#include <unistd.h>
#include "common/common.h"
-#include "common/cpu.h"
#include "macroblock.h"
#include "me.h"
#include "ratecontrol.h"
int i_cost16x16direct;
int i_cost8x8bi;
int i_cost8x8direct[4];
+ int i_satd8x8[3][4]; /* [L0,L1,BI][8x8 0..3] SATD only */
+ int i_cost_est16x8[2]; /* Per-partition estimated cost */
+ int i_cost_est8x16[2];
int i_cost16x8bi;
int i_cost8x16bi;
int i_rd16x16bi;
} x264_mb_analysis_t;
/* lambda = pow(2,qp/6-2) */
-const uint8_t x264_lambda_tab[52] = {
- 1, 1, 1, 1, 1, 1, 1, 1, /* 0-7 */
- 1, 1, 1, 1, /* 8-11 */
- 1, 1, 1, 1, 2, 2, 2, 2, /* 12-19 */
- 3, 3, 3, 4, 4, 4, 5, 6, /* 20-27 */
- 6, 7, 8, 9,10,11,13,14, /* 28-35 */
- 16,18,20,23,25,29,32,36, /* 36-43 */
- 40,45,51,57,64,72,81,91 /* 44-51 */
+const uint16_t x264_lambda_tab[QP_MAX_MAX+1] = {
+ 1, 1, 1, 1, 1, 1, 1, 1, /* 0- 7 */
+ 1, 1, 1, 1, 1, 1, 1, 1, /* 8-15 */
+ 2, 2, 2, 2, 3, 3, 3, 4, /* 16-23 */
+ 4, 4, 5, 6, 6, 7, 8, 9, /* 24-31 */
+ 10, 11, 13, 14, 16, 18, 20, 23, /* 32-39 */
+ 25, 29, 32, 36, 40, 45, 51, 57, /* 40-47 */
+ 64, 72, 81, 91, 102, 114, 128, 144, /* 48-55 */
+ 161, 181, 203, 228, 256, 287, 323, 362, /* 56-63 */
};
/* lambda2 = pow(lambda,2) * .9 * 256 */
-const int x264_lambda2_tab[52] = {
- 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 */
+const int x264_lambda2_tab[QP_MAX_MAX+1] = {
+ 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, 2378021, 2996119, 3774873, 4756042, /* 48-55 */
+5992238,7549747,9512085,11984476,15099494,19024170,23968953,30198988, /* 56-63 */
};
const uint8_t x264_exp2_lut[64] = {
// should the intra and inter lambdas be different?
// I'm just matching the behaviour of deadzone quant.
-static const int x264_trellis_lambda2_tab[2][52] = {
+static const int x264_trellis_lambda2_tab[2][QP_MAX_MAX+1] = {
// inter lambda = .85 * .85 * 2**(qp/3. + 10 - LAMBDA_BITS)
- { 46, 58, 73, 92, 117, 147,
- 185, 233, 294, 370, 466, 587,
- 740, 932, 1174, 1480, 1864, 2349,
- 2959, 3728, 4697, 5918, 7457, 9395,
- 11837, 14914, 18790, 23674, 29828, 37581,
- 47349, 59656, 75163, 94699, 119313, 150326,
- 189399, 238627, 300652, 378798, 477255, 601304,
- 757596, 954511, 1202608, 1515192, 1909022, 2405217,
- 3030384, 3818045, 4810435, 6060769 },
+ { 46, 58, 73, 92, 117, 147,
+ 185, 233, 294, 370, 466, 587,
+ 740, 932, 1174, 1480, 1864, 2349,
+ 2959, 3728, 4697, 5918, 7457, 9395,
+ 11837, 14914, 18790, 23674, 29828, 37581,
+ 47349, 59656, 75163, 94699, 119313, 150326,
+ 189399, 238627, 300652, 378798, 477255, 601304,
+ 757596, 954511, 1202608, 1515192, 1909022, 2405217,
+ 3030384, 3818045, 4810435, 6060769, 7636091, 9620872,
+ 12121539,15272182,19241743,24243077,30544363,38483486,
+ 48486154,61088726,76966972,96972308 },
// intra lambda = .65 * .65 * 2**(qp/3. + 10 - LAMBDA_BITS)
- { 27, 34, 43, 54, 68, 86,
- 108, 136, 172, 216, 273, 343,
- 433, 545, 687, 865, 1090, 1374,
- 1731, 2180, 2747, 3461, 4361, 5494,
- 6922, 8721, 10988, 13844, 17442, 21976,
- 27688, 34885, 43953, 55377, 69771, 87906,
- 110755, 139543, 175813, 221511, 279087, 351627,
- 443023, 558174, 703255, 886046, 1116348, 1406511,
- 1772093, 2232697, 2813022, 3544186 }
+ { 27, 34, 43, 54, 68, 86,
+ 108, 136, 172, 216, 273, 343,
+ 433, 545, 687, 865, 1090, 1374,
+ 1731, 2180, 2747, 3461, 4361, 5494,
+ 6922, 8721, 10988, 13844, 17442, 21976,
+ 27688, 34885, 43953, 55377, 69771, 87906,
+ 110755, 139543, 175813, 221511, 279087, 351627,
+ 443023, 558174, 703255, 886046, 1116348, 1406511,
+ 1772093, 2232697, 2813022, 3544186, 4465396, 5626046,
+ 7088374, 8930791,11252092,14176748,17861583,22504184,
+ 28353495,35723165,45008368,56706990 }
};
static const uint16_t x264_chroma_lambda2_offset_tab[] = {
static void x264_analyse_update_cache( x264_t *h, x264_mb_analysis_t *a );
-static uint16_t x264_cost_ref[92][3][33];
+static uint16_t x264_cost_ref[LAMBDA_MAX+1][3][33];
static UNUSED x264_pthread_mutex_t cost_ref_mutex = X264_PTHREAD_MUTEX_INITIALIZER;
int x264_analyse_init_costs( x264_t *h, int qp )
{
- int i, j;
int lambda = x264_lambda_tab[qp];
if( h->cost_mv[lambda] )
return 0;
/* factor of 4 from qpel, 2 from sign, and 2 because mv can be opposite from mvp */
CHECKED_MALLOC( h->cost_mv[lambda], (4*4*2048 + 1) * sizeof(uint16_t) );
h->cost_mv[lambda] += 2*4*2048;
- for( i = 0; i <= 2*4*2048; i++ )
+ for( int i = 0; i <= 2*4*2048; i++ )
{
h->cost_mv[lambda][-i] =
h->cost_mv[lambda][i] = lambda * (log2f(i+1)*2 + 0.718f + !!i) + .5f;
}
x264_pthread_mutex_lock( &cost_ref_mutex );
- for( i = 0; i < 3; i++ )
- for( j = 0; j < 33; j++ )
+ for( int i = 0; i < 3; i++ )
+ for( int j = 0; j < 33; j++ )
x264_cost_ref[lambda][i][j] = i ? lambda * bs_size_te( i, j ) : 0;
x264_pthread_mutex_unlock( &cost_ref_mutex );
if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->cost_mv_fpel[lambda][0] )
{
- for( j=0; j<4; j++ )
+ for( int j = 0; j < 4; j++ )
{
CHECKED_MALLOC( h->cost_mv_fpel[lambda][j], (4*2048 + 1) * sizeof(uint16_t) );
h->cost_mv_fpel[lambda][j] += 2*2048;
- for( i = -2*2048; i < 2*2048; i++ )
+ for( int i = -2*2048; i < 2*2048; i++ )
h->cost_mv_fpel[lambda][j][i] = h->cost_mv[lambda][i*4+j];
}
}
void x264_analyse_free_costs( x264_t *h )
{
- int i, j;
- for( i = 0; i < 92; i++ )
+ for( int i = 0; i < LAMBDA_MAX+1; i++ )
{
if( h->cost_mv[i] )
x264_free( h->cost_mv[i] - 2*4*2048 );
if( h->cost_mv_fpel[i][0] )
- for( j = 0; j < 4; j++ )
+ for( int j = 0; j < 4; j++ )
x264_free( h->cost_mv_fpel[i][j] - 2*2048 );
}
}
void x264_analyse_weight_frame( x264_t *h, int end )
{
- int j;
- for( j=0; j<h->i_ref0; j++ )
+ for( int j = 0; j < h->i_ref0; j++ )
{
if( h->sh.weight[j][0].weightfn )
{
int width = frame->i_width[0] + 2*PADH;
int i_padv = PADV << h->param.b_interlaced;
int offset, height;
- uint8_t *src = frame->filtered[0] - frame->i_stride[0]*i_padv - PADH;
- int k;
+ pixel *src = frame->filtered[0] - frame->i_stride[0]*i_padv - PADH;
height = X264_MIN( 16 + end + i_padv, h->fref0[j]->i_lines[0] + i_padv*2 ) - h->fenc->i_lines_weighted;
offset = h->fenc->i_lines_weighted*frame->i_stride[0];
h->fenc->i_lines_weighted += height;
if( height )
- {
- for( k = j; k < h->i_ref0; k++ )
+ for( int k = j; k < h->i_ref0; k++ )
if( h->sh.weight[k][0].weightfn )
{
- uint8_t *dst = h->fenc->weighted[k] - h->fenc->i_stride[0]*i_padv - PADH;
+ pixel *dst = h->fenc->weighted[k] - h->fenc->i_stride[0]*i_padv - PADH;
x264_weight_scale_plane( h, dst + offset, frame->i_stride[0],
src + offset, frame->i_stride[0],
width, height, &h->sh.weight[k][0] );
}
- }
break;
}
}
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);
+ int subme = h->param.analyse.i_subpel_refine - (h->sh.i_type == SLICE_TYPE_B);
/* mbrd == 1 -> RD mode decision */
/* mbrd == 2 -> RD refinement */
/* mbrd == 3 -> QPRD */
- a->i_mbrd = (i>=6) + (i>=8) + (h->param.analyse.i_subpel_refine>=10);
+ a->i_mbrd = (subme>=6) + (subme>=8) + (h->param.analyse.i_subpel_refine>=10);
x264_mb_analyse_init_qp( h, a, i_qp );
/* II: Inter part P/B frame */
if( h->sh.i_type != SLICE_TYPE_I )
{
- 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 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-- )
+ for( int 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++ )
+ for( int j = 0; j < i_ref; j++ )
{
x264_frame_cond_wait( fref[j]->orig, thresh );
thread_mvy_range = X264_MIN( thread_mvy_range, fref[j]->orig->i_lines_completed - pix_y );
a->i_cost8x16bi = COST_MAX;
}
else if( h->param.analyse.inter & X264_ANALYSE_PSUB8x8 )
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
a->l0.i_cost4x4[i] =
a->l0.i_cost8x4[i] =
/* For trellis=2, we need to do this for both sizes of DCT, for trellis=1 we only need to use it on the chosen mode. */
static void inline x264_psy_trellis_init( x264_t *h, int do_both_dct )
{
- ALIGNED_16( static uint8_t zero[16*FDEC_STRIDE] ) = {0};
+ ALIGNED_16( static pixel zero[16*FDEC_STRIDE] ) = {0};
if( do_both_dct || h->mb.b_transform_8x8 )
h->dctf.sub16x16_dct8( h->mb.pic.fenc_dct8, h->mb.pic.p_fenc[0], zero );
static void x264_mb_analyse_intra_chroma( x264_t *h, x264_mb_analysis_t *a )
{
- int b_merged_satd = !!h->pixf.intra_mbcmp_x3_8x8c && !h->mb.b_lossless;
-
if( a->i_satd_i8x8chroma < COST_MAX )
return;
const int8_t *predict_mode = predict_8x8chroma_mode_available( h->mb.i_neighbour_intra );
/* 8x8 prediction selection for chroma */
- if( predict_mode[3] >= 0 && b_merged_satd )
+ if( predict_mode[3] >= 0 && !h->mb.b_lossless )
{
int satdu[4], satdv[4];
h->pixf.intra_mbcmp_x3_8x8c( h->mb.pic.p_fenc[1], h->mb.pic.p_fdec[1], satdu );
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;
- uint8_t *p_src = h->mb.pic.p_fenc[0];
- uint8_t *p_dst = h->mb.pic.p_fdec[0];
+ pixel *p_src = h->mb.pic.p_fenc[0];
+ pixel *p_dst = h->mb.pic.p_fdec[0];
+ static const int8_t intra_analysis_shortcut[2][2][5] =
+ {{{I_PRED_4x4_HU, -1},
+ {I_PRED_4x4_DDL, I_PRED_4x4_VL, -1}},
+ {{I_PRED_4x4_DDR, I_PRED_4x4_HD, I_PRED_4x4_HU, -1},
+ {I_PRED_4x4_DDL, I_PRED_4x4_DDR, I_PRED_4x4_VR, I_PRED_4x4_VL, -1}}};
- int i, idx;
- int b_merged_satd = !!h->pixf.intra_mbcmp_x3_16x16 && !h->mb.b_lossless;
+ int idx;
+ int lambda = a->i_lambda;
/*---------------- Try all mode and calculate their score ---------------*/
/* 16x16 prediction selection */
const int8_t *predict_mode = predict_16x16_mode_available( h->mb.i_neighbour_intra );
- if( b_merged_satd && predict_mode[3] >= 0 )
+ /* Not heavily tuned */
+ static const uint8_t i16x16_thresh_lut[11] = { 2, 2, 2, 3, 3, 4, 4, 4, 4, 4, 4 };
+ int i16x16_thresh = a->b_fast_intra ? (i16x16_thresh_lut[h->mb.i_subpel_refine]*i_satd_inter)>>1 : COST_MAX;
+
+ if( !h->mb.b_lossless && predict_mode[3] >= 0 )
{
h->pixf.intra_mbcmp_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_satd_i16x16_dir[0] += lambda * bs_size_ue(0);
+ a->i_satd_i16x16_dir[1] += lambda * bs_size_ue(1);
+ a->i_satd_i16x16_dir[2] += lambda * bs_size_ue(2);
+ COPY2_IF_LT( a->i_satd_i16x16, a->i_satd_i16x16_dir[0], a->i_predict16x16, 0 );
+ COPY2_IF_LT( a->i_satd_i16x16, a->i_satd_i16x16_dir[1], a->i_predict16x16, 1 );
+ COPY2_IF_LT( a->i_satd_i16x16, a->i_satd_i16x16_dir[2], a->i_predict16x16, 2 );
+
+ /* Plane is expensive, so don't check it unless one of the previous modes was useful. */
+ if( a->i_satd_i16x16 <= i16x16_thresh )
{
- 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 );
+ 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 );
+ a->i_satd_i16x16_dir[I_PRED_16x16_P] += lambda * bs_size_ue(3);
+ COPY2_IF_LT( a->i_satd_i16x16, a->i_satd_i16x16_dir[I_PRED_16x16_P], a->i_predict16x16, 3 );
}
}
else
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] );
+ 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;
}
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];
+ a->i_satd_i16x16 += lambda * i_mb_b_cost_table[I_16x16];
- /* Not heavily tuned */
- const uint8_t i16x16_thresh[11] = { 2, 2, 2, 3, 3, 4, 4, 4, 4, 4, 4 };
- int thresh = i16x16_thresh[h->mb.i_subpel_refine];
- if( a->b_fast_intra && a->i_satd_i16x16 > (thresh*i_satd_inter)>>1 )
+ if( a->i_satd_i16x16 > i16x16_thresh )
return;
/* 8x8 prediction selection */
if( flags & X264_ANALYSE_I8x8 )
{
- ALIGNED_ARRAY_16( uint8_t, edge,[33] );
+ ALIGNED_ARRAY_16( pixel, 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->i_mbrd ? COST_MAX : X264_MIN( i_satd_inter, a->i_satd_i16x16 );
// FIXME some bias like in i4x4?
- int i_cost = a->i_lambda * 4; /* base predmode costs */
+ int i_cost = lambda * 4; /* base predmode costs */
h->mb.i_cbp_luma = 0;
- b_merged_satd = h->pixf.intra_mbcmp_x3_8x8 && !h->mb.b_lossless;
if( h->sh.i_type == SLICE_TYPE_B )
- i_cost += a->i_lambda * i_mb_b_cost_table[I_8x8];
+ i_cost += 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;
+ pixel *p_src_by = p_src + 8*x + 8*y*FENC_STRIDE;
+ pixel *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_mode = predict_4x4_mode_available( h->mb.i_neighbour8[idx] );
h->predict_8x8_filter( p_dst_by, edge, h->mb.i_neighbour8[idx], ALL_NEIGHBORS );
- if( b_merged_satd && predict_mode[8] >= 0 )
+ if( !h->mb.b_lossless && predict_mode[5] >= 0 )
{
int satd[9];
h->pixf.intra_mbcmp_x3_8x8( p_src_by, edge, satd );
- satd[i_pred_mode] -= 3 * a->i_lambda;
- for( i=2; i>=0; i-- )
+ int favor_vertical = satd[I_PRED_4x4_H] > satd[I_PRED_4x4_V];
+ satd[i_pred_mode] -= 3 * lambda;
+ for( int i = 2; i >= 0; i-- )
{
- int cost = a->i_satd_i8x8_dir[i][idx] = satd[i];
+ int cost = satd[i];
+ a->i_satd_i8x8_dir[i][idx] = cost + 4 * lambda;
COPY2_IF_LT( i_best, cost, a->i_predict8x8[idx], i );
}
- predict_mode += 3;
+
+ /* Take analysis shortcuts: don't analyse modes that are too
+ * far away direction-wise from the favored mode. */
+ if( a->i_mbrd < 1 + a->b_fast_intra )
+ predict_mode = intra_analysis_shortcut[predict_mode[8] >= 0][favor_vertical];
+ else
+ predict_mode += 3;
}
for( ; *predict_mode >= 0 && (i_best >= 0 || a->i_mbrd >= 2); predict_mode++ )
i_satd = sa8d( p_dst_by, FDEC_STRIDE, p_src_by, FENC_STRIDE );
if( i_pred_mode == x264_mb_pred_mode4x4_fix(i_mode) )
- i_satd -= 3 * a->i_lambda;
+ i_satd -= 3 * lambda;
COPY2_IF_LT( i_best, i_satd, a->i_predict8x8[idx], i_mode );
- a->i_satd_i8x8_dir[i_mode][idx] = i_satd + 4 * a->i_lambda;
+ a->i_satd_i8x8_dir[i_mode][idx] = i_satd + 4 * lambda;
}
- i_cost += i_best + 3 * a->i_lambda;
+ i_cost += i_best + 3 * lambda;
if( idx == 3 || i_cost > i_satd_thresh )
break;
i_cost = (i_cost * cost_div_fix8[idx]) >> 8;
}
/* Not heavily tuned */
- const uint8_t i8x8_thresh[11] = { 4, 4, 4, 5, 5, 5, 6, 6, 6, 6, 6 };
- int thresh = i8x8_thresh[h->mb.i_subpel_refine];
- if( X264_MIN(i_cost, a->i_satd_i16x16) > (i_satd_inter*thresh)>>2 )
+ static const uint8_t i8x8_thresh[11] = { 4, 4, 4, 5, 5, 5, 6, 6, 6, 6, 6 };
+ if( X264_MIN(i_cost, a->i_satd_i16x16) > (i_satd_inter*i8x8_thresh[h->mb.i_subpel_refine])>>2 )
return;
}
/* 4x4 prediction selection */
if( flags & X264_ANALYSE_I4x4 )
{
- int i_cost = a->i_lambda * (24+16); /* 24from JVT (SATD0), 16 from base predmode costs */
+ int i_cost = lambda * (24+16); /* 24from JVT (SATD0), 16 from base predmode costs */
int i_satd_thresh = X264_MIN3( i_satd_inter, a->i_satd_i16x16, a->i_satd_i8x8 );
h->mb.i_cbp_luma = 0;
- b_merged_satd = h->pixf.intra_mbcmp_x3_4x4 && !h->mb.b_lossless;
+
if( a->i_mbrd )
i_satd_thresh = i_satd_thresh * (10-a->b_fast_intra)/8;
if( h->sh.i_type == SLICE_TYPE_B )
- i_cost += a->i_lambda * i_mb_b_cost_table[I_4x4];
+ i_cost += lambda * i_mb_b_cost_table[I_4x4];
for( idx = 0;; idx++ )
{
- uint8_t *p_src_by = p_src + block_idx_xy_fenc[idx];
- uint8_t *p_dst_by = p_dst + block_idx_xy_fdec[idx];
+ pixel *p_src_by = p_src + block_idx_xy_fenc[idx];
+ pixel *p_dst_by = p_dst + block_idx_xy_fdec[idx];
int i_best = COST_MAX;
int i_pred_mode = x264_mb_predict_intra4x4_mode( h, idx );
- const int8_t *predict_mode = predict_4x4_mode_available( h->mb.i_neighbour4[idx] );
+ predict_mode = predict_4x4_mode_available( h->mb.i_neighbour4[idx] );
if( (h->mb.i_neighbour4[idx] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
/* emulate missing topright samples */
- M32( &p_dst_by[4 - FDEC_STRIDE] ) = p_dst_by[3 - FDEC_STRIDE] * 0x01010101U;
+ MPIXEL_X4( &p_dst_by[4 - FDEC_STRIDE] ) = PIXEL_SPLAT_X4( p_dst_by[3 - FDEC_STRIDE] );
- if( b_merged_satd && predict_mode[5] >= 0 )
+ if( !h->mb.b_lossless && predict_mode[5] >= 0 )
{
int satd[9];
h->pixf.intra_mbcmp_x3_4x4( p_src_by, p_dst_by, satd );
- satd[i_pred_mode] -= 3 * a->i_lambda;
- for( i=2; i>=0; i-- )
+ int favor_vertical = satd[I_PRED_4x4_H] > satd[I_PRED_4x4_V];
+ satd[i_pred_mode] -= 3 * lambda;
+ for( int i = 2; i >= 0; i-- )
COPY2_IF_LT( i_best, satd[i], a->i_predict4x4[idx], i );
- predict_mode += 3;
+
+ /* Take analysis shortcuts: don't analyse modes that are too
+ * far away direction-wise from the favored mode. */
+ if( a->i_mbrd < 1 + a->b_fast_intra )
+ predict_mode = intra_analysis_shortcut[predict_mode[8] >= 0][favor_vertical];
+ else
+ predict_mode += 3;
}
if( i_best > 0 )
i_satd = h->pixf.mbcmp[PIXEL_4x4]( p_dst_by, FDEC_STRIDE, p_src_by, FENC_STRIDE );
if( i_pred_mode == x264_mb_pred_mode4x4_fix(i_mode) )
{
- i_satd -= a->i_lambda * 3;
+ i_satd -= lambda * 3;
if( i_satd <= 0 )
{
i_best = i_satd;
COPY2_IF_LT( i_best, i_satd, a->i_predict4x4[idx], i_mode );
}
}
- i_cost += i_best + 3 * a->i_lambda;
+ i_cost += i_best + 3 * lambda;
if( i_cost > i_satd_thresh || idx == 15 )
break;
static void x264_intra_rd_refine( x264_t *h, x264_mb_analysis_t *a )
{
- uint8_t *p_dst = h->mb.pic.p_fdec[0];
-
- int i, idx, x, y;
- int i_mode, i_thresh;
+ pixel *p_dst = h->mb.pic.p_fdec[0];
uint64_t i_satd, i_best;
h->mb.i_skip_intra = 0;
{
int old_pred_mode = a->i_predict16x16;
const int8_t *predict_mode = predict_16x16_mode_available( h->mb.i_neighbour_intra );
- i_thresh = a->i_satd_i16x16_dir[old_pred_mode] * 9/8;
+ int i_thresh = a->i_satd_i16x16_dir[old_pred_mode] * 9/8;
i_best = a->i_satd_i16x16;
for( ; *predict_mode >= 0; predict_mode++ )
{
{
int8_t predict_mode_sorted[4];
int i_max;
- i_thresh = a->i_satd_i8x8chroma * 5/4;
+ int i_thresh = a->i_satd_i8x8chroma * 5/4;
for( i_max = 0; *predict_mode >= 0; predict_mode++ )
{
- i_mode = *predict_mode;
+ int i_mode = *predict_mode;
if( a->i_satd_i8x8chroma_dir[i_mode] < i_thresh && i_mode != a->i_predict8x8chroma )
predict_mode_sorted[i_max++] = i_mode;
}
* 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++ )
+ for( int i = 0; i < i_max; i++ )
{
- i_mode = predict_mode_sorted[i];
+ int i_mode = predict_mode_sorted[i];
if( h->mb.b_lossless )
x264_predict_lossless_8x8_chroma( h, 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
+ pixel4 pels[4] = {0}; // doesn't need initting, just shuts up a gcc warning
int i_nnz = 0;
- for( idx = 0; idx < 16; idx++ )
+ for( int idx = 0; idx < 16; idx++ )
{
- uint8_t *p_dst_by = p_dst + block_idx_xy_fdec[idx];
+ pixel *p_dst_by = p_dst + block_idx_xy_fdec[idx];
i_best = COST_MAX64;
- const int8_t *predict_mode = predict_4x4_mode_available( h->mb.i_neighbour4[idx] );
+ predict_mode = predict_4x4_mode_available( h->mb.i_neighbour4[idx] );
if( (h->mb.i_neighbour4[idx] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
/* emulate missing topright samples */
- M32( &p_dst_by[4 - FDEC_STRIDE] ) = p_dst_by[3 - FDEC_STRIDE] * 0x01010101U;
+ MPIXEL_X4( &p_dst_by[4 - FDEC_STRIDE] ) = PIXEL_SPLAT_X4( p_dst_by[3 - FDEC_STRIDE] );
for( ; *predict_mode >= 0; predict_mode++ )
{
- i_mode = *predict_mode;
+ int i_mode = *predict_mode;
if( h->mb.b_lossless )
x264_predict_lossless_4x4( h, p_dst_by, idx, i_mode );
else
{
a->i_predict4x4[idx] = i_mode;
i_best = i_satd;
- pels[0] = M32( p_dst_by+0*FDEC_STRIDE );
- pels[1] = M32( p_dst_by+1*FDEC_STRIDE );
- pels[2] = M32( p_dst_by+2*FDEC_STRIDE );
- pels[3] = M32( p_dst_by+3*FDEC_STRIDE );
+ pels[0] = MPIXEL_X4( p_dst_by+0*FDEC_STRIDE );
+ pels[1] = MPIXEL_X4( p_dst_by+1*FDEC_STRIDE );
+ pels[2] = MPIXEL_X4( p_dst_by+2*FDEC_STRIDE );
+ pels[3] = MPIXEL_X4( p_dst_by+3*FDEC_STRIDE );
i_nnz = h->mb.cache.non_zero_count[x264_scan8[idx]];
}
}
- M32( p_dst_by+0*FDEC_STRIDE ) = pels[0];
- M32( p_dst_by+1*FDEC_STRIDE ) = pels[1];
- M32( p_dst_by+2*FDEC_STRIDE ) = pels[2];
- M32( p_dst_by+3*FDEC_STRIDE ) = pels[3];
+ MPIXEL_X4( p_dst_by+0*FDEC_STRIDE ) = pels[0];
+ MPIXEL_X4( p_dst_by+1*FDEC_STRIDE ) = pels[1];
+ MPIXEL_X4( p_dst_by+2*FDEC_STRIDE ) = pels[2];
+ MPIXEL_X4( 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 )
{
- ALIGNED_ARRAY_16( uint8_t, edge,[33] );
- for( idx = 0; idx < 4; idx++ )
+ ALIGNED_ARRAY_16( pixel, edge,[33] );
+ for( int idx = 0; idx < 4; idx++ )
{
- uint64_t pels_h = 0;
- uint8_t pels_v[7];
+ pixel4 pels_h[2] = {0};
+ pixel pels_v[7] = {0};
uint16_t i_nnz[2] = {0}; //shut up gcc
- uint8_t *p_dst_by;
- int j;
+ pixel *p_dst_by;
int cbp_luma_new = 0;
- i_thresh = a->i_satd_i8x8_dir[a->i_predict8x8[idx]][idx] * 11/8;
+ int i_thresh = a->i_satd_i8x8_dir[a->i_predict8x8[idx]][idx] * 11/8;
i_best = COST_MAX64;
- x = idx&1;
- y = idx>>1;
+ int x = idx&1;
+ int y = idx>>1;
+ int s8 = X264_SCAN8_0 + 2*x + 16*y;
p_dst_by = p_dst + 8*x + 8*y*FDEC_STRIDE;
- const int8_t *predict_mode = predict_4x4_mode_available( h->mb.i_neighbour8[idx] );
+ predict_mode = predict_4x4_mode_available( h->mb.i_neighbour8[idx] );
h->predict_8x8_filter( p_dst_by, edge, h->mb.i_neighbour8[idx], ALL_NEIGHBORS );
for( ; *predict_mode >= 0; predict_mode++ )
{
- i_mode = *predict_mode;
+ int i_mode = *predict_mode;
if( a->i_satd_i8x8_dir[i_mode][idx] > i_thresh )
continue;
cbp_luma_new = h->mb.i_cbp_luma;
i_best = i_satd;
- pels_h = M64( p_dst_by+7*FDEC_STRIDE );
+ pels_h[0] = MPIXEL_X4( p_dst_by+7*FDEC_STRIDE+0 );
+ pels_h[1] = MPIXEL_X4( p_dst_by+7*FDEC_STRIDE+4 );
if( !(idx&1) )
- for( j=0; j<7; j++ )
+ for( int j = 0; j < 7; j++ )
pels_v[j] = p_dst_by[7+j*FDEC_STRIDE];
- i_nnz[0] = M16( &h->mb.cache.non_zero_count[x264_scan8[4*idx+0]] );
- i_nnz[1] = M16( &h->mb.cache.non_zero_count[x264_scan8[4*idx+2]] );
+ i_nnz[0] = M16( &h->mb.cache.non_zero_count[s8 + 0*8] );
+ i_nnz[1] = M16( &h->mb.cache.non_zero_count[s8 + 1*8] );
}
}
a->i_cbp_i8x8_luma = cbp_luma_new;
- M64( p_dst_by+7*FDEC_STRIDE ) = pels_h;
+ MPIXEL_X4( p_dst_by+7*FDEC_STRIDE+0 ) = pels_h[0];
+ MPIXEL_X4( p_dst_by+7*FDEC_STRIDE+4 ) = pels_h[1];
if( !(idx&1) )
- for( j=0; j<7; j++ )
+ for( int j = 0; j < 7; j++ )
p_dst_by[7+j*FDEC_STRIDE] = pels_v[j];
- M16( &h->mb.cache.non_zero_count[x264_scan8[4*idx+0]] ) = i_nnz[0];
- M16( &h->mb.cache.non_zero_count[x264_scan8[4*idx+2]] ) = i_nnz[1];
+ M16( &h->mb.cache.non_zero_count[s8 + 0*8] ) = i_nnz[0];
+ M16( &h->mb.cache.non_zero_count[s8 + 1*8] ) = i_nnz[1];
x264_macroblock_cache_intra8x8_pred( h, 2*x, 2*y, a->i_predict8x8[idx] );
}
static void x264_mb_analyse_inter_p16x16( x264_t *h, x264_mb_analysis_t *a )
{
x264_me_t m;
- int i_ref, i_mvc;
+ int i_mvc;
ALIGNED_4( int16_t mvc[8][2] );
int i_halfpel_thresh = INT_MAX;
int *p_halfpel_thresh = h->mb.pic.i_fref[0]>1 ? &i_halfpel_thresh : NULL;
LOAD_FENC( &m, h->mb.pic.p_fenc, 0, 0 );
a->l0.me16x16.cost = INT_MAX;
- for( i_ref = 0; i_ref < h->mb.pic.i_fref[0]; i_ref++ )
+ for( int i_ref = 0; i_ref < h->mb.pic.i_fref[0]; i_ref++ )
{
m.i_ref_cost = REF_COST( 0, i_ref );
i_halfpel_thresh -= m.i_ref_cost;
static void x264_mb_analyse_inter_p8x8_mixed_ref( x264_t *h, x264_mb_analysis_t *a )
{
x264_me_t m;
- int i_ref, i;
- uint8_t **p_fenc = h->mb.pic.p_fenc;
+ pixel **p_fenc = h->mb.pic.p_fenc;
int i_maxref = h->mb.pic.i_fref[0]-1;
h->mb.i_partition = D_8x8;
}
#undef CHECK_NEIGHBOUR
- for( i_ref = 0; i_ref <= i_maxref; i_ref++ )
+ for( int i_ref = 0; i_ref <= i_maxref; i_ref++ )
CP32( a->l0.mvc[i_ref][0], h->mb.mvr[0][i_ref][h->mb.i_mb_xy] );
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
x264_me_t *l0m = &a->l0.me8x8[i];
- const int x8 = i%2;
- const int y8 = i/2;
+ int x8 = i&1;
+ int y8 = i>>1;
m.i_pixel = PIXEL_8x8;
LOAD_FENC( &m, p_fenc, 8*x8, 8*y8 );
l0m->cost = INT_MAX;
- for( i_ref = 0; i_ref <= i_maxref || i_ref == h->mb.ref_blind_dupe; )
+ for( int i_ref = 0; i_ref <= i_maxref || i_ref == h->mb.ref_blind_dupe; )
{
m.i_ref_cost = REF_COST( 0, i_ref );
x264_macroblock_cache_mv_ptr( h, 2*x8, 2*y8, 2, 2, 0, l0m->mv );
x264_macroblock_cache_ref( h, 2*x8, 2*y8, 2, 2, 0, l0m->i_ref );
+ a->i_satd8x8[0][i] = l0m->cost - ( l0m->cost_mv + l0m->i_ref_cost );
+
/* If CABAC is on and we're not doing sub-8x8 analysis, the costs
are effectively zero. */
if( !h->param.b_cabac || (h->param.analyse.inter & X264_ANALYSE_PSUB8x8) )
* don't bother analysing the dupes. */
const int i_ref = h->mb.ref_blind_dupe == a->l0.me16x16.i_ref ? 0 : a->l0.me16x16.i_ref;
const int i_ref_cost = h->param.b_cabac || i_ref ? REF_COST( 0, i_ref ) : 0;
- uint8_t **p_fenc = h->mb.pic.p_fenc;
+ pixel **p_fenc = h->mb.pic.p_fenc;
int i_mvc;
int16_t (*mvc)[2] = a->l0.mvc[i_ref];
- int i;
/* XXX Needed for x264_mb_predict_mv */
h->mb.i_partition = D_8x8;
i_mvc = 1;
CP32( mvc[0], a->l0.me16x16.mv );
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
x264_me_t *m = &a->l0.me8x8[i];
- const int x8 = i%2;
- const int y8 = i/2;
+ int x8 = i&1;
+ int y8 = i>>1;
m->i_pixel = PIXEL_8x8;
m->i_ref_cost = i_ref_cost;
CP32( mvc[i_mvc], m->mv );
i_mvc++;
+ a->i_satd8x8[0][i] = m->cost - m->cost_mv;
+
/* mb type cost */
m->cost += i_ref_cost;
if( !h->param.b_cabac || (h->param.analyse.inter & X264_ANALYSE_PSUB8x8) )
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 )
+static void x264_mb_analyse_inter_p16x8( x264_t *h, x264_mb_analysis_t *a, int i_best_satd )
{
x264_me_t m;
- uint8_t **p_fenc = h->mb.pic.p_fenc;
+ pixel **p_fenc = h->mb.pic.p_fenc;
ALIGNED_4( int16_t mvc[3][2] );
- int i, j;
/* XXX Needed for x264_mb_predict_mv */
h->mb.i_partition = D_16x8;
- for( i = 0; i < 2; i++ )
+ for( int i = 0; i < 2; i++ )
{
x264_me_t *l0m = &a->l0.me16x8[i];
const int minref = X264_MIN( a->l0.me8x8[2*i].i_ref, a->l0.me8x8[2*i+1].i_ref );
LOAD_FENC( &m, p_fenc, 0, 8*i );
l0m->cost = INT_MAX;
- for( j = 0; j < i_ref8s; j++ )
+ for( int j = 0; j < i_ref8s; j++ )
{
const int i_ref = ref8[j];
m.i_ref_cost = REF_COST( 0, i_ref );
if( m.cost < l0m->cost )
h->mc.memcpy_aligned( l0m, &m, sizeof(x264_me_t) );
}
+
+ /* Early termination based on the current SATD score of partition[0]
+ plus the estimated SATD score of partition[1] */
+ if( !i && l0m->cost + a->i_cost_est16x8[1] > i_best_satd * (4 + !!a->i_mbrd) / 4 )
+ {
+ a->l0.i_cost16x8 = COST_MAX;
+ return;
+ }
+
x264_macroblock_cache_mv_ptr( h, 0, 2*i, 4, 2, 0, l0m->mv );
x264_macroblock_cache_ref( h, 0, 2*i, 4, 2, 0, l0m->i_ref );
}
a->l0.i_cost16x8 = a->l0.me16x8[0].cost + a->l0.me16x8[1].cost;
}
-static void x264_mb_analyse_inter_p8x16( x264_t *h, x264_mb_analysis_t *a )
+static void x264_mb_analyse_inter_p8x16( x264_t *h, x264_mb_analysis_t *a, int i_best_satd )
{
x264_me_t m;
- uint8_t **p_fenc = h->mb.pic.p_fenc;
+ pixel **p_fenc = h->mb.pic.p_fenc;
ALIGNED_4( int16_t mvc[3][2] );
- int i, j;
/* XXX Needed for x264_mb_predict_mv */
h->mb.i_partition = D_8x16;
- for( i = 0; i < 2; i++ )
+ for( int i = 0; i < 2; i++ )
{
x264_me_t *l0m = &a->l0.me8x16[i];
const int minref = X264_MIN( a->l0.me8x8[i].i_ref, a->l0.me8x8[i+2].i_ref );
LOAD_FENC( &m, p_fenc, 8*i, 0 );
l0m->cost = INT_MAX;
- for( j = 0; j < i_ref8s; j++ )
+ for( int j = 0; j < i_ref8s; j++ )
{
const int i_ref = ref8[j];
m.i_ref_cost = REF_COST( 0, i_ref );
if( m.cost < l0m->cost )
h->mc.memcpy_aligned( l0m, &m, sizeof(x264_me_t) );
}
+
+ /* Early termination based on the current SATD score of partition[0]
+ plus the estimated SATD score of partition[1] */
+ if( !i && l0m->cost + a->i_cost_est8x16[1] > i_best_satd * (4 + !!a->i_mbrd) / 4 )
+ {
+ a->l0.i_cost8x16 = COST_MAX;
+ return;
+ }
+
x264_macroblock_cache_mv_ptr( h, 2*i, 0, 2, 4, 0, l0m->mv );
x264_macroblock_cache_ref( h, 2*i, 0, 2, 4, 0, l0m->i_ref );
}
a->l0.i_cost8x16 = a->l0.me8x16[0].cost + a->l0.me8x16[1].cost;
}
-static int x264_mb_analyse_inter_p4x4_chroma( x264_t *h, x264_mb_analysis_t *a, uint8_t **p_fref, int i8x8, int pixel )
+static int x264_mb_analyse_inter_p4x4_chroma( x264_t *h, x264_mb_analysis_t *a, pixel **p_fref, int i8x8, int size )
{
- ALIGNED_ARRAY_8( uint8_t, pix1,[16*8] );
- uint8_t *pix2 = pix1+8;
+ ALIGNED_ARRAY_8( pixel, pix1,[16*8] );
+ pixel *pix2 = pix1+8;
const int i_stride = h->mb.pic.i_stride[1];
const int or = 4*(i8x8&1) + 2*(i8x8&2)*i_stride;
const int oe = 4*(i8x8&1) + 2*(i8x8&2)*FENC_STRIDE;
weight[1].weightfn[width>>2]( &pix1[x+y*16], 16, &pix1[x+y*16], 16, &weight[1], 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]+mvy_offset, width, height ); \
if( weight[2].weightfn ) \
- weight[1].weightfn[width>>2]( &pix2[x+y*16], 16, &pix2[x+y*16], 16, &weight[2], height );
+ weight[2].weightfn[width>>2]( &pix2[x+y*16], 16, &pix2[x+y*16], 16, &weight[2], height );
- if( pixel == PIXEL_4x4 )
+ if( size == PIXEL_4x4 )
{
x264_me_t *m = a->l0.me4x4[i8x8];
CHROMA4x4MC( 2,2, m[0], 0,0 );
CHROMA4x4MC( 2,2, m[2], 0,2 );
CHROMA4x4MC( 2,2, m[3], 2,2 );
}
- else if( pixel == PIXEL_8x4 )
+ else if( size == PIXEL_8x4 )
{
x264_me_t *m = a->l0.me8x4[i8x8];
CHROMA4x4MC( 4,2, m[0], 0,0 );
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;
+ pixel **p_fref = h->mb.pic.p_fref[0][a->l0.me8x8[i8x8].i_ref];
+ pixel **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 */
h->mb.i_partition = D_8x8;
- for( i4x4 = 0; i4x4 < 4; i4x4++ )
+ for( int i4x4 = 0; i4x4 < 4; i4x4++ )
{
const int idx = 4*i8x8 + i4x4;
const int x4 = block_idx_x[idx];
static void x264_mb_analyse_inter_p8x4( 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;
+ pixel **p_fref = h->mb.pic.p_fref[0][a->l0.me8x8[i8x8].i_ref];
+ pixel **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 */
h->mb.i_partition = D_8x8;
- for( i8x4 = 0; i8x4 < 2; i8x4++ )
+ for( int i8x4 = 0; i8x4 < 2; i8x4++ )
{
const int idx = 4*i8x8 + 2*i8x4;
const int x4 = block_idx_x[idx];
static void x264_mb_analyse_inter_p4x8( 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;
+ pixel **p_fref = h->mb.pic.p_fref[0][a->l0.me8x8[i8x8].i_ref];
+ pixel **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 */
h->mb.i_partition = D_8x8;
- for( i4x8 = 0; i4x8 < 2; i4x8++ )
+ for( int i4x8 = 0; i4x8 < 2; i4x8++ )
{
const int idx = 4*i8x8 + i4x8;
const int x4 = block_idx_x[idx];
/* Assumes that fdec still contains the results of
* x264_mb_predict_mv_direct16x16 and x264_mb_mc */
- uint8_t **p_fenc = h->mb.pic.p_fenc;
- uint8_t **p_fdec = h->mb.pic.p_fdec;
- int i;
+ pixel *p_fenc = h->mb.pic.p_fenc[0];
+ pixel *p_fdec = h->mb.pic.p_fdec[0];
a->i_cost16x16direct = a->i_lambda * i_mb_b_cost_table[B_DIRECT];
- for( i = 0; i < 4; i++ )
- {
- 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][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];
- }
+ if( h->param.analyse.inter & X264_ANALYSE_BSUB16x16 )
+ for( int i = 0; i < 4; i++ )
+ {
+ 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[x+y*FENC_STRIDE], FENC_STRIDE, &p_fdec[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];
+ }
+ else
+ a->i_cost16x16direct += h->pixf.mbcmp[PIXEL_16x16]( p_fenc, FENC_STRIDE, p_fdec, FDEC_STRIDE );
}
static void x264_mb_analyse_inter_b16x16( x264_t *h, x264_mb_analysis_t *a )
{
- ALIGNED_ARRAY_16( uint8_t, pix0,[16*16] );
- ALIGNED_ARRAY_16( uint8_t, pix1,[16*16] );
- uint8_t *src0, *src1;
+ ALIGNED_ARRAY_16( pixel, pix0,[16*16] );
+ ALIGNED_ARRAY_16( pixel, pix1,[16*16] );
+ pixel *src0, *src1;
int stride0 = 16, stride1 = 16;
- int i_ref, i_mvc, l;
+ int i_ref, i_mvc;
ALIGNED_4( int16_t mvc[9][2] );
int try_skip = a->b_try_skip;
int list1_skipped = 0;
/* 16x16 Search on list 0 and list 1 */
a->l0.me16x16.cost = INT_MAX;
a->l1.me16x16.cost = INT_MAX;
- for( l = 1; l >= 0; )
+ for( int l = 1; l >= 0; )
{
x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
/* Fast skip detection. */
if( i_ref == 0 && try_skip )
{
- if( abs(lX->bi16x16.mv[0]-h->mb.cache.direct_mv[l][0][0]) +
- abs(lX->bi16x16.mv[1]-h->mb.cache.direct_mv[l][0][1]) > 1 )
+ if( abs(lX->me16x16.mv[0]-h->mb.cache.direct_mv[l][0][0]) +
+ abs(lX->me16x16.mv[1]-h->mb.cache.direct_mv[l][0][1]) > 1 )
{
try_skip = 0;
}
static inline void x264_mb_cache_mv_p8x8( x264_t *h, x264_mb_analysis_t *a, int i )
{
- const int x = 2*(i%2);
- const int y = 2*(i/2);
+ int x = 2*(i&1);
+ int y = i&2;
switch( h->mb.i_sub_partition[i] )
{
static void x264_mb_load_mv_direct8x8( x264_t *h, int idx )
{
- const int x = 2*(idx&1);
- const int y = 2*(idx>>1);
+ int x = 2*(idx&1);
+ int y = idx&2;
x264_macroblock_cache_ref( h, x, y, 2, 2, 0, h->mb.cache.direct_ref[0][idx] );
x264_macroblock_cache_ref( h, x, y, 2, 2, 1, h->mb.cache.direct_ref[1][idx] );
x264_macroblock_cache_mv_ptr( h, x, y, 2, 2, 0, h->mb.cache.direct_mv[0][idx] );
static inline void x264_mb_cache_mv_b8x8( x264_t *h, x264_mb_analysis_t *a, int i, int b_mvd )
{
- int x = (i%2)*2;
- int y = (i/2)*2;
+ int x = 2*(i&1);
+ int y = i&2;
if( h->mb.i_sub_partition[i] == D_DIRECT_8x8 )
{
x264_mb_load_mv_direct8x8( h, i );
static void x264_mb_analyse_inter_b8x8_mixed_ref( x264_t *h, x264_mb_analysis_t *a )
{
- ALIGNED_ARRAY_8( uint8_t, pix,[2],[8*8] );
- int i_ref, i, l;
+ ALIGNED_ARRAY_8( pixel, pix,[2],[8*8] );
int i_maxref[2] = {h->mb.pic.i_fref[0]-1, h->mb.pic.i_fref[1]-1};
/* early termination: if 16x16 chose ref 0, then evalute no refs older
i_maxref[l] = ref;\
}
- for( l = 0; l < 2; l++ )
+ for( int l = 0; l < 2; l++ )
{
x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
if( i_maxref[l] > 0 && lX->me16x16.i_ref == 0 &&
a->i_cost8x8bi = 0;
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
- int x8 = i%2;
- int y8 = i/2;
+ int x8 = i&1;
+ int y8 = i>>1;
int i_part_cost;
int i_part_cost_bi;
int stride[2] = {8,8};
- uint8_t *src[2];
+ pixel *src[2];
x264_me_t m;
m.i_pixel = PIXEL_8x8;
LOAD_FENC( &m, h->mb.pic.p_fenc, 8*x8, 8*y8 );
- for( l = 0; l < 2; l++ )
+ for( int l = 0; l < 2; l++ )
{
x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
lX->me8x8[i].cost = INT_MAX;
- for( i_ref = 0; i_ref <= i_maxref[l]; i_ref++ )
+ for( int i_ref = 0; i_ref <= i_maxref[l]; i_ref++ )
{
m.i_ref_cost = REF_COST( l, i_ref );;
m.cost += m.i_ref_cost;
if( m.cost < lX->me8x8[i].cost )
+ {
h->mc.memcpy_aligned( &lX->me8x8[i], &m, sizeof(x264_me_t) );
+ a->i_satd8x8[l][i] = m.cost - ( m.cost_mv + m.i_ref_cost );
+ }
/* save mv for predicting other partitions within this MB */
CP32( lX->mvc[i_ref][i+1], m.mv );
h->mc.avg[PIXEL_8x8]( pix[0], 8, src[0], stride[0], src[1], stride[1],
h->mb.bipred_weight[a->l0.me8x8[i].i_ref][a->l1.me8x8[i].i_ref] );
- i_part_cost_bi = h->pixf.mbcmp[PIXEL_8x8]( a->l0.me8x8[i].p_fenc[0], FENC_STRIDE, pix[0], 8 )
- + a->l0.me8x8[i].cost_mv + a->l1.me8x8[i].cost_mv + a->l0.me8x8[i].i_ref_cost
- + a->l1.me8x8[i].i_ref_cost + a->i_lambda * i_sub_mb_b_cost_table[D_BI_8x8];
+ a->i_satd8x8[2][i] = h->pixf.mbcmp[PIXEL_8x8]( a->l0.me8x8[i].p_fenc[0], FENC_STRIDE, pix[0], 8 );
+ i_part_cost_bi = a->i_satd8x8[2][i] + a->l0.me8x8[i].cost_mv + a->l1.me8x8[i].cost_mv
+ + a->l0.me8x8[i].i_ref_cost + a->l1.me8x8[i].i_ref_cost
+ + 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];
static void x264_mb_analyse_inter_b8x8( x264_t *h, x264_mb_analysis_t *a )
{
- uint8_t **p_fref[2] =
+ pixel **p_fref[2] =
{ h->mb.pic.p_fref[0][a->l0.me16x16.i_ref],
h->mb.pic.p_fref[1][a->l1.me16x16.i_ref] };
- ALIGNED_ARRAY_8( uint8_t, pix,[2],[8*8] );
- int i, l;
+ ALIGNED_ARRAY_8( pixel, pix,[2],[8*8] );
/* XXX Needed for x264_mb_predict_mv */
h->mb.i_partition = D_8x8;
a->i_cost8x8bi = 0;
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
- const int x8 = i%2;
- const int y8 = i/2;
+ int x8 = i&1;
+ int y8 = i>>1;
int i_part_cost;
int i_part_cost_bi = 0;
int stride[2] = {8,8};
- uint8_t *src[2];
+ pixel *src[2];
- for( l = 0; l < 2; l++ )
+ for( int l = 0; l < 2; l++ )
{
x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
x264_me_t *m = &lX->me8x8[i];
x264_macroblock_cache_ref( h, x8*2, y8*2, 2, 2, l, lX->me16x16.i_ref );
x264_mb_predict_mv( h, l, 4*i, 2, m->mvp );
x264_me_search( h, m, &lX->me16x16.mv, 1 );
+ a->i_satd8x8[l][i] = m->cost - m->cost_mv;
m->cost += m->i_ref_cost;
x264_macroblock_cache_mv_ptr( h, 2*x8, 2*y8, 2, 2, l, m->mv );
i_part_cost_bi += m->cost_mv + m->i_ref_cost;
}
h->mc.avg[PIXEL_8x8]( pix[0], 8, src[0], stride[0], src[1], stride[1], h->mb.bipred_weight[a->l0.me16x16.i_ref][a->l1.me16x16.i_ref] );
- 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->i_satd8x8[2][i] = h->pixf.mbcmp[PIXEL_8x8]( a->l0.me8x8[i].p_fenc[0], FENC_STRIDE, pix[0], 8 );
+ i_part_cost_bi += a->i_satd8x8[2][i] + 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];
a->i_cost8x8bi += a->i_lambda * i_mb_b_cost_table[B_8x8];
}
-static void x264_mb_analyse_inter_b16x8( x264_t *h, x264_mb_analysis_t *a )
+static void x264_mb_analyse_inter_b16x8( x264_t *h, x264_mb_analysis_t *a, int i_best_satd )
{
- ALIGNED_ARRAY_16( uint8_t, pix,[2],[16*8] );
+ ALIGNED_ARRAY_16( pixel, pix,[2],[16*8] );
ALIGNED_4( int16_t mvc[3][2] );
- int i, j, l, i_ref;
h->mb.i_partition = D_16x8;
a->i_cost16x8bi = 0;
- for( i = 0; i < 2; i++ )
+ for( int i = 0; i < 2; i++ )
{
int i_part_cost;
int i_part_cost_bi = 0;
int stride[2] = {16,16};
- uint8_t *src[2];
+ pixel *src[2];
x264_me_t m;
m.i_pixel = PIXEL_16x8;
LOAD_FENC( &m, h->mb.pic.p_fenc, 0, 8*i );
- for( l = 0; l < 2; l++ )
+ for( int l = 0; l < 2; l++ )
{
x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
int ref8[2] = { lX->me8x8[2*i].i_ref, lX->me8x8[2*i+1].i_ref };
int i_ref8s = ( ref8[0] == ref8[1] ) ? 1 : 2;
lX->me16x8[i].cost = INT_MAX;
- for( j = 0; j < i_ref8s; j++ )
+ for( int j = 0; j < i_ref8s; j++ )
{
- i_ref = ref8[j];
+ int i_ref = ref8[j];
m.i_ref_cost = REF_COST( l, i_ref );;
LOAD_HPELS( &m, h->mb.pic.p_fref[l][i_ref], l, i_ref, 0, 8*i );
}
a->i_cost16x8bi += i_part_cost;
+ /* Early termination based on the current SATD score of partition[0]
+ plus the estimated SATD score of partition[1] */
+ if( !i && i_part_cost + a->i_cost_est16x8[1] > i_best_satd
+ * (16 + (!!a->i_mbrd + !!h->mb.i_psy_rd))/16 )
+ {
+ a->i_cost16x8bi = COST_MAX;
+ return;
+ }
+
x264_mb_cache_mv_b16x8( h, a, i, 0 );
}
a->i_cost16x8bi += a->i_lambda * i_mb_b16x8_cost_table[a->i_mb_type16x8];
}
-static void x264_mb_analyse_inter_b8x16( x264_t *h, x264_mb_analysis_t *a )
+static void x264_mb_analyse_inter_b8x16( x264_t *h, x264_mb_analysis_t *a, int i_best_satd )
{
- ALIGNED_ARRAY_8( uint8_t, pix,[2],[8*16] );
- ALIGNED_4( int16_t mvc[2][2] );
- int i, j, l, i_ref;
+ ALIGNED_ARRAY_8( pixel, pix,[2],[8*16] );
+ ALIGNED_4( int16_t mvc[3][2] );
h->mb.i_partition = D_8x16;
a->i_cost8x16bi = 0;
- for( i = 0; i < 2; i++ )
+ for( int i = 0; i < 2; i++ )
{
int i_part_cost;
int i_part_cost_bi = 0;
int stride[2] = {8,8};
- uint8_t *src[2];
+ pixel *src[2];
x264_me_t m;
m.i_pixel = PIXEL_8x16;
LOAD_FENC( &m, h->mb.pic.p_fenc, 8*i, 0 );
- for( l = 0; l < 2; l++ )
+ for( int l = 0; l < 2; l++ )
{
x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
int ref8[2] = { lX->me8x8[i].i_ref, lX->me8x8[i+2].i_ref };
int i_ref8s = ( ref8[0] == ref8[1] ) ? 1 : 2;
lX->me8x16[i].cost = INT_MAX;
- for( j = 0; j < i_ref8s; j++ )
+ for( int j = 0; j < i_ref8s; j++ )
{
- i_ref = ref8[j];
+ int i_ref = ref8[j];
m.i_ref_cost = REF_COST( l, i_ref );
LOAD_HPELS( &m, h->mb.pic.p_fref[l][i_ref], l, i_ref, 8*i, 0 );
}
a->i_cost8x16bi += i_part_cost;
+ /* Early termination based on the current SATD score of partition[0]
+ plus the estimated SATD score of partition[1] */
+ if( !i && i_part_cost + a->i_cost_est8x16[1] > i_best_satd
+ * (16 + (!!a->i_mbrd + !!h->mb.i_psy_rd))/16 )
+ {
+ a->i_cost8x16bi = COST_MAX;
+ return;
+ }
+
x264_mb_cache_mv_b8x16( h, a, i, 0 );
}
h->mb.i_partition = D_8x8;
if( h->param.analyse.inter & X264_ANALYSE_PSUB8x8 )
{
- int i;
x264_macroblock_cache_ref( h, 0, 0, 2, 2, 0, a->l0.me8x8[0].i_ref );
x264_macroblock_cache_ref( h, 2, 0, 2, 2, 0, a->l0.me8x8[1].i_ref );
x264_macroblock_cache_ref( h, 0, 2, 2, 2, 0, a->l0.me8x8[2].i_ref );
x264_macroblock_cache_ref( h, 2, 2, 2, 2, 0, a->l0.me8x8[3].i_ref );
/* FIXME: In the 8x8 blocks where RDO isn't run, the NNZ values used for context selection
* for future blocks are those left over from previous RDO calls. */
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
int costs[4] = {a->l0.i_cost4x4[i], a->l0.i_cost8x4[i], a->l0.i_cost4x8[i], a->l0.me8x8[i].cost};
- int thresh = X264_MIN4( costs[0], costs[1], costs[2], costs[3] ) * 5 / 4;
+ int sub8x8_thresh = X264_MIN4( costs[0], costs[1], costs[2], costs[3] ) * 5 / 4;
int subtype, btype = D_L0_8x8;
uint64_t bcost = COST_MAX64;
for( subtype = D_L0_4x4; subtype <= D_L0_8x8; subtype++ )
{
uint64_t cost;
- if( costs[subtype] > thresh || (subtype == D_L0_8x8 && bcost == COST_MAX64) )
+ if( costs[subtype] > sub8x8_thresh || (subtype == D_L0_8x8 && bcost == COST_MAX64) )
continue;
h->mb.i_sub_partition[i] = subtype;
x264_mb_cache_mv_p8x8( h, a, i );
static void x264_refine_bidir( x264_t *h, x264_mb_analysis_t *a )
{
int i_biweight;
- int i;
if( IS_INTRA(h->mb.i_type) )
return;
}
break;
case D_16x8:
- for( i=0; i<2; i++ )
+ for( int i = 0; i < 2; i++ )
if( a->i_mb_partition16x8[i] == D_BI_8x8 )
{
i_biweight = h->mb.bipred_weight[a->l0.me16x8[i].i_ref][a->l1.me16x8[i].i_ref];
}
break;
case D_8x16:
- for( i=0; i<2; i++ )
+ for( int i = 0; i < 2; i++ )
if( a->i_mb_partition8x16[i] == D_BI_8x8 )
{
i_biweight = h->mb.bipred_weight[a->l0.me8x16[i].i_ref][a->l1.me8x16[i].i_ref];
}
break;
case D_8x8:
- for( i=0; i<4; i++ )
+ for( int i = 0; i < 4; i++ )
if( h->mb.i_sub_partition[i] == D_BI_8x8 )
{
i_biweight = h->mb.bipred_weight[a->l0.me8x8[i].i_ref][a->l1.me8x8[i].i_ref];
{
if( x264_mb_transform_8x8_allowed( h ) && h->param.analyse.b_transform_8x8 && !h->mb.b_lossless )
{
- int i_cost4, i_cost8;
/* Only luma MC is really needed, but the full MC is re-used in macroblock_encode. */
x264_mb_mc( h );
- i_cost8 = h->pixf.sa8d[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE,
+ int 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,
+ int 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;
{
if( x264_mb_transform_8x8_allowed( h ) && h->param.analyse.b_transform_8x8 )
{
- int i_rd8;
x264_analyse_update_cache( h, a );
h->mb.b_transform_8x8 ^= 1;
/* FIXME only luma is needed, but the score for comparison already includes chroma */
- i_rd8 = x264_rd_cost_mb( h, a->i_lambda2 );
+ int i_rd8 = x264_rd_cost_mb( h, a->i_lambda2 );
if( *i_rd >= i_rd8 )
{
* trick. */
static inline void x264_mb_analyse_qp_rd( x264_t *h, x264_mb_analysis_t *a )
{
- int bcost, cost, direction, failures, prevcost, origcost;
+ int bcost, cost, failures, prevcost, origcost;
int orig_qp = h->mb.i_qp, bqp = h->mb.i_qp;
int last_qp_tried = 0;
origcost = bcost = x264_rd_cost_mb( h, a->i_lambda2 );
int origcbp = h->mb.cbp[h->mb.i_mb_xy];
/* If CBP is already zero, don't raise the quantizer any higher. */
- for( direction = origcbp ? 1 : -1; direction >= -1; direction-=2 )
+ for( int direction = origcbp ? 1 : -1; direction >= -1; direction-=2 )
{
/* Without psy-RD, require monotonicity when moving quant away from previous
* macroblock's quant; allow 1 failure when moving quant towards previous quant.
{
x264_mb_analysis_t analysis;
int i_cost = COST_MAX;
- int i;
- h->mb.i_qp = x264_ratecontrol_qp( h );
- if( h->param.rc.i_aq_mode )
- {
- x264_adaptive_quant( h );
- /* If the QP of this MB is within 1 of the previous MB, code the same QP as the previous MB,
- * to lower the bit cost of the qp_delta. Don't do this if QPRD is enabled. */
- if( h->param.analyse.i_subpel_refine < 10 && abs(h->mb.i_qp - h->mb.i_last_qp) == 1 )
- h->mb.i_qp = h->mb.i_last_qp;
- }
+ h->mb.i_qp = x264_ratecontrol_mb_qp( h );
+ /* If the QP of this MB is within 1 of the previous MB, code the same QP as the previous MB,
+ * to lower the bit cost of the qp_delta. Don't do this if QPRD is enabled. */
+ if( h->param.rc.i_aq_mode && h->param.analyse.i_subpel_refine < 10 && abs(h->mb.i_qp - h->mb.i_last_qp) == 1 )
+ h->mb.i_qp = h->mb.i_last_qp;
x264_mb_analyse_init( h, &analysis, h->mb.i_qp );
h->mb.i_partition = D_16x16;
assert( h->mb.cache.pskip_mv[1] <= h->mb.mv_max_spel[1] || h->i_thread_frames == 1 );
/* Set up MVs for future predictors */
- if( b_skip )
- for( i = 0; i < h->mb.pic.i_fref[0]; i++ )
- M32( h->mb.mvr[0][i][h->mb.i_mb_xy] ) = 0;
+ for( int i = 0; i < h->mb.pic.i_fref[0]; i++ )
+ M32( h->mb.mvr[0][i][h->mb.i_mb_xy] ) = 0;
}
else
{
if( h->mb.i_type == P_SKIP )
{
- for( i = 1; i < h->mb.pic.i_fref[0]; i++ )
+ for( int i = 1; i < h->mb.pic.i_fref[0]; i++ )
M32( h->mb.mvr[0][i][h->mb.i_mb_xy] ) = 0;
return;
}
/* Do sub 8x8 */
if( flags & X264_ANALYSE_PSUB8x8 )
{
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
{
x264_mb_analyse_inter_p4x4( h, &analysis, i );
if( analysis.l0.i_cost4x4[i] < analysis.l0.me8x8[i].cost )
if( ( flags & X264_ANALYSE_PSUB16x16 ) &&
analysis.l0.i_cost8x8 < analysis.l0.me16x16.cost + i_thresh16x8 )
{
- x264_mb_analyse_inter_p16x8( h, &analysis );
+ int i_avg_mv_ref_cost = (analysis.l0.me8x8[2].cost_mv + analysis.l0.me8x8[2].i_ref_cost
+ + analysis.l0.me8x8[3].cost_mv + analysis.l0.me8x8[3].i_ref_cost + 1) >> 1;
+ analysis.i_cost_est16x8[1] = analysis.i_satd8x8[0][2] + analysis.i_satd8x8[0][3] + i_avg_mv_ref_cost;
+
+ x264_mb_analyse_inter_p16x8( h, &analysis, i_cost );
COPY3_IF_LT( i_cost, analysis.l0.i_cost16x8, i_type, P_L0, i_partition, D_16x8 );
- x264_mb_analyse_inter_p8x16( h, &analysis );
+ i_avg_mv_ref_cost = (analysis.l0.me8x8[1].cost_mv + analysis.l0.me8x8[1].i_ref_cost
+ + analysis.l0.me8x8[3].cost_mv + analysis.l0.me8x8[3].i_ref_cost + 1) >> 1;
+ analysis.i_cost_est8x16[1] = analysis.i_satd8x8[0][1] + analysis.i_satd8x8[0][3] + i_avg_mv_ref_cost;
+
+ x264_mb_analyse_inter_p8x16( h, &analysis, i_cost );
COPY3_IF_LT( i_cost, analysis.l0.i_cost8x16, i_type, P_L0, i_partition, D_8x16 );
}
}
else if( i_partition == D_8x8 )
{
- int i8x8;
i_cost = 0;
- for( i8x8 = 0; i8x8 < 4; i8x8++ )
+ for( int i8x8 = 0; i8x8 < 4; i8x8++ )
{
switch( h->mb.i_sub_partition[i8x8] )
{
}
else if( i_partition == D_8x8 )
{
- int i8x8;
x264_analyse_update_cache( h, &analysis );
- for( i8x8 = 0; i8x8 < 4; i8x8++ )
+ for( int i8x8 = 0; i8x8 < 4; i8x8++ )
{
if( h->mb.i_sub_partition[i8x8] == D_L0_8x8 )
{
if( h->mb.b_direct_auto_write )
{
/* direct=auto heuristic: prefer whichever mode allows more Skip macroblocks */
- for( i = 0; i < 2; i++ )
+ for( int i = 0; i < 2; i++ )
{
int b_changed = 1;
h->sh.b_direct_spatial_mv_pred ^= 1;
/* Set up MVs for future predictors */
if( b_skip )
{
- for( i = 0; i < h->mb.pic.i_fref[0]; i++ )
+ for( int i = 0; i < h->mb.pic.i_fref[0]; i++ )
M32( h->mb.mvr[0][i][h->mb.i_mb_xy] ) = 0;
- for( i = 0; i < h->mb.pic.i_fref[1]; i++ )
+ for( int i = 0; i < h->mb.pic.i_fref[1]; i++ )
M32( h->mb.mvr[1][i][h->mb.i_mb_xy] ) = 0;
}
}
if( h->mb.i_type == B_SKIP )
{
- for( i = 1; i < h->mb.pic.i_fref[0]; i++ )
+ for( int i = 1; i < h->mb.pic.i_fref[0]; i++ )
M32( h->mb.mvr[0][i][h->mb.i_mb_xy] ) = 0;
- for( i = 1; i < h->mb.pic.i_fref[1]; i++ )
+ for( int i = 1; i < h->mb.pic.i_fref[1]; i++ )
M32( h->mb.mvr[0][i][h->mb.i_mb_xy] ) = 0;
return;
}
else
x264_mb_analyse_inter_b8x8( h, &analysis );
- if( analysis.i_cost8x8bi < i_cost )
- {
- i_type = B_8x8;
- i_partition = D_8x8;
- i_cost = analysis.i_cost8x8bi;
+ COPY3_IF_LT( i_cost, analysis.i_cost8x8bi, i_type, B_8x8, i_partition, D_8x8 );
- if( h->mb.i_sub_partition[0] == h->mb.i_sub_partition[1] ||
- h->mb.i_sub_partition[2] == h->mb.i_sub_partition[3] )
- {
- x264_mb_analyse_inter_b16x8( h, &analysis );
- 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 );
- COPY3_IF_LT( i_cost, analysis.i_cost8x16bi,
- i_type, analysis.i_mb_type8x16,
- i_partition, D_8x16 );
- }
+ /* Try to estimate the cost of b16x8/b8x16 based on the satd scores of the b8x8 modes */
+ int i_cost_est16x8bi_total = 0, i_cost_est8x16bi_total = 0;
+ int i_mb_type, i_partition16x8[2], i_partition8x16[2];
+ for( int i = 0; i < 2; i++ )
+ {
+ int avg_l0_mv_ref_cost, avg_l1_mv_ref_cost;
+ int i_l0_satd, i_l1_satd, i_bi_satd, i_best_cost;
+ // 16x8
+ i_best_cost = COST_MAX;
+ i_l0_satd = analysis.i_satd8x8[0][i*2] + analysis.i_satd8x8[0][i*2+1];
+ i_l1_satd = analysis.i_satd8x8[1][i*2] + analysis.i_satd8x8[1][i*2+1];
+ i_bi_satd = analysis.i_satd8x8[2][i*2] + analysis.i_satd8x8[2][i*2+1];
+ avg_l0_mv_ref_cost = ( analysis.l0.me8x8[i*2].cost_mv + analysis.l0.me8x8[i*2].i_ref_cost
+ + analysis.l0.me8x8[i*2+1].cost_mv + analysis.l0.me8x8[i*2+1].i_ref_cost + 1 ) >> 1;
+ avg_l1_mv_ref_cost = ( analysis.l1.me8x8[i*2].cost_mv + analysis.l1.me8x8[i*2].i_ref_cost
+ + analysis.l1.me8x8[i*2+1].cost_mv + analysis.l1.me8x8[i*2+1].i_ref_cost + 1 ) >> 1;
+ COPY2_IF_LT( i_best_cost, i_l0_satd + avg_l0_mv_ref_cost, i_partition16x8[i], D_L0_8x8 );
+ COPY2_IF_LT( i_best_cost, i_l1_satd + avg_l1_mv_ref_cost, i_partition16x8[i], D_L1_8x8 );
+ COPY2_IF_LT( i_best_cost, i_bi_satd + avg_l0_mv_ref_cost + avg_l1_mv_ref_cost, i_partition16x8[i], D_BI_8x8 );
+ analysis.i_cost_est16x8[i] = i_best_cost;
+
+ // 8x16
+ i_best_cost = COST_MAX;
+ i_l0_satd = analysis.i_satd8x8[0][i] + analysis.i_satd8x8[0][i+2];
+ i_l1_satd = analysis.i_satd8x8[1][i] + analysis.i_satd8x8[1][i+2];
+ i_bi_satd = analysis.i_satd8x8[2][i] + analysis.i_satd8x8[2][i+2];
+ avg_l0_mv_ref_cost = ( analysis.l0.me8x8[i].cost_mv + analysis.l0.me8x8[i].i_ref_cost
+ + analysis.l0.me8x8[i+2].cost_mv + analysis.l0.me8x8[i+2].i_ref_cost + 1 ) >> 1;
+ avg_l1_mv_ref_cost = ( analysis.l1.me8x8[i].cost_mv + analysis.l1.me8x8[i].i_ref_cost
+ + analysis.l1.me8x8[i+2].cost_mv + analysis.l1.me8x8[i+2].i_ref_cost + 1 ) >> 1;
+ COPY2_IF_LT( i_best_cost, i_l0_satd + avg_l0_mv_ref_cost, i_partition8x16[i], D_L0_8x8 );
+ COPY2_IF_LT( i_best_cost, i_l1_satd + avg_l1_mv_ref_cost, i_partition8x16[i], D_L1_8x8 );
+ COPY2_IF_LT( i_best_cost, i_bi_satd + avg_l0_mv_ref_cost + avg_l1_mv_ref_cost, i_partition8x16[i], D_BI_8x8 );
+ analysis.i_cost_est8x16[i] = i_best_cost;
+ }
+ i_mb_type = B_L0_L0 + (i_partition16x8[0]>>2) * 3 + (i_partition16x8[1]>>2);
+ analysis.i_cost_est16x8[1] += analysis.i_lambda * i_mb_b16x8_cost_table[i_mb_type];
+ i_cost_est16x8bi_total = analysis.i_cost_est16x8[0] + analysis.i_cost_est16x8[1];
+ i_mb_type = B_L0_L0 + (i_partition8x16[0]>>2) * 3 + (i_partition8x16[1]>>2);
+ analysis.i_cost_est8x16[1] += analysis.i_lambda * i_mb_b16x8_cost_table[i_mb_type];
+ i_cost_est8x16bi_total = analysis.i_cost_est8x16[0] + analysis.i_cost_est8x16[1];
+
+ /* We can gain a little speed by checking the mode with the lowest estimated cost first */
+ int try_16x8_first = i_cost_est16x8bi_total < i_cost_est8x16bi_total;
+ if( try_16x8_first && i_cost_est16x8bi_total < i_cost )
+ {
+ x264_mb_analyse_inter_b16x8( h, &analysis, i_cost );
+ COPY3_IF_LT( i_cost, analysis.i_cost16x8bi, i_type, analysis.i_mb_type16x8, i_partition, D_16x8 );
+ }
+ if( i_cost_est8x16bi_total < i_cost )
+ {
+ x264_mb_analyse_inter_b8x16( h, &analysis, i_cost );
+ COPY3_IF_LT( i_cost, analysis.i_cost8x16bi, i_type, analysis.i_mb_type8x16, i_partition, D_8x16 );
+ }
+ if( !try_16x8_first && i_cost_est16x8bi_total < i_cost )
+ {
+ x264_mb_analyse_inter_b16x8( h, &analysis, i_cost );
+ COPY3_IF_LT( i_cost, analysis.i_cost16x8bi, i_type, analysis.i_mb_type16x8, i_partition, D_16x8 );
}
}
}
else if( i_partition == D_16x8 )
{
- for( i=0; i<2; i++ )
+ for( int i = 0; i < 2; i++ )
{
if( analysis.i_mb_partition16x8[i] != D_L1_8x8 )
x264_me_refine_qpel( h, &analysis.l0.me16x8[i] );
}
else if( i_partition == D_8x16 )
{
- for( i=0; i<2; i++ )
+ for( int i = 0; i < 2; i++ )
{
if( analysis.i_mb_partition8x16[i] != D_L1_8x8 )
x264_me_refine_qpel( h, &analysis.l0.me8x16[i] );
}
else if( i_partition == D_8x8 )
{
- for( i=0; i<4; i++ )
+ for( int i = 0; i < 4; i++ )
{
x264_me_t *m;
int i_part_cost_old;
}
else if( i_partition == D_16x8 )
{
- for( i = 0; i < 2; i++ )
+ for( int 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 )
}
else if( i_partition == D_8x16 )
{
- for( i = 0; i < 2; i++ )
+ for( int 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 )
}
else if( i_partition == D_8x8 )
{
- for( i = 0; i < 4; i++ )
+ for( int 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*4, 0 );
/*-------------------- Update MB from the analysis ----------------------*/
static void x264_analyse_update_cache( x264_t *h, x264_mb_analysis_t *a )
{
- int i;
-
switch( h->mb.i_type )
{
case I_4x4:
- for( i = 0; i < 16; i++ )
+ for( int i = 0; i < 16; 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++ )
+ for( int i = 0; i < 4; i++ )
x264_macroblock_cache_intra8x8_pred( h, 2*(i&1), 2*(i>>1), a->i_predict8x8[i] );
x264_mb_analyse_intra_chroma( h, a );
x264_macroblock_cache_ref( h, 2, 0, 2, 2, 0, a->l0.me8x8[1].i_ref );
x264_macroblock_cache_ref( h, 0, 2, 2, 2, 0, a->l0.me8x8[2].i_ref );
x264_macroblock_cache_ref( h, 2, 2, 2, 2, 0, a->l0.me8x8[3].i_ref );
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
x264_mb_cache_mv_p8x8( h, a, i );
break;
case B_8x8:
/* optimize: cache might not need to be rewritten */
- for( i = 0; i < 4; i++ )
+ for( int i = 0; i < 4; i++ )
x264_mb_cache_mv_b8x8( h, a, i, 1 );
break;
#ifndef NDEBUG
if( h->i_thread_frames > 1 && !IS_INTRA(h->mb.i_type) )
{
- int l;
- for( l=0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
+ for( int l = 0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
{
int completed;
int ref = h->mb.cache.ref[l][x264_scan8[0]];
projects / x264 / blobdiff