/*****************************************************************************
* frame.c: h264 encoder library
*****************************************************************************
- * Copyright (C) 2003 Laurent Aimar
- * $Id: frame.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>
*
* 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"
+#define ALIGN(x,a) (((x)+((a)-1))&~((a)-1))
+
x264_frame_t *x264_frame_new( x264_t *h )
{
x264_frame_t *frame = x264_malloc( sizeof(x264_frame_t) );
int i, j;
int i_mb_count = h->mb.i_mb_count;
- int i_stride;
- int i_lines;
+ int i_stride, i_width, i_lines;
+ int i_padv = PADV << h->param.b_interlaced;
+ int luma_plane_size;
+ int align = h->param.cpu&X264_CPU_CACHELINE_64 ? 64 : h->param.cpu&X264_CPU_CACHELINE_32 ? 32 : 16;
if( !frame ) return NULL;
memset( frame, 0, sizeof(x264_frame_t) );
/* allocate frame data (+64 for extra data for me) */
- i_stride = ( ( h->param.i_width + 15 )&0xfffff0 )+ 64;
- i_lines = ( ( h->param.i_height + 15 )&0xfffff0 );
+ i_width = ALIGN( h->param.i_width, 16 );
+ i_stride = ALIGN( i_width + 2*PADH, align );
+ i_lines = ALIGN( h->param.i_height, 16<<h->param.b_interlaced );
frame->i_plane = 3;
for( i = 0; i < 3; i++ )
{
- int i_divh = 1;
- int i_divw = 1;
- if( i > 0 )
- {
- if( h->param.i_csp == X264_CSP_I420 )
- i_divh = i_divw = 2;
- else if( h->param.i_csp == X264_CSP_I422 )
- i_divw = 2;
- }
- frame->i_stride[i] = i_stride / i_divw;
- frame->i_lines[i] = i_lines / i_divh;
- CHECKED_MALLOC( frame->buffer[i],
- frame->i_stride[i] * ( frame->i_lines[i] + 64 / i_divh ) );
- frame->plane[i] = ((uint8_t*)frame->buffer[i]) +
- frame->i_stride[i] * 32 / i_divh + 32 / i_divw;
+ frame->i_stride[i] = i_stride >> !!i;
+ frame->i_width[i] = i_width >> !!i;
+ frame->i_lines[i] = i_lines >> !!i;
}
- frame->i_stride[3] = 0;
- frame->i_lines[3] = 0;
- frame->buffer[3] = NULL;
- frame->plane[3] = NULL;
- frame->filtered[0] = frame->plane[0];
- for( i = 0; i < 3; i++ )
+ luma_plane_size = (frame->i_stride[0] * ( frame->i_lines[0] + 2*i_padv ));
+ for( i = 1; i < 3; i++ )
{
- CHECKED_MALLOC( frame->buffer[4+i],
- frame->i_stride[0] * ( frame->i_lines[0] + 64 ) );
- frame->filtered[i+1] = ((uint8_t*)frame->buffer[4+i]) +
- frame->i_stride[0] * 32 + 32;
+ CHECKED_MALLOC( frame->buffer[i], luma_plane_size/4 );
+ frame->plane[i] = frame->buffer[i] + (frame->i_stride[i] * i_padv + PADH)/2;
}
+ /* all 4 luma planes allocated together, since the cacheline split code
+ * requires them to be in-phase wrt cacheline alignment. */
+ CHECKED_MALLOC( frame->buffer[0], 4*luma_plane_size);
+ for( i = 0; i < 4; i++ )
+ frame->filtered[i] = frame->buffer[0] + i*luma_plane_size + frame->i_stride[0] * i_padv + PADH;
+ frame->plane[0] = frame->filtered[0];
if( h->frames.b_have_lowres )
{
- frame->i_stride_lowres = frame->i_stride[0]/2 + 32;
+ frame->i_width_lowres = frame->i_width[0]/2;
+ frame->i_stride_lowres = ALIGN( frame->i_width_lowres + 2*PADH, align );
frame->i_lines_lowres = frame->i_lines[0]/2;
+
+ luma_plane_size = frame->i_stride_lowres * ( frame->i_lines[0]/2 + 2*i_padv );
+
+ CHECKED_MALLOC( frame->buffer_lowres[0], 4 * luma_plane_size );
for( i = 0; i < 4; i++ )
- {
- CHECKED_MALLOC( frame->buffer[7+i],
- frame->i_stride_lowres * ( frame->i_lines[0]/2 + 64 ) );
- frame->lowres[i] = ((uint8_t*)frame->buffer[7+i]) +
- frame->i_stride_lowres * 32 + 32;
- }
+ frame->lowres[i] = frame->buffer_lowres[0] + (frame->i_stride_lowres * i_padv + PADH) + i * luma_plane_size;
}
- if( h->param.analyse.i_me_method == X264_ME_ESA )
+ if( h->param.analyse.i_me_method >= X264_ME_ESA )
{
- CHECKED_MALLOC( frame->buffer[11],
- frame->i_stride[0] * (frame->i_lines[0] + 64) * sizeof(uint16_t) );
- frame->integral = (uint16_t*)frame->buffer[11] + frame->i_stride[0] * 32 + 32;
+ CHECKED_MALLOC( frame->buffer[3],
+ 2 * frame->i_stride[0] * (frame->i_lines[0] + 2*i_padv) * sizeof(uint16_t) );
+ frame->integral = (uint16_t*)frame->buffer[3] + frame->i_stride[0] * i_padv + PADH;
}
frame->i_poc = -1;
frame->i_pts = -1;
frame->i_frame = -1;
frame->i_frame_num = -1;
+ frame->i_lines_completed = -1;
CHECKED_MALLOC( frame->mb_type, i_mb_count * sizeof(int8_t));
CHECKED_MALLOC( frame->mv[0], 2*16 * i_mb_count * sizeof(int16_t) );
for( j = 0; j < h->param.i_bframe + 2; j++ )
CHECKED_MALLOC( frame->i_row_satds[i][j], i_lines/16 * sizeof(int) );
+ x264_pthread_mutex_init( &frame->mutex, NULL );
+ x264_pthread_cond_init( &frame->cv, NULL );
+
return frame;
fail:
void x264_frame_delete( x264_frame_t *frame )
{
int i, j;
- for( i = 0; i < 12; i++ )
+ for( i = 0; i < 4; i++ )
x264_free( frame->buffer[i] );
+ for( i = 0; i < 4; i++ )
+ x264_free( frame->buffer_lowres[i] );
for( i = 0; i < X264_BFRAME_MAX+2; i++ )
for( j = 0; j < X264_BFRAME_MAX+2; j++ )
x264_free( frame->i_row_satds[i][j] );
x264_free( frame->mv[1] );
x264_free( frame->ref[0] );
x264_free( frame->ref[1] );
+ x264_pthread_mutex_destroy( &frame->mutex );
+ x264_pthread_cond_destroy( &frame->cv );
x264_free( frame );
}
-void x264_frame_copy_picture( x264_t *h, x264_frame_t *dst, x264_picture_t *src )
+int x264_frame_copy_picture( x264_t *h, x264_frame_t *dst, x264_picture_t *src )
{
+ int i_csp = src->img.i_csp & X264_CSP_MASK;
+ int i;
+ if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
+ {
+ x264_log( h, X264_LOG_ERROR, "Arg invalid CSP\n" );
+ return -1;
+ }
+
dst->i_type = src->i_type;
dst->i_qpplus1 = src->i_qpplus1;
dst->i_pts = src->i_pts;
- switch( src->img.i_csp & X264_CSP_MASK )
+ for( i=0; i<3; i++ )
{
- case X264_CSP_I420:
- h->csp.i420( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_YV12:
- h->csp.yv12( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_I422:
- h->csp.i422( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_I444:
- h->csp.i444( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_YUYV:
- h->csp.yuyv( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_RGB:
- h->csp.rgb( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_BGR:
- h->csp.bgr( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
- case X264_CSP_BGRA:
- h->csp.bgra( dst, &src->img, h->param.i_width, h->param.i_height );
- break;
-
- default:
- x264_log( h, X264_LOG_ERROR, "Arg invalid CSP\n" );
- break;
+ int s = (i_csp == X264_CSP_YV12 && i) ? i^3 : i;
+ uint8_t *plane = src->img.plane[s];
+ int stride = src->img.i_stride[s];
+ int width = h->param.i_width >> !!i;
+ int height = h->param.i_height >> !!i;
+ if( src->img.i_csp & X264_CSP_VFLIP )
+ {
+ plane += (height-1)*stride;
+ stride = -stride;
+ }
+ h->mc.plane_copy( dst->plane[i], dst->i_stride[i], plane, stride, width, height );
}
+ return 0;
}
-static void plane_expand_border( uint8_t *pix, int i_stride, int i_height, int i_pad )
+static void plane_expand_border( uint8_t *pix, int i_stride, int i_width, int i_height, int i_padh, int i_padv, int b_pad_top, int b_pad_bottom )
{
#define PPIXEL(x, y) ( pix + (x) + (y)*i_stride )
- const int i_width = i_stride - 2*i_pad;
int y;
-
for( y = 0; y < i_height; y++ )
{
/* left band */
- memset( PPIXEL(-i_pad, y), PPIXEL(0, y)[0], i_pad );
+ memset( PPIXEL(-i_padh, y), PPIXEL(0, y)[0], i_padh );
/* right band */
- memset( PPIXEL(i_width, y), PPIXEL(i_width-1, y)[0], i_pad );
+ memset( PPIXEL(i_width, y), PPIXEL(i_width-1, y)[0], i_padh );
}
/* upper band */
- for( y = 0; y < i_pad; y++ )
- memcpy( PPIXEL(-i_pad, -y-1), PPIXEL(-i_pad, 0), i_stride );
+ if( b_pad_top )
+ for( y = 0; y < i_padv; y++ )
+ memcpy( PPIXEL(-i_padh, -y-1), PPIXEL(-i_padh, 0), i_width+2*i_padh );
/* lower band */
- for( y = 0; y < i_pad; y++ )
- memcpy( PPIXEL(-i_pad, i_height+y), PPIXEL(-i_pad, i_height-1), i_stride );
+ if( b_pad_bottom )
+ for( y = 0; y < i_padv; y++ )
+ memcpy( PPIXEL(-i_padh, i_height+y), PPIXEL(-i_padh, i_height-1), i_width+2*i_padh );
#undef PPIXEL
}
-void x264_frame_expand_border( x264_frame_t *frame )
+void x264_frame_expand_border( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
{
int i;
+ int b_start = !mb_y;
+ if( mb_y & h->sh.b_mbaff )
+ return;
for( i = 0; i < frame->i_plane; i++ )
{
- int i_pad = i ? 16 : 32;
- plane_expand_border( frame->plane[i], frame->i_stride[i], frame->i_lines[i], i_pad );
+ int stride = frame->i_stride[i];
+ int width = 16*h->sps->i_mb_width >> !!i;
+ int height = (b_end ? 16*(h->sps->i_mb_height - mb_y) >> h->sh.b_mbaff : 16) >> !!i;
+ int padh = PADH >> !!i;
+ int padv = PADV >> !!i;
+ // buffer: 2 chroma, 3 luma (rounded to 4) because deblocking goes beyond the top of the mb
+ uint8_t *pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> !!i);
+ if( b_end && !b_start )
+ height += 4 >> (!!i + h->sh.b_mbaff);
+ if( h->sh.b_mbaff )
+ {
+ plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end );
+ plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end );
+ }
+ else
+ {
+ plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end );
+ }
}
}
-void x264_frame_expand_border_filtered( x264_frame_t *frame )
+void x264_frame_expand_border_filtered( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
{
- /* during filtering, 8 extra pixels were filtered on each edge.
+ /* during filtering, 8 extra pixels were filtered on each edge,
+ * but up to 3 of the horizontal ones may be wrong.
we want to expand border from the last filtered pixel */
+ int b_start = !mb_y;
+ int stride = frame->i_stride[0];
+ int width = 16*h->sps->i_mb_width + 8;
+ int height = b_end ? (16*(h->sps->i_mb_height - mb_y) >> h->sh.b_mbaff) + 16 : 16;
+ int padh = PADH - 4;
+ int padv = PADV - 8;
int i;
for( i = 1; i < 4; i++ )
- plane_expand_border( frame->filtered[i] - 8*frame->i_stride[0] - 8, frame->i_stride[0], frame->i_lines[0]+2*8, 24 );
+ {
+ // buffer: 8 luma, to match the hpel filter
+ uint8_t *pix = frame->filtered[i] + (16*mb_y - (8 << h->sh.b_mbaff)) * stride - 4;
+ if( h->sh.b_mbaff )
+ {
+ plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end );
+ plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end );
+ }
+ else
+ {
+ plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end );
+ }
+ }
}
void x264_frame_expand_border_lowres( x264_frame_t *frame )
{
int i;
for( i = 0; i < 4; i++ )
- plane_expand_border( frame->lowres[i], frame->i_stride_lowres, frame->i_lines_lowres, 32 );
+ plane_expand_border( frame->lowres[i], frame->i_stride_lowres, frame->i_stride_lowres - 2*PADH, frame->i_lines_lowres, PADH, PADV, 1, 1 );
}
void x264_frame_expand_border_mod16( x264_t *h, x264_frame_t *frame )
}
if( i_pady )
{
+ //FIXME interlace? or just let it pad using the wrong field
for( y = i_height; y < i_height + i_pady; y++ )
memcpy( &frame->plane[i][y*frame->i_stride[i]],
&frame->plane[i][(i_height-1)*frame->i_stride[i]],
}
+/* cavlc + 8x8 transform stores nnz per 16 coeffs for the purpose of
+ * entropy coding, but per 64 coeffs for the purpose of deblocking */
+void munge_cavlc_nnz_row( x264_t *h, int mb_y, uint8_t (*buf)[16] )
+{
+ uint32_t (*src)[6] = (uint32_t(*)[6])h->mb.non_zero_count + mb_y * h->sps->i_mb_width;
+ int8_t *transform = h->mb.mb_transform_size + mb_y * h->sps->i_mb_width;
+ int x, nnz;
+ for( x=0; x<h->sps->i_mb_width; x++ )
+ {
+ memcpy( buf+x, src+x, 16 );
+ if( transform[x] )
+ {
+ nnz = src[x][0] | src[x][1];
+ src[x][0] = src[x][1] = ((uint16_t)nnz ? 0x0101 : 0) + (nnz>>16 ? 0x01010000 : 0);
+ nnz = src[x][2] | src[x][3];
+ src[x][2] = src[x][3] = ((uint16_t)nnz ? 0x0101 : 0) + (nnz>>16 ? 0x01010000 : 0);
+ }
+ }
+}
+
+static void restore_cavlc_nnz_row( x264_t *h, int mb_y, uint8_t (*buf)[16] )
+{
+ uint8_t (*dst)[24] = h->mb.non_zero_count + mb_y * h->sps->i_mb_width;
+ int x;
+ for( x=0; x<h->sps->i_mb_width; x++ )
+ memcpy( dst+x, buf+x, 16 );
+}
+
+static void munge_cavlc_nnz( x264_t *h, int mb_y, uint8_t (*buf)[16], void (*func)(x264_t*, int, uint8_t (*)[16]) )
+{
+ func( h, mb_y, buf );
+ if( mb_y > 0 )
+ func( h, mb_y-1, buf + h->sps->i_mb_width );
+ if( h->sh.b_mbaff )
+ {
+ func( h, mb_y+1, buf + h->sps->i_mb_width * 2 );
+ if( mb_y > 0 )
+ func( h, mb_y-2, buf + h->sps->i_mb_width * 3 );
+ }
+}
+
+
/* Deblocking filter */
static const int i_alpha_table[52] =
}
}
-void x264_frame_deblocking_filter( x264_t *h, int i_slice_type )
+void x264_frame_deblock_row( x264_t *h, int mb_y )
{
const int s8x8 = 2 * h->mb.i_mb_stride;
const int s4x4 = 4 * h->mb.i_mb_stride;
+ const int b_interlaced = h->sh.b_mbaff;
+ const int mvy_limit = 4 >> b_interlaced;
+ int mb_x;
+
+ int i_stride2[3] = { h->fdec->i_stride[0] << b_interlaced,
+ h->fdec->i_stride[1] << b_interlaced,
+ h->fdec->i_stride[2] << b_interlaced };
- int mb_y, mb_x;
+ if( !h->pps->b_cabac && h->pps->b_transform_8x8_mode )
+ munge_cavlc_nnz( h, mb_y, h->mb.nnz_backup, munge_cavlc_nnz_row );
- for( mb_y = 0, mb_x = 0; mb_y < h->sps->i_mb_height; )
+ for( mb_x = 0; mb_x < h->sps->i_mb_width; )
{
const int mb_xy = mb_y * h->mb.i_mb_stride + mb_x;
const int mb_8x8 = 2 * s8x8 * mb_y + 2 * mb_x;
const int mb_4x4 = 4 * s4x4 * mb_y + 4 * mb_x;
const int b_8x8_transform = h->mb.mb_transform_size[mb_xy];
const int i_edge_end = (h->mb.type[mb_xy] == P_SKIP) ? 1 : 4;
- int i_edge, i_dir;
+ int i_edge;
- /* cavlc + 8x8 transform stores nnz per 16 coeffs for the purpose of
- * entropy coding, but per 64 coeffs for the purpose of deblocking */
- if( !h->param.b_cabac && b_8x8_transform )
+ int i_pix_y[3] = { 16*mb_y*h->fdec->i_stride[0] + 16*mb_x,
+ 8*mb_y*h->fdec->i_stride[1] + 8*mb_x,
+ 8*mb_y*h->fdec->i_stride[2] + 8*mb_x };
+ if( b_interlaced && (mb_y&1) )
{
- uint32_t *nnz = (uint32_t*)h->mb.non_zero_count[mb_xy];
- if( nnz[0] ) nnz[0] = 0x01010101;
- if( nnz[1] ) nnz[1] = 0x01010101;
- if( nnz[2] ) nnz[2] = 0x01010101;
- if( nnz[3] ) nnz[3] = 0x01010101;
+ i_pix_y[0] -= 15*h->fdec->i_stride[0];
+ i_pix_y[1] -= 7*h->fdec->i_stride[1];
+ i_pix_y[2] -= 7*h->fdec->i_stride[2];
}
+ x264_prefetch_fenc( h, h->fdec, mb_x, mb_y );
+
/* i_dir == 0 -> vertical edge
* i_dir == 1 -> horizontal edge */
- for( i_dir = 0; i_dir < 2; i_dir++ )
- {
- int i_start = (i_dir ? mb_y : mb_x) ? 0 : 1;
- int i_qp, i_qpn;
-
- for( i_edge = i_start; i_edge < i_edge_end; i_edge++ )
- {
- int mbn_xy, mbn_8x8, mbn_4x4;
- int bS[4]; /* filtering strength */
-
- if( b_8x8_transform && (i_edge&1) )
- continue;
- mbn_xy = i_edge > 0 ? mb_xy : ( i_dir == 0 ? mb_xy - 1 : mb_xy - h->mb.i_mb_stride );
- mbn_8x8 = i_edge > 0 ? mb_8x8 : ( i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8 );
- mbn_4x4 = i_edge > 0 ? mb_4x4 : ( i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4 );
+ #define deblock_dir(i_dir)\
+ {\
+ int i_start = (i_dir ? (mb_y <= b_interlaced) : (mb_x == 0));\
+ int i_qp, i_qpn;\
+ for( i_edge = i_start; i_edge < i_edge_end; i_edge++ )\
+ {\
+ int mbn_xy, mbn_8x8, mbn_4x4;\
+ int bS[4]; /* filtering strength */\
+ if( b_8x8_transform && (i_edge&1) )\
+ continue;\
+ mbn_xy = i_edge > 0 ? mb_xy : ( i_dir == 0 ? mb_xy - 1 : mb_xy - h->mb.i_mb_stride );\
+ mbn_8x8 = i_edge > 0 ? mb_8x8 : ( i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8 );\
+ mbn_4x4 = i_edge > 0 ? mb_4x4 : ( i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4 );\
+ if( b_interlaced && i_edge == 0 && i_dir == 1 )\
+ {\
+ mbn_xy -= h->mb.i_mb_stride;\
+ mbn_8x8 -= 2 * s8x8;\
+ mbn_4x4 -= 4 * s4x4;\
+ }\
+ /* *** Get bS for each 4px for the current edge *** */\
+ if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy] ) )\
+ bS[0] = bS[1] = bS[2] = bS[3] = ( i_edge == 0 && !(b_interlaced && i_dir) ? 4 : 3 );\
+ else\
+ {\
+ int i;\
+ for( i = 0; i < 4; i++ )\
+ {\
+ int x = i_dir == 0 ? i_edge : i;\
+ int y = i_dir == 0 ? i : i_edge;\
+ int xn = (x - (i_dir == 0 ? 1 : 0 ))&0x03;\
+ int yn = (y - (i_dir == 0 ? 0 : 1 ))&0x03;\
+ if( h->mb.non_zero_count[mb_xy][x+y*4] != 0 ||\
+ h->mb.non_zero_count[mbn_xy][xn+yn*4] != 0 )\
+ {\
+ bS[i] = 2;\
+ }\
+ else\
+ {\
+ /* FIXME: A given frame may occupy more than one position in\
+ * the reference list. So we should compare the frame numbers,\
+ * not the indices in the ref list.\
+ * No harm yet, as we don't generate that case.*/\
+ int i8p= mb_8x8+(x/2)+(y/2)*s8x8;\
+ int i8q= mbn_8x8+(xn/2)+(yn/2)*s8x8;\
+ int i4p= mb_4x4+x+y*s4x4;\
+ int i4q= mbn_4x4+xn+yn*s4x4;\
+ int l;\
+ bS[i] = 0;\
+ for( l = 0; l < 1 + (h->sh.i_type == SLICE_TYPE_B); l++ )\
+ {\
+ if( h->mb.ref[l][i8p] != h->mb.ref[l][i8q] ||\
+ abs( h->mb.mv[l][i4p][0] - h->mb.mv[l][i4q][0] ) >= 4 ||\
+ abs( h->mb.mv[l][i4p][1] - h->mb.mv[l][i4q][1] ) >= mvy_limit )\
+ {\
+ bS[i] = 1;\
+ break;\
+ }\
+ }\
+ }\
+ }\
+ }\
+ /* *** filter *** */\
+ /* Y plane */\
+ i_qp = h->mb.qp[mb_xy];\
+ i_qpn= h->mb.qp[mbn_xy];\
+ if( i_dir == 0 )\
+ {\
+ /* vertical edge */\
+ deblock_edge( h, &h->fdec->plane[0][i_pix_y[0] + 4*i_edge],\
+ i_stride2[0], bS, (i_qp+i_qpn+1) >> 1, 0,\
+ h->loopf.deblock_h_luma, h->loopf.deblock_h_luma_intra );\
+ if( !(i_edge & 1) )\
+ {\
+ /* U/V planes */\
+ int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +\
+ i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;\
+ deblock_edge( h, &h->fdec->plane[1][i_pix_y[1] + 2*i_edge],\
+ i_stride2[1], bS, i_qpc, 1,\
+ h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra );\
+ deblock_edge( h, &h->fdec->plane[2][i_pix_y[2] + 2*i_edge],\
+ i_stride2[2], bS, i_qpc, 1,\
+ h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra );\
+ }\
+ }\
+ else\
+ {\
+ /* horizontal edge */\
+ deblock_edge( h, &h->fdec->plane[0][i_pix_y[0] + 4*i_edge*i_stride2[0]],\
+ i_stride2[0], bS, (i_qp+i_qpn+1) >> 1, 0,\
+ h->loopf.deblock_v_luma, h->loopf.deblock_v_luma_intra );\
+ /* U/V planes */\
+ if( !(i_edge & 1) )\
+ {\
+ int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +\
+ i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;\
+ deblock_edge( h, &h->fdec->plane[1][i_pix_y[1] + 2*i_edge*i_stride2[1]],\
+ i_stride2[1], bS, i_qpc, 1,\
+ h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra );\
+ deblock_edge( h, &h->fdec->plane[2][i_pix_y[2] + 2*i_edge*i_stride2[2]],\
+ i_stride2[2], bS, i_qpc, 1,\
+ h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra );\
+ }\
+ }\
+ }\
+ }
- /* *** Get bS for each 4px for the current edge *** */
- if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy] ) )
- {
- bS[0] = bS[1] = bS[2] = bS[3] = ( i_edge == 0 ? 4 : 3 );
- }
- else
- {
- int i;
- for( i = 0; i < 4; i++ )
- {
- int x = i_dir == 0 ? i_edge : i;
- int y = i_dir == 0 ? i : i_edge;
- int xn = (x - (i_dir == 0 ? 1 : 0 ))&0x03;
- int yn = (y - (i_dir == 0 ? 0 : 1 ))&0x03;
-
- if( h->mb.non_zero_count[mb_xy][block_idx_xy[x][y]] != 0 ||
- h->mb.non_zero_count[mbn_xy][block_idx_xy[xn][yn]] != 0 )
- {
- bS[i] = 2;
- }
- else
- {
- /* FIXME: A given frame may occupy more than one position in
- * the reference list. So we should compare the frame numbers,
- * not the indices in the ref list.
- * No harm yet, as we don't generate that case.*/
-
- int i8p= mb_8x8+(x/2)+(y/2)*s8x8;
- int i8q= mbn_8x8+(xn/2)+(yn/2)*s8x8;
- int i4p= mb_4x4+x+y*s4x4;
- int i4q= mbn_4x4+xn+yn*s4x4;
- int l;
-
- bS[i] = 0;
-
- for( l = 0; l < 1 + (i_slice_type == SLICE_TYPE_B); l++ )
- {
- if( h->mb.ref[l][i8p] != h->mb.ref[l][i8q] ||
- abs( h->mb.mv[l][i4p][0] - h->mb.mv[l][i4q][0] ) >= 4 ||
- abs( h->mb.mv[l][i4p][1] - h->mb.mv[l][i4q][1] ) >= 4 )
- {
- bS[i] = 1;
- break;
- }
- }
- }
- }
- }
+ deblock_dir(0);
+ deblock_dir(1);
- /* *** filter *** */
- /* Y plane */
- i_qp = h->mb.qp[mb_xy];
- i_qpn= h->mb.qp[mbn_xy];
+ /* next mb */
+ if( !b_interlaced || (mb_y&1) )
+ mb_x++;
+ mb_y ^= b_interlaced;
+ }
- if( i_dir == 0 )
- {
- /* vertical edge */
- deblock_edge( h, &h->fdec->plane[0][16*mb_y * h->fdec->i_stride[0] + 16*mb_x + 4*i_edge],
- h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1, 0,
- h->loopf.deblock_h_luma, h->loopf.deblock_h_luma_intra );
- if( !(i_edge & 1) )
- {
- /* U/V planes */
- int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +
- i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;
- deblock_edge( h, &h->fdec->plane[1][8*(mb_y*h->fdec->i_stride[1]+mb_x)+2*i_edge],
- h->fdec->i_stride[1], bS, i_qpc, 1,
- h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra );
- deblock_edge( h, &h->fdec->plane[2][8*(mb_y*h->fdec->i_stride[2]+mb_x)+2*i_edge],
- h->fdec->i_stride[2], bS, i_qpc, 1,
- h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra );
- }
- }
- else
- {
- /* horizontal edge */
- deblock_edge( h, &h->fdec->plane[0][(16*mb_y + 4*i_edge) * h->fdec->i_stride[0] + 16*mb_x],
- h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1, 0,
- h->loopf.deblock_v_luma, h->loopf.deblock_v_luma_intra );
- /* U/V planes */
- if( !(i_edge & 1) )
- {
- int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +
- i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;
- deblock_edge( h, &h->fdec->plane[1][8*(mb_y*h->fdec->i_stride[1]+mb_x)+2*i_edge*h->fdec->i_stride[1]],
- h->fdec->i_stride[1], bS, i_qpc, 1,
- h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra );
- deblock_edge( h, &h->fdec->plane[2][8*(mb_y*h->fdec->i_stride[2]+mb_x)+2*i_edge*h->fdec->i_stride[2]],
- h->fdec->i_stride[2], bS, i_qpc, 1,
- h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra );
- }
- }
- }
- }
+ if( !h->pps->b_cabac && h->pps->b_transform_8x8_mode )
+ munge_cavlc_nnz( h, mb_y, h->mb.nnz_backup, restore_cavlc_nnz_row );
+}
- /* newt mb */
- mb_x++;
- if( mb_x >= h->sps->i_mb_width )
- {
- mb_x = 0;
- mb_y++;
- }
- }
+void x264_frame_deblock( x264_t *h )
+{
+ int mb_y;
+ for( mb_y = 0; mb_y < h->sps->i_mb_height; mb_y += 1 + h->sh.b_mbaff )
+ x264_frame_deblock_row( h, mb_y );
}
-#ifdef HAVE_MMXEXT
+#ifdef HAVE_MMX
void x264_deblock_v_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void x264_deblock_h_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void x264_deblock_v_chroma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta );
void x264_deblock_h_chroma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta );
-#ifdef ARCH_X86_64
void x264_deblock_v_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void x264_deblock_h_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
-#else
+void x264_deblock_v_luma_intra_sse2( uint8_t *pix, int stride, int alpha, int beta );
+void x264_deblock_h_luma_intra_sse2( uint8_t *pix, int stride, int alpha, int beta );
+#ifdef ARCH_X86
void x264_deblock_h_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void x264_deblock_v8_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
+void x264_deblock_h_luma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta );
+void x264_deblock_v8_luma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta );
void x264_deblock_v_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
{
x264_deblock_v8_luma_mmxext( pix, stride, alpha, beta, tc0 );
x264_deblock_v8_luma_mmxext( pix+8, stride, alpha, beta, tc0+2 );
}
+void x264_deblock_v_luma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta )
+{
+ x264_deblock_v8_luma_intra_mmxext( pix, stride, alpha, beta );
+ x264_deblock_v8_luma_intra_mmxext( pix+8, stride, alpha, beta );
+}
#endif
#endif
+#ifdef ARCH_PPC
+void x264_deblock_v_luma_altivec( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
+void x264_deblock_h_luma_altivec( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
+#endif // ARCH_PPC
+
void x264_deblock_init( int cpu, x264_deblock_function_t *pf )
{
pf->deblock_v_luma = deblock_v_luma_c;
pf->deblock_v_chroma_intra = deblock_v_chroma_intra_c;
pf->deblock_h_chroma_intra = deblock_h_chroma_intra_c;
-#ifdef HAVE_MMXEXT
+#ifdef HAVE_MMX
if( cpu&X264_CPU_MMXEXT )
{
pf->deblock_v_chroma = x264_deblock_v_chroma_mmxext;
pf->deblock_h_chroma = x264_deblock_h_chroma_mmxext;
pf->deblock_v_chroma_intra = x264_deblock_v_chroma_intra_mmxext;
pf->deblock_h_chroma_intra = x264_deblock_h_chroma_intra_mmxext;
-
-#ifdef ARCH_X86_64
- if( cpu&X264_CPU_SSE2 )
+#ifdef ARCH_X86
+ pf->deblock_v_luma = x264_deblock_v_luma_mmxext;
+ pf->deblock_h_luma = x264_deblock_h_luma_mmxext;
+ pf->deblock_v_luma_intra = x264_deblock_v_luma_intra_mmxext;
+ pf->deblock_h_luma_intra = x264_deblock_h_luma_intra_mmxext;
+#endif
+ if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_STACK_MOD4) )
{
pf->deblock_v_luma = x264_deblock_v_luma_sse2;
pf->deblock_h_luma = x264_deblock_h_luma_sse2;
+ pf->deblock_v_luma_intra = x264_deblock_v_luma_intra_sse2;
+ pf->deblock_h_luma_intra = x264_deblock_h_luma_intra_sse2;
}
-#else
- pf->deblock_v_luma = x264_deblock_v_luma_mmxext;
- pf->deblock_h_luma = x264_deblock_h_luma_mmxext;
-#endif
}
#endif
+
+#ifdef ARCH_PPC
+ if( cpu&X264_CPU_ALTIVEC )
+ {
+ pf->deblock_v_luma = x264_deblock_v_luma_altivec;
+ pf->deblock_h_luma = x264_deblock_h_luma_altivec;
+ }
+#endif // ARCH_PPC
}
+
+/* threading */
+void x264_frame_cond_broadcast( x264_frame_t *frame, int i_lines_completed )
+{
+ x264_pthread_mutex_lock( &frame->mutex );
+ frame->i_lines_completed = i_lines_completed;
+ x264_pthread_cond_broadcast( &frame->cv );
+ x264_pthread_mutex_unlock( &frame->mutex );
+}
+
+void x264_frame_cond_wait( x264_frame_t *frame, int i_lines_completed )
+{
+ x264_pthread_mutex_lock( &frame->mutex );
+ while( frame->i_lines_completed < i_lines_completed )
+ x264_pthread_cond_wait( &frame->cv, &frame->mutex );
+ x264_pthread_mutex_unlock( &frame->mutex );
+}
+
+/* list operators */
+
+void x264_frame_push( x264_frame_t **list, x264_frame_t *frame )
+{
+ int i = 0;
+ while( list[i] ) i++;
+ list[i] = frame;
+}
+
+x264_frame_t *x264_frame_pop( x264_frame_t **list )
+{
+ x264_frame_t *frame;
+ int i = 0;
+ assert( list[0] );
+ while( list[i+1] ) i++;
+ frame = list[i];
+ list[i] = NULL;
+ return frame;
+}
+
+void x264_frame_unshift( x264_frame_t **list, x264_frame_t *frame )
+{
+ int i = 0;
+ while( list[i] ) i++;
+ while( i-- )
+ list[i+1] = list[i];
+ list[0] = frame;
+}
+
+x264_frame_t *x264_frame_shift( x264_frame_t **list )
+{
+ x264_frame_t *frame = list[0];
+ int i;
+ for( i = 0; list[i]; i++ )
+ list[i] = list[i+1];
+ assert(frame);
+ return frame;
+}
+
+void x264_frame_push_unused( x264_t *h, x264_frame_t *frame )
+{
+ assert( frame->i_reference_count > 0 );
+ frame->i_reference_count--;
+ if( frame->i_reference_count == 0 )
+ x264_frame_push( h->frames.unused, frame );
+ assert( h->frames.unused[ sizeof(h->frames.unused) / sizeof(*h->frames.unused) - 1 ] == NULL );
+}
+
+x264_frame_t *x264_frame_pop_unused( x264_t *h )
+{
+ x264_frame_t *frame;
+ if( h->frames.unused[0] )
+ frame = x264_frame_pop( h->frames.unused );
+ else
+ frame = x264_frame_new( h );
+ assert( frame->i_reference_count == 0 );
+ frame->i_reference_count = 1;
+ return frame;
+}
+
+void x264_frame_sort( x264_frame_t **list, int b_dts )
+{
+ int i, b_ok;
+ do {
+ b_ok = 1;
+ for( i = 0; list[i+1]; i++ )
+ {
+ int dtype = list[i]->i_type - list[i+1]->i_type;
+ int dtime = list[i]->i_frame - list[i+1]->i_frame;
+ int swap = b_dts ? dtype > 0 || ( dtype == 0 && dtime > 0 )
+ : dtime > 0;
+ if( swap )
+ {
+ XCHG( x264_frame_t*, list[i], list[i+1] );
+ b_ok = 0;
+ }
+ }
+ } while( !b_ok );
+}
projects / x264 / blobdiff