/*****************************************************************************
* macroblock.c: h264 encoder library
*****************************************************************************
- * Copyright (C) 2003 Laurent Aimar
- * $Id: macroblock.c,v 1.1 2004/06/03 19:27:06 fenrir Exp $
+ * Copyright (C) 2003-2008 x264 project
*
* Authors: Laurent Aimar <fenrir@via.ecp.fr>
+ * Loren Merritt <lorenm@u.washington.edu>
+ * Fiona Glaser <fiona@x264.com>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
*****************************************************************************/
-#include <stdio.h>
-#include <string.h>
-
#include "common.h"
+#include "encoder/me.h"
-static const int dequant_mf[6][4][4] =
-{
- { {10, 13, 10, 13}, {13, 16, 13, 16}, {10, 13, 10, 13}, {13, 16, 13, 16} },
- { {11, 14, 11, 14}, {14, 18, 14, 18}, {11, 14, 11, 14}, {14, 18, 14, 18} },
- { {13, 16, 13, 16}, {16, 20, 16, 20}, {13, 16, 13, 16}, {16, 20, 16, 20} },
- { {14, 18, 14, 18}, {18, 23, 18, 23}, {14, 18, 14, 18}, {18, 23, 18, 23} },
- { {16, 20, 16, 20}, {20, 25, 20, 25}, {16, 20, 16, 20}, {20, 25, 20, 25} },
- { {18, 23, 18, 23}, {23, 29, 23, 29}, {18, 23, 18, 23}, {23, 29, 23, 29} }
-};
-
-int x264_mb_predict_intra4x4_mode( x264_t *h, int idx )
-{
- const int ma = h->mb.cache.intra4x4_pred_mode[x264_scan8[idx] - 1];
- const int mb = h->mb.cache.intra4x4_pred_mode[x264_scan8[idx] - 8];
- const int m = X264_MIN( x264_mb_pred_mode4x4_fix(ma),
- x264_mb_pred_mode4x4_fix(mb) );
-
- if( m < 0 )
- return I_PRED_4x4_DC;
-
- return m;
-}
-
-int x264_mb_predict_non_zero_code( x264_t *h, int idx )
-{
- const int za = h->mb.cache.non_zero_count[x264_scan8[idx] - 1];
- const int zb = h->mb.cache.non_zero_count[x264_scan8[idx] - 8];
-
- int i_ret = za + zb;
-
- if( i_ret < 0x80 )
- {
- i_ret = ( i_ret + 1 ) >> 1;
- }
- return i_ret & 0x7f;
-}
-
-int x264_mb_transform_8x8_allowed( x264_t *h )
-{
- if( IS_SKIP( h->mb.i_type ) )
- return 0;
- if( h->mb.i_type == P_8x8 || h->mb.i_type == B_8x8 )
- {
- int i;
- for( i = 0; i < 4; i++ )
- if( !IS_SUB8x8(h->mb.i_sub_partition[i])
- || ( h->mb.i_sub_partition[i] == D_DIRECT_8x8 && !h->sps->b_direct8x8_inference ) )
- {
- return 0;
- }
- }
- if( h->mb.i_type == B_DIRECT && !h->sps->b_direct8x8_inference )
- return 0;
-
- return 1;
-}
-
-void x264_mb_predict_mv( x264_t *h, int i_list, int idx, int i_width, int mvp[2] )
+void x264_mb_predict_mv( x264_t *h, int i_list, int idx, int i_width, int16_t mvp[2] )
{
const int i8 = x264_scan8[idx];
const int i_ref= h->mb.cache.ref[i_list][i8];
int i_refc = h->mb.cache.ref[i_list][i8 - 8 + i_width ];
int16_t *mv_c = h->mb.cache.mv[i_list][i8 - 8 + i_width];
- int i_count;
+ int i_count = 0;
- if( (idx&0x03) == 3 || ( i_width == 2 && (idx&0x3) == 2 )|| i_refc == -2 )
+ if( (idx&3) >= 2 + (i_width&1) || i_refc == -2 )
{
i_refc = h->mb.cache.ref[i_list][i8 - 8 - 1];
mv_c = h->mb.cache.mv[i_list][i8 - 8 - 1];
if( h->mb.i_partition == D_16x8 )
{
- if( idx == 0 && i_refb == i_ref )
+ if( idx == 0 )
{
- mvp[0] = mv_b[0];
- mvp[1] = mv_b[1];
- return;
+ if( i_refb == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_b;
+ return;
+ }
}
- else if( idx != 0 && i_refa == i_ref )
+ else
{
- mvp[0] = mv_a[0];
- mvp[1] = mv_a[1];
- return;
+ if( i_refa == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
+ return;
+ }
}
}
else if( h->mb.i_partition == D_8x16 )
{
- if( idx == 0 && i_refa == i_ref )
+ if( idx == 0 )
{
- mvp[0] = mv_a[0];
- mvp[1] = mv_a[1];
- return;
+ if( i_refa == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
+ return;
+ }
}
- else if( idx != 0 && i_refc == i_ref )
+ else
{
- mvp[0] = mv_c[0];
- mvp[1] = mv_c[1];
- return;
+ if( i_refc == i_ref )
+ {
+ *(uint32_t*)mvp = *(uint32_t*)mv_c;
+ return;
+ }
}
}
- i_count = 0;
if( i_refa == i_ref ) i_count++;
if( i_refb == i_ref ) i_count++;
if( i_refc == i_ref ) i_count++;
if( i_count > 1 )
{
- mvp[0] = x264_median( mv_a[0], mv_b[0], mv_c[0] );
- mvp[1] = x264_median( mv_a[1], mv_b[1], mv_c[1] );
+median:
+ x264_median_mv( mvp, mv_a, mv_b, mv_c );
}
else if( i_count == 1 )
{
if( i_refa == i_ref )
- {
- mvp[0] = mv_a[0];
- mvp[1] = mv_a[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
else if( i_refb == i_ref )
- {
- mvp[0] = mv_b[0];
- mvp[1] = mv_b[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_b;
else
- {
- mvp[0] = mv_c[0];
- mvp[1] = mv_c[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_c;
}
else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
- {
- mvp[0] = mv_a[0];
- mvp[1] = mv_a[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
else
- {
- mvp[0] = x264_median( mv_a[0], mv_b[0], mv_c[0] );
- mvp[1] = x264_median( mv_a[1], mv_b[1], mv_c[1] );
- }
+ goto median;
}
-void x264_mb_predict_mv_16x16( x264_t *h, int i_list, int i_ref, int mvp[2] )
+void x264_mb_predict_mv_16x16( x264_t *h, int i_list, int i_ref, int16_t mvp[2] )
{
int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
int16_t *mv_a = h->mb.cache.mv[i_list][X264_SCAN8_0 - 1];
int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
int16_t *mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
- int i_count;
+ int i_count = 0;
if( i_refc == -2 )
{
mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
}
- i_count = 0;
if( i_refa == i_ref ) i_count++;
if( i_refb == i_ref ) i_count++;
if( i_refc == i_ref ) i_count++;
if( i_count > 1 )
{
- mvp[0] = x264_median( mv_a[0], mv_b[0], mv_c[0] );
- mvp[1] = x264_median( mv_a[1], mv_b[1], mv_c[1] );
+median:
+ x264_median_mv( mvp, mv_a, mv_b, mv_c );
}
else if( i_count == 1 )
{
if( i_refa == i_ref )
- {
- mvp[0] = mv_a[0];
- mvp[1] = mv_a[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
else if( i_refb == i_ref )
- {
- mvp[0] = mv_b[0];
- mvp[1] = mv_b[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_b;
else
- {
- mvp[0] = mv_c[0];
- mvp[1] = mv_c[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_c;
}
else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
- {
- mvp[0] = mv_a[0];
- mvp[1] = mv_a[1];
- }
+ *(uint32_t*)mvp = *(uint32_t*)mv_a;
else
- {
- mvp[0] = x264_median( mv_a[0], mv_b[0], mv_c[0] );
- mvp[1] = x264_median( mv_a[1], mv_b[1], mv_c[1] );
- }
+ goto median;
}
-void x264_mb_predict_mv_pskip( x264_t *h, int mv[2] )
+void x264_mb_predict_mv_pskip( x264_t *h, int16_t mv[2] )
{
int i_refa = h->mb.cache.ref[0][X264_SCAN8_0 - 1];
int i_refb = h->mb.cache.ref[0][X264_SCAN8_0 - 8];
int16_t *mv_b = h->mb.cache.mv[0][X264_SCAN8_0 - 8];
if( i_refa == -2 || i_refb == -2 ||
- ( i_refa == 0 && mv_a[0] == 0 && mv_a[1] == 0 ) ||
- ( i_refb == 0 && mv_b[0] == 0 && mv_b[1] == 0 ) )
+ !( i_refa | *(uint32_t*)mv_a ) ||
+ !( i_refb | *(uint32_t*)mv_b ) )
{
- mv[0] = mv[1] = 0;
+ *(uint32_t*)mv = 0;
}
else
{
{
int i_mb_4x4 = 16 * h->mb.i_mb_stride * h->mb.i_mb_y + 4 * h->mb.i_mb_x;
int i_mb_8x8 = 4 * h->mb.i_mb_stride * h->mb.i_mb_y + 2 * h->mb.i_mb_x;
- int i8, i4;
- int b8x8;
+ int i8;
const int type_col = h->fref1[0]->mb_type[ h->mb.i_mb_xy ];
-
+
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
-
+
if( IS_INTRA( type_col ) )
{
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
- x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0, 0 );
- x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0, 0 );
+ x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0 );
+ x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
return 1;
}
- b8x8 = h->sps->b_direct8x8_inference ||
- (type_col != P_8x8 && type_col != B_SKIP && type_col != B_DIRECT && type_col != B_8x8);
for( i8 = 0; i8 < 4; i8++ )
{
if( i_ref >= 0 )
{
const int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
-
+ const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
+ const int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
+ const int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
+ if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_col[1] > h->mb.mv_max_spel[1]) )
+ return 0;
x264_macroblock_cache_ref( h, 2*x8, 2*y8, 2, 2, 0, i_ref );
-
- if( b8x8 )
- {
- const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
- int mv_l0[2];
- mv_l0[0] = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
- mv_l0[1] = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, mv_l0[0], mv_l0[1] );
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, mv_l0[0] - mv_col[0], mv_l0[1] - mv_col[1] );
- }
- else
- {
- for( i4 = 0; i4 < 4; i4++ )
- {
- const int x4 = i4%2 + 2*x8;
- const int y4 = i4/2 + 2*y8;
- const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + x4 + y4 * h->mb.i_b4_stride ];
- int mv_l0[2];
- mv_l0[0] = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
- mv_l0[1] = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, mv_l0[0], mv_l0[1] );
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 1, mv_l0[0] - mv_col[0], mv_l0[1] - mv_col[1] );
- }
- }
+ x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, pack16to32_mask(l0x, l0y) );
+ x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
}
else
{
- /* the colocated ref isn't in the current list0 */
+ /* the collocated ref isn't in the current list0 */
/* FIXME: we might still be able to use direct_8x8 on some partitions */
/* FIXME: with B-pyramid + extensive ref list reordering
* (not currently used), we would also have to check
static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
{
int ref[2];
- int mv[2][2];
+ ALIGNED_8( int16_t mv[2][2] );
int i_list;
- int i8, i4;
- int b8x8;
+ int i8;
const int8_t *l1ref0 = &h->fref1[0]->ref[0][ h->mb.i_b8_xy ];
const int8_t *l1ref1 = &h->fref1[0]->ref[1][ h->mb.i_b8_xy ];
const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->fref1[0]->mv[0][ h->mb.i_b4_xy ];
if( ref[0] < 0 && ref[1] < 0 )
{
- ref[0] =
- ref[1] = 0;
- mv[0][0] =
- mv[0][1] =
- mv[1][0] =
- mv[1][1] = 0;
+ x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
+ x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
+ x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0 );
+ x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
+ return 1;
}
+
+ if( ref[0] >= 0 )
+ x264_mb_predict_mv_16x16( h, 0, ref[0], mv[0] );
else
- {
- for( i_list=0; i_list<2; i_list++ )
- {
- if( ref[i_list] >= 0 )
- x264_mb_predict_mv_16x16( h, i_list, ref[i_list], mv[i_list] );
- else
- mv[i_list][0] = mv[i_list][1] = 0;
- }
- }
+ *(uint32_t*)mv[0] = 0;
+ if( ref[1] >= 0 )
+ x264_mb_predict_mv_16x16( h, 1, ref[1], mv[1] );
+ else
+ *(uint32_t*)mv[1] = 0;
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, ref[0] );
x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, ref[1] );
- x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, mv[0][0], mv[0][1] );
- x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, mv[1][0], mv[1][1] );
+ x264_macroblock_cache_mv_ptr( h, 0, 0, 4, 4, 0, mv[0] );
+ x264_macroblock_cache_mv_ptr( h, 0, 0, 4, 4, 1, mv[1] );
- if( IS_INTRA( type_col ) )
+ if( h->param.i_threads > 1
+ && ( mv[0][1] > h->mb.mv_max_spel[1]
+ || mv[1][1] > h->mb.mv_max_spel[1] ) )
+ {
+#if 0
+ fprintf(stderr, "direct_spatial: (%d,%d) (%d,%d) > %d \n",
+ mv[0][0], mv[0][1], mv[1][0], mv[1][1],
+ h->mb.mv_max_spel[1]);
+#endif
+ return 0;
+ }
+
+ if( IS_INTRA( type_col ) || (ref[0]&&ref[1]) )
return 1;
- b8x8 = h->sps->b_direct8x8_inference ||
- (type_col != P_8x8 && type_col != B_SKIP && type_col != B_DIRECT && type_col != B_8x8);
/* col_zero_flag */
for( i8=0; i8<4; i8++ )
const int x8 = i8%2;
const int y8 = i8/2;
const int o8 = x8 + y8 * h->mb.i_b8_stride;
- if( l1ref0[o8] == 0 || ( l1ref0[o8] < 0 && l1ref1[o8] == 0 ) )
+ const int o4 = 3*(x8 + y8 * h->mb.i_b4_stride);
+ if( l1ref0[o8] == 0 )
{
- const int16_t (*l1mv)[2] = (l1ref0[o8] == 0) ? l1mv0 : l1mv1;
- if( b8x8 )
+ if( abs( l1mv0[o4][0] ) <= 1 && abs( l1mv0[o4][1] ) <= 1 )
{
- const int16_t *mvcol = l1mv[3*x8 + 3*y8 * h->mb.i_b4_stride];
- if( abs( mvcol[0] ) <= 1 && abs( mvcol[1] ) <= 1 )
- {
- if( ref[0] == 0 )
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0, 0 );
- if( ref[1] == 0 )
- x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0, 0 );
- }
+ if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
+ if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
}
- else
+ }
+ else if( l1ref0[o8] < 0 && l1ref1[o8] == 0 )
+ {
+ if( abs( l1mv1[o4][0] ) <= 1 && abs( l1mv1[o4][1] ) <= 1 )
{
- for( i4=0; i4<4; i4++ )
- {
- const int x4 = i4%2 + 2*x8;
- const int y4 = i4/2 + 2*y8;
- const int16_t *mvcol = l1mv[x4 + y4 * h->mb.i_b4_stride];
- if( abs( mvcol[0] ) <= 1 && abs( mvcol[1] ) <= 1 )
- {
- if( ref[0] == 0 )
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, 0, 0 );
- if( ref[1] == 0 )
- x264_macroblock_cache_mv( h, x4, y4, 1, 1, 1, 0, 0 );
- }
- }
+ if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
+ if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
}
}
}
for( l = 0; l < 2; l++ )
for( i = 0; i < 4; i++ )
h->mb.cache.direct_ref[l][i] = h->mb.cache.ref[l][x264_scan8[i*4]];
- memcpy(h->mb.cache.direct_mv, h->mb.cache.mv, sizeof(h->mb.cache.mv));
+ h->mc.memcpy_aligned(h->mb.cache.direct_mv, h->mb.cache.mv, sizeof(h->mb.cache.mv));
}
return b_available;
{
const int x = 2*(idx%2);
const int y = 2*(idx/2);
- int l;
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] );
- for( l = 0; l < 2; l++ )
- {
- *(uint64_t*)h->mb.cache.mv[l][x264_scan8[idx*4]] =
- *(uint64_t*)h->mb.cache.direct_mv[l][x264_scan8[idx*4]];
- *(uint64_t*)h->mb.cache.mv[l][x264_scan8[idx*4]+8] =
- *(uint64_t*)h->mb.cache.direct_mv[l][x264_scan8[idx*4]+8];
- }
+ *(uint64_t*)h->mb.cache.mv[0][x264_scan8[idx*4]] =
+ *(uint64_t*)h->mb.cache.direct_mv[0][x264_scan8[idx*4]];
+ *(uint64_t*)h->mb.cache.mv[0][x264_scan8[idx*4]+8] =
+ *(uint64_t*)h->mb.cache.direct_mv[0][x264_scan8[idx*4]+8];
+ *(uint64_t*)h->mb.cache.mv[1][x264_scan8[idx*4]] =
+ *(uint64_t*)h->mb.cache.direct_mv[1][x264_scan8[idx*4]];
+ *(uint64_t*)h->mb.cache.mv[1][x264_scan8[idx*4]+8] =
+ *(uint64_t*)h->mb.cache.direct_mv[1][x264_scan8[idx*4]+8];
}
/* This just improves encoder performance, it's not part of the spec */
-void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int mvc[8][2], int *i_mvc )
+void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int16_t mvc[9][2], int *i_mvc )
{
int16_t (*mvr)[2] = h->mb.mvr[i_list][i_ref];
int i = 0;
#define SET_MVP(mvp) { \
- mvc[i][0] = mvp[0]; \
- mvc[i][1] = mvp[1]; \
+ *(uint32_t*)mvc[i] = *(uint32_t*)mvp; \
i++; \
}
SET_MVP( h->mb.cache.mv[i_list][x264_scan8[12]] );
}
+ if( i_ref == 0 && h->frames.b_have_lowres )
+ {
+ int16_t (*lowres_mv)[2] = i_list ? h->fenc->lowres_mvs[1][h->fref1[0]->i_frame-h->fenc->i_frame-1]
+ : h->fenc->lowres_mvs[0][h->fenc->i_frame-h->fref0[0]->i_frame-1];
+ if( lowres_mv[0][0] != 0x7fff ) *(uint32_t*)mvc[i++] = (*(uint32_t*)lowres_mv[h->mb.i_mb_xy]*2)&0xfffeffff;
+ }
+
/* spatial predictors */
if( h->mb.i_neighbour & MB_LEFT )
{
}
if( h->mb.i_neighbour & MB_TOP )
{
- int i_mb_t = h->mb.i_mb_xy - h->mb.i_mb_stride;
+ int i_mb_t = h->mb.i_mb_top_xy;
if( !IS_SKIP( h->mb.type[i_mb_t] ) )
SET_MVP( mvr[i_mb_t] );
#undef SET_MVP
/* temporal predictors */
- if( h->fref0[0]->i_ref[0] > 0 )
+ /* FIXME temporal scaling w/ interlace */
+ if( h->fref0[0]->i_ref[0] > 0 && !h->sh.b_mbaff )
{
x264_frame_t *l0 = h->fref0[0];
- int ref_col_cur, ref_col_prev = -1;
- int scale = 0;
#define SET_TMVP(dx, dy) { \
int i_b4 = h->mb.i_b4_xy + dx*4 + dy*4*h->mb.i_b4_stride; \
int i_b8 = h->mb.i_b8_xy + dx*2 + dy*2*h->mb.i_b8_stride; \
- ref_col_cur = l0->ref[0][i_b8]; \
- if( ref_col_cur >= 0 ) \
+ int ref_col = l0->ref[0][i_b8]; \
+ if( ref_col >= 0 ) \
{ \
- /* TODO: calc once per frame and tablize? */\
- if( ref_col_cur != ref_col_prev ) \
- scale = 256 * (h->fenc->i_poc - h->fref0[i_ref]->i_poc) \
- / (l0->i_poc - l0->ref_poc[0][ref_col_cur]); \
- mvc[i][0] = l0->mv[0][i_b4][0] * scale / 256; \
- mvc[i][1] = l0->mv[0][i_b4][1] * scale / 256; \
+ int scale = (h->fdec->i_poc - h->fdec->ref_poc[0][i_ref]) * l0->inv_ref_poc[ref_col];\
+ mvc[i][0] = (l0->mv[0][i_b4][0]*scale + 128) >> 8;\
+ mvc[i][1] = (l0->mv[0][i_b4][1]*scale + 128) >> 8;\
i++; \
- ref_col_prev = ref_col_cur; \
} \
}
*i_mvc = i;
}
+/* Set up a lookup table for delta pocs to reduce an IDIV to an IMUL */
+static void setup_inverse_delta_pocs( x264_t *h )
+{
+ int i;
+ for( i = 0; i < h->i_ref0; i++ )
+ {
+ int delta = h->fdec->i_poc - h->fref0[i]->i_poc;
+ h->fdec->inv_ref_poc[i] = (256 + delta/2) / delta;
+ }
+}
+
static inline void x264_mb_mc_0xywh( x264_t *h, int x, int y, int width, int height )
{
const int i8 = x264_scan8[0]+x+8*y;
const int i_ref = h->mb.cache.ref[0][i8];
const int mvx = x264_clip3( h->mb.cache.mv[0][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
- const int mvy = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+ int mvy = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+
+ h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0],
+ mvx + 4*4*x, mvy + 4*4*y, 4*width, 4*height );
- h->mc.mc_luma( h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0],
- &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
- mvx + 4*4*x, mvy + 4*4*y, 4*width, 4*height );
+ // chroma is offset if MCing from a field of opposite parity
+ if( h->mb.b_interlaced & i_ref )
+ mvy += (h->mb.i_mb_y & 1)*4 - 2;
- h->mc.mc_chroma( &h->mb.pic.p_fref[0][i_ref][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
- &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- mvx, mvy, 2*width, 2*height );
+ h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
+ &h->mb.pic.p_fref[0][i_ref][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
+ mvx, mvy, 2*width, 2*height );
- h->mc.mc_chroma( &h->mb.pic.p_fref[0][i_ref][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
- &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- mvx, mvy, 2*width, 2*height );
+ h->mc.mc_chroma( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
+ &h->mb.pic.p_fref[0][i_ref][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
+ mvx, mvy, 2*width, 2*height );
}
static inline void x264_mb_mc_1xywh( x264_t *h, int x, int y, int width, int height )
{
const int i8 = x264_scan8[0]+x+8*y;
const int i_ref = h->mb.cache.ref[1][i8];
const int mvx = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
- const int mvy = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+ int mvy = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+
+ h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
+ h->mb.pic.p_fref[1][i_ref], h->mb.pic.i_stride[0],
+ mvx + 4*4*x, mvy + 4*4*y, 4*width, 4*height );
- h->mc.mc_luma( h->mb.pic.p_fref[1][i_ref], h->mb.pic.i_stride[0],
- &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
- mvx + 4*4*x, mvy + 4*4*y, 4*width, 4*height );
+ if( h->mb.b_interlaced & i_ref )
+ mvy += (h->mb.i_mb_y & 1)*4 - 2;
- h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
- &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- mvx, mvy, 2*width, 2*height );
+ h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
+ &h->mb.pic.p_fref[1][i_ref][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
+ mvx, mvy, 2*width, 2*height );
- h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
- &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
- mvx, mvy, 2*width, 2*height );
+ h->mc.mc_chroma( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
+ &h->mb.pic.p_fref[1][i_ref][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
+ mvx, mvy, 2*width, 2*height );
}
static inline void x264_mb_mc_01xywh( x264_t *h, int x, int y, int width, int height )
{
const int i8 = x264_scan8[0]+x+8*y;
-
+ const int i_ref0 = h->mb.cache.ref[0][i8];
const int i_ref1 = h->mb.cache.ref[1][i8];
+ const int weight = h->mb.bipred_weight[i_ref0][i_ref1];
+ const int mvx0 = x264_clip3( h->mb.cache.mv[0][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
const int mvx1 = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
- const int mvy1 = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
- DECLARE_ALIGNED( uint8_t, tmp[16*16], 16 );
- int i_mode = x264_size2pixel[height][width];
-
- x264_mb_mc_0xywh( h, x, y, width, height );
-
- h->mc.mc_luma( h->mb.pic.p_fref[1][i_ref1], h->mb.pic.i_stride[0],
- tmp, 16, mvx1 + 4*4*x, mvy1 + 4*4*y, 4*width, 4*height );
-
- if( h->param.analyse.b_weighted_bipred )
- {
- const int i_ref0 = h->mb.cache.ref[0][i8];
- const int weight = h->mb.bipred_weight[i_ref0][i_ref1];
-
- h->mc.avg_weight[i_mode]( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE, tmp, 16, weight );
-
- h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
- tmp, 16, mvx1, mvy1, 2*width, 2*height );
- h->mc.avg_weight[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp, 16, weight );
-
- h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
- tmp, 16, mvx1, mvy1, 2*width, 2*height );
- h->mc.avg_weight[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp, 16, weight );
- }
- else
- {
- h->mc.avg[i_mode]( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE, tmp, 16 );
-
- h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
- tmp, 16, mvx1, mvy1, 2*width, 2*height );
- h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp, 16 );
-
- h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
- tmp, 16, mvx1, mvy1, 2*width, 2*height );
- h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp, 16 );
- }
+ int mvy0 = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+ int mvy1 = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+ int i_mode = x264_size2pixel[height][width];
+ int i_stride0 = 16, i_stride1 = 16;
+ ALIGNED_ARRAY_16( uint8_t, tmp0,[16*16] );
+ ALIGNED_ARRAY_16( uint8_t, tmp1,[16*16] );
+ uint8_t *src0, *src1;
+
+ src0 = h->mc.get_ref( tmp0, &i_stride0, h->mb.pic.p_fref[0][i_ref0], h->mb.pic.i_stride[0],
+ mvx0 + 4*4*x, mvy0 + 4*4*y, 4*width, 4*height );
+ src1 = h->mc.get_ref( tmp1, &i_stride1, h->mb.pic.p_fref[1][i_ref1], h->mb.pic.i_stride[0],
+ mvx1 + 4*4*x, mvy1 + 4*4*y, 4*width, 4*height );
+ h->mc.avg[i_mode]( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
+ src0, i_stride0, src1, i_stride1, weight );
+
+ if( h->mb.b_interlaced & i_ref0 )
+ mvy0 += (h->mb.i_mb_y & 1)*4 - 2;
+ if( h->mb.b_interlaced & i_ref1 )
+ mvy1 += (h->mb.i_mb_y & 1)*4 - 2;
+
+ h->mc.mc_chroma( tmp0, 16, &h->mb.pic.p_fref[0][i_ref0][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
+ mvx0, mvy0, 2*width, 2*height );
+ h->mc.mc_chroma( tmp1, 16, &h->mb.pic.p_fref[1][i_ref1][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
+ mvx1, mvy1, 2*width, 2*height );
+ h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp0, 16, tmp1, 16, weight );
+ h->mc.mc_chroma( tmp0, 16, &h->mb.pic.p_fref[0][i_ref0][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
+ mvx0, mvy0, 2*width, 2*height );
+ h->mc.mc_chroma( tmp1, 16, &h->mb.pic.p_fref[1][i_ref1][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
+ mvx1, mvy1, 2*width, 2*height );
+ h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp0, 16, tmp1, 16, weight );
}
static void x264_mb_mc_direct8x8( x264_t *h, int x, int y )
{
const int i8 = x264_scan8[0] + x + 8*y;
- /* FIXME: optimize based on current block size, not global settings? */
- if( h->sps->b_direct8x8_inference )
- {
- if( h->mb.cache.ref[0][i8] >= 0 )
- if( h->mb.cache.ref[1][i8] >= 0 )
- x264_mb_mc_01xywh( h, x, y, 2, 2 );
- else
- x264_mb_mc_0xywh( h, x, y, 2, 2 );
+ if( h->mb.cache.ref[0][i8] >= 0 )
+ if( h->mb.cache.ref[1][i8] >= 0 )
+ x264_mb_mc_01xywh( h, x, y, 2, 2 );
else
- x264_mb_mc_1xywh( h, x, y, 2, 2 );
- }
+ x264_mb_mc_0xywh( h, x, y, 2, 2 );
else
- {
- if( h->mb.cache.ref[0][i8] >= 0 )
- {
- if( h->mb.cache.ref[1][i8] >= 0 )
- {
- x264_mb_mc_01xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+1, 1, 1 );
- }
- else
- {
- x264_mb_mc_0xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_0xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_0xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_0xywh( h, x+1, y+1, 1, 1 );
- }
- }
- else
- {
- x264_mb_mc_1xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+1, 1, 1 );
- }
- }
+ x264_mb_mc_1xywh( h, x, y, 2, 2 );
}
void x264_mb_mc_8x8( x264_t *h, int i8 )
case D_L1_8x8:
x264_mb_mc_1xywh( h, x, y, 2, 2 );
break;
- case D_L1_8x4:
- x264_mb_mc_1xywh( h, x, y+0, 2, 1 );
- x264_mb_mc_1xywh( h, x, y+1, 2, 1 );
- break;
- case D_L1_4x8:
- x264_mb_mc_1xywh( h, x+0, y, 1, 2 );
- x264_mb_mc_1xywh( h, x+1, y, 1, 2 );
- break;
- case D_L1_4x4:
- x264_mb_mc_1xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_1xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_1xywh( h, x+1, y+1, 1, 1 );
- break;
case D_BI_8x8:
x264_mb_mc_01xywh( h, x, y, 2, 2 );
break;
- case D_BI_8x4:
- x264_mb_mc_01xywh( h, x, y+0, 2, 1 );
- x264_mb_mc_01xywh( h, x, y+1, 2, 1 );
- break;
- case D_BI_4x8:
- x264_mb_mc_01xywh( h, x+0, y, 1, 2 );
- x264_mb_mc_01xywh( h, x+1, y, 1, 2 );
- break;
- case D_BI_4x4:
- x264_mb_mc_01xywh( h, x+0, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+0, 1, 1 );
- x264_mb_mc_01xywh( h, x+0, y+1, 1, 1 );
- x264_mb_mc_01xywh( h, x+1, y+1, 1, 1 );
- break;
case D_DIRECT_8x8:
x264_mb_mc_direct8x8( h, x, y );
break;
}
else /* B_*x* */
{
- int b_list0[2];
- int b_list1[2];
-
- int i;
+ const uint8_t *b_list0 = x264_mb_type_list_table[h->mb.i_type][0];
+ const uint8_t *b_list1 = x264_mb_type_list_table[h->mb.i_type][1];
- /* init ref list utilisations */
- for( i = 0; i < 2; i++ )
- {
- b_list0[i] = x264_mb_type_list0_table[h->mb.i_type][i];
- b_list1[i] = x264_mb_type_list1_table[h->mb.i_type][i];
- }
if( h->mb.i_partition == D_16x16 )
{
if( b_list0[0] && b_list1[0] ) x264_mb_mc_01xywh( h, 0, 0, 4, 4 );
}
}
-void x264_macroblock_cache_init( x264_t *h )
+int x264_macroblock_cache_init( x264_t *h )
{
int i, j;
int i_mb_count = h->mb.i_mb_count;
h->mb.i_b8_stride = h->sps->i_mb_width * 2;
h->mb.i_b4_stride = h->sps->i_mb_width * 4;
- h->mb.qp = x264_malloc( i_mb_count * sizeof(int8_t) );
- h->mb.cbp = x264_malloc( i_mb_count * sizeof(int16_t) );
- h->mb.skipbp = x264_malloc( i_mb_count * sizeof(int8_t) );
- h->mb.mb_transform_size = x264_malloc( i_mb_count * sizeof(int8_t) );
+ h->mb.b_interlaced = h->param.b_interlaced;
+
+ CHECKED_MALLOC( h->mb.qp, i_mb_count * sizeof(int8_t) );
+ CHECKED_MALLOC( h->mb.cbp, i_mb_count * sizeof(int16_t) );
+ CHECKED_MALLOC( h->mb.skipbp, i_mb_count * sizeof(int8_t) );
+ CHECKED_MALLOC( h->mb.mb_transform_size, i_mb_count * sizeof(int8_t) );
/* 0 -> 3 top(4), 4 -> 6 : left(3) */
- h->mb.intra4x4_pred_mode = x264_malloc( i_mb_count * 7 * sizeof( int8_t ) );
+ CHECKED_MALLOC( h->mb.intra4x4_pred_mode, i_mb_count * 8 * sizeof(int8_t) );
/* all coeffs */
- h->mb.non_zero_count = x264_malloc( i_mb_count * 24 * sizeof( uint8_t ) );
+ CHECKED_MALLOC( h->mb.non_zero_count, i_mb_count * 24 * sizeof(uint8_t) );
+ CHECKED_MALLOC( h->mb.nnz_backup, h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t) );
if( h->param.b_cabac )
{
- h->mb.chroma_pred_mode = x264_malloc( i_mb_count * sizeof( int8_t) );
- h->mb.mvd[0] = x264_malloc( 2*16 * i_mb_count * sizeof( int16_t ) );
- h->mb.mvd[1] = x264_malloc( 2*16 * i_mb_count * sizeof( int16_t ) );
+ CHECKED_MALLOC( h->mb.chroma_pred_mode, i_mb_count * sizeof(int8_t) );
+ CHECKED_MALLOC( h->mb.mvd[0], 2*16 * i_mb_count * sizeof(int16_t) );
+ CHECKED_MALLOC( h->mb.mvd[1], 2*16 * i_mb_count * sizeof(int16_t) );
}
for( i=0; i<2; i++ )
{
- int i_refs = (i ? 1 : h->param.i_frame_reference) + h->param.b_bframe_pyramid;
- for( j=0; j < i_refs && j < 16; j++ )
- h->mb.mvr[i][j] = x264_malloc( 2 * i_mb_count * sizeof( int16_t ) );
+ int i_refs = X264_MIN(16, (i ? 1 : h->param.i_frame_reference) + h->param.b_bframe_pyramid) << h->param.b_interlaced;
+ for( j=0; j < i_refs; j++ )
+ CHECKED_MALLOC( h->mb.mvr[i][j], 2 * i_mb_count * sizeof(int16_t) );
}
- /* init with not avaiable (for top right idx=7,15) */
+ for( i=0; i<=h->param.b_interlaced; i++ )
+ for( j=0; j<3; j++ )
+ {
+ /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
+ CHECKED_MALLOCZERO( h->mb.intra_border_backup[i][j], (h->sps->i_mb_width*16+32)>>!!j );
+ h->mb.intra_border_backup[i][j] += 8;
+ }
+
+ /* init with not available (for top right idx=7,15) */
memset( h->mb.cache.ref[0], -2, X264_SCAN8_SIZE * sizeof( int8_t ) );
memset( h->mb.cache.ref[1], -2, X264_SCAN8_SIZE * sizeof( int8_t ) );
+
+ /* fdec: fenc:
+ * yyyyyyy
+ * yYYYY YYYY
+ * yYYYY YYYY
+ * yYYYY YYYY
+ * yYYYY YYYY
+ * uuu vvv UUVV
+ * uUU vVV UUVV
+ * uUU vVV
+ */
+ h->mb.pic.p_fenc[0] = h->mb.pic.fenc_buf;
+ h->mb.pic.p_fenc[1] = h->mb.pic.fenc_buf + 16*FENC_STRIDE;
+ h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 16*FENC_STRIDE + 8;
+ h->mb.pic.p_fdec[0] = h->mb.pic.fdec_buf + 2*FDEC_STRIDE;
+ h->mb.pic.p_fdec[1] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE;
+ h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE + 16;
+
+ h->mb.i_neighbour4[6] =
+ h->mb.i_neighbour4[9] =
+ h->mb.i_neighbour4[12] =
+ h->mb.i_neighbour4[14] = MB_LEFT|MB_TOP|MB_TOPLEFT|MB_TOPRIGHT;
+ h->mb.i_neighbour4[3] =
+ h->mb.i_neighbour4[7] =
+ h->mb.i_neighbour4[11] =
+ h->mb.i_neighbour4[13] =
+ h->mb.i_neighbour4[15] =
+ h->mb.i_neighbour8[3] = MB_LEFT|MB_TOP|MB_TOPLEFT;
+
+ int buf_hpel = (h->param.i_width+48) * sizeof(int16_t);
+ int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
+ int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
+ int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
+ ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
+ int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
+ CHECKED_MALLOC( h->scratch_buffer, X264_MAX4( buf_hpel, buf_ssim, buf_tesa, buf_mbtree ) );
+
+ return 0;
+fail: return -1;
}
void x264_macroblock_cache_end( x264_t *h )
{
int i, j;
+ for( i=0; i<=h->param.b_interlaced; i++ )
+ for( j=0; j<3; j++ )
+ x264_free( h->mb.intra_border_backup[i][j] - 8 );
for( i=0; i<2; i++ )
- {
- int i_refs = i ? 1 + h->param.b_bframe_pyramid : h->param.i_frame_reference;
- for( j=0; j < i_refs; j++ )
+ for( j=0; j<32; j++ )
x264_free( h->mb.mvr[i][j] );
- }
if( h->param.b_cabac )
{
x264_free( h->mb.chroma_pred_mode );
}
x264_free( h->mb.intra4x4_pred_mode );
x264_free( h->mb.non_zero_count );
+ x264_free( h->mb.nnz_backup );
x264_free( h->mb.mb_transform_size );
x264_free( h->mb.skipbp );
x264_free( h->mb.cbp );
x264_free( h->mb.qp );
+ x264_free( h->scratch_buffer );
}
void x264_macroblock_slice_init( x264_t *h )
{
}
if( h->sh.i_type == SLICE_TYPE_P )
memset( h->mb.cache.skip, 0, X264_SCAN8_SIZE * sizeof( int8_t ) );
+
+ setup_inverse_delta_pocs( h );
}
+void x264_prefetch_fenc( x264_t *h, x264_frame_t *fenc, int i_mb_x, int i_mb_y )
+{
+ int stride_y = fenc->i_stride[0];
+ int stride_uv = fenc->i_stride[1];
+ int off_y = 16 * (i_mb_x + i_mb_y * stride_y);
+ int off_uv = 8 * (i_mb_x + i_mb_y * stride_uv);
+ h->mc.prefetch_fenc( fenc->plane[0]+off_y, stride_y,
+ fenc->plane[1+(i_mb_x&1)]+off_uv, stride_uv, i_mb_x );
+}
-void x264_macroblock_cache_load( x264_t *h, int i_mb_x, int i_mb_y )
+static NOINLINE void copy_column8( uint8_t *dst, uint8_t *src )
{
- const int i_mb_4x4 = 4*(i_mb_y * h->mb.i_b4_stride + i_mb_x);
- const int i_mb_8x8 = 2*(i_mb_y * h->mb.i_b8_stride + i_mb_x);
+ // input pointers are offset by 4 rows because that's faster (smaller instruction size on x86)
+ int i;
+ for( i = -4; i < 4; i++ )
+ dst[i*FDEC_STRIDE] = src[i*FDEC_STRIDE];
+}
+static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int i_mb_x, int i_mb_y, int i)
+{
+ const int w = (i == 0 ? 16 : 8);
+ const int i_stride = h->fdec->i_stride[!!i];
+ const int i_stride2 = i_stride << h->mb.b_interlaced;
+ const int i_pix_offset = h->mb.b_interlaced
+ ? w * (i_mb_x + (i_mb_y&~1) * i_stride) + (i_mb_y&1) * i_stride
+ : w * (i_mb_x + i_mb_y * i_stride);
+ int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
+ const uint8_t *intra_fdec = &h->mb.intra_border_backup[i_mb_y & h->sh.b_mbaff][i][i_mb_x*16>>!!i];
+ x264_frame_t **fref[2] = { h->fref0, h->fref1 };
+ int j, k;
+ if( h->mb.b_interlaced )
+ ref_pix_offset[1] += (1-2*(i_mb_y&1)) * i_stride;
+ h->mb.pic.i_stride[i] = i_stride2;
+ h->mb.pic.p_fenc_plane[i] = &h->fenc->plane[i][i_pix_offset];
+ h->mc.copy[i?PIXEL_8x8:PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE,
+ h->mb.pic.p_fenc_plane[i], i_stride2, w );
+ memcpy( &h->mb.pic.p_fdec[i][-1-FDEC_STRIDE], intra_fdec-1, w*3/2+1 );
+ if( h->mb.b_interlaced || h->mb.b_reencode_mb )
+ {
+ const uint8_t *plane_fdec = &h->fdec->plane[i][i_pix_offset];
+ for( j = 0; j < w; j++ )
+ h->mb.pic.p_fdec[i][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
+ }
+ for( j = 0; j < h->mb.pic.i_fref[0]; j++ )
+ {
+ h->mb.pic.p_fref[0][j][i==0 ? 0:i+3] = &fref[0][j >> h->mb.b_interlaced]->plane[i][ref_pix_offset[j&1]];
+ if( i == 0 )
+ for( k = 1; k < 4; k++ )
+ h->mb.pic.p_fref[0][j][k] = &fref[0][j >> h->mb.b_interlaced]->filtered[k][ref_pix_offset[j&1]];
+ }
+ if( h->sh.i_type == SLICE_TYPE_B )
+ for( j = 0; j < h->mb.pic.i_fref[1]; j++ )
+ {
+ h->mb.pic.p_fref[1][j][i==0 ? 0:i+3] = &fref[1][j >> h->mb.b_interlaced]->plane[i][ref_pix_offset[j&1]];
+ if( i == 0 )
+ for( k = 1; k < 4; k++ )
+ h->mb.pic.p_fref[1][j][k] = &fref[1][j >> h->mb.b_interlaced]->filtered[k][ref_pix_offset[j&1]];
+ }
+}
+
+void x264_macroblock_cache_load( x264_t *h, int i_mb_x, int i_mb_y )
+{
int i_mb_xy = i_mb_y * h->mb.i_mb_stride + i_mb_x;
- int i_top_xy = i_mb_xy - h->mb.i_mb_stride;
+ int i_mb_4x4 = 4*(i_mb_y * h->mb.i_b4_stride + i_mb_x);
+ int i_mb_8x8 = 2*(i_mb_y * h->mb.i_b8_stride + i_mb_x);
+ int i_top_y = i_mb_y - (1 << h->mb.b_interlaced);
+ int i_top_xy = i_top_y * h->mb.i_mb_stride + i_mb_x;
+ int i_top_4x4 = (4*i_top_y+3) * h->mb.i_b4_stride + 4*i_mb_x;
+ int i_top_8x8 = (2*i_top_y+1) * h->mb.i_b8_stride + 2*i_mb_x;
int i_left_xy = -1;
int i_top_type = -1; /* gcc warn */
int i_left_type= -1;
h->mb.i_mb_xy = i_mb_xy;
h->mb.i_b8_xy = i_mb_8x8;
h->mb.i_b4_xy = i_mb_4x4;
+ h->mb.i_mb_top_xy = i_top_xy;
h->mb.i_neighbour = 0;
- /* fdec: fenc:
- * yyyyyyy
- * yYYYY YYYY
- * yYYYY YYYY
- * yYYYY YYYY
- * yYYYY YYYY
- * uuu vvv UUVV
- * uUU vVV UUVV
- * uUU vVV
- */
- h->mb.pic.p_fenc[0] = h->mb.pic.fenc_buf;
- h->mb.pic.p_fenc[1] = h->mb.pic.fenc_buf + 16*FENC_STRIDE;
- h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 16*FENC_STRIDE + 8;
- h->mb.pic.p_fdec[0] = h->mb.pic.fdec_buf + 2*FDEC_STRIDE;
- h->mb.pic.p_fdec[1] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE;
- h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE + 16;
-
- /* load picture pointers */
- for( i = 0; i < 3; i++ )
- {
- const int w = (i == 0 ? 16 : 8);
- const int i_stride = h->fdec->i_stride[i];
- const uint8_t *plane_fdec = &h->fdec->plane[i][ w * (i_mb_x + i_mb_y * i_stride) ];
- int j;
-
- h->mb.pic.i_stride[i] = i_stride;
-
- h->mc.copy[i?PIXEL_8x8:PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE,
- &h->fenc->plane[i][ w * (i_mb_x + i_mb_y * i_stride) ], i_stride, w );
- memcpy( &h->mb.pic.p_fdec[i][-1-FDEC_STRIDE], &plane_fdec[-1-i_stride], w*3/2+1 );
- for( j = 0; j < w; j++ )
- h->mb.pic.p_fdec[i][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride];
-
- for( j = 0; j < h->i_ref0; j++ )
- {
- h->mb.pic.p_fref[0][j][i==0 ? 0:i+3] = &h->fref0[j]->plane[i][ w * ( i_mb_x + i_mb_y * i_stride )];
- h->mb.pic.p_fref[0][j][i+1] = &h->fref0[j]->filtered[i+1][ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
- }
- for( j = 0; j < h->i_ref1; j++ )
- {
- h->mb.pic.p_fref[1][j][i==0 ? 0:i+3] = &h->fref1[j]->plane[i][ w * ( i_mb_x + i_mb_y * i_stride )];
- h->mb.pic.p_fref[1][j][i+1] = &h->fref1[j]->filtered[i+1][ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
- }
- }
-
- if( h->fdec->integral )
- {
- for( i = 0; i < h->i_ref0; i++ )
- h->mb.pic.p_integral[0][i] = &h->fref0[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
- for( i = 0; i < h->i_ref1; i++ )
- h->mb.pic.p_integral[1][i] = &h->fref1[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
- }
-
/* load cache */
- if( i_mb_xy >= h->sh.i_first_mb + h->mb.i_mb_stride )
+ if( i_top_xy >= h->sh.i_first_mb )
{
h->mb.i_mb_type_top =
i_top_type= h->mb.type[i_top_xy];
+ h->mb.cache.i_cbp_top = h->mb.cbp[i_top_xy];
h->mb.i_neighbour |= MB_TOP;
/* load intra4x4 */
- h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] = h->mb.intra4x4_pred_mode[i_top_xy][0];
- h->mb.cache.intra4x4_pred_mode[x264_scan8[1] - 8] = h->mb.intra4x4_pred_mode[i_top_xy][1];
- h->mb.cache.intra4x4_pred_mode[x264_scan8[4] - 8] = h->mb.intra4x4_pred_mode[i_top_xy][2];
- h->mb.cache.intra4x4_pred_mode[x264_scan8[5] - 8] = h->mb.intra4x4_pred_mode[i_top_xy][3];
+ *(uint32_t*)&h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] = *(uint32_t*)&h->mb.intra4x4_pred_mode[i_top_xy][0];
/* load non_zero_count */
- h->mb.cache.non_zero_count[x264_scan8[0] - 8] = h->mb.non_zero_count[i_top_xy][10];
- h->mb.cache.non_zero_count[x264_scan8[1] - 8] = h->mb.non_zero_count[i_top_xy][11];
- h->mb.cache.non_zero_count[x264_scan8[4] - 8] = h->mb.non_zero_count[i_top_xy][14];
- h->mb.cache.non_zero_count[x264_scan8[5] - 8] = h->mb.non_zero_count[i_top_xy][15];
-
- h->mb.cache.non_zero_count[x264_scan8[16+0] - 8] = h->mb.non_zero_count[i_top_xy][16+2];
- h->mb.cache.non_zero_count[x264_scan8[16+1] - 8] = h->mb.non_zero_count[i_top_xy][16+3];
-
- h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 8] = h->mb.non_zero_count[i_top_xy][16+4+2];
- h->mb.cache.non_zero_count[x264_scan8[16+4+1] - 8] = h->mb.non_zero_count[i_top_xy][16+4+3];
+ *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0] - 8] = *(uint32_t*)&h->mb.non_zero_count[i_top_xy][12];
+ /* shift because x264_scan8[16] is misaligned */
+ *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16] - 9] = *(uint16_t*)&h->mb.non_zero_count[i_top_xy][18] << 8;
+ *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] = *(uint16_t*)&h->mb.non_zero_count[i_top_xy][22] << 8;
}
else
{
h->mb.i_mb_type_top = -1;
-
+ h->mb.cache.i_cbp_top = -1;
+
/* load intra4x4 */
h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] =
h->mb.cache.intra4x4_pred_mode[x264_scan8[1] - 8] =
h->mb.cache.non_zero_count[x264_scan8[16+1] - 8] =
h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 8] =
h->mb.cache.non_zero_count[x264_scan8[16+4+1] - 8] = 0x80;
-
}
if( i_mb_x > 0 && i_mb_xy > h->sh.i_first_mb )
i_left_xy = i_mb_xy - 1;
h->mb.i_mb_type_left =
i_left_type = h->mb.type[i_left_xy];
+ h->mb.cache.i_cbp_left = h->mb.cbp[h->mb.i_mb_xy - 1];
h->mb.i_neighbour |= MB_LEFT;
h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][3];
/* load non_zero_count */
- h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = h->mb.non_zero_count[i_left_xy][5];
+ h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = h->mb.non_zero_count[i_left_xy][3];
h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = h->mb.non_zero_count[i_left_xy][7];
- h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = h->mb.non_zero_count[i_left_xy][13];
+ h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = h->mb.non_zero_count[i_left_xy][11];
h->mb.cache.non_zero_count[x264_scan8[10] - 1] = h->mb.non_zero_count[i_left_xy][15];
h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = h->mb.non_zero_count[i_left_xy][16+1];
else
{
h->mb.i_mb_type_left = -1;
+ h->mb.cache.i_cbp_left = -1;
h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] =
h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] =
else
h->mb.i_mb_type_topleft = -1;
- if( h->param.analyse.b_transform_8x8 )
+ if( h->pps->b_transform_8x8_mode )
{
h->mb.cache.i_neighbour_transform_size =
( i_left_type >= 0 && h->mb.mb_transform_size[i_left_xy] )
+ ( i_top_type >= 0 && h->mb.mb_transform_size[i_top_xy] );
}
+ if( h->sh.b_mbaff )
+ {
+ h->mb.pic.i_fref[0] = h->i_ref0 << h->mb.b_interlaced;
+ h->mb.pic.i_fref[1] = h->i_ref1 << h->mb.b_interlaced;
+ h->mb.cache.i_neighbour_interlaced =
+ !!(h->mb.i_neighbour & MB_LEFT)
+ + !!(h->mb.i_neighbour & MB_TOP);
+ }
+
+ if( !h->mb.b_interlaced && !h->mb.b_reencode_mb )
+ {
+ copy_column8( h->mb.pic.p_fdec[0]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+ 4*FDEC_STRIDE );
+ copy_column8( h->mb.pic.p_fdec[0]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+12*FDEC_STRIDE );
+ copy_column8( h->mb.pic.p_fdec[1]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[1]+ 7+ 4*FDEC_STRIDE );
+ copy_column8( h->mb.pic.p_fdec[2]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[2]+ 7+ 4*FDEC_STRIDE );
+ }
+
+ /* load picture pointers */
+ x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 0 );
+ x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 1 );
+ x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 2 );
+
+ if( h->fdec->integral )
+ {
+ assert( !h->mb.b_interlaced );
+ for( i = 0; i < h->mb.pic.i_fref[0]; i++ )
+ h->mb.pic.p_integral[0][i] = &h->fref0[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
+ for( i = 0; i < h->mb.pic.i_fref[1]; i++ )
+ h->mb.pic.p_integral[1][i] = &h->fref1[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
+ }
+
+ x264_prefetch_fenc( h, h->fenc, i_mb_x, i_mb_y );
+
/* load ref/mv/mvd */
if( h->sh.i_type != SLICE_TYPE_I )
{
if( h->mb.i_neighbour & MB_TOPLEFT )
{
const int i8 = x264_scan8[0] - 1 - 1*8;
- const int ir = i_mb_8x8 - s8x8 - 1;
- const int iv = i_mb_4x4 - s4x4 - 1;
+ const int ir = i_top_8x8 - 1;
+ const int iv = i_top_4x4 - 1;
h->mb.cache.ref[i_list][i8] = h->mb.ref[i_list][ir];
- h->mb.cache.mv[i_list][i8][0] = h->mb.mv[i_list][iv][0];
- h->mb.cache.mv[i_list][i8][1] = h->mb.mv[i_list][iv][1];
+ *(uint32_t*)h->mb.cache.mv[i_list][i8] = *(uint32_t*)h->mb.mv[i_list][iv];
}
else
{
const int i8 = x264_scan8[0] - 1 - 1*8;
h->mb.cache.ref[i_list][i8] = -2;
- h->mb.cache.mv[i_list][i8][0] = 0;
- h->mb.cache.mv[i_list][i8][1] = 0;
+ *(uint32_t*)h->mb.cache.mv[i_list][i8] = 0;
}
if( h->mb.i_neighbour & MB_TOP )
{
const int i8 = x264_scan8[0] - 8;
- const int ir = i_mb_8x8 - s8x8;
- const int iv = i_mb_4x4 - s4x4;
-
+ const int ir = i_top_8x8;
+ const int iv = i_top_4x4;
h->mb.cache.ref[i_list][i8+0] =
h->mb.cache.ref[i_list][i8+1] = h->mb.ref[i_list][ir + 0];
h->mb.cache.ref[i_list][i8+2] =
h->mb.cache.ref[i_list][i8+3] = h->mb.ref[i_list][ir + 1];
-
- for( i = 0; i < 4; i++ )
- {
- h->mb.cache.mv[i_list][i8+i][0] = h->mb.mv[i_list][iv + i][0];
- h->mb.cache.mv[i_list][i8+i][1] = h->mb.mv[i_list][iv + i][1];
- }
+ *(uint64_t*)h->mb.cache.mv[i_list][i8+0] = *(uint64_t*)h->mb.mv[i_list][iv+0];
+ *(uint64_t*)h->mb.cache.mv[i_list][i8+2] = *(uint64_t*)h->mb.mv[i_list][iv+2];
}
else
{
const int i8 = x264_scan8[0] - 8;
- for( i = 0; i < 4; i++ )
- {
- h->mb.cache.ref[i_list][i8+i] = -2;
- h->mb.cache.mv[i_list][i8+i][0] =
- h->mb.cache.mv[i_list][i8+i][1] = 0;
- }
+ *(uint64_t*)h->mb.cache.mv[i_list][i8+0] = 0;
+ *(uint64_t*)h->mb.cache.mv[i_list][i8+2] = 0;
+ *(uint32_t*)&h->mb.cache.ref[i_list][i8] = (uint8_t)(-2) * 0x01010101U;
}
if( h->mb.i_neighbour & MB_TOPRIGHT )
{
const int i8 = x264_scan8[0] + 4 - 1*8;
- const int ir = i_mb_8x8 - s8x8 + 2;
- const int iv = i_mb_4x4 - s4x4 + 4;
-
+ const int ir = i_top_8x8 + 2;
+ const int iv = i_top_4x4 + 4;
h->mb.cache.ref[i_list][i8] = h->mb.ref[i_list][ir];
- h->mb.cache.mv[i_list][i8][0] = h->mb.mv[i_list][iv][0];
- h->mb.cache.mv[i_list][i8][1] = h->mb.mv[i_list][iv][1];
+ *(uint32_t*)h->mb.cache.mv[i_list][i8] = *(uint32_t*)h->mb.mv[i_list][iv];
}
else
{
const int i8 = x264_scan8[0] + 4 - 1*8;
h->mb.cache.ref[i_list][i8] = -2;
- h->mb.cache.mv[i_list][i8][0] = 0;
- h->mb.cache.mv[i_list][i8][1] = 0;
+ *(uint32_t*)h->mb.cache.mv[i_list][i8] = 0;
}
if( h->mb.i_neighbour & MB_LEFT )
const int i8 = x264_scan8[0] - 1;
const int ir = i_mb_8x8 - 1;
const int iv = i_mb_4x4 - 1;
-
h->mb.cache.ref[i_list][i8+0*8] =
h->mb.cache.ref[i_list][i8+1*8] = h->mb.ref[i_list][ir + 0*s8x8];
h->mb.cache.ref[i_list][i8+2*8] =
h->mb.cache.ref[i_list][i8+3*8] = h->mb.ref[i_list][ir + 1*s8x8];
- for( i = 0; i < 4; i++ )
- {
- h->mb.cache.mv[i_list][i8+i*8][0] = h->mb.mv[i_list][iv + i*s4x4][0];
- h->mb.cache.mv[i_list][i8+i*8][1] = h->mb.mv[i_list][iv + i*s4x4][1];
- }
+ *(uint32_t*)h->mb.cache.mv[i_list][i8+0*8] = *(uint32_t*)h->mb.mv[i_list][iv + 0*s4x4];
+ *(uint32_t*)h->mb.cache.mv[i_list][i8+1*8] = *(uint32_t*)h->mb.mv[i_list][iv + 1*s4x4];
+ *(uint32_t*)h->mb.cache.mv[i_list][i8+2*8] = *(uint32_t*)h->mb.mv[i_list][iv + 2*s4x4];
+ *(uint32_t*)h->mb.cache.mv[i_list][i8+3*8] = *(uint32_t*)h->mb.mv[i_list][iv + 3*s4x4];
}
else
{
for( i = 0; i < 4; i++ )
{
h->mb.cache.ref[i_list][i8+i*8] = -2;
- h->mb.cache.mv[i_list][i8+i*8][0] =
- h->mb.cache.mv[i_list][i8+i*8][1] = 0;
+ *(uint32_t*)h->mb.cache.mv[i_list][i8+i*8] = 0;
}
}
if( i_top_type >= 0 )
{
const int i8 = x264_scan8[0] - 8;
- const int iv = i_mb_4x4 - s4x4;
- for( i = 0; i < 4; i++ )
- {
- h->mb.cache.mvd[i_list][i8+i][0] = h->mb.mvd[i_list][iv + i][0];
- h->mb.cache.mvd[i_list][i8+i][1] = h->mb.mvd[i_list][iv + i][1];
- }
+ const int iv = i_top_4x4;
+ *(uint64_t*)h->mb.cache.mvd[i_list][i8+0] = *(uint64_t*)h->mb.mvd[i_list][iv+0];
+ *(uint64_t*)h->mb.cache.mvd[i_list][i8+2] = *(uint64_t*)h->mb.mvd[i_list][iv+2];
}
else
{
const int i8 = x264_scan8[0] - 8;
- for( i = 0; i < 4; i++ )
- {
- h->mb.cache.mvd[i_list][i8+i][0] =
- h->mb.cache.mvd[i_list][i8+i][1] = 0;
- }
+ *(uint64_t*)h->mb.cache.mvd[i_list][i8+0] =
+ *(uint64_t*)h->mb.cache.mvd[i_list][i8+2] = 0;
}
if( i_left_type >= 0 )
{
const int i8 = x264_scan8[0] - 1;
const int iv = i_mb_4x4 - 1;
- for( i = 0; i < 4; i++ )
- {
- h->mb.cache.mvd[i_list][i8+i*8][0] = h->mb.mvd[i_list][iv + i*s4x4][0];
- h->mb.cache.mvd[i_list][i8+i*8][1] = h->mb.mvd[i_list][iv + i*s4x4][1];
- }
+ *(uint32_t*)h->mb.cache.mvd[i_list][i8+0*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 0*s4x4];
+ *(uint32_t*)h->mb.cache.mvd[i_list][i8+1*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 1*s4x4];
+ *(uint32_t*)h->mb.cache.mvd[i_list][i8+2*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 2*s4x4];
+ *(uint32_t*)h->mb.cache.mvd[i_list][i8+3*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 3*s4x4];
}
else
{
const int i8 = x264_scan8[0] - 1;
for( i = 0; i < 4; i++ )
- {
- h->mb.cache.mvd[i_list][i8+i*8][0] =
- h->mb.cache.mvd[i_list][i8+i*8][1] = 0;
- }
+ *(uint32_t*)h->mb.cache.mvd[i_list][i8+i*8] = 0;
}
}
}
/* load skip */
if( h->sh.i_type == SLICE_TYPE_B && h->param.b_cabac )
{
- memset( h->mb.cache.skip, 0, X264_SCAN8_SIZE * sizeof( int8_t ) );
- if( i_left_type >= 0 )
- {
- h->mb.cache.skip[x264_scan8[0] - 1] = h->mb.skipbp[i_left_xy] & 0x2;
- h->mb.cache.skip[x264_scan8[8] - 1] = h->mb.skipbp[i_left_xy] & 0x8;
- }
- if( i_top_type >= 0 )
- {
- h->mb.cache.skip[x264_scan8[0] - 8] = h->mb.skipbp[i_top_xy] & 0x4;
- h->mb.cache.skip[x264_scan8[4] - 8] = h->mb.skipbp[i_top_xy] & 0x8;
- }
+ uint8_t skipbp;
+ x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
+ skipbp = i_left_type >= 0 ? h->mb.skipbp[i_left_xy] : 0;
+ h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
+ h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
+ skipbp = i_top_type >= 0 ? h->mb.skipbp[i_top_xy] : 0;
+ h->mb.cache.skip[x264_scan8[0] - 8] = skipbp & 0x4;
+ h->mb.cache.skip[x264_scan8[4] - 8] = skipbp & 0x8;
}
+
+ if( h->sh.i_type == SLICE_TYPE_P )
+ x264_mb_predict_mv_pskip( h, h->mb.cache.pskip_mv );
}
h->mb.i_neighbour4[0] =
h->mb.i_neighbour4[8] =
h->mb.i_neighbour4[10] =
h->mb.i_neighbour8[2] = MB_TOP|MB_TOPRIGHT | ((h->mb.i_neighbour & MB_LEFT) ? (MB_LEFT|MB_TOPLEFT) : 0);
- h->mb.i_neighbour4[3] =
- h->mb.i_neighbour4[7] =
- h->mb.i_neighbour4[11] =
- h->mb.i_neighbour4[13] =
- h->mb.i_neighbour4[15] =
- h->mb.i_neighbour8[3] = MB_LEFT|MB_TOP|MB_TOPLEFT;
h->mb.i_neighbour4[5] =
h->mb.i_neighbour8[1] = MB_LEFT | (h->mb.i_neighbour & MB_TOPRIGHT)
| ((h->mb.i_neighbour & MB_TOP) ? MB_TOP|MB_TOPLEFT : 0);
- h->mb.i_neighbour4[6] =
- h->mb.i_neighbour4[9] =
- h->mb.i_neighbour4[12] =
- h->mb.i_neighbour4[14] = MB_LEFT|MB_TOP|MB_TOPLEFT|MB_TOPRIGHT;
+}
+
+static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int i)
+{
+ int w = i ? 8 : 16;
+ int i_stride = h->fdec->i_stride[!!i];
+ int i_stride2 = i_stride << h->mb.b_interlaced;
+ int i_pix_offset = h->mb.b_interlaced
+ ? w * (h->mb.i_mb_x + (h->mb.i_mb_y&~1) * i_stride) + (h->mb.i_mb_y&1) * i_stride
+ : w * (h->mb.i_mb_x + h->mb.i_mb_y * i_stride);
+ h->mc.copy[i?PIXEL_8x8:PIXEL_16x16](
+ &h->fdec->plane[i][i_pix_offset], i_stride2,
+ h->mb.pic.p_fdec[i], FDEC_STRIDE, w );
}
void x264_macroblock_cache_save( x264_t *h )
const int i_mb_4x4 = h->mb.i_b4_xy;
const int i_mb_8x8 = h->mb.i_b8_xy;
- int i;
+ /* GCC pessimizes direct stores to heap-allocated 8-bit arrays due to aliasing.*/
+ /* By only dereferencing them once, we avoid this issue. */
+ int8_t *intra4x4_pred_mode = h->mb.intra4x4_pred_mode[i_mb_xy];
+ uint8_t *non_zero_count = h->mb.non_zero_count[i_mb_xy];
- for( i = 0; i < 3; i++ )
- {
- int w = i ? 8 : 16;
- h->mc.copy[i?PIXEL_8x8:PIXEL_16x16](
- &h->fdec->plane[i][ w * (h->mb.i_mb_x + h->mb.i_mb_y * h->fdec->i_stride[i]) ],
- h->fdec->i_stride[i],
- h->mb.pic.p_fdec[i], FDEC_STRIDE, w );
- }
+ int i, y;
- h->mb.type[i_mb_xy] = i_mb_type;
+ x264_macroblock_store_pic( h, 0 );
+ x264_macroblock_store_pic( h, 1 );
+ x264_macroblock_store_pic( h, 2 );
- if( h->mb.i_type != I_16x16 && h->mb.i_cbp_luma == 0 && h->mb.i_cbp_chroma == 0 )
- h->mb.i_qp = h->mb.i_last_qp;
- h->mb.qp[i_mb_xy] = h->mb.i_qp;
+ x264_prefetch_fenc( h, h->fdec, h->mb.i_mb_x, h->mb.i_mb_y );
- h->mb.i_last_dqp = h->mb.i_qp - h->mb.i_last_qp;
- h->mb.i_last_qp = h->mb.i_qp;
+ h->mb.type[i_mb_xy] = i_mb_type;
+ h->mb.i_mb_prev_xy = i_mb_xy;
/* save intra4x4 */
if( i_mb_type == I_4x4 )
{
- h->mb.intra4x4_pred_mode[i_mb_xy][0] = h->mb.cache.intra4x4_pred_mode[x264_scan8[10] ];
- h->mb.intra4x4_pred_mode[i_mb_xy][1] = h->mb.cache.intra4x4_pred_mode[x264_scan8[11] ];
- h->mb.intra4x4_pred_mode[i_mb_xy][2] = h->mb.cache.intra4x4_pred_mode[x264_scan8[14] ];
- h->mb.intra4x4_pred_mode[i_mb_xy][3] = h->mb.cache.intra4x4_pred_mode[x264_scan8[15] ];
- h->mb.intra4x4_pred_mode[i_mb_xy][4] = h->mb.cache.intra4x4_pred_mode[x264_scan8[5] ];
- h->mb.intra4x4_pred_mode[i_mb_xy][5] = h->mb.cache.intra4x4_pred_mode[x264_scan8[7] ];
- h->mb.intra4x4_pred_mode[i_mb_xy][6] = h->mb.cache.intra4x4_pred_mode[x264_scan8[13] ];
+ *(uint32_t*)&intra4x4_pred_mode[0] = *(uint32_t*)&h->mb.cache.intra4x4_pred_mode[x264_scan8[10] ];
+ *(uint32_t*)&intra4x4_pred_mode[4] = pack8to32(h->mb.cache.intra4x4_pred_mode[x264_scan8[5] ],
+ h->mb.cache.intra4x4_pred_mode[x264_scan8[7] ],
+ h->mb.cache.intra4x4_pred_mode[x264_scan8[13] ], 0);
}
else
- {
- h->mb.intra4x4_pred_mode[i_mb_xy][0] =
- h->mb.intra4x4_pred_mode[i_mb_xy][1] =
- h->mb.intra4x4_pred_mode[i_mb_xy][2] =
- h->mb.intra4x4_pred_mode[i_mb_xy][3] =
- h->mb.intra4x4_pred_mode[i_mb_xy][4] =
- h->mb.intra4x4_pred_mode[i_mb_xy][5] =
- h->mb.intra4x4_pred_mode[i_mb_xy][6] = I_PRED_4x4_DC;
- }
+ *(uint64_t*)intra4x4_pred_mode = I_PRED_4x4_DC * 0x0101010101010101ULL;
if( i_mb_type == I_PCM )
{
+ h->mb.qp[i_mb_xy] = 0;
+ h->mb.i_last_dqp = 0;
+ h->mb.i_cbp_chroma = 2;
+ h->mb.i_cbp_luma = 0xf;
h->mb.cbp[i_mb_xy] = 0x72f; /* all set */
+ h->mb.b_transform_8x8 = 0;
for( i = 0; i < 16 + 2*4; i++ )
- {
- h->mb.non_zero_count[i_mb_xy][i] = 16;
- }
+ non_zero_count[i] = 16;
}
else
{
/* save non zero count */
- for( i = 0; i < 16 + 2*4; i++ )
- {
- h->mb.non_zero_count[i_mb_xy][i] = h->mb.cache.non_zero_count[x264_scan8[i]];
- }
+ *(uint32_t*)&non_zero_count[0*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+0*8];
+ *(uint32_t*)&non_zero_count[1*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+1*8];
+ *(uint32_t*)&non_zero_count[2*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+2*8];
+ *(uint32_t*)&non_zero_count[3*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+3*8];
+ *(uint16_t*)&non_zero_count[16+0*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+0*2]-1] >> 8;
+ *(uint16_t*)&non_zero_count[16+1*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+1*2]-1] >> 8;
+ *(uint16_t*)&non_zero_count[16+2*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+2*2]-1] >> 8;
+ *(uint16_t*)&non_zero_count[16+3*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+3*2]-1] >> 8;
+
+ if( h->mb.i_type != I_16x16 && h->mb.i_cbp_luma == 0 && h->mb.i_cbp_chroma == 0 )
+ h->mb.i_qp = h->mb.i_last_qp;
+ h->mb.qp[i_mb_xy] = h->mb.i_qp;
+ h->mb.i_last_dqp = h->mb.i_qp - h->mb.i_last_qp;
+ h->mb.i_last_qp = h->mb.i_qp;
}
if( h->mb.i_cbp_luma == 0 && h->mb.i_type != I_8x8 )
h->mb.b_transform_8x8 = 0;
h->mb.mb_transform_size[i_mb_xy] = h->mb.b_transform_8x8;
- if( !IS_INTRA( i_mb_type ) )
+ if( h->sh.i_type != SLICE_TYPE_I )
{
- int i_list;
- for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
+ if( !IS_INTRA( i_mb_type ) )
{
- int y,x;
-
- h->mb.ref[i_list][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[i_list][x264_scan8[0]];
- h->mb.ref[i_list][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[i_list][x264_scan8[4]];
- h->mb.ref[i_list][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[i_list][x264_scan8[8]];
- h->mb.ref[i_list][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[i_list][x264_scan8[12]];
-
+ h->mb.ref[0][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[0][x264_scan8[0]];
+ h->mb.ref[0][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[0][x264_scan8[4]];
+ h->mb.ref[0][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[0][x264_scan8[8]];
+ h->mb.ref[0][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[0][x264_scan8[12]];
for( y = 0; y < 4; y++ )
{
- for( x = 0; x < 4; x++ )
+ *(uint64_t*)h->mb.mv[0][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mv[0][x264_scan8[0]+8*y+0];
+ *(uint64_t*)h->mb.mv[0][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mv[0][x264_scan8[0]+8*y+2];
+ }
+ if( h->sh.i_type == SLICE_TYPE_B )
+ {
+ h->mb.ref[1][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[1][x264_scan8[0]];
+ h->mb.ref[1][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[1][x264_scan8[4]];
+ h->mb.ref[1][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[1][x264_scan8[8]];
+ h->mb.ref[1][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[1][x264_scan8[12]];
+ for( y = 0; y < 4; y++ )
{
- h->mb.mv[i_list][i_mb_4x4+x+y*s4x4][0] = h->mb.cache.mv[i_list][x264_scan8[0]+x+8*y][0];
- h->mb.mv[i_list][i_mb_4x4+x+y*s4x4][1] = h->mb.cache.mv[i_list][x264_scan8[0]+x+8*y][1];
+ *(uint64_t*)h->mb.mv[1][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mv[1][x264_scan8[0]+8*y+0];
+ *(uint64_t*)h->mb.mv[1][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mv[1][x264_scan8[0]+8*y+2];
}
}
}
- }
- else
- {
- int i_list;
- for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
+ else
{
- int y,x;
-
- h->mb.ref[i_list][i_mb_8x8+0+0*s8x8] =
- h->mb.ref[i_list][i_mb_8x8+1+0*s8x8] =
- h->mb.ref[i_list][i_mb_8x8+0+1*s8x8] =
- h->mb.ref[i_list][i_mb_8x8+1+1*s8x8] = -1;
-
- for( y = 0; y < 4; y++ )
+ int i_list;
+ for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
{
- for( x = 0; x < 4; x++ )
+ *(uint16_t*)&h->mb.ref[i_list][i_mb_8x8+0*s8x8] = (uint8_t)(-1) * 0x0101;
+ *(uint16_t*)&h->mb.ref[i_list][i_mb_8x8+1*s8x8] = (uint8_t)(-1) * 0x0101;
+ for( y = 0; y < 4; y++ )
{
- h->mb.mv[i_list][i_mb_4x4+x+y*s4x4][0] = 0;
- h->mb.mv[i_list][i_mb_4x4+x+y*s4x4][1] = 0;
+ *(uint64_t*)h->mb.mv[i_list][i_mb_4x4+y*s4x4+0] = 0;
+ *(uint64_t*)h->mb.mv[i_list][i_mb_4x4+y*s4x4+2] = 0;
}
}
}
if( h->param.b_cabac )
{
- if( i_mb_type == I_4x4 || i_mb_type == I_16x16 )
+ if( IS_INTRA(i_mb_type) && i_mb_type != I_PCM )
h->mb.chroma_pred_mode[i_mb_xy] = x264_mb_pred_mode8x8c_fix[ h->mb.i_chroma_pred_mode ];
else
h->mb.chroma_pred_mode[i_mb_xy] = I_PRED_CHROMA_DC;
if( !IS_INTRA( i_mb_type ) && !IS_SKIP( i_mb_type ) && !IS_DIRECT( i_mb_type ) )
{
- int i_list;
- for( i_list = 0; i_list < 2; i_list++ )
+ for( y = 0; y < 4; y++ )
{
- const int s4x4 = 4 * h->mb.i_mb_stride;
- int y,x;
+ *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mvd[0][x264_scan8[0]+8*y+0];
+ *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mvd[0][x264_scan8[0]+8*y+2];
+ }
+ if( h->sh.i_type == SLICE_TYPE_B )
for( y = 0; y < 4; y++ )
{
- for( x = 0; x < 4; x++ )
- {
- h->mb.mvd[i_list][i_mb_4x4+x+y*s4x4][0] = h->mb.cache.mvd[i_list][x264_scan8[0]+x+8*y][0];
- h->mb.mvd[i_list][i_mb_4x4+x+y*s4x4][1] = h->mb.cache.mvd[i_list][x264_scan8[0]+x+8*y][1];
- }
+ *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mvd[1][x264_scan8[0]+8*y+0];
+ *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mvd[1][x264_scan8[0]+8*y+2];
}
- }
}
else
{
- int i_list;
- for( i_list = 0; i_list < 2; i_list++ )
+ for( y = 0; y < 4; y++ )
{
- const int s4x4 = 4 * h->mb.i_mb_stride;
- int y,x;
+ *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+0] = 0;
+ *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+2] = 0;
+ }
+ if( h->sh.i_type == SLICE_TYPE_B )
for( y = 0; y < 4; y++ )
{
- for( x = 0; x < 4; x++ )
- {
- h->mb.mvd[i_list][i_mb_4x4+x+y*s4x4][0] = 0;
- h->mb.mvd[i_list][i_mb_4x4+x+y*s4x4][1] = 0;
- }
+ *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+0] = 0;
+ *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+2] = 0;
}
- }
}
+
if( h->sh.i_type == SLICE_TYPE_B )
{
if( i_mb_type == B_SKIP || i_mb_type == B_DIRECT )
h->mb.skipbp[i_mb_xy] = 0xf;
else if( i_mb_type == B_8x8 )
{
- int skipbp = 0;
- for( i = 0; i < 4; i++ )
- skipbp |= ( h->mb.i_sub_partition[i] == D_DIRECT_8x8 ) << i;
+ int skipbp = ( h->mb.i_sub_partition[0] == D_DIRECT_8x8 ) << 0;
+ skipbp |= ( h->mb.i_sub_partition[1] == D_DIRECT_8x8 ) << 1;
+ skipbp |= ( h->mb.i_sub_partition[2] == D_DIRECT_8x8 ) << 2;
+ skipbp |= ( h->mb.i_sub_partition[3] == D_DIRECT_8x8 ) << 3;
h->mb.skipbp[i_mb_xy] = skipbp;
}
else
if( h->param.analyse.b_weighted_bipred
&& dist_scale_factor >= -64
&& dist_scale_factor <= 128 )
+ {
h->mb.bipred_weight[i_ref0][i_ref1] = 64 - dist_scale_factor;
+ // ssse3 implementation of biweight doesn't support the extrema.
+ // if we ever generate them, we'll have to drop that optimization.
+ assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
+ }
else
h->mb.bipred_weight[i_ref0][i_ref1] = 32;
}
}
+ if( h->sh.b_mbaff )
+ {
+ for( i_ref0 = 2*h->i_ref0-1; i_ref0 >= 0; i_ref0-- )
+ for( i_ref1 = 2*h->i_ref1-1; i_ref1 >= 0; i_ref1-- )
+ h->mb.bipred_weight[i_ref0][i_ref1] = h->mb.bipred_weight[i_ref0>>1][i_ref1>>1];
+ }
}
projects / x264 / blobdiff