} 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 )
void x264_analyse_free_costs( x264_t *h )
{
- for( int 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 );
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 );
const unsigned int flags = h->sh.i_type == SLICE_TYPE_I ? h->param.analyse.intra : h->param.analyse.inter;
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 idx;
- int b_merged_satd = !!h->pixf.intra_mbcmp_x3_16x16 && !h->mb.b_lossless;
+ 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( int 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 };
- if( a->b_fast_intra && a->i_satd_i16x16 > (i16x16_thresh[h->mb.i_subpel_refine]*i_satd_inter)>>1 )
+ if( a->i_satd_i16x16 > i16x16_thresh )
return;
/* 8x8 prediction selection */
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++ )
{
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;
+ 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 };
+ 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++ )
{
/* emulate missing topright samples */
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;
+ 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;
for( int idx = 0; idx < 4; idx++ )
{
pixel4 pels_h[2] = {0};
- pixel pels_v[7];
+ pixel pels_v[7] = {0};
uint16_t i_nnz[2] = {0}; //shut up gcc
pixel *p_dst_by;
int cbp_luma_new = 0;
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;
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;
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 );
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};
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};