* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
*****************************************************************************/
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
#include <math.h>
#include <limits.h>
+#ifndef _MSC_VER
+#include <unistd.h>
+#endif
#include "common/common.h"
#include "macroblock.h"
{
/* 16x16 */
int i_ref;
+ int i_rd16x16;
x264_me_t me16x16;
/* 8x8 */
int i_cost8x8;
- int mvc[16][5][2]; /* [ref][0] is 16x16 mv,
- [ref][1..4] are 8x8 mv from partition [0..3] */
+ /* [ref][0] is 16x16 mv, [ref][1..4] are 8x8 mv from partition [0..3] */
+ DECLARE_ALIGNED_8( int mvc[32][5][2] );
x264_me_t me8x8[4];
/* Sub 4x4 */
/* Sub 4x8 */
int i_cost4x8[4]; /* cost per 8x8 partition */
- x264_me_t me4x8[4][4];
+ x264_me_t me4x8[4][2];
/* 16x8 */
int i_cost16x8;
/* I: Intra part */
/* Take some shortcuts in intra search if intra is deemed unlikely */
int b_fast_intra;
- int i_best_satd;
+ int b_try_pskip;
/* Luma part */
- int i_sad_i16x16;
+ int i_satd_i16x16;
+ int i_satd_i16x16_dir[7];
int i_predict16x16;
- int i_sad_i8x8;
- int i_predict8x8[2][2];
+ int i_satd_i8x8;
+ int i_satd_i8x8_dir[12][4];
+ int i_predict8x8[4];
- int i_sad_i4x4;
- int i_predict4x4[4][4];
+ int i_satd_i4x4;
+ int i_predict4x4[16];
/* Chroma part */
- int i_sad_i8x8chroma;
+ int i_satd_i8x8chroma;
+ int i_satd_i8x8chroma_dir[4];
int i_predict8x8chroma;
/* II: Inter part P/B frame */
int i_cost8x8direct[4];
int i_cost16x8bi;
int i_cost8x16bi;
+ int i_rd16x16bi;
+ int i_rd16x16direct;
+ int i_rd16x8bi;
+ int i_rd8x16bi;
+ int i_rd8x8bi;
int i_mb_partition16x8[2]; /* mb_partition_e */
int i_mb_partition8x16[2];
40,45,51,57,64,72,81,91 /* 44-51 */
};
-/* pow(lambda,2) * .9 */
+/* lambda2 = pow(lambda,2) * .9 * 256 */
static const int i_qp0_cost2_table[52] = {
- 1, 1, 1, 1, 1, 1, /* 0-5 */
- 1, 1, 1, 1, 1, 1, /* 6-11 */
- 1, 1, 1, 2, 2, 3, /* 12-17 */
- 4, 5, 6, 7, 9, 11, /* 18-23 */
- 14, 18, 23, 29, 36, 46, /* 24-29 */
- 58, 73, 91, 115, 145, 183, /* 30-35 */
- 230, 290, 366, 461, 581, 731, /* 36-41 */
- 922,1161,1463,1843,2322,2926, /* 42-47 */
-3686,4645,5852,7373
+ 14, 18, 22, 28, 36, 45, 57, 72, /* 0 - 7 */
+ 91, 115, 145, 182, 230, 290, 365, 460, /* 8 - 15 */
+ 580, 731, 921, 1161, 1462, 1843, 2322, 2925, /* 16 - 23 */
+ 3686, 4644, 5851, 7372, 9289, 11703, 14745, 18578, /* 24 - 31 */
+ 23407, 29491, 37156, 46814, 58982, 74313, 93628, 117964, /* 32 - 39 */
+148626, 187257, 235929, 297252, 374514, 471859, 594505, 749029, /* 40 - 47 */
+943718, 1189010, 1498059, 1887436 /* 48 - 51 */
};
/* TODO: calculate CABAC costs */
-static const int i_mb_b_cost_table[19] = {
+static const int i_mb_b_cost_table[X264_MBTYPE_MAX] = {
9, 9, 9, 9, 0, 0, 0, 1, 3, 7, 7, 7, 3, 7, 7, 7, 5, 9, 0
};
static const int i_mb_b16x8_cost_table[17] = {
static void x264_analyse_update_cache( x264_t *h, x264_mb_analysis_t *a );
+uint16_t *x264_cost_mv_fpel[52][4];
+
/* initialize an array of lambda*nbits for all possible mvs */
static void x264_mb_analyse_load_costs( x264_t *h, x264_mb_analysis_t *a )
{
static int16_t *p_cost_mv[52];
+ int i, j;
if( !p_cost_mv[a->i_qp] )
{
/* could be faster, but isn't called many times */
/* factor of 4 from qpel, 2 from sign, and 2 because mv can be opposite from mvp */
- int i;
- p_cost_mv[a->i_qp] = x264_malloc( (4*4*h->param.analyse.i_mv_range + 1) * sizeof(int16_t) );
- p_cost_mv[a->i_qp] += 2*4*h->param.analyse.i_mv_range;
- for( i = 0; i <= 2*4*h->param.analyse.i_mv_range; i++ )
+ p_cost_mv[a->i_qp] = x264_malloc( (4*4*2048 + 1) * sizeof(int16_t) );
+ p_cost_mv[a->i_qp] += 2*4*2048;
+ for( i = 0; i <= 2*4*2048; i++ )
{
p_cost_mv[a->i_qp][-i] =
p_cost_mv[a->i_qp][i] = a->i_lambda * bs_size_se( i );
}
}
-
a->p_cost_mv = p_cost_mv[a->i_qp];
+
+ /* FIXME is this useful for all me methods? */
+ if( h->param.analyse.i_me_method >= X264_ME_ESA && !x264_cost_mv_fpel[a->i_qp][0] )
+ {
+ for( j=0; j<4; j++ )
+ {
+ x264_cost_mv_fpel[a->i_qp][j] = x264_malloc( (4*2048 + 1) * sizeof(int16_t) );
+ x264_cost_mv_fpel[a->i_qp][j] += 2*2048;
+ for( i = -2*2048; i < 2*2048; i++ )
+ x264_cost_mv_fpel[a->i_qp][j][i] = p_cost_mv[a->i_qp][i*4+j];
+ }
+ }
}
static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int i_qp )
{
- memset( a, 0, sizeof( x264_mb_analysis_t ) );
-
/* conduct the analysis using this lamda and QP */
a->i_qp = h->mb.i_qp = i_qp;
+ h->mb.i_chroma_qp = i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];
a->i_lambda = i_qp0_cost_table[i_qp];
a->i_lambda2 = i_qp0_cost2_table[i_qp];
a->b_mbrd = h->param.analyse.i_subpel_refine >= 6 &&
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;
+ h->mb.b_trellis = h->param.analyse.i_trellis > 1 && a->b_mbrd;
h->mb.b_transform_8x8 = 0;
+ h->mb.b_noise_reduction = 0;
/* I: Intra part */
- a->i_sad_i16x16 =
- a->i_sad_i8x8 =
- a->i_sad_i4x4 =
- a->i_sad_i8x8chroma = COST_MAX;
+ a->i_satd_i16x16 =
+ a->i_satd_i8x8 =
+ a->i_satd_i4x4 =
+ a->i_satd_i8x8chroma = COST_MAX;
a->b_fast_intra = 0;
- a->i_best_satd = COST_MAX;
+ h->mb.i_skip_intra =
+ h->mb.b_lossless ? 0 :
+ a->b_mbrd ? 2 :
+ !h->param.analyse.i_trellis && !h->param.analyse.i_noise_reduction;
/* II: Inter part P/B frame */
if( h->sh.i_type != SLICE_TYPE_I )
{
- int i;
- int i_fmv_range = h->param.analyse.i_mv_range - 16;
+ int i, j;
+ int i_fmv_range = 4 * h->param.analyse.i_mv_range;
+ // limit motion search to a slightly smaller range than the theoretical limit,
+ // since the search may go a few iterations past its given range
+ int i_fpel_border = 5; // umh unconditional radius
+ int i_spel_border = 8; // 1.5 for subpel_satd, 1.5 for subpel_rd, 2 for bime, round up
/* Calculate max allowed MV range */
-#define CLIP_FMV(mv) x264_clip3( mv, -i_fmv_range, i_fmv_range )
- h->mb.mv_min_fpel[0] = CLIP_FMV( -16*h->mb.i_mb_x - 8 );
- h->mb.mv_max_fpel[0] = CLIP_FMV( 16*( h->sps->i_mb_width - h->mb.i_mb_x ) - 8 );
- h->mb.mv_min[0] = 4*( h->mb.mv_min_fpel[0] - 16 );
- h->mb.mv_max[0] = 4*( h->mb.mv_max_fpel[0] + 16 );
+#define CLIP_FMV(mv) x264_clip3( mv, -i_fmv_range, i_fmv_range-1 )
+ h->mb.mv_min[0] = 4*( -16*h->mb.i_mb_x - 24 );
+ h->mb.mv_max[0] = 4*( 16*( h->sps->i_mb_width - h->mb.i_mb_x - 1 ) + 24 );
+ h->mb.mv_min_spel[0] = CLIP_FMV( h->mb.mv_min[0] );
+ h->mb.mv_max_spel[0] = CLIP_FMV( h->mb.mv_max[0] );
+ h->mb.mv_min_fpel[0] = (h->mb.mv_min_spel[0]>>2) + i_fpel_border;
+ h->mb.mv_max_fpel[0] = (h->mb.mv_max_spel[0]>>2) - i_fpel_border;
if( h->mb.i_mb_x == 0)
{
- h->mb.mv_min_fpel[1] = CLIP_FMV( -16*h->mb.i_mb_y - 8 );
- h->mb.mv_max_fpel[1] = CLIP_FMV( 16*( h->sps->i_mb_height - h->mb.i_mb_y ) - 8 );
- h->mb.mv_min[1] = 4*( h->mb.mv_min_fpel[1] - 16 );
- h->mb.mv_max[1] = 4*( h->mb.mv_max_fpel[1] + 16 );
+ int mb_y = h->mb.i_mb_y >> h->sh.b_mbaff;
+ int mb_height = h->sps->i_mb_height >> h->sh.b_mbaff;
+ int thread_mvy_range = i_fmv_range;
+
+ if( h->param.i_threads > 1 )
+ {
+ int pix_y = (h->mb.i_mb_y | h->mb.b_interlaced) * 16;
+ int thresh = pix_y + h->param.analyse.i_mv_range_thread;
+ for( i = (h->sh.i_type == SLICE_TYPE_B); i >= 0; i-- )
+ {
+ x264_frame_t **fref = i ? h->fref1 : h->fref0;
+ int i_ref = i ? h->i_ref1 : h->i_ref0;
+ for( j=0; j<i_ref; j++ )
+ {
+ x264_frame_cond_wait( fref[j], thresh );
+ thread_mvy_range = X264_MIN( thread_mvy_range, fref[j]->i_lines_completed - pix_y );
+ }
+ }
+ if( h->param.b_deterministic )
+ thread_mvy_range = h->param.analyse.i_mv_range_thread;
+ if( h->mb.b_interlaced )
+ thread_mvy_range >>= 1;
+ }
+
+ h->mb.mv_min[1] = 4*( -16*mb_y - 24 );
+ h->mb.mv_max[1] = 4*( 16*( mb_height - mb_y - 1 ) + 24 );
+ h->mb.mv_min_spel[1] = x264_clip3( h->mb.mv_min[1], X264_MAX(4*(-512+i_spel_border), -i_fmv_range), i_fmv_range );
+ h->mb.mv_max_spel[1] = CLIP_FMV( h->mb.mv_max[1] );
+ h->mb.mv_max_spel[1] = X264_MIN( h->mb.mv_max_spel[1], thread_mvy_range*4 );
+ h->mb.mv_min_fpel[1] = (h->mb.mv_min_spel[1]>>2) + i_fpel_border;
+ h->mb.mv_max_fpel[1] = (h->mb.mv_max_spel[1]>>2) - i_fpel_border;
}
#undef CLIP_FMV
a->l0.me16x16.cost =
+ a->l0.i_rd16x16 =
a->l0.i_cost8x8 = COST_MAX;
for( i = 0; i < 4; i++ )
if( h->sh.i_type == SLICE_TYPE_B )
{
a->l1.me16x16.cost =
+ a->l1.i_rd16x16 =
a->l1.i_cost8x8 = COST_MAX;
for( i = 0; i < 4; i++ )
a->l1.i_cost16x8 =
a->l1.i_cost8x16 =
-
+ a->i_rd16x16bi =
+ a->i_rd16x16direct =
+ a->i_rd8x8bi =
+ a->i_rd16x8bi =
+ a->i_rd8x16bi =
a->i_cost16x16bi =
a->i_cost16x16direct =
a->i_cost8x8bi =
/* Fast intra decision */
if( h->mb.i_mb_xy - h->sh.i_first_mb > 4 )
{
- if( a->b_mbrd
- || IS_INTRA( h->mb.i_mb_type_left )
+ if( IS_INTRA( h->mb.i_mb_type_left )
|| IS_INTRA( h->mb.i_mb_type_top )
|| IS_INTRA( h->mb.i_mb_type_topleft )
|| IS_INTRA( h->mb.i_mb_type_topright )
{
if( i_neighbour & MB_TOPLEFT )
{
- /* top and left avaible */
+ /* top and left available */
*mode++ = I_PRED_16x16_V;
*mode++ = I_PRED_16x16_H;
*mode++ = I_PRED_16x16_DC;
}
else
{
- /* none avaible */
+ /* none available */
*mode = I_PRED_16x16_DC_128;
*pi_count = 1;
}
{
if( i_neighbour & MB_TOPLEFT )
{
- /* top and left avaible */
+ /* top and left available */
*mode++ = I_PRED_CHROMA_V;
*mode++ = I_PRED_CHROMA_H;
*mode++ = I_PRED_CHROMA_DC;
}
else
{
- /* none avaible */
+ /* none available */
*mode = I_PRED_CHROMA_DC_128;
*pi_count = 1;
}
if( b_l && b_t )
{
+ *pi_count = 6;
*mode++ = I_PRED_4x4_DC;
*mode++ = I_PRED_4x4_H;
*mode++ = I_PRED_4x4_V;
*mode++ = I_PRED_4x4_DDL;
- *mode++ = I_PRED_4x4_DDR;
- *mode++ = I_PRED_4x4_VR;
- *mode++ = I_PRED_4x4_HD;
+ if( i_neighbour & MB_TOPLEFT )
+ {
+ *mode++ = I_PRED_4x4_DDR;
+ *mode++ = I_PRED_4x4_VR;
+ *mode++ = I_PRED_4x4_HD;
+ *pi_count += 3;
+ }
*mode++ = I_PRED_4x4_VL;
*mode++ = I_PRED_4x4_HU;
- *pi_count = 9;
}
else if( b_l )
{
int i;
int i_max;
- int predict_mode[9];
+ int predict_mode[4];
uint8_t *p_dstc[2], *p_srcc[2];
- int i_stride[2];
- if( a->i_sad_i8x8chroma < COST_MAX )
+ if( a->i_satd_i8x8chroma < COST_MAX )
return;
/* 8x8 prediction selection for chroma */
p_srcc[0] = h->mb.pic.p_fenc[1];
p_srcc[1] = h->mb.pic.p_fenc[2];
- i_stride[0] = h->mb.pic.i_stride[1];
- i_stride[1] = h->mb.pic.i_stride[2];
-
predict_8x8chroma_mode_available( h->mb.i_neighbour, predict_mode, &i_max );
- a->i_sad_i8x8chroma = COST_MAX;
- for( i = 0; i < i_max; i++ )
+ a->i_satd_i8x8chroma = COST_MAX;
+ if( i_max == 4 && h->pixf.intra_satd_x3_8x8c && h->pixf.mbcmp[0] == h->pixf.satd[0] )
{
- int i_sad;
- int i_mode;
-
- i_mode = predict_mode[i];
-
- /* we do the prediction */
- h->predict_8x8c[i_mode]( p_dstc[0], i_stride[0] );
- h->predict_8x8c[i_mode]( p_dstc[1], i_stride[1] );
-
- /* we calculate the cost */
- i_sad = h->pixf.mbcmp[PIXEL_8x8]( p_dstc[0], i_stride[0],
- p_srcc[0], i_stride[0] ) +
- h->pixf.mbcmp[PIXEL_8x8]( p_dstc[1], i_stride[1],
- p_srcc[1], i_stride[1] ) +
- a->i_lambda * bs_size_ue( x264_mb_pred_mode8x8c_fix[i_mode] );
+ int satdu[4], satdv[4];
+ h->pixf.intra_satd_x3_8x8c( p_srcc[0], p_dstc[0], satdu );
+ h->pixf.intra_satd_x3_8x8c( p_srcc[1], p_dstc[1], satdv );
+ h->predict_8x8c[I_PRED_CHROMA_P]( p_dstc[0] );
+ h->predict_8x8c[I_PRED_CHROMA_P]( p_dstc[1] );
+ satdu[I_PRED_CHROMA_P] =
+ h->pixf.mbcmp[PIXEL_8x8]( p_dstc[0], FDEC_STRIDE, p_srcc[0], FENC_STRIDE );
+ satdv[I_PRED_CHROMA_P] =
+ h->pixf.mbcmp[PIXEL_8x8]( p_dstc[1], FDEC_STRIDE, p_srcc[1], FENC_STRIDE );
+
+ for( i=0; i<i_max; i++ )
+ {
+ int i_mode = predict_mode[i];
+ int i_satd = satdu[i_mode] + satdv[i_mode]
+ + a->i_lambda * bs_size_ue(i_mode);
- /* if i_score is lower it is better */
- if( a->i_sad_i8x8chroma > i_sad )
+ a->i_satd_i8x8chroma_dir[i] = i_satd;
+ COPY2_IF_LT( a->i_satd_i8x8chroma, i_satd, a->i_predict8x8chroma, i_mode );
+ }
+ }
+ else
+ {
+ for( i=0; i<i_max; i++ )
{
- a->i_predict8x8chroma = i_mode;
- a->i_sad_i8x8chroma = i_sad;
+ int i_satd;
+ int i_mode = predict_mode[i];
+
+ /* we do the prediction */
+ h->predict_8x8c[i_mode]( p_dstc[0] );
+ h->predict_8x8c[i_mode]( p_dstc[1] );
+
+ /* we calculate the cost */
+ i_satd = h->pixf.mbcmp[PIXEL_8x8]( p_dstc[0], FDEC_STRIDE,
+ p_srcc[0], FENC_STRIDE ) +
+ h->pixf.mbcmp[PIXEL_8x8]( p_dstc[1], FDEC_STRIDE,
+ p_srcc[1], FENC_STRIDE ) +
+ a->i_lambda * bs_size_ue( x264_mb_pred_mode8x8c_fix[i_mode] );
+
+ a->i_satd_i8x8chroma_dir[i] = i_satd;
+ COPY2_IF_LT( a->i_satd_i8x8chroma, i_satd, a->i_predict8x8chroma, i_mode );
}
}
h->mb.i_chroma_pred_mode = a->i_predict8x8chroma;
}
-static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_cost_inter )
+static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_inter )
{
const unsigned int flags = h->sh.i_type == SLICE_TYPE_I ? h->param.analyse.intra : h->param.analyse.inter;
- const int i_stride = h->mb.pic.i_stride[0];
uint8_t *p_src = h->mb.pic.p_fenc[0];
uint8_t *p_dst = h->mb.pic.p_fdec[0];
- int f8_satd_rd_ratio = 0;
int i, idx;
int i_max;
int predict_mode[9];
- int i_satd_thresh;
-
- if( h->sh.i_type == SLICE_TYPE_B )
- i_satd_thresh = a->i_best_satd * 9/8;
- else
- i_satd_thresh = a->i_best_satd * 5/4 + a->i_lambda * 10;
+ int b_merged_satd = h->pixf.intra_satd_x3_16x16 && h->pixf.mbcmp[0] == h->pixf.satd[0];
/*---------------- Try all mode and calculate their score ---------------*/
/* 16x16 prediction selection */
predict_16x16_mode_available( h->mb.i_neighbour, predict_mode, &i_max );
- for( i = 0; i < i_max; i++ )
- {
- int i_sad;
- int i_mode;
- i_mode = predict_mode[i];
- h->predict_16x16[i_mode]( p_dst, i_stride );
-
- i_sad = h->pixf.mbcmp[PIXEL_16x16]( p_dst, i_stride, p_src, i_stride ) +
- a->i_lambda * bs_size_ue( x264_mb_pred_mode16x16_fix[i_mode] );
- if( a->i_sad_i16x16 > i_sad )
+ if( b_merged_satd && i_max == 4 )
+ {
+ h->pixf.intra_satd_x3_16x16( p_src, p_dst, a->i_satd_i16x16_dir );
+ h->predict_16x16[I_PRED_16x16_P]( p_dst );
+ a->i_satd_i16x16_dir[I_PRED_16x16_P] =
+ h->pixf.mbcmp[PIXEL_16x16]( p_dst, FDEC_STRIDE, p_src, FENC_STRIDE );
+ for( i=0; i<4; i++ )
{
- a->i_predict16x16 = i_mode;
- a->i_sad_i16x16 = i_sad;
+ int cost = a->i_satd_i16x16_dir[i] += a->i_lambda * bs_size_ue(i);
+ COPY2_IF_LT( a->i_satd_i16x16, cost, a->i_predict16x16, i );
}
}
-
- if( a->b_mbrd )
+ else
{
- f8_satd_rd_ratio = ((unsigned)i_cost_inter << 8) / a->i_best_satd + 1;
- x264_mb_analyse_intra_chroma( h, a );
- if( h->mb.b_chroma_me )
- a->i_sad_i16x16 += a->i_sad_i8x8chroma;
- if( a->i_sad_i16x16 < i_satd_thresh )
+ for( i = 0; i < i_max; i++ )
{
- h->mb.i_type = I_16x16;
- h->mb.i_intra16x16_pred_mode = a->i_predict16x16;
- a->i_sad_i16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
+ int i_satd;
+ int i_mode = predict_mode[i];
+ h->predict_16x16[i_mode]( p_dst );
+
+ i_satd = h->pixf.mbcmp[PIXEL_16x16]( p_dst, FDEC_STRIDE, p_src, FENC_STRIDE ) +
+ a->i_lambda * bs_size_ue( x264_mb_pred_mode16x16_fix[i_mode] );
+ COPY2_IF_LT( a->i_satd_i16x16, i_satd, a->i_predict16x16, i_mode );
+ a->i_satd_i16x16_dir[i_mode] = i_satd;
}
- else
- a->i_sad_i16x16 = a->i_sad_i16x16 * f8_satd_rd_ratio >> 8;
}
- else
+
+ if( h->sh.i_type == SLICE_TYPE_B )
+ /* cavlc mb type prefix */
+ a->i_satd_i16x16 += a->i_lambda * i_mb_b_cost_table[I_16x16];
+ if( a->b_fast_intra && a->i_satd_i16x16 > 2*i_satd_inter )
+ return;
+
+ /* 8x8 prediction selection */
+ if( flags & X264_ANALYSE_I8x8 )
{
+ DECLARE_ALIGNED_16( uint8_t edge[33] );
+ x264_pixel_cmp_t sa8d = (*h->pixf.mbcmp == *h->pixf.sad) ? h->pixf.sad[PIXEL_8x8] : h->pixf.sa8d[PIXEL_8x8];
+ int i_satd_thresh = a->b_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];
+
+ // FIXME some bias like in i4x4?
if( h->sh.i_type == SLICE_TYPE_B )
- /* cavlc mb type prefix */
- a->i_sad_i16x16 += a->i_lambda * i_mb_b_cost_table[I_16x16];
- if( a->b_fast_intra && a->i_sad_i16x16 > 2*i_cost_inter )
+ i_cost += a->i_lambda * i_mb_b_cost_table[I_8x8];
+
+ for( idx = 0;; idx++ )
+ {
+ int x = idx&1;
+ int y = idx>>1;
+ uint8_t *p_src_by = p_src + 8*x + 8*y*FENC_STRIDE;
+ uint8_t *p_dst_by = p_dst + 8*x + 8*y*FDEC_STRIDE;
+ int i_best = COST_MAX;
+ int i_pred_mode = x264_mb_predict_intra4x4_mode( h, 4*idx );
+
+ predict_4x4_mode_available( h->mb.i_neighbour8[idx], predict_mode, &i_max );
+ x264_predict_8x8_filter( p_dst_by, edge, h->mb.i_neighbour8[idx], ALL_NEIGHBORS );
+
+ if( b_merged_satd && i_max == 9 )
+ {
+ int satd[3];
+ h->pixf.intra_sa8d_x3_8x8( p_src_by, edge, satd );
+ if( i_pred_mode < 3 )
+ satd[i_pred_mode] -= 3 * a->i_lambda;
+ for( i=2; i>=0; i-- )
+ {
+ int cost = a->i_satd_i8x8_dir[i][idx] = satd[i] + 4 * a->i_lambda;
+ COPY2_IF_LT( i_best, cost, a->i_predict8x8[idx], i );
+ }
+ i = 3;
+ }
+ else
+ i = 0;
+
+ for( ; i<i_max; i++ )
+ {
+ int i_satd;
+ int i_mode = predict_mode[i];
+
+ h->predict_8x8[i_mode]( p_dst_by, edge );
+
+ i_satd = sa8d( p_dst_by, FDEC_STRIDE, p_src_by, FENC_STRIDE )
+ + a->i_lambda * (i_pred_mode == x264_mb_pred_mode4x4_fix(i_mode) ? 1 : 4);
+
+ COPY2_IF_LT( i_best, i_satd, a->i_predict8x8[idx], i_mode );
+ a->i_satd_i8x8_dir[i_mode][idx] = i_satd;
+ }
+ i_cost += i_best;
+
+ if( idx == 3 || i_cost > i_satd_thresh )
+ break;
+
+ /* we need to encode this block now (for next ones) */
+ h->predict_8x8[a->i_predict8x8[idx]]( p_dst_by, edge );
+ x264_mb_encode_i8x8( h, idx, a->i_qp );
+
+ x264_macroblock_cache_intra8x8_pred( h, 2*x, 2*y, a->i_predict8x8[idx] );
+ }
+
+ if( idx == 3 )
+ {
+ a->i_satd_i8x8 = i_cost;
+ if( h->mb.i_skip_intra )
+ {
+ h->mc.copy[PIXEL_16x16]( h->mb.pic.i8x8_fdec_buf, 16, p_dst, FDEC_STRIDE, 16 );
+ if( h->mb.i_skip_intra == 2 )
+ h->mc.memcpy_aligned( h->mb.pic.i8x8_dct_buf, h->dct.luma8x8, sizeof(h->mb.pic.i8x8_dct_buf) );
+ }
+ }
+ else
+ {
+ 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 )
return;
}
/* 4x4 prediction selection */
if( flags & X264_ANALYSE_I4x4 )
{
- a->i_sad_i4x4 = 0;
- for( idx = 0; idx < 16; idx++ )
- {
- uint8_t *p_src_by;
- uint8_t *p_dst_by;
- int i_best;
- int x, y;
- int i_pred_mode;
+ 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 )
+ i_satd_thresh = i_satd_thresh * (10-a->b_fast_intra)/8;
- i_pred_mode= x264_mb_predict_intra4x4_mode( h, idx );
- x = block_idx_x[idx];
- y = block_idx_y[idx];
+ i_cost = a->i_lambda * 24; /* from JVT (SATD0) */
+ if( h->sh.i_type == SLICE_TYPE_B )
+ i_cost += a->i_lambda * i_mb_b_cost_table[I_4x4];
- p_src_by = p_src + 4 * x + 4 * y * i_stride;
- p_dst_by = p_dst + 4 * x + 4 * y * i_stride;
+ for( idx = 0;; idx++ )
+ {
+ int x = block_idx_x[idx];
+ int y = block_idx_y[idx];
+ uint8_t *p_src_by = p_src + 4*x + 4*y*FENC_STRIDE;
+ uint8_t *p_dst_by = p_dst + 4*x + 4*y*FDEC_STRIDE;
+ int i_best = COST_MAX;
+ int i_pred_mode = x264_mb_predict_intra4x4_mode( h, idx );
- i_best = COST_MAX;
predict_4x4_mode_available( h->mb.i_neighbour4[idx], predict_mode, &i_max );
if( (h->mb.i_neighbour4[idx] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
/* emulate missing topright samples */
- *(uint32_t*) &p_dst_by[4 - i_stride] = p_dst_by[3 - i_stride] * 0x01010101U;
+ *(uint32_t*) &p_dst_by[4 - FDEC_STRIDE] = p_dst_by[3 - FDEC_STRIDE] * 0x01010101U;
- for( i = 0; i < i_max; i++ )
+ if( b_merged_satd && i_max >= 6 )
{
- int i_sad;
- int i_mode;
+ int satd[3];
+ h->pixf.intra_satd_x3_4x4( p_src_by, p_dst_by, satd );
+ if( i_pred_mode < 3 )
+ satd[i_pred_mode] -= 3 * a->i_lambda;
+ for( i=2; i>=0; i-- )
+ COPY2_IF_LT( i_best, satd[i] + 4 * a->i_lambda,
+ a->i_predict4x4[idx], i );
+ i = 3;
+ }
+ else
+ i = 0;
- i_mode = predict_mode[i];
- h->predict_4x4[i_mode]( p_dst_by, i_stride );
+ for( ; i<i_max; i++ )
+ {
+ int i_satd;
+ int i_mode = predict_mode[i];
- i_sad = h->pixf.mbcmp[PIXEL_4x4]( p_dst_by, i_stride,
- p_src_by, i_stride )
- + a->i_lambda * (i_pred_mode == x264_mb_pred_mode4x4_fix(i_mode) ? 1 : 4);
+ h->predict_4x4[i_mode]( p_dst_by );
- if( i_best > i_sad )
- {
- a->i_predict4x4[x][y] = i_mode;
- i_best = i_sad;
- }
+ i_satd = h->pixf.mbcmp[PIXEL_4x4]( p_dst_by, FDEC_STRIDE,
+ p_src_by, FENC_STRIDE )
+ + a->i_lambda * (i_pred_mode == x264_mb_pred_mode4x4_fix(i_mode) ? 1 : 4);
+
+ COPY2_IF_LT( i_best, i_satd, a->i_predict4x4[idx], i_mode );
}
- a->i_sad_i4x4 += i_best;
+ i_cost += i_best;
+
+ if( i_cost > i_satd_thresh || idx == 15 )
+ break;
/* we need to encode this block now (for next ones) */
- h->predict_4x4[a->i_predict4x4[x][y]]( p_dst_by, i_stride );
+ h->predict_4x4[a->i_predict4x4[idx]]( p_dst_by );
x264_mb_encode_i4x4( h, idx, a->i_qp );
- h->mb.cache.intra4x4_pred_mode[x264_scan8[idx]] = a->i_predict4x4[x][y];
+ h->mb.cache.intra4x4_pred_mode[x264_scan8[idx]] = a->i_predict4x4[idx];
}
-
- a->i_sad_i4x4 += a->i_lambda * 24; /* from JVT (SATD0) */
- if( a->b_mbrd )
+ if( idx == 15 )
{
- if( h->mb.b_chroma_me )
- a->i_sad_i4x4 += a->i_sad_i8x8chroma;
- if( a->i_sad_i4x4 < i_satd_thresh )
+ a->i_satd_i4x4 = i_cost;
+ if( h->mb.i_skip_intra )
{
- h->mb.i_type = I_4x4;
- a->i_sad_i4x4 = x264_rd_cost_mb( h, a->i_lambda2 );
+ h->mc.copy[PIXEL_16x16]( h->mb.pic.i4x4_fdec_buf, 16, p_dst, FDEC_STRIDE, 16 );
+ if( h->mb.i_skip_intra == 2 )
+ h->mc.memcpy_aligned( h->mb.pic.i4x4_dct_buf, h->dct.luma4x4, sizeof(h->mb.pic.i4x4_dct_buf) );
}
- else
- a->i_sad_i4x4 = a->i_sad_i4x4 * f8_satd_rd_ratio >> 8;
}
else
+ a->i_satd_i4x4 = COST_MAX;
+ }
+}
+
+static void x264_intra_rd( x264_t *h, x264_mb_analysis_t *a, int i_satd_thresh )
+{
+ if( a->i_satd_i16x16 <= i_satd_thresh )
+ {
+ h->mb.i_type = I_16x16;
+ x264_analyse_update_cache( h, a );
+ a->i_satd_i16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+ else
+ a->i_satd_i16x16 = COST_MAX;
+
+ if( a->i_satd_i4x4 <= i_satd_thresh && a->i_satd_i4x4 < COST_MAX )
+ {
+ h->mb.i_type = I_4x4;
+ x264_analyse_update_cache( h, a );
+ a->i_satd_i4x4 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+ else
+ a->i_satd_i4x4 = COST_MAX;
+
+ if( a->i_satd_i8x8 <= i_satd_thresh && a->i_satd_i8x8 < COST_MAX )
+ {
+ h->mb.i_type = I_8x8;
+ x264_analyse_update_cache( h, a );
+ a->i_satd_i8x8 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+ else
+ a->i_satd_i8x8 = COST_MAX;
+}
+
+static void x264_intra_rd_refine( x264_t *h, x264_mb_analysis_t *a )
+{
+ uint8_t *p_src = h->mb.pic.p_fenc[0];
+ uint8_t *p_dst = h->mb.pic.p_fdec[0];
+
+ int i, j, idx, x, y;
+ int i_max, i_satd, i_best, i_mode, i_thresh;
+ int i_pred_mode;
+ int predict_mode[9];
+ h->mb.i_skip_intra = 0;
+
+ if( h->mb.i_type == I_16x16 )
+ {
+ int old_pred_mode = a->i_predict16x16;
+ i_thresh = a->i_satd_i16x16_dir[old_pred_mode] * 9/8;
+ i_best = a->i_satd_i16x16;
+ predict_16x16_mode_available( h->mb.i_neighbour, predict_mode, &i_max );
+ for( i = 0; i < i_max; i++ )
{
- if( h->sh.i_type == SLICE_TYPE_B )
- a->i_sad_i4x4 += a->i_lambda * i_mb_b_cost_table[I_4x4];
+ int i_mode = predict_mode[i];
+ if( i_mode == old_pred_mode || a->i_satd_i16x16_dir[i_mode] > i_thresh )
+ continue;
+ h->mb.i_intra16x16_pred_mode = i_mode;
+ i_satd = x264_rd_cost_mb( h, a->i_lambda2 );
+ COPY2_IF_LT( i_best, i_satd, a->i_predict16x16, i_mode );
}
}
+ else if( h->mb.i_type == I_4x4 )
+ {
+ uint32_t pels[4] = {0}; // doesn't need initting, just shuts up a gcc warning
+ int i_nnz = 0;
+ for( idx = 0; idx < 16; idx++ )
+ {
+ uint8_t *p_src_by;
+ uint8_t *p_dst_by;
+ i_best = COST_MAX;
- /* 8x8 prediction selection */
- if( flags & X264_ANALYSE_I8x8 )
+ i_pred_mode = x264_mb_predict_intra4x4_mode( h, idx );
+ x = block_idx_x[idx];
+ y = block_idx_y[idx];
+
+ p_src_by = p_src + 4*x + 4*y*FENC_STRIDE;
+ p_dst_by = p_dst + 4*x + 4*y*FDEC_STRIDE;
+ predict_4x4_mode_available( h->mb.i_neighbour4[idx], predict_mode, &i_max );
+
+ if( (h->mb.i_neighbour4[idx] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
+ /* emulate missing topright samples */
+ *(uint32_t*) &p_dst_by[4 - FDEC_STRIDE] = p_dst_by[3 - FDEC_STRIDE] * 0x01010101U;
+
+ for( i = 0; i < i_max; i++ )
+ {
+ i_mode = predict_mode[i];
+ h->predict_4x4[i_mode]( p_dst_by );
+ i_satd = x264_rd_cost_i4x4( h, a->i_lambda2, idx, i_mode );
+
+ if( i_best > i_satd )
+ {
+ a->i_predict4x4[idx] = i_mode;
+ i_best = i_satd;
+ pels[0] = *(uint32_t*)(p_dst_by+0*FDEC_STRIDE);
+ pels[1] = *(uint32_t*)(p_dst_by+1*FDEC_STRIDE);
+ pels[2] = *(uint32_t*)(p_dst_by+2*FDEC_STRIDE);
+ pels[3] = *(uint32_t*)(p_dst_by+3*FDEC_STRIDE);
+ i_nnz = h->mb.cache.non_zero_count[x264_scan8[idx]];
+ }
+ }
+
+ *(uint32_t*)(p_dst_by+0*FDEC_STRIDE) = pels[0];
+ *(uint32_t*)(p_dst_by+1*FDEC_STRIDE) = pels[1];
+ *(uint32_t*)(p_dst_by+2*FDEC_STRIDE) = pels[2];
+ *(uint32_t*)(p_dst_by+3*FDEC_STRIDE) = pels[3];
+ h->mb.cache.non_zero_count[x264_scan8[idx]] = i_nnz;
+
+ h->mb.cache.intra4x4_pred_mode[x264_scan8[idx]] = a->i_predict4x4[idx];
+ }
+ }
+ else if( h->mb.i_type == I_8x8 )
{
- a->i_sad_i8x8 = 0;
+ DECLARE_ALIGNED_16( uint8_t edge[33] );
for( idx = 0; idx < 4; idx++ )
{
+ uint64_t pels_h = 0;
+ uint8_t pels_v[7];
+ int i_nnz[3];
uint8_t *p_src_by;
uint8_t *p_dst_by;
- int i_best;
- int x, y;
- int i_pred_mode;
+ int j;
+ i_thresh = a->i_satd_i8x8_dir[a->i_predict8x8[idx]][idx] * 11/8;
- i_pred_mode= x264_mb_predict_intra4x4_mode( h, 4*idx );
+ i_best = COST_MAX;
+ i_pred_mode = x264_mb_predict_intra4x4_mode( h, 4*idx );
x = idx&1;
y = idx>>1;
- p_src_by = p_src + 8 * x + 8 * y * i_stride;
- p_dst_by = p_dst + 8 * x + 8 * y * i_stride;
-
- i_best = COST_MAX;
+ p_src_by = p_src + 8*x + 8*y*FENC_STRIDE;
+ p_dst_by = p_dst + 8*x + 8*y*FDEC_STRIDE;
predict_4x4_mode_available( h->mb.i_neighbour8[idx], predict_mode, &i_max );
+ x264_predict_8x8_filter( p_dst_by, edge, h->mb.i_neighbour8[idx], ALL_NEIGHBORS );
+
for( i = 0; i < i_max; i++ )
{
- int i_sad;
- int i_mode;
-
i_mode = predict_mode[i];
- h->predict_8x8[i_mode]( p_dst_by, i_stride, h->mb.i_neighbour8[idx] );
+ if( a->i_satd_i8x8_dir[i_mode][idx] > i_thresh )
+ continue;
+ h->predict_8x8[i_mode]( p_dst_by, edge );
+ i_satd = x264_rd_cost_i8x8( h, a->i_lambda2, idx, i_mode );
- /* could use sa8d, but it doesn't seem worth the speed cost (without mmx at least) */
- i_sad = h->pixf.mbcmp[PIXEL_8x8]( p_dst_by, i_stride,
- p_src_by, i_stride )
- + a->i_lambda * (i_pred_mode == x264_mb_pred_mode4x4_fix(i_mode) ? 1 : 4);
-
- if( i_best > i_sad )
+ if( i_best > i_satd )
{
- a->i_predict8x8[x][y] = i_mode;
- i_best = i_sad;
+ a->i_predict8x8[idx] = i_mode;
+ i_best = i_satd;
+
+ pels_h = *(uint64_t*)(p_dst_by+7*FDEC_STRIDE);
+ if( !(idx&1) )
+ for( j=0; j<7; j++ )
+ pels_v[j] = p_dst_by[7+j*FDEC_STRIDE];
+ for( j=0; j<3; j++ )
+ i_nnz[j] = h->mb.cache.non_zero_count[x264_scan8[4*idx+j+1]];
}
}
- a->i_sad_i8x8 += i_best;
- /* we need to encode this block now (for next ones) */
- h->predict_8x8[a->i_predict8x8[x][y]]( p_dst_by, i_stride, h->mb.i_neighbour );
- x264_mb_encode_i8x8( h, idx, a->i_qp );
+ *(uint64_t*)(p_dst_by+7*FDEC_STRIDE) = pels_h;
+ if( !(idx&1) )
+ for( j=0; j<7; j++ )
+ p_dst_by[7+j*FDEC_STRIDE] = pels_v[j];
+ for( j=0; j<3; j++ )
+ h->mb.cache.non_zero_count[x264_scan8[4*idx+j+1]] = i_nnz[j];
- x264_macroblock_cache_intra8x8_pred( h, 2*x, 2*y, a->i_predict8x8[x][y] );
+ x264_macroblock_cache_intra8x8_pred( h, 2*x, 2*y, a->i_predict8x8[idx] );
}
+ }
- if( a->b_mbrd )
- {
- if( h->mb.b_chroma_me )
- a->i_sad_i8x8 += a->i_sad_i8x8chroma;
- if( a->i_sad_i8x8 < i_satd_thresh )
+ /* RD selection for chroma prediction */
+ predict_8x8chroma_mode_available( h->mb.i_neighbour, predict_mode, &i_max );
+ if( i_max > 1 )
+ {
+ i_thresh = a->i_satd_i8x8chroma * 5/4;
+
+ for( i = j = 0; i < i_max; i++ )
+ if( a->i_satd_i8x8chroma_dir[i] < i_thresh &&
+ predict_mode[i] != a->i_predict8x8chroma )
{
- h->mb.i_type = I_8x8;
- a->i_sad_i8x8 = x264_rd_cost_mb( h, a->i_lambda2 );
+ predict_mode[j++] = predict_mode[i];
}
- else
- a->i_sad_i8x8 = a->i_sad_i8x8 * f8_satd_rd_ratio >> 8;
- }
- else
+ i_max = j;
+
+ if( i_max > 0 )
{
- // FIXME some bias like in i4x4?
- if( h->sh.i_type == SLICE_TYPE_B )
- a->i_sad_i8x8 += a->i_lambda * i_mb_b_cost_table[I_8x8];
+ int i_chroma_lambda = i_qp0_cost2_table[h->mb.i_chroma_qp];
+ /* the previous thing encoded was x264_intra_rd(), so the pixels and
+ * coefs for the current chroma mode are still around, so we only
+ * have to recount the bits. */
+ i_best = x264_rd_cost_i8x8_chroma( h, i_chroma_lambda, a->i_predict8x8chroma, 0 );
+ for( i = 0; i < i_max; i++ )
+ {
+ i_mode = predict_mode[i];
+ h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[1] );
+ h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[2] );
+ /* if we've already found a mode that needs no residual, then
+ * probably any mode with a residual will be worse.
+ * so avoid dct on the remaining modes to improve speed. */
+ i_satd = x264_rd_cost_i8x8_chroma( h, i_chroma_lambda, i_mode, h->mb.i_cbp_chroma != 0x00 );
+ COPY2_IF_LT( i_best, i_satd, a->i_predict8x8chroma, i_mode );
+ }
+ h->mb.i_chroma_pred_mode = a->i_predict8x8chroma;
}
}
}
#define LOAD_FENC( m, src, xoff, yoff) \
(m)->i_stride[0] = h->mb.pic.i_stride[0]; \
(m)->i_stride[1] = h->mb.pic.i_stride[1]; \
- (m)->p_fenc[0] = &(src)[0][(xoff)+(yoff)*(m)->i_stride[0]]; \
- (m)->p_fenc[1] = &(src)[1][((xoff)>>1)+((yoff)>>1)*(m)->i_stride[1]]; \
- (m)->p_fenc[2] = &(src)[2][((xoff)>>1)+((yoff)>>1)*(m)->i_stride[1]];
+ (m)->p_fenc[0] = &(src)[0][(xoff)+(yoff)*FENC_STRIDE]; \
+ (m)->p_fenc[1] = &(src)[1][((xoff)>>1)+((yoff)>>1)*FENC_STRIDE]; \
+ (m)->p_fenc[2] = &(src)[2][((xoff)>>1)+((yoff)>>1)*FENC_STRIDE];
-#define LOAD_HPELS(m, src, xoff, yoff) \
+#define LOAD_HPELS(m, src, list, ref, xoff, yoff) \
(m)->p_fref[0] = &(src)[0][(xoff)+(yoff)*(m)->i_stride[0]]; \
(m)->p_fref[1] = &(src)[1][(xoff)+(yoff)*(m)->i_stride[0]]; \
(m)->p_fref[2] = &(src)[2][(xoff)+(yoff)*(m)->i_stride[0]]; \
(m)->p_fref[3] = &(src)[3][(xoff)+(yoff)*(m)->i_stride[0]]; \
(m)->p_fref[4] = &(src)[4][((xoff)>>1)+((yoff)>>1)*(m)->i_stride[1]]; \
- (m)->p_fref[5] = &(src)[5][((xoff)>>1)+((yoff)>>1)*(m)->i_stride[1]];
+ (m)->p_fref[5] = &(src)[5][((xoff)>>1)+((yoff)>>1)*(m)->i_stride[1]]; \
+ (m)->integral = &h->mb.pic.p_integral[list][ref][(xoff)+(yoff)*(m)->i_stride[0]];
#define REF_COST(list, ref) \
(a->i_lambda * bs_size_te( h->sh.i_num_ref_idx_l##list##_active - 1, ref ))
int i_ref;
int mvc[7][2], i_mvc;
int i_halfpel_thresh = INT_MAX;
- int *p_halfpel_thresh = h->i_ref0>1 ? &i_halfpel_thresh : NULL;
+ int *p_halfpel_thresh = h->mb.pic.i_fref[0]>1 ? &i_halfpel_thresh : NULL;
/* 16x16 Search on all ref frame */
m.i_pixel = PIXEL_16x16;
LOAD_FENC( &m, h->mb.pic.p_fenc, 0, 0 );
a->l0.me16x16.cost = INT_MAX;
- for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
+ for( i_ref = 0; i_ref < h->mb.pic.i_fref[0]; i_ref++ )
{
const int i_ref_cost = REF_COST( 0, i_ref );
i_halfpel_thresh -= i_ref_cost;
m.i_ref = i_ref;
/* search with ref */
- LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, 0 );
+ LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, i_ref, 0, 0 );
x264_mb_predict_mv_16x16( h, 0, i_ref, m.mvp );
x264_mb_predict_mv_ref16x16( h, 0, i_ref, mvc, &i_mvc );
x264_me_search_ref( h, &m, mvc, i_mvc, p_halfpel_thresh );
+ /* early termination
+ * SSD threshold would probably be better than SATD */
+ if( i_ref == 0
+ && a->b_try_pskip
+ && m.cost-m.cost_mv < 300*a->i_lambda
+ && abs(m.mv[0]-h->mb.cache.pskip_mv[0])
+ + abs(m.mv[1]-h->mb.cache.pskip_mv[1]) <= 1
+ && x264_macroblock_probe_pskip( h ) )
+ {
+ h->mb.i_type = P_SKIP;
+ x264_analyse_update_cache( h, a );
+ assert( h->mb.cache.pskip_mv[1] <= h->mb.mv_max_spel[1] || h->param.i_threads == 1 );
+ return;
+ }
+
m.cost += i_ref_cost;
i_halfpel_thresh += i_ref_cost;
}
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, a->l0.me16x16.i_ref );
+ assert( a->l0.me16x16.mv[1] <= h->mb.mv_max_spel[1] || h->param.i_threads == 1 );
- if( a->b_mbrd )
+ h->mb.i_type = P_L0;
+ if( a->b_mbrd && a->l0.me16x16.i_ref == 0
+ && a->l0.me16x16.mv[0] == h->mb.cache.pskip_mv[0]
+ && a->l0.me16x16.mv[1] == h->mb.cache.pskip_mv[1] )
{
- a->i_best_satd = a->l0.me16x16.cost;
- h->mb.i_type = P_L0;
h->mb.i_partition = D_16x16;
- x264_macroblock_cache_mv ( h, 0, 0, 4, 4, 0, a->l0.me16x16.mv[0], a->l0.me16x16.mv[1] );
- a->l0.me16x16.cost = x264_rd_cost_mb( h, a->i_lambda2 );
+ x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, a->l0.me16x16.mv[0], a->l0.me16x16.mv[1] );
+ a->l0.i_rd16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
}
}
int i_ref;
uint8_t **p_fenc = h->mb.pic.p_fenc;
int i_halfpel_thresh = INT_MAX;
- int *p_halfpel_thresh = /*h->i_ref0>1 ? &i_halfpel_thresh : */NULL;
+ int *p_halfpel_thresh = /*h->mb.pic.i_fref[0]>1 ? &i_halfpel_thresh : */NULL;
int i;
- int i_maxref = h->i_ref0-1;
+ int i_maxref = h->mb.pic.i_fref[0]-1;
h->mb.i_partition = D_8x8;
m.i_ref_cost = i_ref_cost;
m.i_ref = i_ref;
- LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 8*x8, 8*y8 );
+ LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, i_ref, 8*x8, 8*y8 );
x264_macroblock_cache_ref( h, 2*x8, 2*y8, 2, 2, 0, i_ref );
x264_mb_predict_mv( h, 0, 4*i, 2, m.mvp );
x264_me_search_ref( h, &m, a->l0.mvc[i_ref], i+1, p_halfpel_thresh );
a->l0.i_cost8x8 = a->l0.me8x8[0].cost + a->l0.me8x8[1].cost +
a->l0.me8x8[2].cost + a->l0.me8x8[3].cost;
- if( a->b_mbrd )
- {
- if( a->i_best_satd > a->l0.i_cost8x8 )
- a->i_best_satd = a->l0.i_cost8x8;
- h->mb.i_type = P_8x8;
- h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
- h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
- a->l0.i_cost8x8 = x264_rd_cost_mb( h, a->i_lambda2 );
- }
+ h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
+ h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
}
static void x264_mb_analyse_inter_p8x8( x264_t *h, x264_mb_analysis_t *a )
m->i_ref = i_ref;
LOAD_FENC( m, p_fenc, 8*x8, 8*y8 );
- LOAD_HPELS( m, p_fref, 8*x8, 8*y8 );
+ LOAD_HPELS( m, p_fref, 0, i_ref, 8*x8, 8*y8 );
x264_mb_predict_mv( h, 0, 4*i, 2, m->mvp );
x264_me_search( h, m, mvc, i_mvc );
a->l0.i_cost8x8 = a->l0.me8x8[0].cost + a->l0.me8x8[1].cost +
a->l0.me8x8[2].cost + a->l0.me8x8[3].cost -
REF_COST( 0, a->l0.me16x16.i_ref );
- if( a->b_mbrd )
- {
- if( a->i_best_satd > a->l0.i_cost8x8 )
- a->i_best_satd = a->l0.i_cost8x8;
- h->mb.i_type = P_8x8;
- h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
- h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
- a->l0.i_cost8x8 = x264_rd_cost_mb( h, a->i_lambda2 );
- }
+ h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
+ h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
}
static void x264_mb_analyse_inter_p16x8( x264_t *h, x264_mb_analysis_t *a )
{
x264_me_t m;
uint8_t **p_fenc = h->mb.pic.p_fenc;
- int mvc[3][2];
+ DECLARE_ALIGNED_8( int mvc[3][2] );
int i, j;
/* XXX Needed for x264_mb_predict_mv */
*(uint64_t*)mvc[1] = *(uint64_t*)a->l0.mvc[i_ref][2*i+1];
*(uint64_t*)mvc[2] = *(uint64_t*)a->l0.mvc[i_ref][2*i+2];
- LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, 8*i );
+ LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, i_ref, 0, 8*i );
x264_macroblock_cache_ref( h, 0, 2*i, 4, 2, 0, i_ref );
x264_mb_predict_mv( h, 0, 8*i, 4, m.mvp );
x264_me_search( h, &m, mvc, 3 );
}
a->l0.i_cost16x8 = a->l0.me16x8[0].cost + a->l0.me16x8[1].cost;
- if( a->b_mbrd )
- {
- if( a->i_best_satd > a->l0.i_cost16x8 )
- a->i_best_satd = a->l0.i_cost16x8;
- h->mb.i_type = P_L0;
- a->l0.i_cost16x8 = x264_rd_cost_mb( h, a->i_lambda2 );
- }
}
static void x264_mb_analyse_inter_p8x16( x264_t *h, x264_mb_analysis_t *a )
{
x264_me_t m;
uint8_t **p_fenc = h->mb.pic.p_fenc;
- int mvc[3][2];
+ DECLARE_ALIGNED_8( int mvc[3][2] );
int i, j;
/* XXX Needed for x264_mb_predict_mv */
*(uint64_t*)mvc[1] = *(uint64_t*)a->l0.mvc[i_ref][i+1];
*(uint64_t*)mvc[2] = *(uint64_t*)a->l0.mvc[i_ref][i+3];
- LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 8*i, 0 );
+ LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, i_ref, 8*i, 0 );
x264_macroblock_cache_ref( h, 2*i, 0, 2, 4, 0, i_ref );
x264_mb_predict_mv( h, 0, 4*i, 2, m.mvp );
x264_me_search( h, &m, mvc, 3 );
}
a->l0.i_cost8x16 = a->l0.me8x16[0].cost + a->l0.me8x16[1].cost;
- if( a->b_mbrd )
- {
- if( a->i_best_satd > a->l0.i_cost8x16 )
- a->i_best_satd = a->l0.i_cost8x16;
- h->mb.i_type = P_L0;
- a->l0.i_cost8x16 = x264_rd_cost_mb( h, a->i_lambda2 );
- }
}
static int x264_mb_analyse_inter_p4x4_chroma( x264_t *h, x264_mb_analysis_t *a, uint8_t **p_fref, int i8x8, int pixel )
{
- uint8_t pix1[8*8], pix2[8*8];
+ DECLARE_ALIGNED_8( uint8_t pix1[16*8] );
+ uint8_t *pix2 = pix1+8;
const int i_stride = h->mb.pic.i_stride[1];
- const int off = 4*(i8x8&1) + 2*(i8x8&2)*i_stride;
+ const int or = 4*(i8x8&1) + 2*(i8x8&2)*i_stride;
+ const int oe = 4*(i8x8&1) + 2*(i8x8&2)*FENC_STRIDE;
#define CHROMA4x4MC( width, height, me, x, y ) \
- h->mc.mc_chroma( &p_fref[4][off+x+y*i_stride], i_stride, &pix1[x+y*8], 8, (me).mv[0], (me).mv[1], width, height ); \
- h->mc.mc_chroma( &p_fref[5][off+x+y*i_stride], i_stride, &pix2[x+y*8], 8, (me).mv[0], (me).mv[1], width, height );
+ h->mc.mc_chroma( &pix1[x+y*16], 16, &p_fref[4][or+x+y*i_stride], i_stride, (me).mv[0], (me).mv[1], width, height ); \
+ h->mc.mc_chroma( &pix2[x+y*16], 16, &p_fref[5][or+x+y*i_stride], i_stride, (me).mv[0], (me).mv[1], width, height );
if( pixel == PIXEL_4x4 )
{
CHROMA4x4MC( 2,2, a->l0.me4x4[i8x8][0], 0,0 );
- CHROMA4x4MC( 2,2, a->l0.me4x4[i8x8][1], 0,2 );
- CHROMA4x4MC( 2,2, a->l0.me4x4[i8x8][2], 2,0 );
+ CHROMA4x4MC( 2,2, a->l0.me4x4[i8x8][1], 2,0 );
+ CHROMA4x4MC( 2,2, a->l0.me4x4[i8x8][2], 0,2 );
CHROMA4x4MC( 2,2, a->l0.me4x4[i8x8][3], 2,2 );
}
else if( pixel == PIXEL_8x4 )
CHROMA4x4MC( 2,4, a->l0.me4x8[i8x8][1], 2,0 );
}
- return h->pixf.mbcmp[PIXEL_4x4]( &h->mb.pic.p_fenc[1][off], i_stride, pix1, 8 )
- + h->pixf.mbcmp[PIXEL_4x4]( &h->mb.pic.p_fenc[2][off], i_stride, pix2, 8 );
+ return h->pixf.mbcmp[PIXEL_4x4]( &h->mb.pic.p_fenc[1][oe], FENC_STRIDE, pix1, 16 )
+ + h->pixf.mbcmp[PIXEL_4x4]( &h->mb.pic.p_fenc[2][oe], FENC_STRIDE, pix2, 16 );
}
static void x264_mb_analyse_inter_p4x4( x264_t *h, x264_mb_analysis_t *a, int i8x8 )
{
uint8_t **p_fref = h->mb.pic.p_fref[0][a->l0.me8x8[i8x8].i_ref];
uint8_t **p_fenc = h->mb.pic.p_fenc;
-
+ const int i_ref = a->l0.me8x8[i8x8].i_ref;
int i4x4;
/* XXX Needed for x264_mb_predict_mv */
m->p_cost_mv = a->p_cost_mv;
LOAD_FENC( m, p_fenc, 4*x4, 4*y4 );
- LOAD_HPELS( m, p_fref, 4*x4, 4*y4 );
+ LOAD_HPELS( m, p_fref, 0, i_ref, 4*x4, 4*y4 );
x264_mb_predict_mv( h, 0, idx, 1, m->mvp );
x264_me_search( h, m, &a->l0.me8x8[i8x8].mv, i_mvc );
a->l0.me4x4[i8x8][1].cost +
a->l0.me4x4[i8x8][2].cost +
a->l0.me4x4[i8x8][3].cost +
- REF_COST( 0, a->l0.me8x8[i8x8].i_ref ) +
+ REF_COST( 0, i_ref ) +
a->i_lambda * i_sub_mb_p_cost_table[D_L0_4x4];
if( h->mb.b_chroma_me )
a->l0.i_cost4x4[i8x8] += x264_mb_analyse_inter_p4x4_chroma( h, a, p_fref, i8x8, PIXEL_4x4 );
{
uint8_t **p_fref = h->mb.pic.p_fref[0][a->l0.me8x8[i8x8].i_ref];
uint8_t **p_fenc = h->mb.pic.p_fenc;
-
+ const int i_ref = a->l0.me8x8[i8x8].i_ref;
int i8x4;
/* XXX Needed for x264_mb_predict_mv */
m->p_cost_mv = a->p_cost_mv;
LOAD_FENC( m, p_fenc, 4*x4, 4*y4 );
- LOAD_HPELS( m, p_fref, 4*x4, 4*y4 );
+ LOAD_HPELS( m, p_fref, 0, i_ref, 4*x4, 4*y4 );
x264_mb_predict_mv( h, 0, idx, 2, m->mvp );
x264_me_search( h, m, &a->l0.me4x4[i8x8][0].mv, i_mvc );
x264_macroblock_cache_mv( h, x4, y4, 2, 1, 0, m->mv[0], m->mv[1] );
}
a->l0.i_cost8x4[i8x8] = a->l0.me8x4[i8x8][0].cost + a->l0.me8x4[i8x8][1].cost +
- REF_COST( 0, a->l0.me8x8[i8x8].i_ref ) +
+ REF_COST( 0, i_ref ) +
a->i_lambda * i_sub_mb_p_cost_table[D_L0_8x4];
if( h->mb.b_chroma_me )
a->l0.i_cost8x4[i8x8] += x264_mb_analyse_inter_p4x4_chroma( h, a, p_fref, i8x8, PIXEL_8x4 );
{
uint8_t **p_fref = h->mb.pic.p_fref[0][a->l0.me8x8[i8x8].i_ref];
uint8_t **p_fenc = h->mb.pic.p_fenc;
-
+ const int i_ref = a->l0.me8x8[i8x8].i_ref;
int i4x8;
/* XXX Needed for x264_mb_predict_mv */
m->p_cost_mv = a->p_cost_mv;
LOAD_FENC( m, p_fenc, 4*x4, 4*y4 );
- LOAD_HPELS( m, p_fref, 4*x4, 4*y4 );
+ LOAD_HPELS( m, p_fref, 0, i_ref, 4*x4, 4*y4 );
x264_mb_predict_mv( h, 0, idx, 1, m->mvp );
x264_me_search( h, m, &a->l0.me4x4[i8x8][0].mv, i_mvc );
x264_macroblock_cache_mv( h, x4, y4, 1, 2, 0, m->mv[0], m->mv[1] );
}
a->l0.i_cost4x8[i8x8] = a->l0.me4x8[i8x8][0].cost + a->l0.me4x8[i8x8][1].cost +
- REF_COST( 0, a->l0.me8x8[i8x8].i_ref ) +
+ REF_COST( 0, i_ref ) +
a->i_lambda * i_sub_mb_p_cost_table[D_L0_4x8];
if( h->mb.b_chroma_me )
a->l0.i_cost4x8[i8x8] += x264_mb_analyse_inter_p4x4_chroma( h, a, p_fref, i8x8, PIXEL_4x8 );
uint8_t **p_fenc = h->mb.pic.p_fenc;
uint8_t **p_fdec = h->mb.pic.p_fdec;
- int i_stride= h->mb.pic.i_stride[0];
int i;
- a->i_cost16x16direct = 0;
+ a->i_cost16x16direct = a->i_lambda * i_mb_b_cost_table[B_DIRECT];
for( i = 0; i < 4; i++ )
{
- const int x8 = i%2;
- const int y8 = i/2;
- const int off = 8 * x8 + 8 * i_stride * y8;
+ const int x = (i&1)*8;
+ const int y = (i>>1)*8;
a->i_cost16x16direct +=
a->i_cost8x8direct[i] =
- h->pixf.mbcmp[PIXEL_8x8]( &p_fenc[0][off], i_stride, &p_fdec[0][off], i_stride );
+ h->pixf.mbcmp[PIXEL_8x8]( &p_fenc[0][x+y*FENC_STRIDE], FENC_STRIDE, &p_fdec[0][x+y*FDEC_STRIDE], FDEC_STRIDE );
/* mb type cost */
a->i_cost8x8direct[i] += a->i_lambda * i_sub_mb_b_cost_table[D_DIRECT_8x8];
}
- a->i_cost16x16direct += a->i_lambda * i_mb_b_cost_table[B_DIRECT];
-
- if( a->b_mbrd )
- {
- if( a->i_cost16x16direct < a->i_best_satd )
- a->i_best_satd = a->i_cost16x16direct;
-
- h->mb.i_type = B_DIRECT;
- a->i_cost16x16direct = x264_rd_cost_mb( h, a->i_lambda2 );
- }
}
#define WEIGHTED_AVG( size, pix1, stride1, src2, stride2 ) \
static void x264_mb_analyse_inter_b16x16( x264_t *h, x264_mb_analysis_t *a )
{
- uint8_t pix1[16*16], pix2[16*16];
+ DECLARE_ALIGNED_16( uint8_t pix1[16*16] );
+ DECLARE_ALIGNED_16( uint8_t pix2[16*16] );
uint8_t *src2;
int stride2 = 16;
- int src2_ref, pix1_ref;
+ int weight;
x264_me_t m;
int i_ref;
int mvc[8][2], i_mvc;
int i_halfpel_thresh = INT_MAX;
- int *p_halfpel_thresh = h->i_ref0>1 ? &i_halfpel_thresh : NULL;
+ int *p_halfpel_thresh = h->mb.pic.i_fref[0]>1 ? &i_halfpel_thresh : NULL;
/* 16x16 Search on all ref frame */
m.i_pixel = PIXEL_16x16;
/* ME for List 0 */
a->l0.me16x16.cost = INT_MAX;
- for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
+ for( i_ref = 0; i_ref < h->mb.pic.i_fref[0]; i_ref++ )
{
/* search with ref */
- LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, 0 );
+ LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, i_ref, 0, 0 );
x264_mb_predict_mv_16x16( h, 0, i_ref, m.mvp );
x264_mb_predict_mv_ref16x16( h, 0, i_ref, mvc, &i_mvc );
x264_me_search_ref( h, &m, mvc, i_mvc, p_halfpel_thresh );
/* ME for list 1 */
i_halfpel_thresh = INT_MAX;
- p_halfpel_thresh = h->i_ref1>1 ? &i_halfpel_thresh : NULL;
+ p_halfpel_thresh = h->mb.pic.i_fref[1]>1 ? &i_halfpel_thresh : NULL;
a->l1.me16x16.cost = INT_MAX;
- for( i_ref = 0; i_ref < h->i_ref1; i_ref++ )
+ for( i_ref = 0; i_ref < h->mb.pic.i_fref[1]; i_ref++ )
{
/* search with ref */
- LOAD_HPELS( &m, h->mb.pic.p_fref[1][i_ref], 0, 0 );
+ LOAD_HPELS( &m, h->mb.pic.p_fref[1][i_ref], 1, i_ref, 0, 0 );
x264_mb_predict_mv_16x16( h, 1, i_ref, m.mvp );
x264_mb_predict_mv_ref16x16( h, 1, i_ref, mvc, &i_mvc );
x264_me_search_ref( h, &m, mvc, i_mvc, p_halfpel_thresh );
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, a->l1.i_ref );
/* get cost of BI mode */
+ weight = h->mb.bipred_weight[a->l0.i_ref][a->l1.i_ref];
if ( ((a->l0.me16x16.mv[0] | a->l0.me16x16.mv[1]) & 1) == 0 )
{
/* l0 reference is halfpel, so get_ref on it will make it faster */
- src2 = h->mc.get_ref( h->mb.pic.p_fref[0][a->l0.i_ref], h->mb.pic.i_stride[0],
- pix2, &stride2,
- a->l0.me16x16.mv[0], a->l0.me16x16.mv[1],
- 16, 16 );
- h->mc.mc_luma( h->mb.pic.p_fref[1][a->l1.i_ref], h->mb.pic.i_stride[0],
- pix1, 16,
- a->l1.me16x16.mv[0], a->l1.me16x16.mv[1],
- 16, 16 );
- src2_ref = a->l0.i_ref;
- pix1_ref = a->l1.i_ref;
+ src2 =
+ h->mc.get_ref( pix2, &stride2,
+ h->mb.pic.p_fref[0][a->l0.i_ref], h->mb.pic.i_stride[0],
+ a->l0.me16x16.mv[0], a->l0.me16x16.mv[1],
+ 16, 16 );
+ h->mc.mc_luma( pix1, 16,
+ h->mb.pic.p_fref[1][a->l1.i_ref], h->mb.pic.i_stride[0],
+ a->l1.me16x16.mv[0], a->l1.me16x16.mv[1],
+ 16, 16 );
+ weight = 64 - weight;
}
else
{
/* if l0 was qpel, we'll use get_ref on l1 instead */
- h->mc.mc_luma( h->mb.pic.p_fref[0][a->l0.i_ref], h->mb.pic.i_stride[0],
- pix1, 16,
- a->l0.me16x16.mv[0], a->l0.me16x16.mv[1],
- 16, 16 );
- src2 = h->mc.get_ref( h->mb.pic.p_fref[1][a->l1.i_ref], h->mb.pic.i_stride[0],
- pix2, &stride2,
- a->l1.me16x16.mv[0], a->l1.me16x16.mv[1],
- 16, 16 );
- src2_ref = a->l1.i_ref;
- pix1_ref = a->l0.i_ref;
+ h->mc.mc_luma( pix1, 16,
+ h->mb.pic.p_fref[0][a->l0.i_ref], h->mb.pic.i_stride[0],
+ a->l0.me16x16.mv[0], a->l0.me16x16.mv[1],
+ 16, 16 );
+ src2 =
+ h->mc.get_ref( pix2, &stride2,
+ h->mb.pic.p_fref[1][a->l1.i_ref], h->mb.pic.i_stride[0],
+ a->l1.me16x16.mv[0], a->l1.me16x16.mv[1],
+ 16, 16 );
}
if( h->param.analyse.b_weighted_bipred )
- h->mc.avg_weight[PIXEL_16x16]( pix1, 16, src2, stride2,
- h->mb.bipred_weight[pix1_ref][src2_ref] );
+ h->mc.avg_weight[PIXEL_16x16]( pix1, 16, src2, stride2, weight );
else
h->mc.avg[PIXEL_16x16]( pix1, 16, src2, stride2 );
- a->i_cost16x16bi = h->pixf.mbcmp[PIXEL_16x16]( h->mb.pic.p_fenc[0], h->mb.pic.i_stride[0], pix1, 16 )
+ a->i_cost16x16bi = h->pixf.mbcmp[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE, pix1, 16 )
+ REF_COST( 0, a->l0.i_ref )
+ REF_COST( 1, a->l1.i_ref )
+ a->l0.me16x16.cost_mv
a->i_cost16x16bi += a->i_lambda * i_mb_b_cost_table[B_BI_BI];
a->l0.me16x16.cost += a->i_lambda * i_mb_b_cost_table[B_L0_L0];
a->l1.me16x16.cost += a->i_lambda * i_mb_b_cost_table[B_L1_L1];
-
- if( a->b_mbrd )
- {
- int i_satd_thresh;
-
- if( a->l0.me16x16.cost < a->i_best_satd )
- a->i_best_satd = a->l0.me16x16.cost;
- if( a->l1.me16x16.cost < a->i_best_satd )
- a->i_best_satd = a->l1.me16x16.cost;
- if( a->i_cost16x16bi < a->i_best_satd )
- a->i_best_satd = a->i_cost16x16bi;
-
- i_satd_thresh = a->i_best_satd * 3/2;
-
- h->mb.i_partition = D_16x16;
- /* L0 */
- if( a->l0.me16x16.cost < i_satd_thresh )
- {
- h->mb.i_type = B_L0_L0;
- x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, a->l0.me16x16.mv[0], a->l0.me16x16.mv[1] );
- a->l0.me16x16.cost = x264_rd_cost_mb( h, a->i_lambda2 );
- }
- else
- a->l0.me16x16.cost = COST_MAX;
-
- /* L1 */
- if( a->l1.me16x16.cost < i_satd_thresh )
- {
- h->mb.i_type = B_L1_L1;
- x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, a->l1.me16x16.mv[0], a->l1.me16x16.mv[1] );
- a->l1.me16x16.cost = x264_rd_cost_mb( h, a->i_lambda2 );
- }
- else
- a->l1.me16x16.cost = COST_MAX;
-
- /* BI */
- if( a->i_cost16x16bi < i_satd_thresh )
- {
- h->mb.i_type = B_BI_BI;
- a->i_cost16x16bi = x264_rd_cost_mb( h, a->i_lambda2 );
- }
- else
- a->i_cost16x16bi = COST_MAX;
- }
}
static inline void x264_mb_cache_mv_p8x8( x264_t *h, x264_mb_analysis_t *a, int i )
uint8_t **p_fref[2] =
{ h->mb.pic.p_fref[0][a->l0.i_ref],
h->mb.pic.p_fref[1][a->l1.i_ref] };
- uint8_t pix[2][8*8];
+ DECLARE_ALIGNED_8( uint8_t pix[2][8*8] );
int i, l;
/* XXX Needed for x264_mb_predict_mv */
m->p_cost_mv = a->p_cost_mv;
LOAD_FENC( m, h->mb.pic.p_fenc, 8*x8, 8*y8 );
- LOAD_HPELS( m, p_fref[l], 8*x8, 8*y8 );
+ LOAD_HPELS( m, p_fref[l], l, lX->i_ref, 8*x8, 8*y8 );
x264_mb_predict_mv( h, l, 4*i, 2, m->mvp );
x264_me_search( h, m, &lX->me16x16.mv, 1 );
x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, l, m->mv[0], m->mv[1] );
/* BI mode */
- h->mc.mc_luma( m->p_fref, m->i_stride[0], pix[l], 8,
- m->mv[0], m->mv[1], 8, 8 );
+ h->mc.mc_luma( pix[l], 8, m->p_fref, m->i_stride[0],
+ m->mv[0], m->mv[1], 8, 8 );
i_part_cost_bi += m->cost_mv;
/* FIXME: ref cost */
}
WEIGHTED_AVG( PIXEL_8x8, pix[0], 8, pix[1], 8 );
- i_part_cost_bi += h->pixf.mbcmp[PIXEL_8x8]( a->l0.me8x8[i].p_fenc[0], h->mb.pic.i_stride[0], pix[0], 8 )
+ i_part_cost_bi += h->pixf.mbcmp[PIXEL_8x8]( a->l0.me8x8[i].p_fenc[0], FENC_STRIDE, pix[0], 8 )
+ a->i_lambda * i_sub_mb_b_cost_table[D_BI_8x8];
a->l0.me8x8[i].cost += a->i_lambda * i_sub_mb_b_cost_table[D_L0_8x8];
a->l1.me8x8[i].cost += a->i_lambda * i_sub_mb_b_cost_table[D_L1_8x8];
i_part_cost = a->l0.me8x8[i].cost;
h->mb.i_sub_partition[i] = D_L0_8x8;
- if( a->l1.me8x8[i].cost < i_part_cost )
- {
- i_part_cost = a->l1.me8x8[i].cost;
- h->mb.i_sub_partition[i] = D_L1_8x8;
- }
- if( i_part_cost_bi < i_part_cost )
- {
- i_part_cost = i_part_cost_bi;
- h->mb.i_sub_partition[i] = D_BI_8x8;
- }
- if( a->i_cost8x8direct[i] < i_part_cost )
- {
- i_part_cost = a->i_cost8x8direct[i];
- h->mb.i_sub_partition[i] = D_DIRECT_8x8;
- }
+ COPY2_IF_LT( i_part_cost, a->l1.me8x8[i].cost, h->mb.i_sub_partition[i], D_L1_8x8 );
+ COPY2_IF_LT( i_part_cost, i_part_cost_bi, h->mb.i_sub_partition[i], D_BI_8x8 );
+ COPY2_IF_LT( i_part_cost, a->i_cost8x8direct[i], h->mb.i_sub_partition[i], D_DIRECT_8x8 );
a->i_cost8x8bi += i_part_cost;
/* XXX Needed for x264_mb_predict_mv */
/* mb type cost */
a->i_cost8x8bi += a->i_lambda * i_mb_b_cost_table[B_8x8];
-
- if( a->b_mbrd )
- {
- if( a->i_cost8x8bi < a->i_best_satd )
- a->i_best_satd = a->i_cost8x8bi;
-
- if( a->i_cost8x8bi < a->i_best_satd * 3/2 )
- {
- h->mb.i_type = B_8x8;
- h->mb.i_partition = D_8x8;
- a->i_cost8x8bi = x264_rd_cost_mb( h, a->i_lambda2 );
- }
- else
- a->i_cost8x8bi = COST_MAX;
- }
}
static void x264_mb_analyse_inter_b16x8( x264_t *h, x264_mb_analysis_t *a )
uint8_t **p_fref[2] =
{ h->mb.pic.p_fref[0][a->l0.i_ref],
h->mb.pic.p_fref[1][a->l1.i_ref] };
- uint8_t pix[2][16*8];
- int mvc[2][2];
+ DECLARE_ALIGNED_16( uint8_t pix[2][16*8] );
+ DECLARE_ALIGNED_8( int mvc[2][2] );
int i, l;
h->mb.i_partition = D_16x8;
m->p_cost_mv = a->p_cost_mv;
LOAD_FENC( m, h->mb.pic.p_fenc, 0, 8*i );
- LOAD_HPELS( m, p_fref[l], 0, 8*i );
+ LOAD_HPELS( m, p_fref[l], l, lX->i_ref, 0, 8*i );
- mvc[0][0] = lX->me8x8[2*i].mv[0];
- mvc[0][1] = lX->me8x8[2*i].mv[1];
- mvc[1][0] = lX->me8x8[2*i+1].mv[0];
- mvc[1][1] = lX->me8x8[2*i+1].mv[1];
+ *(uint64_t*)mvc[0] = *(uint64_t*)lX->me8x8[2*i].mv;
+ *(uint64_t*)mvc[1] = *(uint64_t*)lX->me8x8[2*i+1].mv;
- x264_mb_predict_mv( h, 0, 8*i, 2, m->mvp );
+ x264_mb_predict_mv( h, l, 8*i, 2, m->mvp );
x264_me_search( h, m, mvc, 2 );
/* BI mode */
- h->mc.mc_luma( m->p_fref, m->i_stride[0], pix[l], 16,
- m->mv[0], m->mv[1], 16, 8 );
+ h->mc.mc_luma( pix[l], 16, m->p_fref, m->i_stride[0],
+ m->mv[0], m->mv[1], 16, 8 );
/* FIXME: ref cost */
i_part_cost_bi += m->cost_mv;
}
WEIGHTED_AVG( PIXEL_16x8, pix[0], 16, pix[1], 16 );
- i_part_cost_bi += h->pixf.mbcmp[PIXEL_16x8]( a->l0.me16x8[i].p_fenc[0], h->mb.pic.i_stride[0], pix[0], 16 );
+ i_part_cost_bi += h->pixf.mbcmp[PIXEL_16x8]( a->l0.me16x8[i].p_fenc[0], FENC_STRIDE, pix[0], 16 );
i_part_cost = a->l0.me16x8[i].cost;
a->i_mb_partition16x8[i] = D_L0_8x8; /* not actually 8x8, only the L0 matters */
+ (a->i_mb_partition16x8[0]>>2) * 3
+ (a->i_mb_partition16x8[1]>>2);
a->i_cost16x8bi += a->i_lambda * i_mb_b16x8_cost_table[a->i_mb_type16x8];
-
- if( a->b_mbrd )
- {
- if( a->i_cost16x8bi < a->i_best_satd )
- a->i_best_satd = a->i_cost16x8bi;
-
- if( a->i_cost16x8bi < a->i_best_satd * 3/2 )
- {
- h->mb.i_type = a->i_mb_type16x8;
- h->mb.i_partition = D_16x8;
- a->i_cost16x8bi = x264_rd_cost_mb( h, a->i_lambda2 );
- }
- else
- a->i_cost16x8bi = COST_MAX;
- }
}
+
static void x264_mb_analyse_inter_b8x16( x264_t *h, x264_mb_analysis_t *a )
{
uint8_t **p_fref[2] =
{ h->mb.pic.p_fref[0][a->l0.i_ref],
h->mb.pic.p_fref[1][a->l1.i_ref] };
- uint8_t pix[2][8*16];
- int mvc[2][2];
+ DECLARE_ALIGNED_8( uint8_t pix[2][8*16] );
+ DECLARE_ALIGNED_8( int mvc[2][2] );
int i, l;
h->mb.i_partition = D_8x16;
m->p_cost_mv = a->p_cost_mv;
LOAD_FENC( m, h->mb.pic.p_fenc, 8*i, 0 );
- LOAD_HPELS( m, p_fref[l], 8*i, 0 );
+ LOAD_HPELS( m, p_fref[l], l, lX->i_ref, 8*i, 0 );
- mvc[0][0] = lX->me8x8[i].mv[0];
- mvc[0][1] = lX->me8x8[i].mv[1];
- mvc[1][0] = lX->me8x8[i+2].mv[0];
- mvc[1][1] = lX->me8x8[i+2].mv[1];
+ *(uint64_t*)mvc[0] = *(uint64_t*)lX->me8x8[i].mv;
+ *(uint64_t*)mvc[1] = *(uint64_t*)lX->me8x8[i+2].mv;
- x264_mb_predict_mv( h, 0, 4*i, 2, m->mvp );
+ x264_mb_predict_mv( h, l, 4*i, 2, m->mvp );
x264_me_search( h, m, mvc, 2 );
/* BI mode */
- h->mc.mc_luma( m->p_fref, m->i_stride[0], pix[l], 8,
- m->mv[0], m->mv[1], 8, 16 );
+ h->mc.mc_luma( pix[l], 8, m->p_fref, m->i_stride[0],
+ m->mv[0], m->mv[1], 8, 16 );
/* FIXME: ref cost */
i_part_cost_bi += m->cost_mv;
}
WEIGHTED_AVG( PIXEL_8x16, pix[0], 8, pix[1], 8 );
- i_part_cost_bi += h->pixf.mbcmp[PIXEL_8x16]( a->l0.me8x16[i].p_fenc[0], h->mb.pic.i_stride[0], pix[0], 8 );
+ i_part_cost_bi += h->pixf.mbcmp[PIXEL_8x16]( a->l0.me8x16[i].p_fenc[0], FENC_STRIDE, pix[0], 8 );
i_part_cost = a->l0.me8x16[i].cost;
a->i_mb_partition8x16[i] = D_L0_8x8;
+ (a->i_mb_partition8x16[0]>>2) * 3
+ (a->i_mb_partition8x16[1]>>2);
a->i_cost8x16bi += a->i_lambda * i_mb_b16x8_cost_table[a->i_mb_type8x16];
+}
- if( a->b_mbrd )
+static void x264_mb_analyse_p_rd( x264_t *h, x264_mb_analysis_t *a, int i_satd )
+{
+ int thresh = i_satd * 5/4;
+
+ h->mb.i_type = P_L0;
+ if( a->l0.i_rd16x16 == COST_MAX && a->l0.me16x16.cost <= i_satd * 3/2 )
{
- if( a->i_cost8x16bi < a->i_best_satd )
- a->i_best_satd = a->i_cost8x16bi;
+ h->mb.i_partition = D_16x16;
+ x264_analyse_update_cache( h, a );
+ a->l0.i_rd16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+ a->l0.me16x16.cost = a->l0.i_rd16x16;
- if( a->i_cost8x16bi < a->i_best_satd * 3/2 )
+ if( a->l0.i_cost16x8 <= thresh )
+ {
+ h->mb.i_partition = D_16x8;
+ x264_analyse_update_cache( h, a );
+ a->l0.i_cost16x8 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+ else
+ a->l0.i_cost16x8 = COST_MAX;
+
+ if( a->l0.i_cost8x16 <= thresh )
+ {
+ h->mb.i_partition = D_8x16;
+ x264_analyse_update_cache( h, a );
+ a->l0.i_cost8x16 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+ else
+ a->l0.i_cost8x16 = COST_MAX;
+
+ if( a->l0.i_cost8x8 <= thresh )
+ {
+ h->mb.i_type = P_8x8;
+ x264_analyse_update_cache( h, a );
+ a->l0.i_cost8x8 = x264_rd_cost_mb( h, a->i_lambda2 );
+
+ if( h->param.analyse.inter & X264_ANALYSE_PSUB8x8 )
{
- h->mb.i_type = a->i_mb_type8x16;
- h->mb.i_partition = D_8x16;
- a->i_cost8x16bi = x264_rd_cost_mb( h, a->i_lambda2 );
+ /* FIXME: RD per subpartition */
+ int part_bak[4];
+ int i, i_cost;
+ int b_sub8x8 = 0;
+ for( i=0; i<4; i++ )
+ {
+ part_bak[i] = h->mb.i_sub_partition[i];
+ b_sub8x8 |= (part_bak[i] != D_L0_8x8);
+ }
+ if( b_sub8x8 )
+ {
+ h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
+ h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
+ i_cost = x264_rd_cost_mb( h, a->i_lambda2 );
+ if( a->l0.i_cost8x8 < i_cost )
+ {
+ for( i=0; i<4; i++ )
+ h->mb.i_sub_partition[i] = part_bak[i];
+ }
+ else
+ a->l0.i_cost8x8 = i_cost;
+ }
}
- else
- a->i_cost8x16bi = COST_MAX;
}
+ else
+ a->l0.i_cost8x8 = COST_MAX;
}
-static inline void x264_mb_analyse_transform( x264_t *h )
+static void x264_mb_analyse_b_rd( x264_t *h, x264_mb_analysis_t *a, int i_satd_inter )
{
- h->mb.cache.b_transform_8x8_allowed =
- h->param.analyse.b_transform_8x8
- && !IS_INTRA( h->mb.i_type ) && x264_mb_transform_8x8_allowed( h );
+ int thresh = i_satd_inter * 17/16;
+
+ if( a->b_direct_available && a->i_rd16x16direct == COST_MAX )
+ {
+ h->mb.i_type = B_DIRECT;
+ x264_analyse_update_cache( h, a );
+ a->i_rd16x16direct = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+
+ //FIXME not all the update_cache calls are needed
+ h->mb.i_partition = D_16x16;
+ /* L0 */
+ if( a->l0.me16x16.cost <= thresh && a->l0.i_rd16x16 == COST_MAX )
+ {
+ h->mb.i_type = B_L0_L0;
+ x264_analyse_update_cache( h, a );
+ a->l0.i_rd16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
- if( h->mb.cache.b_transform_8x8_allowed )
+ /* L1 */
+ if( a->l1.me16x16.cost <= thresh && a->l1.i_rd16x16 == COST_MAX )
+ {
+ h->mb.i_type = B_L1_L1;
+ x264_analyse_update_cache( h, a );
+ a->l1.i_rd16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+
+ /* BI */
+ if( a->i_cost16x16bi <= thresh && a->i_rd16x16bi == COST_MAX )
+ {
+ h->mb.i_type = B_BI_BI;
+ x264_analyse_update_cache( h, a );
+ a->i_rd16x16bi = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+
+ /* 8x8 */
+ if( a->i_cost8x8bi <= thresh && a->i_rd8x8bi == COST_MAX )
+ {
+ h->mb.i_type = B_8x8;
+ h->mb.i_partition = D_8x8;
+ x264_analyse_update_cache( h, a );
+ a->i_rd8x8bi = x264_rd_cost_mb( h, a->i_lambda2 );
+ x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
+ }
+
+ /* 16x8 */
+ if( a->i_cost16x8bi <= thresh && a->i_rd16x8bi == COST_MAX )
+ {
+ h->mb.i_type = a->i_mb_type16x8;
+ h->mb.i_partition = D_16x8;
+ x264_analyse_update_cache( h, a );
+ a->i_rd16x8bi = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+
+ /* 8x16 */
+ if( a->i_cost8x16bi <= thresh && a->i_rd8x16bi == COST_MAX )
+ {
+ h->mb.i_type = a->i_mb_type8x16;
+ h->mb.i_partition = D_8x16;
+ x264_analyse_update_cache( h, a );
+ a->i_rd8x16bi = x264_rd_cost_mb( h, a->i_lambda2 );
+ }
+}
+
+static void 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;
+
+ 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 );
+ 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 );
+ 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 );
+ 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 );
+ break;
+ }
+}
+
+static inline void x264_mb_analyse_transform( x264_t *h )
+{
+ if( x264_mb_transform_8x8_allowed( h ) && h->param.analyse.b_transform_8x8 )
{
int i_cost4, i_cost8;
/* FIXME only luma mc is needed */
x264_mb_mc( h );
- i_cost8 = h->pixf.sa8d[PIXEL_16x16]( h->mb.pic.p_fenc[0], h->mb.pic.i_stride[0],
- h->mb.pic.p_fdec[0], h->mb.pic.i_stride[0] );
- i_cost4 = h->pixf.satd[PIXEL_16x16]( h->mb.pic.p_fenc[0], h->mb.pic.i_stride[0],
- h->mb.pic.p_fdec[0], h->mb.pic.i_stride[0] );
+ i_cost8 = h->pixf.sa8d[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE,
+ h->mb.pic.p_fdec[0], FDEC_STRIDE );
+ i_cost4 = h->pixf.satd[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE,
+ h->mb.pic.p_fdec[0], FDEC_STRIDE );
h->mb.b_transform_8x8 = i_cost8 < i_cost4;
}
}
-static inline void x264_mb_analyse_transform_rd( x264_t *h, x264_mb_analysis_t *a, int *i_cost )
+static inline void x264_mb_analyse_transform_rd( x264_t *h, x264_mb_analysis_t *a, int *i_satd, int *i_rd )
{
- h->mb.cache.b_transform_8x8_allowed =
- h->param.analyse.b_transform_8x8 && x264_mb_transform_8x8_allowed( h );
-
- if( h->mb.cache.b_transform_8x8_allowed )
+ if( x264_mb_transform_8x8_allowed( h ) && h->param.analyse.b_transform_8x8 )
{
- int i_cost8;
+ int i_rd8;
x264_analyse_update_cache( h, a );
h->mb.b_transform_8x8 = !h->mb.b_transform_8x8;
/* FIXME only luma is needed, but the score for comparison already includes chroma */
- i_cost8 = x264_rd_cost_mb( h, a->i_lambda2 );
+ i_rd8 = x264_rd_cost_mb( h, a->i_lambda2 );
- if( *i_cost >= i_cost8 )
+ if( *i_rd >= i_rd8 )
{
- if( *i_cost > 0 )
- a->i_best_satd = (int64_t)a->i_best_satd * i_cost8 / *i_cost;
- /* prevent a rare division by zero in x264_mb_analyse_intra */
- if( a->i_best_satd == 0 )
- a->i_best_satd = 1;
+ if( *i_rd > 0 )
+ *i_satd = (int64_t)(*i_satd) * i_rd8 / *i_rd;
+ /* prevent a rare division by zero in estimated intra cost */
+ if( *i_satd == 0 )
+ *i_satd = 1;
- *i_cost = i_cost8;
+ *i_rd = i_rd8;
}
else
h->mb.b_transform_8x8 = !h->mb.b_transform_8x8;
int i_cost = COST_MAX;
int i;
- /* init analysis */
- x264_mb_analyse_init( h, &analysis, x264_ratecontrol_qp( h ) );
+ h->mb.i_qp = x264_ratecontrol_qp( h );
+ if( h->param.rc.i_aq_mode )
+ x264_adaptive_quant( h );
+
+ x264_mb_analyse_init( h, &analysis, h->mb.i_qp );
/*--------------------------- Do the analysis ---------------------------*/
if( h->sh.i_type == SLICE_TYPE_I )
{
x264_mb_analyse_intra( h, &analysis, COST_MAX );
+ if( analysis.b_mbrd )
+ x264_intra_rd( h, &analysis, COST_MAX );
- i_cost = analysis.i_sad_i16x16;
+ i_cost = analysis.i_satd_i16x16;
h->mb.i_type = I_16x16;
- if( analysis.i_sad_i4x4 < i_cost )
+ if( analysis.i_satd_i4x4 < i_cost )
{
- i_cost = analysis.i_sad_i4x4;
+ i_cost = analysis.i_satd_i4x4;
h->mb.i_type = I_4x4;
}
- if( analysis.i_sad_i8x8 < i_cost )
+ if( analysis.i_satd_i8x8 < i_cost )
h->mb.i_type = I_8x8;
+
+ if( h->mb.i_subpel_refine >= 7 )
+ x264_intra_rd_refine( h, &analysis );
}
else if( h->sh.i_type == SLICE_TYPE_P )
{
int b_skip = 0;
int i_intra_cost, i_intra_type;
+ h->mc.prefetch_ref( h->mb.pic.p_fref[0][0][h->mb.i_mb_x&3], h->mb.pic.i_stride[0], 0 );
+
/* Fast P_SKIP detection */
- if( h->param.analyse.b_fast_pskip &&
- (( h->mb.i_mb_type_left == P_SKIP ) ||
- ( h->mb.i_mb_type_top == P_SKIP ) ||
- ( h->mb.i_mb_type_topleft == P_SKIP ) ||
- ( h->mb.i_mb_type_topright == P_SKIP )))
+ analysis.b_try_pskip = 0;
+ if( h->param.analyse.b_fast_pskip )
{
- b_skip = x264_macroblock_probe_pskip( h );
+ if( h->param.i_threads > 1 && h->mb.cache.pskip_mv[1] > h->mb.mv_max_spel[1] )
+ // FIXME don't need to check this if the reference frame is done
+ {}
+ else if( h->param.analyse.i_subpel_refine >= 3 )
+ analysis.b_try_pskip = 1;
+ else if( h->mb.i_mb_type_left == P_SKIP ||
+ h->mb.i_mb_type_top == P_SKIP ||
+ h->mb.i_mb_type_topleft == P_SKIP ||
+ h->mb.i_mb_type_topright == P_SKIP )
+ b_skip = x264_macroblock_probe_pskip( h );
}
+ h->mc.prefetch_ref( h->mb.pic.p_fref[0][0][h->mb.i_mb_x&3], h->mb.pic.i_stride[0], 1 );
+
if( b_skip )
{
h->mb.i_type = P_SKIP;
h->mb.i_partition = D_16x16;
+ assert( h->mb.cache.pskip_mv[1] <= h->mb.mv_max_spel[1] || h->param.i_threads == 1 );
}
else
{
int i_type;
int i_partition;
int i_thresh16x8;
+ int i_satd_inter, i_satd_intra;
x264_mb_analyse_load_costs( h, &analysis );
x264_mb_analyse_inter_p16x16( h, &analysis );
- if( analysis.b_mbrd && h->mb.i_type == P_SKIP )
+ if( h->mb.i_type == P_SKIP )
return;
if( flags & X264_ANALYSE_PSUB16x16 )
if( ( flags & X264_ANALYSE_PSUB16x16 ) &&
analysis.l0.i_cost8x8 < analysis.l0.me16x16.cost )
{
- int i;
-
i_type = P_8x8;
i_partition = D_8x8;
- h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
- h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
-
i_cost = analysis.l0.i_cost8x8;
/* Do sub 8x8 */
if( flags & X264_ANALYSE_PSUB8x8 )
{
- int i_cost_bak = i_cost;
- int b_sub8x8 = 0;
for( i = 0; i < 4; i++ )
{
x264_mb_analyse_inter_p4x4( h, &analysis, i );
h->mb.i_sub_partition[i] = D_L0_4x4;
x264_mb_analyse_inter_p8x4( h, &analysis, i );
- if( analysis.l0.i_cost8x4[i] < i_cost8x8 )
- {
- h->mb.i_sub_partition[i] = D_L0_8x4;
- i_cost8x8 = analysis.l0.i_cost8x4[i];
- }
+ COPY2_IF_LT( i_cost8x8, analysis.l0.i_cost8x4[i],
+ h->mb.i_sub_partition[i], D_L0_8x4 );
x264_mb_analyse_inter_p4x8( h, &analysis, i );
- if( analysis.l0.i_cost4x8[i] < i_cost8x8 )
- {
- h->mb.i_sub_partition[i] = D_L0_4x8;
- i_cost8x8 = analysis.l0.i_cost4x8[i];
- }
+ COPY2_IF_LT( i_cost8x8, analysis.l0.i_cost4x8[i],
+ h->mb.i_sub_partition[i], D_L0_4x8 );
i_cost += i_cost8x8 - analysis.l0.me8x8[i].cost;
- b_sub8x8 = 1;
}
x264_mb_cache_mv_p8x8( h, &analysis, i );
}
- /* TODO: RD per subpartition */
- if( b_sub8x8 && analysis.b_mbrd )
- {
- i_cost = x264_rd_cost_mb( h, analysis.i_lambda2 );
- if( i_cost > i_cost_bak )
- {
- i_cost = i_cost_bak;
- h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
- h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
- }
- }
+ analysis.l0.i_cost8x8 = i_cost;
}
}
/* Now do 16x8/8x16 */
i_thresh16x8 = analysis.l0.me8x8[1].cost_mv + analysis.l0.me8x8[2].cost_mv;
- if( analysis.b_mbrd )
- i_thresh16x8 = i_thresh16x8 * analysis.i_lambda2 / analysis.i_lambda;
if( ( flags & X264_ANALYSE_PSUB16x16 ) &&
analysis.l0.i_cost8x8 < analysis.l0.me16x16.cost + i_thresh16x8 )
{
x264_mb_analyse_inter_p16x8( h, &analysis );
- if( analysis.l0.i_cost16x8 < i_cost )
- {
- i_type = P_L0;
- i_partition = D_16x8;
- i_cost = analysis.l0.i_cost16x8;
- }
+ COPY3_IF_LT( i_cost, analysis.l0.i_cost16x8, i_type, P_L0, i_partition, D_16x8 );
x264_mb_analyse_inter_p8x16( h, &analysis );
- if( analysis.l0.i_cost8x16 < i_cost )
- {
- i_type = P_L0;
- i_partition = D_8x16;
- i_cost = analysis.l0.i_cost8x16;
- }
+ COPY3_IF_LT( i_cost, analysis.l0.i_cost8x16, i_type, P_L0, i_partition, D_8x16 );
}
h->mb.i_partition = i_partition;
//FIXME mb_type costs?
if( analysis.b_mbrd )
{
- h->mb.i_type = i_type;
- x264_mb_analyse_transform_rd( h, &analysis, &i_cost );
+ /* refine later */
}
else if( i_partition == D_16x16 )
{
}
}
- x264_mb_analyse_intra( h, &analysis, i_cost );
- if( h->mb.b_chroma_me && !analysis.b_mbrd &&
- ( analysis.i_sad_i16x16 < i_cost
- || analysis.i_sad_i8x8 < i_cost
- || analysis.i_sad_i4x4 < i_cost ))
+ if( h->mb.b_chroma_me )
{
x264_mb_analyse_intra_chroma( h, &analysis );
- analysis.i_sad_i16x16 += analysis.i_sad_i8x8chroma;
- analysis.i_sad_i8x8 += analysis.i_sad_i8x8chroma;
- analysis.i_sad_i4x4 += analysis.i_sad_i8x8chroma;
+ x264_mb_analyse_intra( h, &analysis, i_cost - analysis.i_satd_i8x8chroma );
+ analysis.i_satd_i16x16 += analysis.i_satd_i8x8chroma;
+ analysis.i_satd_i8x8 += analysis.i_satd_i8x8chroma;
+ analysis.i_satd_i4x4 += analysis.i_satd_i8x8chroma;
}
+ else
+ x264_mb_analyse_intra( h, &analysis, i_cost );
- i_intra_type = I_16x16;
- i_intra_cost = analysis.i_sad_i16x16;
+ i_satd_inter = i_cost;
+ i_satd_intra = X264_MIN3( analysis.i_satd_i16x16,
+ analysis.i_satd_i8x8,
+ analysis.i_satd_i4x4 );
- if( analysis.i_sad_i8x8 < i_intra_cost )
- {
- i_intra_type = I_8x8;
- i_intra_cost = analysis.i_sad_i8x8;
- }
- if( analysis.i_sad_i4x4 < i_intra_cost )
+ if( analysis.b_mbrd )
{
- i_intra_type = I_4x4;
- i_intra_cost = analysis.i_sad_i4x4;
+ x264_mb_analyse_p_rd( h, &analysis, X264_MIN(i_satd_inter, i_satd_intra) );
+ i_type = P_L0;
+ i_partition = D_16x16;
+ i_cost = analysis.l0.me16x16.cost;
+ COPY2_IF_LT( i_cost, analysis.l0.i_cost16x8, i_partition, D_16x8 );
+ COPY2_IF_LT( i_cost, analysis.l0.i_cost8x16, i_partition, D_8x16 );
+ COPY3_IF_LT( i_cost, analysis.l0.i_cost8x8, i_partition, D_8x8, i_type, P_8x8 );
+ h->mb.i_type = i_type;
+ h->mb.i_partition = i_partition;
+ if( i_cost < COST_MAX )
+ x264_mb_analyse_transform_rd( h, &analysis, &i_satd_inter, &i_cost );
+ x264_intra_rd( h, &analysis, i_satd_inter * 5/4 );
}
- if( i_intra_cost < i_cost )
- {
- i_type = i_intra_type;
- i_cost = i_intra_cost;
- }
+ i_intra_type = I_16x16;
+ i_intra_cost = analysis.i_satd_i16x16;
+ COPY2_IF_LT( i_intra_cost, analysis.i_satd_i8x8, i_intra_type, I_8x8 );
+ COPY2_IF_LT( i_intra_cost, analysis.i_satd_i4x4, i_intra_type, I_4x4 );
+ COPY2_IF_LT( i_cost, i_intra_cost, i_type, i_intra_type );
+
+ if( i_intra_cost == COST_MAX )
+ i_intra_cost = i_cost * i_satd_intra / i_satd_inter + 1;
h->mb.i_type = i_type;
h->stat.frame.i_intra_cost += i_intra_cost;
h->stat.frame.i_inter_cost += i_cost;
+ h->stat.frame.i_mbs_analysed++;
+
+ if( h->mb.i_subpel_refine >= 7 )
+ {
+ if( IS_INTRA( h->mb.i_type ) )
+ {
+ x264_intra_rd_refine( h, &analysis );
+ }
+ 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 );
+ }
+ else if( i_partition == D_16x8 )
+ {
+ 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 );
+ }
+ else if( i_partition == D_8x16 )
+ {
+ 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 );
+ }
+ else if( i_partition == D_8x8 )
+ {
+ int i8x8;
+ 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 );
+ }
+ }
}
}
else if( h->sh.i_type == SLICE_TYPE_B )
int i_bskip_cost = COST_MAX;
int b_skip = 0;
- analysis.b_direct_available = x264_mb_predict_mv_direct16x16( h );
+ h->mb.i_type = B_SKIP;
+ if( h->mb.b_direct_auto_write )
+ {
+ /* direct=auto heuristic: prefer whichever mode allows more Skip macroblocks */
+ for( i = 0; i < 2; i++ )
+ {
+ int b_changed = 1;
+ h->sh.b_direct_spatial_mv_pred ^= 1;
+ analysis.b_direct_available = x264_mb_predict_mv_direct16x16( h, i && analysis.b_direct_available ? &b_changed : NULL );
+ if( analysis.b_direct_available )
+ {
+ if( b_changed )
+ {
+ x264_mb_mc( h );
+ b_skip = x264_macroblock_probe_bskip( h );
+ }
+ h->stat.frame.i_direct_score[ h->sh.b_direct_spatial_mv_pred ] += b_skip;
+ }
+ else
+ b_skip = 0;
+ }
+ }
+ else
+ analysis.b_direct_available = x264_mb_predict_mv_direct16x16( h, NULL );
+
if( analysis.b_direct_available )
{
- h->mb.i_type = B_SKIP;
- x264_mb_mc( h );
- if( analysis.b_mbrd )
+ if( !h->mb.b_direct_auto_write )
+ x264_mb_mc( h );
+ if( h->mb.b_lossless )
+ {
+ /* chance of skip is too small to bother */
+ }
+ else if( analysis.b_mbrd )
{
- i_bskip_cost = h->pixf.ssd[PIXEL_16x16]( h->mb.pic.p_fenc[0], h->mb.pic.i_stride[0],
- h->mb.pic.p_fdec[0], h->mb.pic.i_stride[0] )
- + h->pixf.ssd[PIXEL_8x8]( h->mb.pic.p_fenc[1], h->mb.pic.i_stride[1],
- h->mb.pic.p_fdec[1], h->mb.pic.i_stride[1] )
- + h->pixf.ssd[PIXEL_8x8]( h->mb.pic.p_fenc[2], h->mb.pic.i_stride[2],
- h->mb.pic.p_fdec[2], h->mb.pic.i_stride[2] );
+ i_bskip_cost = ssd_mb( h );
/* 6 = minimum cavlc cost of a non-skipped MB */
- if( i_bskip_cost <= 6 * analysis.i_lambda2 )
+ if( i_bskip_cost <= ((6 * analysis.i_lambda2 + 128) >> 8) )
{
h->mb.i_type = B_SKIP;
x264_analyse_update_cache( h, &analysis );
return;
}
}
- else
+ else if( !h->mb.b_direct_auto_write )
{
/* Conditioning the probe on neighboring block types
* doesn't seem to help speed or quality. */
- b_skip = !h->mb.b_lossless && x264_macroblock_probe_bskip( h );
+ b_skip = x264_macroblock_probe_bskip( h );
}
}
i_type = B_L0_L0;
i_partition = D_16x16;
i_cost = analysis.l0.me16x16.cost;
- if( analysis.l1.me16x16.cost < i_cost )
- {
- i_type = B_L1_L1;
- i_cost = analysis.l1.me16x16.cost;
- }
- if( analysis.i_cost16x16bi < i_cost )
- {
- i_type = B_BI_BI;
- i_cost = analysis.i_cost16x16bi;
- }
- if( analysis.i_cost16x16direct < i_cost )
- {
- i_type = B_DIRECT;
- i_cost = analysis.i_cost16x16direct;
- }
+ COPY2_IF_LT( i_cost, analysis.l1.me16x16.cost, i_type, B_L1_L1 );
+ 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( i_bskip_cost <= i_cost )
+ if( analysis.b_mbrd && analysis.i_cost16x16direct <= i_cost * 33/32 )
{
- h->mb.i_type = B_SKIP;
- x264_analyse_update_cache( h, &analysis );
- return;
+ x264_mb_analyse_b_rd( h, &analysis, i_cost );
+ if( i_bskip_cost < analysis.i_rd16x16direct &&
+ i_bskip_cost < analysis.i_rd16x16bi &&
+ i_bskip_cost < analysis.l0.i_rd16x16 &&
+ i_bskip_cost < analysis.l1.i_rd16x16 )
+ {
+ h->mb.i_type = B_SKIP;
+ x264_analyse_update_cache( h, &analysis );
+ return;
+ }
}
if( flags & X264_ANALYSE_BSUB16x16 )
h->mb.i_sub_partition[2] == h->mb.i_sub_partition[3] )
{
x264_mb_analyse_inter_b16x8( h, &analysis );
- if( analysis.i_cost16x8bi < i_cost )
- {
- i_partition = D_16x8;
- i_cost = analysis.i_cost16x8bi;
- i_type = analysis.i_mb_type16x8;
- }
+ COPY3_IF_LT( i_cost, analysis.i_cost16x8bi,
+ i_type, analysis.i_mb_type16x8,
+ i_partition, D_16x8 );
}
if( h->mb.i_sub_partition[0] == h->mb.i_sub_partition[2] ||
h->mb.i_sub_partition[1] == h->mb.i_sub_partition[3] )
{
x264_mb_analyse_inter_b8x16( h, &analysis );
- if( analysis.i_cost8x16bi < i_cost )
- {
- i_partition = D_8x16;
- i_cost = analysis.i_cost8x16bi;
- i_type = analysis.i_mb_type8x16;
- }
+ COPY3_IF_LT( i_cost, analysis.i_cost8x16bi,
+ i_type, analysis.i_mb_type8x16,
+ i_partition, D_8x16 );
}
}
}
- h->mb.i_partition = i_partition;
-
if( analysis.b_mbrd )
{
- h->mb.i_type = i_type;
- x264_mb_analyse_transform_rd( h, &analysis, &i_cost );
+ /* refine later */
}
/* refine qpel */
else if( i_partition == D_16x16 )
}
}
- /* best intra mode */
x264_mb_analyse_intra( h, &analysis, i_cost );
- if( analysis.i_sad_i16x16 < i_cost )
- {
- i_type = I_16x16;
- i_cost = analysis.i_sad_i16x16;
- }
- if( analysis.i_sad_i8x8 < i_cost )
- {
- i_type = I_8x8;
- i_cost = analysis.i_sad_i8x8;
- }
- if( analysis.i_sad_i4x4 < i_cost )
+ if( analysis.b_mbrd )
{
- i_type = I_4x4;
- i_cost = analysis.i_sad_i4x4;
+ int i_satd_inter = i_cost;
+ x264_mb_analyse_b_rd( h, &analysis, i_satd_inter );
+ i_type = B_SKIP;
+ i_cost = i_bskip_cost;
+ i_partition = D_16x16;
+ COPY2_IF_LT( i_cost, analysis.l0.i_rd16x16, i_type, B_L0_L0 );
+ COPY2_IF_LT( i_cost, analysis.l1.i_rd16x16, i_type, B_L1_L1 );
+ COPY2_IF_LT( i_cost, analysis.i_rd16x16bi, i_type, B_BI_BI );
+ COPY2_IF_LT( i_cost, analysis.i_rd16x16direct, i_type, B_DIRECT );
+ COPY3_IF_LT( i_cost, analysis.i_rd16x8bi, i_type, analysis.i_mb_type16x8, i_partition, D_16x8 );
+ COPY3_IF_LT( i_cost, analysis.i_rd8x16bi, i_type, analysis.i_mb_type8x16, i_partition, D_8x16 );
+ COPY3_IF_LT( i_cost, analysis.i_rd8x8bi, i_type, B_8x8, i_partition, D_8x8 );
+
+ h->mb.i_type = i_type;
+ h->mb.i_partition = i_partition;
+ x264_mb_analyse_transform_rd( h, &analysis, &i_satd_inter, &i_cost );
+ x264_intra_rd( h, &analysis, i_satd_inter * 17/16 );
}
+ COPY2_IF_LT( i_cost, analysis.i_satd_i16x16, i_type, I_16x16 );
+ COPY2_IF_LT( i_cost, analysis.i_satd_i8x8, i_type, I_8x8 );
+ COPY2_IF_LT( i_cost, analysis.i_satd_i4x4, i_type, I_4x4 );
+
h->mb.i_type = i_type;
+ h->mb.i_partition = i_partition;
+
+ if( h->mb.i_subpel_refine >= 7 && IS_INTRA( i_type ) )
+ x264_intra_rd_refine( h, &analysis );
+ else if( h->param.analyse.b_bidir_me )
+ refine_bidir( h, &analysis );
}
}
x264_mb_analyse_transform( h );
h->mb.b_trellis = h->param.analyse.i_trellis;
+ h->mb.b_noise_reduction = h->param.analyse.i_noise_reduction;
+ if( h->mb.b_trellis == 1 || h->mb.b_noise_reduction )
+ h->mb.i_skip_intra = 0;
}
/*-------------------- Update MB from the analysis ----------------------*/
{
case I_4x4:
for( i = 0; i < 16; i++ )
- {
- h->mb.cache.intra4x4_pred_mode[x264_scan8[i]] =
- a->i_predict4x4[block_idx_x[i]][block_idx_y[i]];
- }
+ h->mb.cache.intra4x4_pred_mode[x264_scan8[i]] = a->i_predict4x4[i];
x264_mb_analyse_intra_chroma( h, a );
break;
case I_8x8:
for( i = 0; i < 4; i++ )
- x264_macroblock_cache_intra8x8_pred( h, 2*(i&1), 2*(i>>1),
- a->i_predict8x8[i&1][i>>1] );
+ x264_macroblock_cache_intra8x8_pred( h, 2*(i&1), 2*(i>>1), a->i_predict8x8[i] );
x264_mb_analyse_intra_chroma( h, a );
break;
case P_SKIP:
{
- int mvp[2];
- x264_mb_predict_mv_pskip( h, mvp );
- /* */
h->mb.i_partition = D_16x16;
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
- x264_macroblock_cache_mv ( h, 0, 0, 4, 4, 0, mvp[0], mvp[1] );
+ x264_macroblock_cache_mv ( h, 0, 0, 4, 4, 0, h->mb.cache.pskip_mv[0],
+ h->mb.cache.pskip_mv[1] );
break;
}
break;
}
}
+
+#ifndef NDEBUG
+ if( h->param.i_threads > 1 && !IS_INTRA(h->mb.i_type) )
+ {
+ int l;
+ for( l=0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
+ {
+ int completed;
+ int ref = h->mb.cache.ref[l][x264_scan8[0]];
+ if( ref < 0 )
+ continue;
+ completed = (l ? h->fref1 : h->fref0)[ ref >> h->mb.b_interlaced ]->i_lines_completed;
+ if( (h->mb.cache.mv[l][x264_scan8[15]][1] >> (2 - h->mb.b_interlaced)) + h->mb.i_mb_y*16 > completed )
+ {
+ x264_log( h, X264_LOG_WARNING, "internal error (MV out of thread range)\n");
+ fprintf(stderr, "mb type: %d \n", h->mb.i_type);
+ fprintf(stderr, "mv: l%dr%d (%d,%d) \n", l, ref,
+ h->mb.cache.mv[l][x264_scan8[15]][0],
+ h->mb.cache.mv[l][x264_scan8[15]][1] );
+ fprintf(stderr, "limit: %d \n", h->mb.mv_max_spel[1]);
+ fprintf(stderr, "mb_xy: %d,%d \n", h->mb.i_mb_x, h->mb.i_mb_y);
+ fprintf(stderr, "completed: %d \n", completed );
+ x264_log( h, X264_LOG_WARNING, "recovering by using intra mode\n");
+ x264_mb_analyse_intra( h, a, COST_MAX );
+ h->mb.i_type = I_16x16;
+ h->mb.i_intra16x16_pred_mode = a->i_predict16x16;
+ x264_mb_analyse_intra_chroma( h, a );
+ }
+ }
+ }
+#endif
}
-#include "slicetype_decision.c"
+#include "slicetype.c"
projects / x264 / blobdiff