/*
* Motion estimation
- * Copyright (c) 2000,2001 Fabrice Bellard.
+ * Copyright (c) 2000,2001 Fabrice Bellard
* Copyright (c) 2002-2004 Michael Niedermayer
*
+ * new motion estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
*
- * This library is free software; you can redistribute it and/or
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
*
- * This library is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
- *
- * new Motion Estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
- * @file motion_est.c
+ * @file
* Motion estimation.
*/
#include <stdlib.h>
#include <stdio.h>
#include <limits.h>
+
#include "avcodec.h"
-#include "dsputil.h"
+#include "mathops.h"
#include "mpegvideo.h"
#undef NDEBUG
#include <assert.h>
-#define SQ(a) ((a)*(a))
-
#define P_LEFT P[1]
#define P_TOP P[2]
#define P_TOPRIGHT P[3]
#define P_MEDIAN P[4]
#define P_MV1 P[9]
-static inline int sad_hpel_motion_search(MpegEncContext * s,
+static int sad_hpel_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin,
int src_index, int ref_index,
int size, int h);
-static inline int update_map_generation(MotionEstContext *c)
+static inline unsigned update_map_generation(MotionEstContext *c)
{
c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
if(c->map_generation==0){
+ (chroma ? FLAG_CHROMA : 0);
}
-static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+static av_always_inline int cmp_direct_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
const int size, const int h, int ref_index, int src_index,
- me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel){
MotionEstContext * const c= &s->me;
const int stride= c->stride;
- const int uvstride= c->uvstride;
- const int qpel= flags&FLAG_QPEL;
- const int chroma= flags&FLAG_CHROMA;
- const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
const int hx= subx + (x<<(1+qpel));
const int hy= suby + (y<<(1+qpel));
uint8_t * const * const ref= c->ref[ref_index];
uint8_t * const * const src= c->src[src_index];
int d;
//FIXME check chroma 4mv, (no crashes ...)
- if(flags&FLAG_DIRECT){
+ assert(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));
if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
const int time_pp= s->pp_time;
const int time_pb= s->pb_time;
d = cmp_func(s, c->temp, src[0], stride, 16);
}else
d= 256*256*256*32;
- }else{
+ return d;
+}
+
+static av_always_inline int cmp_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+ const int size, const int h, int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel, int chroma){
+ MotionEstContext * const c= &s->me;
+ const int stride= c->stride;
+ const int uvstride= c->uvstride;
+ const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
+ const int hx= subx + (x<<(1+qpel));
+ const int hy= suby + (y<<(1+qpel));
+ uint8_t * const * const ref= c->ref[ref_index];
+ uint8_t * const * const src= c->src[src_index];
+ int d;
+ //FIXME check chroma 4mv, (no crashes ...)
int uvdxy; /* no, it might not be used uninitialized */
if(dxy){
if(qpel){
d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1);
d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
}
+ return d;
+}
+
+static int cmp_simple(MpegEncContext *s, const int x, const int y,
+ int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){
+ return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0);
+}
+
+static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y,
+ const int size, const int h, int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
+ if(flags&FLAG_DIRECT){
+ return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
+ }else{
+ return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
}
-#if 0
- if(full_pel){
- const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);
- score_map[index]= d;
+}
+
+static int cmp_internal(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+ const int size, const int h, int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
+ if(flags&FLAG_DIRECT){
+ return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
+ }else{
+ return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL, flags&FLAG_CHROMA);
}
+}
- d += (c->mv_penalty[hx - c->pred_x] + c->mv_penalty[hy - c->pred_y])*c->penalty_factor;
-#endif
- return d;
+/** @brief compares a block (either a full macroblock or a partition thereof)
+ against a proposed motion-compensated prediction of that block
+ */
+static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+ const int size, const int h, int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
+ if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size)
+ && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
+ && flags==0 && h==16 && size==0 && subx==0 && suby==0){
+ return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func);
+ }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
+ && subx==0 && suby==0){
+ return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags);
+ }else{
+ return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags);
+ }
+}
+
+static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+ const int size, const int h, int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
+ if(flags&FLAG_DIRECT){
+ return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0);
+ }else{
+ return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
+ }
+}
+
+static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
+ const int size, const int h, int ref_index, int src_index,
+ me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
+ if(flags&FLAG_DIRECT){
+ return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1);
+ }else{
+ return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA);
+ }
}
#include "motion_est_template.c"
return 0;
}
-static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){
+static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){
}
-void ff_init_me(MpegEncContext *s){
+int ff_init_me(MpegEncContext *s){
MotionEstContext * const c= &s->me;
+ int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
+ int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
+
+ if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -ME_MAP_SIZE){
+ av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n");
+ return -1;
+ }
+ if (s->me_method != ME_ZERO &&
+ s->me_method != ME_EPZS &&
+ s->me_method != ME_X1) {
+ av_log(s->avctx, AV_LOG_ERROR, "me_method is only allowed to be set to zero and epzs; for hex,umh,full and others see dia_size\n");
+ return -1;
+ }
+
c->avctx= s->avctx;
+ if(cache_size < 2*dia_size && !c->stride){
+ av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
+ }
+
ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp);
ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp);
ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp);
else
c->sub_motion_search= hpel_motion_search;
}
- c->hpel_avg= s->dsp.avg_pixels_tab;
- if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab;
- else c->hpel_put= s->dsp.put_pixels_tab;
+ c->hpel_avg = s->hdsp.avg_pixels_tab;
+ if (s->no_rounding)
+ c->hpel_put = s->hdsp.put_no_rnd_pixels_tab;
+ else
+ c->hpel_put = s->hdsp.put_pixels_tab;
if(s->linesize){
c->stride = s->linesize;
c->uvstride= 8*s->mb_width + 16;
}
- // 8x8 fullpel search would need a 4x4 chroma compare, which we dont have yet, and even if we had the motion estimation code doesnt expect it
- if(s->codec_id != CODEC_ID_SNOW){
- if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){
- s->dsp.me_cmp[2]= zero_cmp;
- }
- if((c->avctx->me_sub_cmp&FF_CMP_CHROMA) && !s->dsp.me_sub_cmp[2]){
- s->dsp.me_sub_cmp[2]= zero_cmp;
- }
- c->hpel_put[2][0]= c->hpel_put[2][1]=
- c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
+ /* 8x8 fullpel search would need a 4x4 chroma compare, which we do
+ * not have yet, and even if we had, the motion estimation code
+ * does not expect it. */
+ if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){
+ s->dsp.me_cmp[2]= zero_cmp;
}
-
- if(s->codec_id == CODEC_ID_H261){
- c->sub_motion_search= no_sub_motion_search;
- }
-
- c->temp= c->scratchpad;
-}
-
-#if 0
-static int pix_dev(uint8_t * pix, int line_size, int mean)
-{
- int s, i, j;
-
- s = 0;
- for (i = 0; i < 16; i++) {
- for (j = 0; j < 16; j += 8) {
- s += ABS(pix[0]-mean);
- s += ABS(pix[1]-mean);
- s += ABS(pix[2]-mean);
- s += ABS(pix[3]-mean);
- s += ABS(pix[4]-mean);
- s += ABS(pix[5]-mean);
- s += ABS(pix[6]-mean);
- s += ABS(pix[7]-mean);
- pix += 8;
- }
- pix += line_size - 16;
- }
- return s;
-}
-#endif
-
-static inline void no_motion_search(MpegEncContext * s,
- int *mx_ptr, int *my_ptr)
-{
- *mx_ptr = 16 * s->mb_x;
- *my_ptr = 16 * s->mb_y;
-}
-
-#if 0 /* the use of these functions is inside #if 0 */
-static int full_motion_search(MpegEncContext * s,
- int *mx_ptr, int *my_ptr, int range,
- int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
-{
- int x1, y1, x2, y2, xx, yy, x, y;
- int mx, my, dmin, d;
- uint8_t *pix;
-
- xx = 16 * s->mb_x;
- yy = 16 * s->mb_y;
- x1 = xx - range + 1; /* we loose one pixel to avoid boundary pb with half pixel pred */
- if (x1 < xmin)
- x1 = xmin;
- x2 = xx + range - 1;
- if (x2 > xmax)
- x2 = xmax;
- y1 = yy - range + 1;
- if (y1 < ymin)
- y1 = ymin;
- y2 = yy + range - 1;
- if (y2 > ymax)
- y2 = ymax;
- pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
- dmin = 0x7fffffff;
- mx = 0;
- my = 0;
- for (y = y1; y <= y2; y++) {
- for (x = x1; x <= x2; x++) {
- d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x,
- s->linesize, 16);
- if (d < dmin ||
- (d == dmin &&
- (abs(x - xx) + abs(y - yy)) <
- (abs(mx - xx) + abs(my - yy)))) {
- dmin = d;
- mx = x;
- my = y;
- }
- }
+ if((c->avctx->me_sub_cmp&FF_CMP_CHROMA) && !s->dsp.me_sub_cmp[2]){
+ s->dsp.me_sub_cmp[2]= zero_cmp;
}
+ c->hpel_put[2][0]= c->hpel_put[2][1]=
+ c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
- *mx_ptr = mx;
- *my_ptr = my;
-
-#if 0
- if (*mx_ptr < -(2 * range) || *mx_ptr >= (2 * range) ||
- *my_ptr < -(2 * range) || *my_ptr >= (2 * range)) {
- fprintf(stderr, "error %d %d\n", *mx_ptr, *my_ptr);
+ if(s->codec_id == AV_CODEC_ID_H261){
+ c->sub_motion_search= no_sub_motion_search;
}
-#endif
- return dmin;
-}
-
-
-static int log_motion_search(MpegEncContext * s,
- int *mx_ptr, int *my_ptr, int range,
- int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
-{
- int x1, y1, x2, y2, xx, yy, x, y;
- int mx, my, dmin, d;
- uint8_t *pix;
-
- xx = s->mb_x << 4;
- yy = s->mb_y << 4;
-
- /* Left limit */
- x1 = xx - range;
- if (x1 < xmin)
- x1 = xmin;
-
- /* Right limit */
- x2 = xx + range;
- if (x2 > xmax)
- x2 = xmax;
-
- /* Upper limit */
- y1 = yy - range;
- if (y1 < ymin)
- y1 = ymin;
-
- /* Lower limit */
- y2 = yy + range;
- if (y2 > ymax)
- y2 = ymax;
-
- pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
- dmin = 0x7fffffff;
- mx = 0;
- my = 0;
-
- do {
- for (y = y1; y <= y2; y += range) {
- for (x = x1; x <= x2; x += range) {
- d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);
- if (d < dmin || (d == dmin && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {
- dmin = d;
- mx = x;
- my = y;
- }
- }
- }
-
- range = range >> 1;
-
- x1 = mx - range;
- if (x1 < xmin)
- x1 = xmin;
-
- x2 = mx + range;
- if (x2 > xmax)
- x2 = xmax;
-
- y1 = my - range;
- if (y1 < ymin)
- y1 = ymin;
-
- y2 = my + range;
- if (y2 > ymax)
- y2 = ymax;
-
- } while (range >= 1);
-
-#ifdef DEBUG
- av_log(s->avctx, AV_LOG_DEBUG, "log - MX: %d\tMY: %d\n", mx, my);
-#endif
- *mx_ptr = mx;
- *my_ptr = my;
- return dmin;
-}
-static int phods_motion_search(MpegEncContext * s,
- int *mx_ptr, int *my_ptr, int range,
- int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
-{
- int x1, y1, x2, y2, xx, yy, x, y, lastx, d;
- int mx, my, dminx, dminy;
- uint8_t *pix;
-
- xx = s->mb_x << 4;
- yy = s->mb_y << 4;
-
- /* Left limit */
- x1 = xx - range;
- if (x1 < xmin)
- x1 = xmin;
-
- /* Right limit */
- x2 = xx + range;
- if (x2 > xmax)
- x2 = xmax;
-
- /* Upper limit */
- y1 = yy - range;
- if (y1 < ymin)
- y1 = ymin;
-
- /* Lower limit */
- y2 = yy + range;
- if (y2 > ymax)
- y2 = ymax;
-
- pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
- mx = 0;
- my = 0;
-
- x = xx;
- y = yy;
- do {
- dminx = 0x7fffffff;
- dminy = 0x7fffffff;
-
- lastx = x;
- for (x = x1; x <= x2; x += range) {
- d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);
- if (d < dminx || (d == dminx && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {
- dminx = d;
- mx = x;
- }
- }
-
- x = lastx;
- for (y = y1; y <= y2; y += range) {
- d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);
- if (d < dminy || (d == dminy && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {
- dminy = d;
- my = y;
- }
- }
-
- range = range >> 1;
-
- x = mx;
- y = my;
- x1 = mx - range;
- if (x1 < xmin)
- x1 = xmin;
-
- x2 = mx + range;
- if (x2 > xmax)
- x2 = xmax;
-
- y1 = my - range;
- if (y1 < ymin)
- y1 = ymin;
-
- y2 = my + range;
- if (y2 > ymax)
- y2 = ymax;
-
- } while (range >= 1);
-
-#ifdef DEBUG
- av_log(s->avctx, AV_LOG_DEBUG, "phods - MX: %d\tMY: %d\n", mx, my);
-#endif
-
- /* half pixel search */
- *mx_ptr = mx;
- *my_ptr = my;
- return dminy;
+ return 0;
}
-#endif /* 0 */
-
-#define Z_THRESHOLD 256
#define CHECK_SAD_HALF_MV(suffix, x, y) \
{\
COPY3_IF_LT(dminh, d, dx, x, dy, y)\
}
-static inline int sad_hpel_motion_search(MpegEncContext * s,
+static int sad_hpel_motion_search(MpegEncContext * s,
int *mx_ptr, int *my_ptr, int dmin,
int src_index, int ref_index,
int size, int h)
assert(flags == 0);
if(c->skip){
-// printf("S");
*mx_ptr = 0;
*my_ptr = 0;
return dmin;
}
-// printf("N");
pix = c->src[src_index][0];
if(mv4){
int mot_xy= s->block_index[0];
- s->current_picture.motion_val[0][mot_xy ][0]= mx;
- s->current_picture.motion_val[0][mot_xy ][1]= my;
- s->current_picture.motion_val[0][mot_xy+1][0]= mx;
- s->current_picture.motion_val[0][mot_xy+1][1]= my;
+ s->current_picture.motion_val[0][mot_xy ][0] = mx;
+ s->current_picture.motion_val[0][mot_xy ][1] = my;
+ s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
+ s->current_picture.motion_val[0][mot_xy + 1][1] = my;
mot_xy += s->b8_stride;
- s->current_picture.motion_val[0][mot_xy ][0]= mx;
- s->current_picture.motion_val[0][mot_xy ][1]= my;
- s->current_picture.motion_val[0][mot_xy+1][0]= mx;
- s->current_picture.motion_val[0][mot_xy+1][1]= my;
+ s->current_picture.motion_val[0][mot_xy ][0] = mx;
+ s->current_picture.motion_val[0][mot_xy ][1] = my;
+ s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
+ s->current_picture.motion_val[0][mot_xy + 1][1] = my;
}
}
static inline void get_limits(MpegEncContext *s, int x, int y)
{
MotionEstContext * const c= &s->me;
+ int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
/*
if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
else c->range= 16;
c->xmax = - x + s->mb_width *16 - 16;
c->ymax = - y + s->mb_height*16 - 16;
}
+ if(range){
+ c->xmin = FFMAX(c->xmin,-range);
+ c->xmax = FFMIN(c->xmax, range);
+ c->ymin = FFMAX(c->ymin,-range);
+ c->ymax = FFMIN(c->ymax, range);
+ }
}
static inline void init_mv4_ref(MotionEstContext *c){
dxy = ((my4 & 1) << 1) | (mx4 & 1);
if(s->no_rounding)
- s->dsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , stride, h);
+ s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , stride, h);
else
- s->dsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
+ s->hdsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
}
dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
}else
my4_sum+= my4;
}
- s->current_picture.motion_val[0][ s->block_index[block] ][0]= mx4;
- s->current_picture.motion_val[0][ s->block_index[block] ][1]= my4;
+ s->current_picture.motion_val[0][s->block_index[block]][0] = mx4;
+ s->current_picture.motion_val[0][s->block_index[block]][1] = my4;
if(mx4 != mx || my4 != my) same=0;
}
return INT_MAX;
if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
- dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);
+ dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);
}
if(c->avctx->mb_cmp&FF_CMP_CHROMA){
offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
if(s->no_rounding){
- s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8);
- s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
+ s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.f.data[1] + offset, s->uvlinesize, 8);
+ s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8);
}else{
- s->dsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8);
- s->dsp.put_pixels_tab [1][dxy](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
+ s->hdsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.f.data[1] + offset, s->uvlinesize, 8);
+ s->hdsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8);
}
- dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad , s->uvlinesize, 8);
- dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8);
+ dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.f.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad , s->uvlinesize, 8);
+ dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.f.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8);
}
c->pred_x= mx;
int16_t (*mv_table)[2]= mv_tables[block][field_select];
if(user_field_select){
+ assert(field_select==0 || field_select==1);
+ assert(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);
if(field_select_tables[block][xy] != field_select)
continue;
}
dxy = ((my_i & 1) << 1) | (mx_i & 1);
if(s->no_rounding){
- s->dsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref , stride, h);
+ s->hdsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref , stride, h);
}else{
- s->dsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
+ s->hdsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
}
dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
}
}
-static void clip_input_mv(MpegEncContext * s, int16_t *mv, int interlaced){
- int ymax= s->me.ymax>>interlaced;
- int ymin= s->me.ymin>>interlaced;
-
- if(mv[0] < s->me.xmin) mv[0] = s->me.xmin;
- if(mv[0] > s->me.xmax) mv[0] = s->me.xmax;
- if(mv[1] < ymin) mv[1] = ymin;
- if(mv[1] > ymax) mv[1] = ymax;
-}
-
-static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int p_type){
- MotionEstContext * const c= &s->me;
- Picture *p= s->current_picture_ptr;
- int mb_xy= mb_x + mb_y*s->mb_stride;
- int xy= 2*mb_x + 2*mb_y*s->b8_stride;
- int mb_type= s->current_picture.mb_type[mb_xy];
- int flags= c->flags;
- int shift= (flags&FLAG_QPEL) + 1;
- int mask= (1<<shift)-1;
- int x, y, i;
- int d=0;
- me_cmp_func cmpf= s->dsp.sse[0];
- me_cmp_func chroma_cmpf= s->dsp.sse[1];
-
- if(p_type && USES_LIST(mb_type, 1)){
- av_log(c->avctx, AV_LOG_ERROR, "backward motion vector in P frame\n");
- return INT_MAX/2;
- }
- assert(IS_INTRA(mb_type) || USES_LIST(mb_type,0) || USES_LIST(mb_type,1));
-
- for(i=0; i<4; i++){
- int xy= s->block_index[i];
- clip_input_mv(s, p->motion_val[0][xy], !!IS_INTERLACED(mb_type));
- clip_input_mv(s, p->motion_val[1][xy], !!IS_INTERLACED(mb_type));
- }
-
- if(IS_INTERLACED(mb_type)){
- int xy2= xy + s->b8_stride;
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
- c->stride<<=1;
- c->uvstride<<=1;
-
- if(!(s->flags & CODEC_FLAG_INTERLACED_ME)){
- av_log(c->avctx, AV_LOG_ERROR, "Interlaced macroblock selected but interlaced motion estimation disabled\n");
- return INT_MAX/2;
- }
-
- if(USES_LIST(mb_type, 0)){
- int field_select0= p->ref_index[0][xy ];
- int field_select1= p->ref_index[0][xy2];
- assert(field_select0==0 ||field_select0==1);
- assert(field_select1==0 ||field_select1==1);
- init_interlaced_ref(s, 0);
-
- if(p_type){
- s->p_field_select_table[0][mb_xy]= field_select0;
- s->p_field_select_table[1][mb_xy]= field_select1;
- *(uint32_t*)s->p_field_mv_table[0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[0][xy ];
- *(uint32_t*)s->p_field_mv_table[1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[0][xy2];
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER_I;
- }else{
- s->b_field_select_table[0][0][mb_xy]= field_select0;
- s->b_field_select_table[0][1][mb_xy]= field_select1;
- *(uint32_t*)s->b_field_mv_table[0][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[0][xy ];
- *(uint32_t*)s->b_field_mv_table[0][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[0][xy2];
- s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_FORWARD_I;
- }
-
- x= p->motion_val[0][xy ][0];
- y= p->motion_val[0][xy ][1];
- d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0, 0, cmpf, chroma_cmpf, flags);
- x= p->motion_val[0][xy2][0];
- y= p->motion_val[0][xy2][1];
- d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1, 1, cmpf, chroma_cmpf, flags);
- }
- if(USES_LIST(mb_type, 1)){
- int field_select0= p->ref_index[1][xy ];
- int field_select1= p->ref_index[1][xy2];
- assert(field_select0==0 ||field_select0==1);
- assert(field_select1==0 ||field_select1==1);
- init_interlaced_ref(s, 2);
-
- s->b_field_select_table[1][0][mb_xy]= field_select0;
- s->b_field_select_table[1][1][mb_xy]= field_select1;
- *(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[1][xy ];
- *(uint32_t*)s->b_field_mv_table[1][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[1][xy2];
- if(USES_LIST(mb_type, 0)){
- s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BIDIR_I;
- }else{
- s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BACKWARD_I;
- }
-
- x= p->motion_val[1][xy ][0];
- y= p->motion_val[1][xy ][1];
- d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0+2, 0, cmpf, chroma_cmpf, flags);
- x= p->motion_val[1][xy2][0];
- y= p->motion_val[1][xy2][1];
- d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1+2, 1, cmpf, chroma_cmpf, flags);
- //FIXME bidir scores
- }
- c->stride>>=1;
- c->uvstride>>=1;
- }else if(IS_8X8(mb_type)){
- if(!(s->flags & CODEC_FLAG_4MV)){
- av_log(c->avctx, AV_LOG_ERROR, "4MV macroblock selected but 4MV encoding disabled\n");
- return INT_MAX/2;
- }
- cmpf= s->dsp.sse[1];
- chroma_cmpf= s->dsp.sse[1];
- init_mv4_ref(c);
- for(i=0; i<4; i++){
- xy= s->block_index[i];
- x= p->motion_val[0][xy][0];
- y= p->motion_val[0][xy][1];
- d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 1, 8, i, i, cmpf, chroma_cmpf, flags);
- }
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER4V;
- }else{
- if(USES_LIST(mb_type, 0)){
- if(p_type){
- *(uint32_t*)s->p_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER;
- }else if(USES_LIST(mb_type, 1)){
- *(uint32_t*)s->b_bidir_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
- *(uint32_t*)s->b_bidir_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BIDIR;
- }else{
- *(uint32_t*)s->b_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_FORWARD;
- }
- x= p->motion_val[0][xy][0];
- y= p->motion_val[0][xy][1];
- d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 0, 0, cmpf, chroma_cmpf, flags);
- }else if(USES_LIST(mb_type, 1)){
- *(uint32_t*)s->b_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BACKWARD;
-
- x= p->motion_val[1][xy][0];
- y= p->motion_val[1][xy][1];
- d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 2, 0, cmpf, chroma_cmpf, flags);
- }else
- s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
+static inline int get_penalty_factor(int lambda, int lambda2, int type){
+ switch(type&0xFF){
+ default:
+ case FF_CMP_SAD:
+ return lambda>>FF_LAMBDA_SHIFT;
+ case FF_CMP_DCT:
+ return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
+ case FF_CMP_SATD:
+ case FF_CMP_DCT264:
+ return (2*lambda)>>FF_LAMBDA_SHIFT;
+ case FF_CMP_RD:
+ case FF_CMP_PSNR:
+ case FF_CMP_SSE:
+ case FF_CMP_NSSE:
+ return lambda2>>FF_LAMBDA_SHIFT;
+ case FF_CMP_BIT:
+ return 1;
}
- return d;
}
void ff_estimate_p_frame_motion(MpegEncContext * s,
{
MotionEstContext * const c= &s->me;
uint8_t *pix, *ppix;
- int sum, varc, vard, mx, my, dmin;
+ int sum, mx, my, dmin;
+ int varc; ///< the variance of the block (sum of squared (p[y][x]-average))
+ int vard; ///< sum of squared differences with the estimated motion vector
int P[10][2];
const int shift= 1+s->quarter_sample;
int mb_type=0;
Picture * const pic= &s->current_picture;
- init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
+ init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);
assert(s->quarter_sample==0 || s->quarter_sample==1);
assert(s->linesize == c->stride);
/* intra / predictive decision */
pix = c->src[0][0];
sum = s->dsp.pix_sum(pix, s->linesize);
- varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
+ varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500;
pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
- pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
- c->mb_var_sum_temp += varc;
-
- if(c->avctx->me_threshold){
- vard= (check_input_motion(s, mb_x, mb_y, 1)+128)>>8;
-
- if(vard<c->avctx->me_threshold){
- pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
- c->mc_mb_var_sum_temp += vard;
- if (vard <= 64 || vard < varc) { //FIXME
- c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
- }else{
- c->scene_change_score+= s->qscale;
- }
- return;
- }
- if(vard<c->avctx->mb_threshold)
- mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];
- }
+ pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
+ c->mb_var_sum_temp += (varc+128)>>8;
switch(s->me_method) {
case ME_ZERO:
default:
- no_motion_search(s, &mx, &my);
- mx-= mb_x*16;
- my-= mb_y*16;
+ mx = 0;
+ my = 0;
dmin = 0;
break;
-#if 0
- case ME_FULL:
- dmin = full_motion_search(s, &mx, &my, range, ref_picture);
- mx-= mb_x*16;
- my-= mb_y*16;
- break;
- case ME_LOG:
- dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);
- mx-= mb_x*16;
- my-= mb_y*16;
- break;
- case ME_PHODS:
- dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);
- mx-= mb_x*16;
- my-= mb_y*16;
- break;
-#endif
case ME_X1:
case ME_EPZS:
{
const int mot_stride = s->b8_stride;
const int mot_xy = s->block_index[0];
- P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
- P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
+ P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
+ P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
/* At this point (mx,my) are full-pell and the relative displacement */
ppix = c->ref[0][0] + (my * s->linesize) + mx;
- vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;
+ vard = s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16);
- pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
-// pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
- c->mc_mb_var_sum_temp += vard;
+ pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
+ c->mc_mb_var_sum_temp += (vard+128)>>8;
-#if 0
- printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",
- varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);
-#endif
if(mb_type){
- if (vard <= 64 || vard < varc)
- c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
- else
- c->scene_change_score+= s->qscale;
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
if(mb_type == CANDIDATE_MB_TYPE_INTER){
c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);
}
}else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
- if (vard <= 64 || vard < varc)
- c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
- else
- c->scene_change_score+= s->qscale;
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
- if (vard*2 + 200 > varc)
+ if (vard*2 + 200*256 > varc)
mb_type|= CANDIDATE_MB_TYPE_INTRA;
- if (varc*2 + 200 > vard){
+ if (varc*2 + 200*256 > vard || s->qscale > 24){
+// if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){
mb_type|= CANDIDATE_MB_TYPE_INTER;
c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
if(s->flags&CODEC_FLAG_MV0)
my <<=shift;
}
if((s->flags&CODEC_FLAG_4MV)
- && !c->skip && varc>50 && vard>10){
+ && !c->skip && varc>50<<8 && vard>10<<8){
if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
mb_type|=CANDIDATE_MB_TYPE_INTER4V;
dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
- dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
+ dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
if((s->flags&CODEC_FLAG_4MV)
- && !c->skip && varc>50 && vard>10){
+ && !c->skip && varc>50<<8 && vard>10<<8){
int dmin4= h263_mv4_search(s, mx, my, shift);
if(dmin4 < dmin){
mb_type= CANDIDATE_MB_TYPE_INTER4V;
}
}
-// pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
/* get intra luma score */
if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
- intra_score= (varc<<8) - 500; //FIXME dont scale it down so we dont have to fix it
+ intra_score= varc - 500;
}else{
- int mean= (sum+128)>>8;
+ unsigned mean = (sum+128)>>8;
mean*= 0x01010101;
for(i=0; i<16; i++){
intra_score= s->dsp.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
}
-#if 0 //FIXME
- /* get chroma score */
- if(c->avctx->mb_cmp&FF_CMP_CHROMA){
- for(i=1; i<3; i++){
- uint8_t *dest_c;
- int mean;
-
- if(s->out_format == FMT_H263){
- mean= (s->dc_val[i][mb_x + mb_y*s->b8_stride] + 4)>>3; //FIXME not exact but simple ;)
- }else{
- mean= (s->last_dc[i] + 4)>>3;
- }
- dest_c = s->new_picture.data[i] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
-
- mean*= 0x01010101;
- for(i=0; i<8; i++){
- *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 0]) = mean;
- *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 4]) = mean;
- }
-
- intra_score+= s->dsp.mb_cmp[1](s, c->scratchpad, dest_c, s->uvlinesize);
- }
- }
-#endif
intra_score += c->mb_penalty_factor*16;
if(intra_score < dmin){
mb_type= CANDIDATE_MB_TYPE_INTRA;
- s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
+ s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
}else
- s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;
+ s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;
- if (vard <= 64 || vard < varc) { //FIXME
- c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
- }else{
- c->scene_change_score+= s->qscale;
+ {
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
}
}
int P[10][2];
const int shift= 1+s->quarter_sample;
const int xy= mb_x + mb_y*s->mb_stride;
- init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
+ init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);
assert(s->quarter_sample==0 || s->quarter_sample==1);
return dmin;
}
-static int ff_estimate_motion_b(MpegEncContext * s,
- int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)
+static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,
+ int16_t (*mv_table)[2], int ref_index, int f_code)
{
MotionEstContext * const c= &s->me;
int mx, my, dmin;
switch(s->me_method) {
case ME_ZERO:
default:
- no_motion_search(s, &mx, &my);
+ mx = 0;
+ my = 0;
dmin = 0;
- mx-= mb_x*16;
- my-= mb_y*16;
- break;
-#if 0
- case ME_FULL:
- dmin = full_motion_search(s, &mx, &my, range, ref_picture);
- mx-= mb_x*16;
- my-= mb_y*16;
break;
- case ME_LOG:
- dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);
- mx-= mb_x*16;
- my-= mb_y*16;
- break;
- case ME_PHODS:
- dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);
- mx-= mb_x*16;
- my-= mb_y*16;
- break;
-#endif
case ME_X1:
case ME_EPZS:
- {
- P_LEFT[0] = mv_table[mot_xy - 1][0];
- P_LEFT[1] = mv_table[mot_xy - 1][1];
+ P_LEFT[0] = mv_table[mot_xy - 1][0];
+ P_LEFT[1] = mv_table[mot_xy - 1][1];
- if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
-
- /* special case for first line */
- if (!s->first_slice_line) {
- P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
- P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
- P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1 ][0];
- P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1 ][1];
- if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1]= (c->ymax<<shift);
- if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
- if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
+ if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);
- P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
- P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
- }
- c->pred_x= P_LEFT[0];
- c->pred_y= P_LEFT[1];
+ /* special case for first line */
+ if (!s->first_slice_line) {
+ P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
+ P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
+ P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
+ P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
+ if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);
+ if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);
+ if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);
+
+ P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
+ P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
}
+ c->pred_x = P_LEFT[0];
+ c->pred_y = P_LEFT[1];
if(mv_table == s->b_forw_mv_table){
mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);
dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
- dmin= ff_get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
+ dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
-//printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);
// s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
mv_table[mot_xy][0]= mx;
mv_table[mot_xy][1]= my;
src_y = motion_fy >> 1;
ptr = ref_data[0] + (src_y * stride) + src_x;
- s->dsp.put_pixels_tab[size][dxy](dest_y , ptr , stride, h);
+ s->hdsp.put_pixels_tab[size][dxy](dest_y , ptr , stride, h);
dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
src_x = motion_bx >> 1;
src_y = motion_by >> 1;
ptr = ref2_data[0] + (src_y * stride) + src_x;
- s->dsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
+ s->hdsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
}
fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
const int ymin= c->ymin<<shift;
const int xmax= c->xmax<<shift;
const int ymax= c->ymax<<shift;
+#define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
+#define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
+ int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);
+ uint8_t map[256] = { 0 };
+
+ map[hashidx&255] = 1;
fbmin= check_bidir_mv(s, motion_fx, motion_fy,
motion_bx, motion_by,
0, 16);
if(s->avctx->bidir_refine){
- int score, end;
+ int end;
+ static const uint8_t limittab[5]={0,8,32,64,80};
+ const int limit= limittab[s->avctx->bidir_refine];
+ static const int8_t vect[][4]={
+{ 0, 0, 0, 1}, { 0, 0, 0,-1}, { 0, 0, 1, 0}, { 0, 0,-1, 0}, { 0, 1, 0, 0}, { 0,-1, 0, 0}, { 1, 0, 0, 0}, {-1, 0, 0, 0},
+
+{ 0, 0, 1, 1}, { 0, 0,-1,-1}, { 0, 1, 1, 0}, { 0,-1,-1, 0}, { 1, 1, 0, 0}, {-1,-1, 0, 0}, { 1, 0, 0, 1}, {-1, 0, 0,-1},
+{ 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
+{ 0, 0,-1, 1}, { 0, 0, 1,-1}, { 0,-1, 1, 0}, { 0, 1,-1, 0}, {-1, 1, 0, 0}, { 1,-1, 0, 0}, { 1, 0, 0,-1}, {-1, 0, 0, 1},
+{ 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
+
+{ 0, 1, 1, 1}, { 0,-1,-1,-1}, { 1, 1, 1, 0}, {-1,-1,-1, 0}, { 1, 1, 0, 1}, {-1,-1, 0,-1}, { 1, 0, 1, 1}, {-1, 0,-1,-1},
+{ 0,-1, 1, 1}, { 0, 1,-1,-1}, {-1, 1, 1, 0}, { 1,-1,-1, 0}, { 1, 1, 0,-1}, {-1,-1, 0, 1}, { 1, 0,-1, 1}, {-1, 0, 1,-1},
+{ 0, 1,-1, 1}, { 0,-1, 1,-1}, { 1,-1, 1, 0}, {-1, 1,-1, 0}, {-1, 1, 0, 1}, { 1,-1, 0,-1}, { 1, 0, 1,-1}, {-1, 0,-1, 1},
+{ 0, 1, 1,-1}, { 0,-1,-1, 1}, { 1, 1,-1, 0}, {-1,-1, 1, 0}, { 1,-1, 0, 1}, {-1, 1, 0,-1}, {-1, 0, 1, 1}, { 1, 0,-1,-1},
+
+{ 1, 1, 1, 1}, {-1,-1,-1,-1},
+{ 1, 1, 1,-1}, {-1,-1,-1, 1}, { 1, 1,-1, 1}, {-1,-1, 1,-1}, { 1,-1, 1, 1}, {-1, 1,-1,-1}, {-1, 1, 1, 1}, { 1,-1,-1,-1},
+{ 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},
+ };
+ static const uint8_t hash[]={
+HASH8( 0, 0, 0, 1), HASH8( 0, 0, 0,-1), HASH8( 0, 0, 1, 0), HASH8( 0, 0,-1, 0), HASH8( 0, 1, 0, 0), HASH8( 0,-1, 0, 0), HASH8( 1, 0, 0, 0), HASH8(-1, 0, 0, 0),
+
+HASH8( 0, 0, 1, 1), HASH8( 0, 0,-1,-1), HASH8( 0, 1, 1, 0), HASH8( 0,-1,-1, 0), HASH8( 1, 1, 0, 0), HASH8(-1,-1, 0, 0), HASH8( 1, 0, 0, 1), HASH8(-1, 0, 0,-1),
+HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),
+HASH8( 0, 0,-1, 1), HASH8( 0, 0, 1,-1), HASH8( 0,-1, 1, 0), HASH8( 0, 1,-1, 0), HASH8(-1, 1, 0, 0), HASH8( 1,-1, 0, 0), HASH8( 1, 0, 0,-1), HASH8(-1, 0, 0, 1),
+HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),
+
+HASH8( 0, 1, 1, 1), HASH8( 0,-1,-1,-1), HASH8( 1, 1, 1, 0), HASH8(-1,-1,-1, 0), HASH8( 1, 1, 0, 1), HASH8(-1,-1, 0,-1), HASH8( 1, 0, 1, 1), HASH8(-1, 0,-1,-1),
+HASH8( 0,-1, 1, 1), HASH8( 0, 1,-1,-1), HASH8(-1, 1, 1, 0), HASH8( 1,-1,-1, 0), HASH8( 1, 1, 0,-1), HASH8(-1,-1, 0, 1), HASH8( 1, 0,-1, 1), HASH8(-1, 0, 1,-1),
+HASH8( 0, 1,-1, 1), HASH8( 0,-1, 1,-1), HASH8( 1,-1, 1, 0), HASH8(-1, 1,-1, 0), HASH8(-1, 1, 0, 1), HASH8( 1,-1, 0,-1), HASH8( 1, 0, 1,-1), HASH8(-1, 0,-1, 1),
+HASH8( 0, 1, 1,-1), HASH8( 0,-1,-1, 1), HASH8( 1, 1,-1, 0), HASH8(-1,-1, 1, 0), HASH8( 1,-1, 0, 1), HASH8(-1, 1, 0,-1), HASH8(-1, 0, 1, 1), HASH8( 1, 0,-1,-1),
+
+HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),
+HASH8( 1, 1, 1,-1), HASH8(-1,-1,-1, 1), HASH8( 1, 1,-1, 1), HASH8(-1,-1, 1,-1), HASH8( 1,-1, 1, 1), HASH8(-1, 1,-1,-1), HASH8(-1, 1, 1, 1), HASH8( 1,-1,-1,-1),
+HASH8( 1, 1,-1,-1), HASH8(-1,-1, 1, 1), HASH8( 1,-1,-1, 1), HASH8(-1, 1, 1,-1), HASH8( 1,-1, 1,-1), HASH8(-1, 1,-1, 1),
+};
+
#define CHECK_BIDIR(fx,fy,bx,by)\
- if( (fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
+ if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
+ &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
&&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
+ int score;\
+ map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
score= check_bidir_mv(s, motion_fx+fx, motion_fy+fy, motion_bx+bx, motion_by+by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);\
if(score < fbmin){\
+ hashidx += HASH(fx,fy,bx,by);\
fbmin= score;\
motion_fx+=fx;\
motion_fy+=fy;\
}
#define CHECK_BIDIR2(a,b,c,d)\
CHECK_BIDIR(a,b,c,d)\
-CHECK_BIDIR(-a,-b,-c,-d)
-
-#define CHECK_BIDIRR(a,b,c,d)\
-CHECK_BIDIR2(a,b,c,d)\
-CHECK_BIDIR2(b,c,d,a)\
-CHECK_BIDIR2(c,d,a,b)\
-CHECK_BIDIR2(d,a,b,c)
+CHECK_BIDIR(-(a),-(b),-(c),-(d))
do{
+ int i;
+ int borderdist=0;
end=1;
- CHECK_BIDIRR( 0, 0, 0, 1)
- if(s->avctx->bidir_refine > 1){
- CHECK_BIDIRR( 0, 0, 1, 1)
- CHECK_BIDIR2( 0, 1, 0, 1)
- CHECK_BIDIR2( 1, 0, 1, 0)
- CHECK_BIDIRR( 0, 0,-1, 1)
- CHECK_BIDIR2( 0,-1, 0, 1)
- CHECK_BIDIR2(-1, 0, 1, 0)
- if(s->avctx->bidir_refine > 2){
- CHECK_BIDIRR( 0, 1, 1, 1)
- CHECK_BIDIRR( 0,-1, 1, 1)
- CHECK_BIDIRR( 0, 1,-1, 1)
- CHECK_BIDIRR( 0, 1, 1,-1)
- if(s->avctx->bidir_refine > 3){
- CHECK_BIDIR2( 1, 1, 1, 1)
- CHECK_BIDIRR( 1, 1, 1,-1)
- CHECK_BIDIR2( 1, 1,-1,-1)
- CHECK_BIDIR2( 1,-1,-1, 1)
- CHECK_BIDIR2( 1,-1, 1,-1)
+ CHECK_BIDIR2(0,0,0,1)
+ CHECK_BIDIR2(0,0,1,0)
+ CHECK_BIDIR2(0,1,0,0)
+ CHECK_BIDIR2(1,0,0,0)
+
+ for(i=8; i<limit; i++){
+ int fx= motion_fx+vect[i][0];
+ int fy= motion_fy+vect[i][1];
+ int bx= motion_bx+vect[i][2];
+ int by= motion_by+vect[i][3];
+ if(borderdist<=0){
+ int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);
+ int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);
+ if((a|b) < 0)
+ map[(hashidx+hash[i])&255] = 1;
+ }
+ if(!map[(hashidx+hash[i])&255]){
+ int score;
+ map[(hashidx+hash[i])&255] = 1;
+ score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
+ if(score < fbmin){
+ hashidx += hash[i];
+ fbmin= score;
+ motion_fx=fx;
+ motion_fy=fy;
+ motion_bx=bx;
+ motion_by=by;
+ end=0;
+ borderdist--;
+ if(borderdist<=0){
+ int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);
+ int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);
+ borderdist= FFMIN(a,b);
+ }
}
}
}
}while(!end);
}
+ s->b_bidir_forw_mv_table[xy][0]= motion_fx;
+ s->b_bidir_forw_mv_table[xy][1]= motion_fy;
+ s->b_bidir_back_mv_table[xy][0]= motion_bx;
+ s->b_bidir_back_mv_table[xy][1]= motion_by;
+
return fbmin;
}
ymin= xmin=(-32)>>shift;
ymax= xmax= 31>>shift;
- if(IS_8X8(s->next_picture.mb_type[mot_xy])){
+ if (IS_8X8(s->next_picture.mb_type[mot_xy])) {
s->mv_type= MV_TYPE_8X8;
}else{
s->mv_type= MV_TYPE_16X16;
int index= s->block_index[i];
int min, max;
- c->co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];
- c->co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];
+ c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];
+ c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];
c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
// c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
c->pred_x=0;
c->pred_y=0;
- P_LEFT[0] = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
- P_LEFT[1] = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
+ P_LEFT[0] = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
+ P_LEFT[1] = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
/* special case for first line */
- if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped
- P_TOP[0] = clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
- P_TOP[1] = clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
- P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
- P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, ymax<<shift);
+ if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped
+ P_TOP[0] = av_clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
+ P_TOP[1] = av_clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
+ P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
+ P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, ymax<<shift);
P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
- dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
+ dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
- s->b_direct_mv_table[mot_xy][0]= mx;
- s->b_direct_mv_table[mot_xy][1]= my;
+ mv_table[mot_xy][0]= mx;
+ mv_table[mot_xy][1]= my;
c->flags &= ~FLAG_DIRECT;
c->sub_flags &= ~FLAG_DIRECT;
int fmin, bmin, dmin, fbmin, bimin, fimin;
int type=0;
const int xy = mb_y*s->mb_stride + mb_x;
- init_ref(c, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
+ init_ref(c, s->new_picture.f.data, s->last_picture.f.data,
+ s->next_picture.f.data, 16 * mb_x, 16 * mb_y, 2);
get_limits(s, 16*mb_x, 16*mb_y);
c->skip=0;
- if(c->avctx->me_threshold){
- int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;
-
- if(vard<c->avctx->me_threshold){
-// pix = c->src[0][0];
-// sum = s->dsp.pix_sum(pix, s->linesize);
-// varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
-
-// pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
- s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
-/* pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
- c->mb_var_sum_temp += varc;*/
- c->mc_mb_var_sum_temp += vard;
-/* if (vard <= 64 || vard < varc) {
- c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
- }else{
- c->scene_change_score+= s->qscale;
- }*/
- return;
- }
- if(vard<c->avctx->mb_threshold){
- type= s->mb_type[mb_y*s->mb_stride + mb_x];
- if(type == CANDIDATE_MB_TYPE_DIRECT){
- direct_search(s, mb_x, mb_y);
- }
- if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){
- c->skip=0;
- ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code);
- }
- if(type == CANDIDATE_MB_TYPE_BACKWARD || type == CANDIDATE_MB_TYPE_BIDIR){
- c->skip=0;
- ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code);
- }
- if(type == CANDIDATE_MB_TYPE_FORWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
- c->skip=0;
- c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
- interlaced_search(s, 0,
- s->b_field_mv_table[0], s->b_field_select_table[0],
- s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 1);
- }
- if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
- c->skip=0;
- c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
- interlaced_search(s, 2,
- s->b_field_mv_table[1], s->b_field_select_table[1],
- s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 1);
- }
- return;
- }
+
+ if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {
+ int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
+
+ score= ((unsigned)(score*score + 128*256))>>16;
+ c->mc_mb_var_sum_temp += score;
+ s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
+ s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
+
+ return;
}
- if (s->codec_id == CODEC_ID_MPEG4)
+ if (s->codec_id == AV_CODEC_ID_MPEG4)
dmin= direct_search(s, mb_x, mb_y);
else
dmin= INT_MAX;
//FIXME penalty stuff for non mpeg4
c->skip=0;
- fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;
+ fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +
+ 3 * penalty_factor;
c->skip=0;
- bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor;
-//printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
+ bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +
+ 2 * penalty_factor;
+ av_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
c->skip=0;
fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
-//printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);
+ av_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);
if(s->flags & CODEC_FLAG_INTERLACED_ME){
//FIXME mb type penalty
type |= CANDIDATE_MB_TYPE_BIDIR_I;
}
//FIXME something smarter
- if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB
-#if 0
- if(s->out_format == FMT_MPEG1)
- type |= CANDIDATE_MB_TYPE_INTRA;
-#endif
+ if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB
+ if(s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT && s->flags&CODEC_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
+ type |= CANDIDATE_MB_TYPE_DIRECT0;
}
s->mb_type[mb_y*s->mb_stride + mb_x]= type;
if(s->msmpeg4_version)
range= FFMIN(range, 16);
- else if(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
+ else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
range= FFMIN(range, 256);
for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);
continue;
for(j=0; j<fcode && j<8; j++){
- if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
+ if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
score[j]-= 170;
}
}
best_score= score[i];
best_fcode= i;
}
-// printf("%d %d\n", i, score[i]);
}
-// printf("fcode: %d type: %d\n", i, s->pict_type);
return best_fcode;
-/* for(i=0; i<=MAX_FCODE; i++){
- printf("%d ", mv_num[i]);
- }
- printf("\n");*/
}else{
return 1;
}
MotionEstContext * const c= &s->me;
const int f_code= s->f_code;
int y, range;
- assert(s->pict_type==P_TYPE);
+ assert(s->pict_type==AV_PICTURE_TYPE_P);
range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
assert(range <= 16 || !s->msmpeg4_version);
- assert(range <=256 || !(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
+ assert(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
-//printf("%d no:%d %d//\n", clip, noclip, f_code);
if(s->flags&CODEC_FLAG_4MV){
const int wrap= s->b8_stride;
int block;
for(block=0; block<4; block++){
int off= (block& 1) + (block>>1)*wrap;
- int mx= s->current_picture.motion_val[0][ xy + off ][0];
- int my= s->current_picture.motion_val[0][ xy + off ][1];
+ int mx = s->current_picture.motion_val[0][ xy + off ][0];
+ int my = s->current_picture.motion_val[0][ xy + off ][1];
if( mx >=range || mx <-range
|| my >=range || my <-range){
s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;
s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;
- s->current_picture.mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
+ s->current_picture.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;
}
}
}