8[i], i_biweight );
x264_me_refine_bidir_satd( h, &a->l0.me8x8[i], &a->l1.me8x8[i], i_biweight );
break;
}
}
......@@ -2120,10 +2123,10 @@ void x264_macroblock_analyse( x264_t *h )
/*--------------------------- Do the analysis ---------------------------*/
if( h->sh.i_type == SLICE_TYPE_I )
{
if( analysis.b_mbrd )
if( analysis.i_mbrd )
x264_mb_cache_fenc_satd( h );
x264_mb_analyse_intra( h, &analysis, COST_MAX );
if( analysis.b_mbrd )
if( analysis.i_mbrd )
x264_intra_rd( h, &analysis, COST_MAX );
i_cost = analysis.i_satd_i16x16;
......@@ -2133,7 +2136,7 @@ void x264_macroblock_analyse( x264_t *h )
if( analysis.i_satd_pcm < i_cost )
h->mb.i_type = I_PCM;
else if( h->mb.i_subpel_refine >= 7 )
else if( analysis.i_mbrd >= 2 )
x264_intra_rd_refine( h, &analysis );
}
else if( h->sh.i_type == SLICE_TYPE_P )
......@@ -2245,7 +2248,7 @@ void x264_macroblock_analyse( x264_t *h )
/* refine qpel */
//FIXME mb_type costs?
if( analysis.b_mbrd )
if( analysis.i_mbrd )
{
/* refine later */
}
......@@ -2324,7 +2327,7 @@ void x264_macroblock_analyse( x264_t *h )
analysis.i_satd_i8x8,
analysis.i_satd_i4x4 );
if( analysis.b_mbrd )
if( analysis.i_mbrd )
{
x264_mb_analyse_p_rd( h, &analysis, X264_MIN(i_satd_inter, i_satd_intra) );
i_type = P_L0;
......@@ -2355,7 +2358,7 @@ void x264_macroblock_analyse( x264_t *h )
h->stat.frame.i_inter_cost += i_cost;
h->stat.frame.i_mbs_analysed++;
if( h->mb.i_subpel_refine >= 7 && h->mb.i_type != I_PCM )
if( analysis.i_mbrd >= 2 && h->mb.i_type != I_PCM )
{
if( IS_INTRA( h->mb.i_type ) )
{
......@@ -2364,7 +2367,7 @@ void x264_macroblock_analyse( x264_t *h )
else if( i_partition == D_16x16 )
{
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, analysis.l0.me16x16.i_ref );
x264_me_refine_qpel_rd( h, &analysis.l0.me16x16, analysis.i_lambda2, 0 );
x264_me_refine_qpel_rd( h, &analysis.l0.me16x16, analysis.i_lambda2, 0, 0 );
}
else if( i_partition == D_16x8 )
{
......@@ -2372,8 +2375,8 @@ void x264_macroblock_analyse( x264_t *h )
h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
x264_macroblock_cache_ref( h, 0, 0, 4, 2, 0, analysis.l0.me16x8[0].i_ref );
x264_macroblock_cache_ref( h, 0, 2, 4, 2, 0, analysis.l0.me16x8[1].i_ref );
x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[0], analysis.i_lambda2, 0 );
x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[1], analysis.i_lambda2, 2 );
x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[0], analysis.i_lambda2, 0, 0 );
x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[1], analysis.i_lambda2, 2, 0 );
}
else if( i_partition == D_8x16 )
{
......@@ -2381,8 +2384,8 @@ void x264_macroblock_analyse( x264_t *h )
h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
x264_macroblock_cache_ref( h, 0, 0, 2, 4, 0, analysis.l0.me8x16[0].i_ref );
x264_macroblock_cache_ref( h, 2, 0, 2, 4, 0, analysis.l0.me8x16[1].i_ref );
x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[0], analysis.i_lambda2, 0 );
x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[1], analysis.i_lambda2, 1 );
x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[0], analysis.i_lambda2, 0, 0 );
x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[1], analysis.i_lambda2, 1, 0 );
}
else if( i_partition == D_8x8 )
{
......@@ -2390,7 +2393,7 @@ void x264_macroblock_analyse( x264_t *h )
x264_analyse_update_cache( h, &analysis );
for( i8x8 = 0; i8x8 < 4; i8x8++ )
if( h->mb.i_sub_partition[i8x8] == D_L0_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i8x8], analysis.i_lambda2, i8x8 );
x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i8x8], analysis.i_lambda2, i8x8, 0 );
}
}
}
......@@ -2400,7 +2403,7 @@ void x264_macroblock_analyse( x264_t *h )
int i_bskip_cost = COST_MAX;
int b_skip = 0;
if( analysis.b_mbrd )
if( analysis.i_mbrd )
x264_mb_cache_fenc_satd( h );
h->mb.i_type = B_SKIP;
......@@ -2436,7 +2439,7 @@ void x264_macroblock_analyse( x264_t *h )
{
/* chance of skip is too small to bother */
}
else if( analysis.b_mbrd )
else if( analysis.i_mbrd )
{
i_bskip_cost = ssd_mb( h );
/* 6 = minimum cavlc cost of a non-skipped MB */
......@@ -2474,7 +2477,7 @@ void x264_macroblock_analyse( x264_t *h )
COPY2_IF_LT( i_cost, analysis.i_cost16x16bi, i_type, B_BI_BI );
COPY2_IF_LT( i_cost, analysis.i_cost16x16direct, i_type, B_DIRECT );
if( analysis.b_mbrd && analysis.i_cost16x16direct <= i_cost * 33/32 )
if( analysis.i_mbrd && analysis.i_cost16x16direct <= i_cost * 33/32 )
{
x264_mb_analyse_b_rd( h, &analysis, i_cost );
if( i_bskip_cost < analysis.i_rd16x16direct &&
......@@ -2516,7 +2519,7 @@ void x264_macroblock_analyse( x264_t *h )
}
}
if( analysis.b_mbrd )
if( analysis.i_mbrd )
{
/* refine later */
}
......@@ -2599,7 +2602,7 @@ void x264_macroblock_analyse( x264_t *h )
}
}
if( analysis.b_mbrd )
if( analysis.i_mbrd )
{
i_satd_inter = i_cost;
x264_mb_analyse_b_rd( h, &analysis, i_satd_inter );
......@@ -2620,7 +2623,7 @@ void x264_macroblock_analyse( x264_t *h )
x264_mb_analyse_intra( h, &analysis, i_satd_inter );
if( analysis.b_mbrd )
if( analysis.i_mbrd )
{
x264_mb_analyse_transform_rd( h, &analysis, &i_satd_inter, &i_cost );
x264_intra_rd( h, &analysis, i_satd_inter * 17/16 );
......@@ -2634,16 +2637,70 @@ void x264_macroblock_analyse( x264_t *h )
h->mb.i_type = i_type;
h->mb.i_partition = i_partition;
if( analysis.b_mbrd && h->mb.i_subpel_refine >= 7 && IS_INTRA( i_type ) && i_type != I_PCM )
if( analysis.i_mbrd >= 2 && IS_INTRA( i_type ) && i_type != I_PCM )
x264_intra_rd_refine( h, &analysis );
else if( h->param.analyse.b_bidir_me )
refine_bidir( h, &analysis );
if( h->mb.i_subpel_refine >= 5 )
x264_refine_bidir( h, &analysis );
if( analysis.i_mbrd >= 2 && i_type > B_DIRECT && i_type < B_SKIP )
{
const int i_biweight = h->mb.bipred_weight[analysis.l0.i_ref][analysis.l1.i_ref];
x264_analyse_update_cache( h, &analysis );
if( i_partition == D_16x16 )
{
if( i_type == B_L0_L0 )
x264_me_refine_qpel_rd( h, &analysis.l0.me16x16, analysis.i_lambda2, 0, 0 );
else if( i_type == B_L1_L1 )
x264_me_refine_qpel_rd( h, &analysis.l1.me16x16, analysis.i_lambda2, 0, 1 );
else if( i_type == B_BI_BI )
x264_me_refine_bidir_rd( h, &analysis.l0.me16x16, &analysis.l1.me16x16, i_biweight, 0, analysis.i_lambda2 );
}
else if( i_partition == D_16x8 )
{
for( i = 0; i < 2; i++ )
{
h->mb.i_sub_partition[i*2] = h->mb.i_sub_partition[i*2+1] = analysis.i_mb_partition16x8[i];
if( analysis.i_mb_partition16x8[i] == D_L0_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[i], analysis.i_lambda2, i*2, 0 );
else if( analysis.i_mb_partition16x8[i] == D_L1_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l1.me16x8[i], analysis.i_lambda2, i*2, 1 );
else if( analysis.i_mb_partition16x8[i] == D_BI_8x8 )
x264_me_refine_bidir_rd( h, &analysis.l0.me16x8[i], &analysis.l1.me16x8[i], i_biweight, i*2, analysis.i_lambda2 );
}
}
else if( i_partition == D_8x16 )
{
for( i = 0; i < 2; i++ )
{
h->mb.i_sub_partition[i] = h->mb.i_sub_partition[i+2] = analysis.i_mb_partition8x16[i];
if( analysis.i_mb_partition8x16[i] == D_L0_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[i], analysis.i_lambda2, i, 0 );
else if( analysis.i_mb_partition8x16[i] == D_L1_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l1.me8x16[i], analysis.i_lambda2, i, 1 );
else if( analysis.i_mb_partition8x16[i] == D_BI_8x8 )
x264_me_refine_bidir_rd( h, &analysis.l0.me8x16[i], &analysis.l1.me8x16[i], i_biweight, i, analysis.i_lambda2 );
}
}
else if( i_partition == D_8x8 )
{
for( i = 0; i < 4; i++ )
{
if( h->mb.i_sub_partition[i] == D_L0_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i], analysis.i_lambda2, i, 0 );
else if( h->mb.i_sub_partition[i] == D_L1_8x8 )
x264_me_refine_qpel_rd( h, &analysis.l1.me8x8[i], analysis.i_lambda2, i, 1 );
else if( h->mb.i_sub_partition[i] == D_BI_8x8 )
x264_me_refine_bidir_rd( h, &analysis.l0.me8x8[i], &analysis.l1.me8x8[i], i_biweight, i, analysis.i_lambda2 );
}
}
}
}
}
x264_analyse_update_cache( h, &analysis );
if( !analysis.b_mbrd )
if( !analysis.i_mbrd )
x264_mb_analyse_transform( h );
h->mb.b_trellis = h->param.analyse.i_trellis;
......
......@@ -1026,47 +1026,34 @@ void x264_macroblock_write_cabac( x264_t *h, x264_cabac_t *cb )
/*****************************************************************************
* RD only; doesn't generate a valid bitstream
* doesn't write cbp or chroma dc (I don't know how much this matters)
* doesn't write ref or subpartition (never varies between calls, so no point in doing so)
* works on all partition sizes except 16x16
* for sub8x8, call once per 8x8 block
*****************************************************************************/
static void x264_partition_size_cabac( x264_t *h, x264_cabac_t *cb, int i8, int i_pixel )
{
const int i_mb_type = h->mb.i_type;
int b_8x16 = h->mb.i_partition == D_8x16;
int j;
if( i_mb_type == P_8x8 )
{
x264_cabac_mb_sub_p_partition( cb, h->mb.i_sub_partition[i8] );
if( h->mb.pic.i_fref[0] > 1 )
x264_cabac_mb_ref( h, cb, 0, 4*i8 );
x264_cabac_mb8x8_mvd( h, cb, 0, i8 );
}
else if( i_mb_type == P_L0 )
x264_cabac_mb_mvd( h, cb, 0, 4*i8, 4>>b_8x16, 2<<b_8x16 );
else if( i_mb_type > B_DIRECT && i_mb_type < B_8x8 )
{
if( h->mb.pic.i_fref[0] > 1 )
x264_cabac_mb_ref( h, cb, 0, 4*i8 );
if( h->mb.i_partition == D_16x8 )
x264_cabac_mb_mvd( h, cb, 0, 4*i8, 4, 2 );
else //8x16
x264_cabac_mb_mvd( h, cb, 0, 4*i8, 2, 4 );
if( x264_mb_type_list0_table[ i_mb_type ][!!i8] ) x264_cabac_mb_mvd( h, cb, 0, 4*i8, 4>>b_8x16, 2<<b_8x16 );
if( x264_mb_type_list1_table[ i_mb_type ][!!i8] ) x264_cabac_mb_mvd( h, cb, 1, 4*i8, 4>>b_8x16, 2<<b_8x16 );
}
else if( i_mb_type == B_8x8 )
{
x264_cabac_mb_sub_b_partition( cb, h->mb.i_sub_partition[i8] );
if( h->mb.pic.i_fref[0] > 1
&& x264_mb_partition_listX_table[0][ h->mb.i_sub_partition[i8] ] )
x264_cabac_mb_ref( h, cb, 0, 4*i8 );
if( h->mb.pic.i_fref[1] > 1
&& x264_mb_partition_listX_table[1][ h->mb.i_sub_partition[i8] ] )
x264_cabac_mb_ref( h, cb, 1, 4*i8 );
x264_cabac_mb8x8_mvd( h, cb, 0, i8 );
x264_cabac_mb8x8_mvd( h, cb, 1, i8 );
}
else
{
x264_log(h, X264_LOG_ERROR, "invalid/unhandled mb_type\n" );
x264_log(h, X264_LOG_ERROR, "invalid/unhandled mb_type %d\n",i_mb_type );
return;
}
......
......@@ -639,6 +639,7 @@ void x264_macroblock_write_cavlc( x264_t *h, bs_t *s )
/*****************************************************************************
* RD only; doesn't generate a valid bitstream
* doesn't write cbp or chroma dc (I don't know how much this matters)
* doesn't write ref or subpartition (never varies between calls, so no point in doing so)
* works on all partition sizes except 16x16
* for sub8x8, call once per 8x8 block
*****************************************************************************/
......@@ -646,37 +647,22 @@ static int x264_partition_size_cavlc( x264_t *h, int i8, int i_pixel )
{
bs_t s;
const int i_mb_type = h->mb.i_type;
int b_8x16 = h->mb.i_partition == D_8x16;
int j;
s.i_bits_encoded = 0;
if( i_mb_type == P_8x8 )
{
bs_write_ue( &s, sub_mb_type_p_to_golomb[ h->mb.i_sub_partition[i8] ] );
if( h->mb.pic.i_fref[0] > 1 )
bs_write_te( &s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[4*i8]] );
cavlc_mb8x8_mvd( h, &s, 0, i8 );
}
else if( i_mb_type == P_L0 )
cavlc_mb_mvd( h, &s, 0, 4*i8, 4>>b_8x16 );
else if( i_mb_type > B_DIRECT && i_mb_type < B_8x8 )
{
if( h->mb.pic.i_fref[0] > 1 )
bs_write_te( &s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[4*i8]] );
if( h->mb.i_partition == D_16x8 )
cavlc_mb_mvd( h, &s, 0, 4*i8, 4 );
else //8x16
cavlc_mb_mvd( h, &s, 0, 4*i8, 2 );
if( x264_mb_type_list0_table[ i_mb_type ][!!i8] ) cavlc_mb_mvd( h, &s, 0, 4*i8, 4>>b_8x16 );
if( x264_mb_type_list1_table[ i_mb_type ][!!i8] ) cavlc_mb_mvd( h, &s, 1, 4*i8, 4>>b_8x16 );
}
else if( i_mb_type == B_8x8 )
{
bs_write_ue( &s, sub_mb_type_b_to_golomb[ h->mb.i_sub_partition[i8] ] );
if( h->mb.pic.i_fref[0] > 1
&& x264_mb_partition_listX_table[0][ h->mb.i_sub_partition[i8] ] )
bs_write_te( &s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[4*i8]] );
if( h->mb.pic.i_fref[1] > 1
&& x264_mb_partition_listX_table[1][ h->mb.i_sub_partition[i8] ] )
bs_write_te( &s, h->mb.pic.i_fref[1] - 1, h->mb.cache.ref[1][x264_scan8[4*i8]] );
cavlc_mb8x8_mvd( h, &s, 0, i8 );
cavlc_mb8x8_mvd( h, &s, 1, i8 );
}
......
......@@ -474,8 +474,7 @@ static int x264_validate_parameters( x264_t *h )
if( h->param.analyse.i_me_method == X264_ME_TESA &&
(h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
h->param.analyse.i_me_method = X264_ME_ESA;
h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 7 );
h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 9 );
h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
......@@ -591,7 +590,6 @@ static int x264_validate_parameters( x264_t *h )
BOOLIFY( b_interlaced );
BOOLIFY( analyse.b_transform_8x8 );
BOOLIFY( analyse.i_direct_8x8_inference );
BOOLIFY( analyse.b_bidir_me );
BOOLIFY( analyse.b_chroma_me );
BOOLIFY( analyse.b_fast_pskip );
BOOLIFY( rc.b_stat_write );
......@@ -816,8 +814,6 @@ int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
COPY( analyse.i_noise_reduction );
COPY( analyse.i_subpel_refine );
COPY( analyse.i_trellis );
COPY( analyse.b_bidir_me );
COPY( analyse.b_bframe_rdo );
COPY( analyse.b_chroma_me );
COPY( analyse.b_dct_decimate );
COPY( analyse.b_fast_pskip );
......
......@@ -29,8 +29,8 @@
* subpel_iters[i_subpel_refine] = { refine_hpel, refine_qpel, me_hpel, me_qpel }
* where me_* are the number of EPZS iterations run on all candidate block types,
* and refine_* are run only on the winner.
* the subme=7 values are much higher because any amount of satd search makes
* up its time by reducing the number of rd iterations. */
* the subme=8,9 values are much higher because any amount of satd search makes
* up its time by reducing the number of qpel-rd iterations. */
static const int subpel_iterations[][4] =
{{1,0,0,0},
{1,1,0,0},
......@@ -39,6 +39,8 @@ static const int subpel_iterations[][4] =
{0,2,1,1},
{0,2,1,2},
{0,0,2,2},
{0,0,2,2},
{0,0,4,10},
{0,0,4,10}};
/* (x-1)%6 */
......@@ -797,6 +799,8 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite
BIME_CACHE(a,b) \
BIME_CACHE(-(a),-(b))
#define SATD_THRESH 17/16
#define COST_BIMV_SATD( m0x, m0y, m1x, m1y ) \
if( pass == 0 || !((visited[(m0x)&7][(m0y)&7][(m1x)&7] & (1<<((m1y)&7)))) ) \
{ \
......@@ -808,7 +812,27 @@ if( pass == 0 || !((visited[(m0x)&7][(m0y)&7][(m1x)&7] & (1<<((m1y)&7)))) ) \
cost = h->pixf.mbcmp[i_pixel]( m0->p_fenc[0], FENC_STRIDE, pix, bw ) \
+ p_cost_m0x[ m0x ] + p_cost_m0y[ m0y ] \
+ p_cost_m1x[ m1x ] + p_cost_m1y[ m1y ]; \
if( cost < bcost ) \
if( rd ) \
{ \
if( cost < bcost * SATD_THRESH ) \
{ \
uint64_t costrd; \
if( cost < bcost ) \
bcost = cost; \
*(uint32_t*)cache0_mv = *(uint32_t*)cache0_mv2 = pack16to32_mask(m0x,m0y); \
*(uint32_t*)cache1_mv = *(uint32_t*)cache1_mv2 = pack16to32_mask(m1x,m1y); \
costrd = x264_rd_cost_part( h, i_lambda2, i8, m0->i_pixel ); \
if( costrd < bcostrd ) \
{\
bcostrd = costrd;\
bm0x = m0x; \
bm0y = m0y; \
bm1x = m1x; \
bm1y = m1y; \
}\
} \
} \
else if( cost < bcost ) \
{ \
bcost = cost; \
bm0x = m0x; \
......@@ -831,8 +855,13 @@ if( pass == 0 || !((visited[(m0x)&7][(m0y)&7][(m1x)&7] & (1<<((m1y)&7)))) ) \
CHECK_BIDIR2(c,d,a,b) \
CHECK_BIDIR2(d,a,b,c)
int x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight )
static void ALWAYS_INLINE x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight, int i8, int i_lambda2, int rd )
{
static const int pixel_mv_offs[] = { 0, 4, 4*8, 0 };
int16_t *cache0_mv = h->mb.cache.mv[0][x264_scan8[i8*4]];
int16_t *cache0_mv2 = cache0_mv + pixel_mv_offs[m0->i_pixel];
int16_t *cache1_mv = h->mb.cache.mv[1][x264_scan8[i8*4]];
int16_t *cache1_mv2 = cache1_mv + pixel_mv_offs[m0->i_pixel];
const int i_pixel = m0->i_pixel;
const int bw = x264_pixel_size[i_pixel].w;
const int bh = x264_pixel_size[i_pixel].h;
......@@ -853,17 +882,20 @@ int x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight
int bm1y = m1->mv[1], om1y = bm1y;
int bcost = COST_MAX;
int pass = 0;
uint64_t bcostrd = COST_MAX64;
/* each byte of visited represents 8 possible m1y positions, so a 4D array isn't needed */
DECLARE_ALIGNED_16( uint8_t visited[8][8][8] );
if( bm0y > h->mb.mv_max_spel[1] - 8 ||
bm1y > h->mb.mv_max_spel[1] - 8 )
return;
h->mc.memzero_aligned( visited, sizeof(visited) );
BIME_CACHE( 0, 0 );
CHECK_BIDIR( 0, 0, 0, 0 );
if( bm0y > h->mb.mv_max_spel[1] - 8 ||
bm1y > h->mb.mv_max_spel[1] - 8 )
return bcost;
for( pass = 0; pass < 8; pass++ )
{
/* check all mv pairs that differ in at most 2 components from the current mvs. */
......@@ -897,7 +929,16 @@ int x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight
m0->mv[1] = bm0y;
m1->mv[0] = bm1x;
m1->mv[1] = bm1y;
return bcost;
}
void x264_me_refine_bidir_satd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight )
{
x264_me_refine_bidir( h, m0, m1, i_weight, 0, 0, 0 );
}
void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight, int i8, int i_lambda2 )
{
x264_me_refine_bidir( h, m0, m1, i_weight, i8, i_lambda2, 1 );
}
#undef COST_MV_SATD
......@@ -921,13 +962,11 @@ int x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight
} \
}
#define SATD_THRESH 17/16
void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8 )
void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8, int i_list )
{
// don't have to fill the whole mv cache rectangle
static const int pixel_mv_offs[] = { 0, 4, 4*8, 0 };
int16_t *cache_mv = h->mb.cache.mv[0][x264_scan8[i8*4]];
int16_t *cache_mv = h->mb.cache.mv[i_list][x264_scan8[i8*4]];
int16_t *cache_mv2 = cache_mv + pixel_mv_offs[m->i_pixel];
const int16_t *p_cost_mvx, *p_cost_mvy;
const int bw = x264_pixel_size[m->i_pixel].w>>2;
......@@ -947,7 +986,7 @@ void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8 )
int satds[8];
if( m->i_pixel != PIXEL_16x16 && i8 != 0 )
x264_mb_predict_mv( h, 0, i8*4, bw, m->mvp );
x264_mb_predict_mv( h, i_list, i8*4, bw, m->mvp );
pmx = m->mvp[0];
pmy = m->mvp[1];
p_cost_mvx = m->p_cost_mv - pmx;
......@@ -999,7 +1038,7 @@ void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8 )
m->cost = bcost;
m->mv[0] = bmx;
m->mv[1] = bmy;
x264_macroblock_cache_mv ( h, 2*(i8&1), i8&2, bw, bh, 0, pack16to32_mask(bmx, bmy) );
x264_macroblock_cache_mvd( h, 2*(i8&1), i8&2, bw, bh, 0, pack16to32_mask(bmx - pmx, bmy - pmy) );
x264_macroblock_cache_mv ( h, 2*(i8&1), i8&2, bw, bh, i_list, pack16to32_mask(bmx, bmy) );
x264_macroblock_cache_mvd( h, 2*(i8&1), i8&2, bw, bh, i_list, pack16to32_mask(bmx - pmx, bmy - pmy) );
}
......@@ -53,8 +53,9 @@ static inline void x264_me_search( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], i
{ x264_me_search_ref( h, m, mvc, i_mvc, NULL ); }
void x264_me_refine_qpel( x264_t *h, x264_me_t *m );
void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8 );
int x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight );
void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8, int i_list );
void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight, int i8, int i_lambda2 );
void x264_me_refine_bidir_satd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight );
uint64_t x264_rd_cost_part( x264_t *h, int i_lambda2, int i8, int i_pixel );
extern uint16_t *x264_cost_mv_fpel[52][4];
......
......@@ -248,16 +248,21 @@ static void Help( x264_param_t *defaults, int b_longhelp )
H0( " --merange <integer> Maximum motion vector search range [%d]\n", defaults->analyse.i_me_range );
H1( " --mvrange <integer> Maximum motion vector length [-1 (auto)]\n" );
H1( " --mvrange-thread <int> Minimum buffer between threads [-1 (auto)]\n" );
H0( " -m, --subme <integer> Subpixel motion estimation and partition\n"
" decision quality: 1=fast, 7=best. [%d]\n", defaults->analyse.i_subpel_refine );
H0( " --b-rdo RD based mode decision for B-frames. Requires subme 6.\n" );
H0( " -m, --subme <integer> Subpixel motion estimation and mode decision [%d]\n", defaults->analyse.i_subpel_refine );
H1( " - 1: SAD mode decision, one qpel iteration\n"
" - 2: SATD mode decision\n"
" - 3-5: Progressively more qpel\n"
" - 6: RD mode decision for I/P-frames\n"
" - 7: RD mode decision for all frames\n"
" - 8: RD refinement for I/P-frames\n"
" - 9: RD refinement for all frames\n" );
else H0( " decision quality: 1=fast, 9=best.\n" );
H0( " --psy-rd Strength of psychovisual optimization [\"%.1f:%.1f\"]\n"
" #1: RDO (requires subme>=6)\n"
" #1: RD (requires subme>=6)\n"
" #2: Trellis (requires trellis, experimental)\n",