/*
- * Motion estimation
+ * Motion estimation
* Copyright (c) 2000,2001 Fabrice Bellard.
- * Copyright (c) 2002-2003 Michael Niedermayer
- *
+ * Copyright (c) 2002-2004 Michael Niedermayer
*
- * This library is free software; you can redistribute it and/or
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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,
+ * FFmpeg 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
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
* new Motion Estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
*/
-
+
/**
* @file motion_est.c
* Motion estimation.
*/
-
+
#include <stdlib.h>
#include <stdio.h>
#include <limits.h>
#include "dsputil.h"
#include "mpegvideo.h"
-//#undef NDEBUG
-//#include <assert.h>
+#undef NDEBUG
+#include <assert.h>
#define SQ(a) ((a)*(a))
#define P_MV1 P[9]
static inline int sad_hpel_motion_search(MpegEncContext * s,
- int *mx_ptr, int *my_ptr, int dmin,
- int xmin, int ymin, int xmax, int ymax,
- int pred_x, int pred_y, Picture *picture,
- int n, int size, uint8_t * const mv_penalty);
+ int *mx_ptr, int *my_ptr, int dmin,
+ int src_index, int ref_index,
+ int size, int h);
-static inline int update_map_generation(MpegEncContext * s)
+static inline int update_map_generation(MotionEstContext *c)
{
- s->me.map_generation+= 1<<(ME_MAP_MV_BITS*2);
- if(s->me.map_generation==0){
- s->me.map_generation= 1<<(ME_MAP_MV_BITS*2);
- memset(s->me.map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
+ c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
+ if(c->map_generation==0){
+ c->map_generation= 1<<(ME_MAP_MV_BITS*2);
+ memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
}
- return s->me.map_generation;
+ return c->map_generation;
}
/* shape adaptive search stuff */
static int minima_cmp(const void *a, const void *b){
const Minima *da = (const Minima *) a;
const Minima *db = (const Minima *) b;
-
+
return da->height - db->height;
}
-
-/* SIMPLE */
-#define RENAME(a) simple_ ## a
-#define CMP(d, x, y, size)\
-d = cmp(s, src_y, (ref_y) + (x) + (y)*(stride), stride);
+#define FLAG_QPEL 1 //must be 1
+#define FLAG_CHROMA 2
+#define FLAG_DIRECT 4
-#define CMP_HPEL(d, dx, dy, x, y, size)\
-{\
- const int dxy= (dx) + 2*(dy);\
- hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, (16>>size));\
- d = cmp_sub(s, s->me.scratchpad, src_y, stride);\
-}
-
-#define CMP_QPEL(d, dx, dy, x, y, size)\
-{\
- const int dxy= (dx) + 4*(dy);\
- qpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride);\
- d = cmp_sub(s, s->me.scratchpad, src_y, stride);\
+static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
+ const int offset[3]= {
+ y*c-> stride + x,
+ ((y*c->uvstride + x)>>1),
+ ((y*c->uvstride + x)>>1),
+ };
+ int i;
+ for(i=0; i<3; i++){
+ c->src[0][i]= src [i] + offset[i];
+ c->ref[0][i]= ref [i] + offset[i];
+ }
+ if(ref_index){
+ for(i=0; i<3; i++){
+ c->ref[ref_index][i]= ref2[i] + offset[i];
+ }
+ }
}
-#include "motion_est_template.c"
-#undef RENAME
-#undef CMP
-#undef CMP_HPEL
-#undef CMP_QPEL
-#undef INIT
-
-/* SIMPLE CHROMA */
-#define RENAME(a) simple_chroma_ ## a
-
-#define CMP(d, x, y, size)\
-d = cmp(s, src_y, (ref_y) + (x) + (y)*(stride), stride);\
-if(chroma_cmp){\
- int dxy= ((x)&1) + 2*((y)&1);\
- int c= ((x)>>1) + ((y)>>1)*uvstride;\
-\
- chroma_hpel_put[0][dxy](s->me.scratchpad, ref_u + c, uvstride, 8);\
- d += chroma_cmp(s, s->me.scratchpad, src_u, uvstride);\
- chroma_hpel_put[0][dxy](s->me.scratchpad, ref_v + c, uvstride, 8);\
- d += chroma_cmp(s, s->me.scratchpad, src_v, uvstride);\
+static int get_flags(MotionEstContext *c, int direct, int chroma){
+ return ((c->avctx->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
+ + (direct ? FLAG_DIRECT : 0)
+ + (chroma ? FLAG_CHROMA : 0);
}
-#define CMP_HPEL(d, dx, dy, x, y, size)\
-{\
- const int dxy= (dx) + 2*(dy);\
- hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, (16>>size));\
- d = cmp_sub(s, s->me.scratchpad, src_y, stride);\
- if(chroma_cmp_sub){\
- int cxy= (dxy) | ((x)&1) | (2*((y)&1));\
- int c= ((x)>>1) + ((y)>>1)*uvstride;\
- chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, 8);\
- d += chroma_cmp_sub(s, s->me.scratchpad, src_u, uvstride);\
- chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, 8);\
- d += chroma_cmp_sub(s, s->me.scratchpad, src_v, uvstride);\
- }\
-}
+/*! \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){
+ 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;
+ const int mask= 2*qpel+1;
+ if(s->mv_type==MV_TYPE_8X8){
+ int i;
+ for(i=0; i<4; i++){
+ int fx = c->direct_basis_mv[i][0] + hx;
+ int fy = c->direct_basis_mv[i][1] + hy;
+ int bx = hx ? fx - c->co_located_mv[i][0] : c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(qpel+4));
+ int by = hy ? fy - c->co_located_mv[i][1] : c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(qpel+4));
+ int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
+ int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
+
+ uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);
+ if(qpel){
+ c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);
+ c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);
+ }else{
+ c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);
+ c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);
+ }
+ }
+ }else{
+ int fx = c->direct_basis_mv[0][0] + hx;
+ int fy = c->direct_basis_mv[0][1] + hy;
+ int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
+ int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
+ int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
+ int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
+
+ if(qpel){
+ c->qpel_put[1][fxy](c->temp , ref[0] + (fx>>2) + (fy>>2)*stride , stride);
+ c->qpel_put[1][fxy](c->temp + 8 , ref[0] + (fx>>2) + (fy>>2)*stride + 8 , stride);
+ c->qpel_put[1][fxy](c->temp + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8*stride, stride);
+ c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);
+ c->qpel_avg[1][bxy](c->temp , ref[8] + (bx>>2) + (by>>2)*stride , stride);
+ c->qpel_avg[1][bxy](c->temp + 8 , ref[8] + (bx>>2) + (by>>2)*stride + 8 , stride);
+ c->qpel_avg[1][bxy](c->temp + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8*stride, stride);
+ c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);
+ }else{
+ assert((fx>>1) + 16*s->mb_x >= -16);
+ assert((fy>>1) + 16*s->mb_y >= -16);
+ assert((fx>>1) + 16*s->mb_x <= s->width);
+ assert((fy>>1) + 16*s->mb_y <= s->height);
+ assert((bx>>1) + 16*s->mb_x >= -16);
+ assert((by>>1) + 16*s->mb_y >= -16);
+ assert((bx>>1) + 16*s->mb_x <= s->width);
+ assert((by>>1) + 16*s->mb_y <= s->height);
+
+ c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);
+ c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);
+ }
+ }
+ d = cmp_func(s, c->temp, src[0], stride, 16);
+ }else
+ d= 256*256*256*32;
+ }else{
+ int uvdxy; /* no, it might not be used uninitialized */
+ if(dxy){
+ if(qpel){
+ c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
+ if(chroma){
+ int cx= hx/2;
+ int cy= hy/2;
+ cx= (cx>>1)|(cx&1);
+ cy= (cy>>1)|(cy&1);
+ uvdxy= (cx&1) + 2*(cy&1);
+ //FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264
+ }
+ }else{
+ c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
+ if(chroma)
+ uvdxy= dxy | (x&1) | (2*(y&1));
+ }
+ d = cmp_func(s, c->temp, src[0], stride, h);
+ }else{
+ d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
+ if(chroma)
+ uvdxy= (x&1) + 2*(y&1);
+ }
+ if(chroma){
+ uint8_t * const uvtemp= c->temp + 16*stride;
+ c->hpel_put[size+1][uvdxy](uvtemp , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
+ c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
+ d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1);
+ d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
+ }
+ }
+#if 0
+ if(full_pel){
+ const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);
+ score_map[index]= d;
+ }
-#define CMP_QPEL(d, dx, dy, x, y, size)\
-{\
- const int dxy= (dx) + 4*(dy);\
- qpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride);\
- d = cmp_sub(s, s->me.scratchpad, src_y, stride);\
- if(chroma_cmp_sub){\
- int cxy, c;\
- int cx= (4*(x) + (dx))/2;\
- int cy= (4*(y) + (dy))/2;\
- cx= (cx>>1)|(cx&1);\
- cy= (cy>>1)|(cy&1);\
- cxy= (cx&1) + 2*(cy&1);\
- c= ((cx)>>1) + ((cy)>>1)*uvstride;\
- chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, 8);\
- d += chroma_cmp_sub(s, s->me.scratchpad, src_u, uvstride);\
- chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, 8);\
- d += chroma_cmp_sub(s, s->me.scratchpad, src_v, uvstride);\
- }\
+ d += (c->mv_penalty[hx - c->pred_x] + c->mv_penalty[hy - c->pred_y])*c->penalty_factor;
+#endif
+ return d;
}
#include "motion_est_template.c"
-#undef RENAME
-#undef CMP
-#undef CMP_HPEL
-#undef CMP_QPEL
-#undef INIT
-
-/* SIMPLE DIRECT HPEL */
-#define RENAME(a) simple_direct_hpel_ ## a
-//FIXME precalc divisions stuff
-
-#define CMP_DIRECT(d, dx, dy, x, y, size, cmp_func)\
-if((x) >= xmin && 2*(x) + (dx) <= 2*xmax && (y) >= ymin && 2*(y) + (dy) <= 2*ymax){\
- const int hx= 2*(x) + (dx);\
- const int hy= 2*(y) + (dy);\
- if(s->mv_type==MV_TYPE_8X8){\
- int i;\
- for(i=0; i<4; i++){\
- int fx = s->me.direct_basis_mv[i][0] + hx;\
- int fy = s->me.direct_basis_mv[i][1] + hy;\
- int bx = hx ? fx - s->me.co_located_mv[i][0] : s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + (i &1)*16;\
- int by = hy ? fy - s->me.co_located_mv[i][1] : s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + (i>>1)*16;\
- int fxy= (fx&1) + 2*(fy&1);\
- int bxy= (bx&1) + 2*(by&1);\
-\
- uint8_t *dst= s->me.scratchpad + 8*(i&1) + 8*stride*(i>>1);\
- hpel_put[1][fxy](dst, (ref_y ) + (fx>>1) + (fy>>1)*(stride), stride, 8);\
- hpel_avg[1][bxy](dst, (ref2_y) + (bx>>1) + (by>>1)*(stride), stride, 8);\
- }\
- }else{\
- int fx = s->me.direct_basis_mv[0][0] + hx;\
- int fy = s->me.direct_basis_mv[0][1] + hy;\
- int bx = hx ? fx - s->me.co_located_mv[0][0] : (s->me.co_located_mv[0][0]*(time_pb - time_pp)/time_pp);\
- int by = hy ? fy - s->me.co_located_mv[0][1] : (s->me.co_located_mv[0][1]*(time_pb - time_pp)/time_pp);\
- int fxy= (fx&1) + 2*(fy&1);\
- int bxy= (bx&1) + 2*(by&1);\
- \
- assert((fx>>1) + 16*s->mb_x >= -16);\
- assert((fy>>1) + 16*s->mb_y >= -16);\
- assert((fx>>1) + 16*s->mb_x <= s->width);\
- assert((fy>>1) + 16*s->mb_y <= s->height);\
- assert((bx>>1) + 16*s->mb_x >= -16);\
- assert((by>>1) + 16*s->mb_y >= -16);\
- assert((bx>>1) + 16*s->mb_x <= s->width);\
- assert((by>>1) + 16*s->mb_y <= s->height);\
-\
- hpel_put[0][fxy](s->me.scratchpad, (ref_y ) + (fx>>1) + (fy>>1)*(stride), stride, 16);\
- hpel_avg[0][bxy](s->me.scratchpad, (ref2_y) + (bx>>1) + (by>>1)*(stride), stride, 16);\
- }\
- d = cmp_func(s, s->me.scratchpad, src_y, stride);\
-}else\
- d= 256*256*256*32;
-
-
-#define CMP_HPEL(d, dx, dy, x, y, size)\
- CMP_DIRECT(d, dx, dy, x, y, size, cmp_sub)
-
-#define CMP(d, x, y, size)\
- CMP_DIRECT(d, 0, 0, x, y, size, cmp)
-
-#include "motion_est_template.c"
-#undef RENAME
-#undef CMP
-#undef CMP_HPEL
-#undef CMP_QPEL
-#undef INIT
-#undef CMP_DIRECT
-
-/* SIMPLE DIRECT QPEL */
-#define RENAME(a) simple_direct_qpel_ ## a
-
-#define CMP_DIRECT(d, dx, dy, x, y, size, cmp_func)\
-if((x) >= xmin && 4*(x) + (dx) <= 4*xmax && (y) >= ymin && 4*(y) + (dy) <= 4*ymax){\
- const int qx= 4*(x) + (dx);\
- const int qy= 4*(y) + (dy);\
- if(s->mv_type==MV_TYPE_8X8){\
- int i;\
- for(i=0; i<4; i++){\
- int fx = s->me.direct_basis_mv[i][0] + qx;\
- int fy = s->me.direct_basis_mv[i][1] + qy;\
- int bx = qx ? fx - s->me.co_located_mv[i][0] : s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + (i &1)*16;\
- int by = qy ? fy - s->me.co_located_mv[i][1] : s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + (i>>1)*16;\
- int fxy= (fx&3) + 4*(fy&3);\
- int bxy= (bx&3) + 4*(by&3);\
-\
- uint8_t *dst= s->me.scratchpad + 8*(i&1) + 8*stride*(i>>1);\
- qpel_put[1][fxy](dst, (ref_y ) + (fx>>2) + (fy>>2)*(stride), stride);\
- qpel_avg[1][bxy](dst, (ref2_y) + (bx>>2) + (by>>2)*(stride), stride);\
- }\
- }else{\
- int fx = s->me.direct_basis_mv[0][0] + qx;\
- int fy = s->me.direct_basis_mv[0][1] + qy;\
- int bx = qx ? fx - s->me.co_located_mv[0][0] : s->me.co_located_mv[0][0]*(time_pb - time_pp)/time_pp;\
- int by = qy ? fy - s->me.co_located_mv[0][1] : s->me.co_located_mv[0][1]*(time_pb - time_pp)/time_pp;\
- int fxy= (fx&3) + 4*(fy&3);\
- int bxy= (bx&3) + 4*(by&3);\
-\
- qpel_put[1][fxy](s->me.scratchpad , (ref_y ) + (fx>>2) + (fy>>2)*(stride) , stride);\
- qpel_put[1][fxy](s->me.scratchpad + 8 , (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8 , stride);\
- qpel_put[1][fxy](s->me.scratchpad + 8*stride, (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8*stride, stride);\
- qpel_put[1][fxy](s->me.scratchpad + 8 + 8*stride, (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8 + 8*stride, stride);\
- qpel_avg[1][bxy](s->me.scratchpad , (ref2_y) + (bx>>2) + (by>>2)*(stride) , stride);\
- qpel_avg[1][bxy](s->me.scratchpad + 8 , (ref2_y) + (bx>>2) + (by>>2)*(stride) + 8 , stride);\
- qpel_avg[1][bxy](s->me.scratchpad + 8*stride, (ref2_y) + (bx>>2) + (by>>2)*(stride) + 8*stride, stride);\
- qpel_avg[1][bxy](s->me.scratchpad + 8 + 8*stride, (ref2_y) + (bx>>2) + (by>>2)*(stride) + 8 + 8*stride, stride);\
- }\
- d = cmp_func(s, s->me.scratchpad, src_y, stride);\
-}else\
- d= 256*256*256*32;
-
-
-#define CMP_QPEL(d, dx, dy, x, y, size)\
- CMP_DIRECT(d, dx, dy, x, y, size, cmp_sub)
-
-#define CMP(d, x, y, size)\
- CMP_DIRECT(d, 0, 0, x, y, size, cmp)
-
-#include "motion_est_template.c"
-#undef RENAME
-#undef CMP
-#undef CMP_HPEL
-#undef CMP_QPEL
-#undef INIT
-#undef CMP__DIRECT
-
-static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride){
+static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){
return 0;
}
-static void set_cmp(MpegEncContext *s, me_cmp_func *cmp, int type){
- DSPContext* c= &s->dsp;
- int i;
-
- memset(cmp, 0, sizeof(void*)*11);
-
- switch(type&0xFF){
- case FF_CMP_SAD:
- cmp[0]= c->sad[0];
- cmp[1]= c->sad[1];
- break;
- case FF_CMP_SATD:
- cmp[0]= c->hadamard8_diff[0];
- cmp[1]= c->hadamard8_diff[1];
- break;
- case FF_CMP_SSE:
- cmp[0]= c->sse[0];
- cmp[1]= c->sse[1];
- break;
- case FF_CMP_DCT:
- cmp[0]= c->dct_sad[0];
- cmp[1]= c->dct_sad[1];
- break;
- case FF_CMP_PSNR:
- cmp[0]= c->quant_psnr[0];
- cmp[1]= c->quant_psnr[1];
- break;
- case FF_CMP_BIT:
- cmp[0]= c->bit[0];
- cmp[1]= c->bit[1];
- break;
- case FF_CMP_RD:
- cmp[0]= c->rd[0];
- cmp[1]= c->rd[1];
- break;
- case FF_CMP_ZERO:
- for(i=0; i<7; i++){
- cmp[i]= zero_cmp;
- }
- break;
- default:
- fprintf(stderr,"internal error in cmp function selection\n");
- }
+static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){
}
-static inline int get_penalty_factor(MpegEncContext *s, int type){
- switch(type&0xFF){
- default:
- case FF_CMP_SAD:
- return s->qscale*2;
- case FF_CMP_DCT:
- return s->qscale*3;
- case FF_CMP_SATD:
- return s->qscale*6;
- case FF_CMP_SSE:
- return s->qscale*s->qscale*2;
- case FF_CMP_BIT:
- return 1;
- case FF_CMP_RD:
- case FF_CMP_PSNR:
- return (s->qscale*s->qscale*185 + 64)>>7;
+void 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);
+ 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");
}
-}
-void ff_init_me(MpegEncContext *s){
- set_cmp(s, s->dsp.me_pre_cmp, s->avctx->me_pre_cmp);
- set_cmp(s, s->dsp.me_cmp, s->avctx->me_cmp);
- set_cmp(s, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
- set_cmp(s, s->dsp.mb_cmp, s->avctx->mb_cmp);
+ 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);
+ ff_set_cmp(&s->dsp, s->dsp.mb_cmp, c->avctx->mb_cmp);
+ c->flags = get_flags(c, 0, c->avctx->me_cmp &FF_CMP_CHROMA);
+ c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
+ c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp &FF_CMP_CHROMA);
+
+/*FIXME s->no_rounding b_type*/
if(s->flags&CODEC_FLAG_QPEL){
- if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)
- s->me.sub_motion_search= simple_chroma_qpel_motion_search;
- else
- s->me.sub_motion_search= simple_qpel_motion_search;
+ c->sub_motion_search= qpel_motion_search;
+ c->qpel_avg= s->dsp.avg_qpel_pixels_tab;
+ if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab;
+ else c->qpel_put= s->dsp.put_qpel_pixels_tab;
}else{
- if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)
- s->me.sub_motion_search= simple_chroma_hpel_motion_search;
- else if( s->avctx->me_sub_cmp == FF_CMP_SAD
- && s->avctx-> me_cmp == FF_CMP_SAD
- && s->avctx-> mb_cmp == FF_CMP_SAD)
- s->me.sub_motion_search= sad_hpel_motion_search;
+ if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
+ c->sub_motion_search= hpel_motion_search;
+ else if( c->avctx->me_sub_cmp == FF_CMP_SAD
+ && c->avctx-> me_cmp == FF_CMP_SAD
+ && c->avctx-> mb_cmp == FF_CMP_SAD)
+ c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
else
- s->me.sub_motion_search= simple_hpel_motion_search;
+ 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;
- if(s->avctx->me_cmp&FF_CMP_CHROMA){
- s->me.motion_search[0]= simple_chroma_epzs_motion_search;
- s->me.motion_search[1]= simple_chroma_epzs_motion_search4;
+ if(s->linesize){
+ c->stride = s->linesize;
+ c->uvstride= s->uvlinesize;
}else{
- s->me.motion_search[0]= simple_epzs_motion_search;
- s->me.motion_search[1]= simple_epzs_motion_search4;
+ c->stride = 16*s->mb_width + 32;
+ c->uvstride= 8*s->mb_width + 16;
}
-
- if(s->avctx->me_pre_cmp&FF_CMP_CHROMA){
- s->me.pre_motion_search= simple_chroma_epzs_motion_search;
- }else{
- s->me.pre_motion_search= simple_epzs_motion_search;
+
+ /* 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(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;
}
-
- if(s->flags&CODEC_FLAG_QPEL){
- if(s->avctx->mb_cmp&FF_CMP_CHROMA)
- s->me.get_mb_score= simple_chroma_qpel_get_mb_score;
- else
- s->me.get_mb_score= simple_qpel_get_mb_score;
- }else{
- if(s->avctx->mb_cmp&FF_CMP_CHROMA)
- s->me.get_mb_score= simple_chroma_hpel_get_mb_score;
- else
- s->me.get_mb_score= simple_hpel_get_mb_score;
+
+ 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)
{
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;
+ for (j = 0; j < 16; j += 8) {
+ s += FFABS(pix[0]-mean);
+ s += FFABS(pix[1]-mean);
+ s += FFABS(pix[2]-mean);
+ s += FFABS(pix[3]-mean);
+ s += FFABS(pix[4]-mean);
+ s += FFABS(pix[5]-mean);
+ s += FFABS(pix[6]-mean);
+ s += FFABS(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)
+ 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)
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 */
+ x1 = xx - range + 1; /* we loose one pixel to avoid boundary pb with half pixel pred */
if (x1 < xmin)
- x1 = xmin;
+ x1 = xmin;
x2 = xx + range - 1;
if (x2 > xmax)
- x2 = xmax;
+ x2 = xmax;
y1 = yy - range + 1;
if (y1 < ymin)
- y1 = ymin;
+ y1 = ymin;
y2 = yy + range - 1;
if (y2 > ymax)
- 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_abs16x16(pix, ref_picture + (y * s->linesize) + x,
- s->linesize);
- if (d < dmin ||
- (d == dmin &&
- (abs(x - xx) + abs(y - yy)) <
- (abs(mx - xx) + abs(my - yy)))) {
- dmin = d;
- mx = x;
- my = 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;
+ }
+ }
}
*mx_ptr = mx;
#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);
+ *my_ptr < -(2 * range) || *my_ptr >= (2 * range)) {
+ av_log(NULL, AV_LOG_ERROR, "error %d %d\n", *mx_ptr, *my_ptr);
}
#endif
return dmin;
/* Left limit */
x1 = xx - range;
if (x1 < xmin)
- x1 = xmin;
+ x1 = xmin;
/* Right limit */
x2 = xx + range;
if (x2 > xmax)
- x2 = xmax;
+ x2 = xmax;
/* Upper limit */
y1 = yy - range;
if (y1 < ymin)
- y1 = ymin;
+ y1 = ymin;
/* Lower limit */
y2 = yy + range;
if (y2 > ymax)
- y2 = ymax;
+ y2 = ymax;
pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
dmin = 0x7fffffff;
my = 0;
do {
- for (y = y1; y <= y2; y += range) {
- for (x = x1; x <= x2; x += range) {
- d = s->dsp.pix_abs16x16(pix, ref_picture + (y * s->linesize) + x, s->linesize);
- 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;
+ 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
- fprintf(stderr, "log - MX: %d\tMY: %d\n", mx, my);
+ av_log(s->avctx, AV_LOG_DEBUG, "log - MX: %d\tMY: %d\n", mx, my);
#endif
*mx_ptr = mx;
*my_ptr = my;
/* Left limit */
x1 = xx - range;
if (x1 < xmin)
- x1 = xmin;
+ x1 = xmin;
/* Right limit */
x2 = xx + range;
if (x2 > xmax)
- x2 = xmax;
+ x2 = xmax;
/* Upper limit */
y1 = yy - range;
if (y1 < ymin)
- y1 = ymin;
+ y1 = ymin;
/* Lower limit */
y2 = yy + range;
if (y2 > ymax)
- y2 = ymax;
+ y2 = ymax;
pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
mx = 0;
dminx = 0x7fffffff;
dminy = 0x7fffffff;
- lastx = x;
- for (x = x1; x <= x2; x += range) {
- d = s->dsp.pix_abs16x16(pix, ref_picture + (y * s->linesize) + x, s->linesize);
- 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_abs16x16(pix, ref_picture + (y * s->linesize) + x, s->linesize);
- 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;
+ 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
- fprintf(stderr, "phods - MX: %d\tMY: %d\n", mx, my);
+ av_log(s->avctx, AV_LOG_DEBUG, "phods - MX: %d\tMY: %d\n", mx, my);
#endif
/* half pixel search */
*my_ptr = my;
return dminy;
}
-
+#endif /* 0 */
#define Z_THRESHOLD 256
#define CHECK_SAD_HALF_MV(suffix, x, y) \
{\
- d= pix_abs_ ## suffix(pix, ptr+((x)>>1), s->linesize);\
+ d= s->dsp.pix_abs[size][(x?1:0)+(y?2:0)](NULL, pix, ptr+((x)>>1), stride, h);\
d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
COPY3_IF_LT(dminh, d, dx, x, dy, y)\
}
static inline int sad_hpel_motion_search(MpegEncContext * s,
- int *mx_ptr, int *my_ptr, int dmin,
- int xmin, int ymin, int xmax, int ymax,
- int pred_x, int pred_y, Picture *picture,
- int n, int size, uint8_t * const mv_penalty)
+ int *mx_ptr, int *my_ptr, int dmin,
+ int src_index, int ref_index,
+ int size, int h)
{
- uint8_t *ref_picture= picture->data[0];
- uint32_t *score_map= s->me.score_map;
- const int penalty_factor= s->me.sub_penalty_factor;
- int mx, my, xx, yy, dminh;
+ MotionEstContext * const c= &s->me;
+ const int penalty_factor= c->sub_penalty_factor;
+ int mx, my, dminh;
uint8_t *pix, *ptr;
- op_pixels_abs_func pix_abs_x2;
- op_pixels_abs_func pix_abs_y2;
- op_pixels_abs_func pix_abs_xy2;
-
- if(size==0){
- pix_abs_x2 = s->dsp.pix_abs16x16_x2;
- pix_abs_y2 = s->dsp.pix_abs16x16_y2;
- pix_abs_xy2= s->dsp.pix_abs16x16_xy2;
- }else{
- pix_abs_x2 = s->dsp.pix_abs8x8_x2;
- pix_abs_y2 = s->dsp.pix_abs8x8_y2;
- pix_abs_xy2= s->dsp.pix_abs8x8_xy2;
- }
+ int stride= c->stride;
+ const int flags= c->sub_flags;
+ LOAD_COMMON
- if(s->me.skip){
+ assert(flags == 0);
+
+ if(c->skip){
// printf("S");
*mx_ptr = 0;
*my_ptr = 0;
return dmin;
}
// printf("N");
-
- xx = 16 * s->mb_x + 8*(n&1);
- yy = 16 * s->mb_y + 8*(n>>1);
- pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
+
+ pix = c->src[src_index][0];
mx = *mx_ptr;
my = *my_ptr;
- ptr = ref_picture + ((yy + my) * s->linesize) + (xx + mx);
-
+ ptr = c->ref[ref_index][0] + (my * stride) + mx;
+
dminh = dmin;
- if (mx > xmin && mx < xmax &&
+ if (mx > xmin && mx < xmax &&
my > ymin && my < ymax) {
int dx=0, dy=0;
- int d, pen_x, pen_y;
+ int d, pen_x, pen_y;
const int index= (my<<ME_MAP_SHIFT) + mx;
const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
mx<<=1;
my<<=1;
-
+
pen_x= pred_x + mx;
pen_y= pred_y + my;
- ptr-= s->linesize;
+ ptr-= stride;
if(t<=b){
CHECK_SAD_HALF_MV(y2 , 0, -1)
if(l<=r){
CHECK_SAD_HALF_MV(xy2, -1, -1)
if(t+r<=b+l){
CHECK_SAD_HALF_MV(xy2, +1, -1)
- ptr+= s->linesize;
+ ptr+= stride;
}else{
- ptr+= s->linesize;
+ ptr+= stride;
CHECK_SAD_HALF_MV(xy2, -1, +1)
}
CHECK_SAD_HALF_MV(x2 , -1, 0)
CHECK_SAD_HALF_MV(xy2, +1, -1)
if(t+l<=b+r){
CHECK_SAD_HALF_MV(xy2, -1, -1)
- ptr+= s->linesize;
+ ptr+= stride;
}else{
- ptr+= s->linesize;
+ ptr+= stride;
CHECK_SAD_HALF_MV(xy2, +1, +1)
}
CHECK_SAD_HALF_MV(x2 , +1, 0)
if(l<=r){
if(t+l<=b+r){
CHECK_SAD_HALF_MV(xy2, -1, -1)
- ptr+= s->linesize;
+ ptr+= stride;
}else{
- ptr+= s->linesize;
+ ptr+= stride;
CHECK_SAD_HALF_MV(xy2, +1, +1)
}
CHECK_SAD_HALF_MV(x2 , -1, 0)
}else{
if(t+r<=b+l){
CHECK_SAD_HALF_MV(xy2, +1, -1)
- ptr+= s->linesize;
+ ptr+= stride;
}else{
- ptr+= s->linesize;
+ ptr+= stride;
CHECK_SAD_HALF_MV(xy2, -1, +1)
}
CHECK_SAD_HALF_MV(x2 , +1, 0)
static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
{
const int xy= s->mb_x + s->mb_y*s->mb_stride;
-
+
s->p_mv_table[xy][0] = mx;
s->p_mv_table[xy][1] = my;
- /* has allready been set to the 4 MV if 4MV is done */
+ /* has already been set to the 4 MV if 4MV is done */
if(mv4){
int mot_xy= s->block_index[0];
- s->motion_val[mot_xy ][0]= mx;
- s->motion_val[mot_xy ][1]= my;
- s->motion_val[mot_xy+1][0]= mx;
- s->motion_val[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->block_wrap[0];
- s->motion_val[mot_xy ][0]= mx;
- s->motion_val[mot_xy ][1]= my;
- s->motion_val[mot_xy+1][0]= mx;
- s->motion_val[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;
}
}
/**
* get fullpel ME search limits.
- * @param range the approximate search range for the old ME code, unused for EPZS and newer
*/
-static inline void get_limits(MpegEncContext *s, int *range, int *xmin, int *ymin, int *xmax, int *ymax)
+static inline void get_limits(MpegEncContext *s, int x, int y)
{
- if(s->avctx->me_range) *range= s->avctx->me_range >> 1;
- else *range= 16;
-
+ 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;
+*/
if (s->unrestricted_mv) {
- *xmin = -16;
- *ymin = -16;
- *xmax = s->mb_width*16;
- *ymax = s->mb_height*16;
+ c->xmin = - x - 16;
+ c->ymin = - y - 16;
+ c->xmax = - x + s->mb_width *16;
+ c->ymax = - y + s->mb_height*16;
+ } else if (s->out_format == FMT_H261){
+ // Search range of H261 is different from other codec standards
+ c->xmin = (x > 15) ? - 15 : 0;
+ c->ymin = (y > 15) ? - 15 : 0;
+ c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
+ c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
} else {
- *xmin = 0;
- *ymin = 0;
- *xmax = s->mb_width*16 - 16;
- *ymax = s->mb_height*16 - 16;
+ c->xmin = - x;
+ c->ymin = - y;
+ 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);
}
-
- //FIXME try to limit x/y min/max if me_range is set
}
-static inline int h263_mv4_search(MpegEncContext *s, int xmin, int ymin, int xmax, int ymax, int mx, int my, int shift)
+static inline void init_mv4_ref(MotionEstContext *c){
+ const int stride= c->stride;
+
+ c->ref[1][0] = c->ref[0][0] + 8;
+ c->ref[2][0] = c->ref[0][0] + 8*stride;
+ c->ref[3][0] = c->ref[2][0] + 8;
+ c->src[1][0] = c->src[0][0] + 8;
+ c->src[2][0] = c->src[0][0] + 8*stride;
+ c->src[3][0] = c->src[2][0] + 8;
+}
+
+static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
{
+ MotionEstContext * const c= &s->me;
+ const int size= 1;
+ const int h=8;
int block;
int P[10][2];
int dmin_sum=0, mx4_sum=0, my4_sum=0;
- uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
int same=1;
+ const int stride= c->stride;
+ uint8_t *mv_penalty= c->current_mv_penalty;
+
+ init_mv4_ref(c);
for(block=0; block<4; block++){
int mx4, my4;
int pred_x4, pred_y4;
int dmin4;
static const int off[4]= {2, 1, 1, -1};
- const int mot_stride = s->block_wrap[0];
+ const int mot_stride = s->b8_stride;
const int mot_xy = s->block_index[block];
-// const int block_x= (block&1);
-// const int block_y= (block>>1);
-#if 1 // this saves us a bit of cliping work and shouldnt affect compression in a negative way
- const int rel_xmin4= xmin;
- const int rel_xmax4= xmax;
- const int rel_ymin4= ymin;
- const int rel_ymax4= ymax;
-#else
- const int rel_xmin4= xmin - block_x*8;
- const int rel_xmax4= xmax - block_x*8 + 8;
- const int rel_ymin4= ymin - block_y*8;
- const int rel_ymax4= ymax - block_y*8 + 8;
-#endif
- P_LEFT[0] = s->motion_val[mot_xy - 1][0];
- P_LEFT[1] = s->motion_val[mot_xy - 1][1];
- if(P_LEFT[0] > (rel_xmax4<<shift)) P_LEFT[0] = (rel_xmax4<<shift);
+ 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);
/* special case for first line */
- if (s->mb_y == 0 && block<2) {
- pred_x4= P_LEFT[0];
- pred_y4= P_LEFT[1];
+ if (s->first_slice_line && block<2) {
+ c->pred_x= pred_x4= P_LEFT[0];
+ c->pred_y= pred_y4= P_LEFT[1];
} else {
- P_TOP[0] = s->motion_val[mot_xy - mot_stride ][0];
- P_TOP[1] = s->motion_val[mot_xy - mot_stride ][1];
- P_TOPRIGHT[0] = s->motion_val[mot_xy - mot_stride + off[block]][0];
- P_TOPRIGHT[1] = s->motion_val[mot_xy - mot_stride + off[block]][1];
- if(P_TOP[1] > (rel_ymax4<<shift)) P_TOP[1] = (rel_ymax4<<shift);
- if(P_TOPRIGHT[0] < (rel_xmin4<<shift)) P_TOPRIGHT[0]= (rel_xmin4<<shift);
- if(P_TOPRIGHT[0] > (rel_xmax4<<shift)) P_TOPRIGHT[0]= (rel_xmax4<<shift);
- if(P_TOPRIGHT[1] > (rel_ymax4<<shift)) P_TOPRIGHT[1]= (rel_ymax4<<shift);
-
+ P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
+ P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
+ P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
+ P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][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[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<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]);
-// if(s->out_format == FMT_H263){
- pred_x4 = P_MEDIAN[0];
- pred_y4 = P_MEDIAN[1];
-#if 0
- }else { /* mpeg1 at least */
- pred_x4= P_LEFT[0];
- pred_y4= P_LEFT[1];
- }
-#endif
+ c->pred_x= pred_x4 = P_MEDIAN[0];
+ c->pred_y= pred_y4 = P_MEDIAN[1];
}
P_MV1[0]= mx;
P_MV1[1]= my;
- dmin4 = s->me.motion_search[1](s, block, &mx4, &my4, P, pred_x4, pred_y4, rel_xmin4, rel_ymin4, rel_xmax4, rel_ymax4,
- &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);
+ dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
+
+ dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
- dmin4= s->me.sub_motion_search(s, &mx4, &my4, dmin4, rel_xmin4, rel_ymin4, rel_xmax4, rel_ymax4,
- pred_x4, pred_y4, &s->last_picture, block, 1, mv_penalty);
-
if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
int dxy;
- const int offset= ((block&1) + (block>>1)*s->linesize)*8;
- uint8_t *dest_y = s->me.scratchpad + offset;
-
+ const int offset= ((block&1) + (block>>1)*stride)*8;
+ uint8_t *dest_y = c->scratchpad + offset;
if(s->quarter_sample){
- uint8_t *ref= s->last_picture.data[0] + (s->mb_x*16 + (mx4>>2)) + (s->mb_y*16 + (my4>>2))*s->linesize + offset;
+ uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
dxy = ((my4 & 3) << 2) | (mx4 & 3);
if(s->no_rounding)
- s->dsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y , ref , s->linesize);
+ s->dsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y , ref , stride);
else
- s->dsp.put_qpel_pixels_tab [1][dxy](dest_y , ref , s->linesize);
+ s->dsp.put_qpel_pixels_tab [1][dxy](dest_y , ref , stride);
}else{
- uint8_t *ref= s->last_picture.data[0] + (s->mb_x*16 + (mx4>>1)) + (s->mb_y*16 + (my4>>1))*s->linesize + offset;
+ uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
dxy = ((my4 & 1) << 1) | (mx4 & 1);
if(s->no_rounding)
- s->dsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , s->linesize, 8);
+ s->dsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , stride, h);
else
- s->dsp.put_pixels_tab [1][dxy](dest_y , ref , s->linesize, 8);
+ s->dsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
}
- dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*s->me.mb_penalty_factor;
+ dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
}else
dmin_sum+= dmin4;
mx4_sum+= mx4;
my4_sum+= my4;
}
-
- s->motion_val[ s->block_index[block] ][0]= mx4;
- s->motion_val[ 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;
}
-
+
if(same)
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*s->linesize, s->me.scratchpad, s->linesize);
+ 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);
}
-
- if(s->avctx->mb_cmp&FF_CMP_CHROMA){
+
+ if(c->avctx->mb_cmp&FF_CMP_CHROMA){
int dxy;
int mx, my;
int offset;
mx= ff_h263_round_chroma(mx4_sum);
my= ff_h263_round_chroma(my4_sum);
dxy = ((my & 1) << 1) | (mx & 1);
-
+
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](s->me.scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8);
- s->dsp.put_no_rnd_pixels_tab[1][dxy](s->me.scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
+ 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);
}else{
- s->dsp.put_pixels_tab [1][dxy](s->me.scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8);
- s->dsp.put_pixels_tab [1][dxy](s->me.scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8);
+ 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);
}
- dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, s->me.scratchpad , s->uvlinesize);
- dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, s->me.scratchpad+8, s->uvlinesize);
+ 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);
}
- switch(s->avctx->mb_cmp&0xFF){
+ c->pred_x= mx;
+ c->pred_y= my;
+
+ switch(c->avctx->mb_cmp&0xFF){
/*case FF_CMP_SSE:
return dmin_sum+ 32*s->qscale*s->qscale;*/
case FF_CMP_RD:
return dmin_sum;
default:
- return dmin_sum+ 11*s->me.mb_penalty_factor;
+ return dmin_sum+ 11*c->mb_penalty_factor;
}
}
+static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
+ MotionEstContext * const c= &s->me;
+
+ c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
+ c->src[1][0] = c->src[0][0] + s->linesize;
+ if(c->flags & FLAG_CHROMA){
+ c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
+ c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
+ c->src[1][1] = c->src[0][1] + s->uvlinesize;
+ c->src[1][2] = c->src[0][2] + s->uvlinesize;
+ }
+}
+
+static int interlaced_search(MpegEncContext *s, int ref_index,
+ int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
+{
+ MotionEstContext * const c= &s->me;
+ const int size=0;
+ const int h=8;
+ int block;
+ int P[10][2];
+ uint8_t * const mv_penalty= c->current_mv_penalty;
+ int same=1;
+ const int stride= 2*s->linesize;
+ int dmin_sum= 0;
+ const int mot_stride= s->mb_stride;
+ const int xy= s->mb_x + s->mb_y*mot_stride;
+
+ c->ymin>>=1;
+ c->ymax>>=1;
+ c->stride<<=1;
+ c->uvstride<<=1;
+ init_interlaced_ref(s, ref_index);
+
+ for(block=0; block<2; block++){
+ int field_select;
+ int best_dmin= INT_MAX;
+ int best_field= -1;
+
+ for(field_select=0; field_select<2; field_select++){
+ int dmin, mx_i, my_i;
+ 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;
+ }
+
+ P_LEFT[0] = mv_table[xy - 1][0];
+ P_LEFT[1] = mv_table[xy - 1][1];
+ if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
+
+ c->pred_x= P_LEFT[0];
+ c->pred_y= P_LEFT[1];
+
+ if(!s->first_slice_line){
+ P_TOP[0] = mv_table[xy - mot_stride][0];
+ P_TOP[1] = mv_table[xy - mot_stride][1];
+ P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
+ P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
+ if(P_TOP[1] > (c->ymax<<1)) P_TOP[1] = (c->ymax<<1);
+ if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
+ if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
+ if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
+
+ 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]);
+ }
+ P_MV1[0]= mx; //FIXME not correct if block != field_select
+ P_MV1[1]= my / 2;
+
+ dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);
+
+ dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
+
+ mv_table[xy][0]= mx_i;
+ mv_table[xy][1]= my_i;
+
+ if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
+ int dxy;
+
+ //FIXME chroma ME
+ uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
+ 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);
+ }else{
+ s->dsp.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;
+ }else
+ dmin+= c->mb_penalty_factor; //field_select bits
+
+ dmin += field_select != block; //slightly prefer same field
+
+ if(dmin < best_dmin){
+ best_dmin= dmin;
+ best_field= field_select;
+ }
+ }
+ {
+ int16_t (*mv_table)[2]= mv_tables[block][best_field];
+
+ if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all
+ if(mv_table[xy][1]&1) same=0;
+ if(mv_table[xy][1]*2 != my) same=0;
+ if(best_field != block) same=0;
+ }
+
+ field_select_tables[block][xy]= best_field;
+ dmin_sum += best_dmin;
+ }
+
+ c->ymin<<=1;
+ c->ymax<<=1;
+ c->stride>>=1;
+ c->uvstride>>=1;
+
+ if(same)
+ return INT_MAX;
+
+ switch(c->avctx->mb_cmp&0xFF){
+ /*case FF_CMP_SSE:
+ return dmin_sum+ 32*s->qscale*s->qscale;*/
+ case FF_CMP_RD:
+ return dmin_sum;
+ default:
+ return dmin_sum+ 11*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;
+ }
+ return d;
+}
+
void ff_estimate_p_frame_motion(MpegEncContext * s,
int mb_x, int mb_y)
{
+ MotionEstContext * const c= &s->me;
uint8_t *pix, *ppix;
- int sum, varc, vard, mx, my, range, dmin, xx, yy;
- int xmin, ymin, xmax, ymax;
- int rel_xmin, rel_ymin, rel_xmax, rel_ymax;
- int pred_x=0, pred_y=0;
+ 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;
- uint8_t *ref_picture= s->last_picture.data[0];
Picture * const pic= &s->current_picture;
- uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
-
+
+ init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
+
assert(s->quarter_sample==0 || s->quarter_sample==1);
+ assert(s->linesize == c->stride);
+ assert(s->uvlinesize == c->uvstride);
+
+ c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
+ c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
+ c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
- s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp);
- s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);
- s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);
+ get_limits(s, 16*mb_x, 16*mb_y);
+ c->skip=0;
- get_limits(s, &range, &xmin, &ymin, &xmax, &ymax);
- rel_xmin= xmin - mb_x*16;
- rel_xmax= xmax - mb_x*16;
- rel_ymin= ymin - mb_y*16;
- rel_ymax= ymax - mb_y*16;
- s->me.skip=0;
+ /* 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;
+
+ pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
+ pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
+ c->mb_var_sum_temp += (varc+128)>>8;
+
+ if(c->avctx->me_threshold){
+ vard= check_input_motion(s, mb_x, mb_y, 1);
+
+ if((vard+128)>>8 < c->avctx->me_threshold){
+ int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
+ int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
+ pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
+ c->mc_mb_var_sum_temp += (vard+128)>>8;
+ c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
+ return;
+ }
+ if((vard+128)>>8 < c->avctx->mb_threshold)
+ mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];
+ }
switch(s->me_method) {
case ME_ZERO:
default:
- no_motion_search(s, &mx, &my);
+ no_motion_search(s, &mx, &my);
mx-= mb_x*16;
my-= mb_y*16;
dmin = 0;
break;
+#if 0
case ME_FULL:
- dmin = full_motion_search(s, &mx, &my, range, xmin, ymin, xmax, ymax, ref_picture);
+ 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, xmin, ymin, xmax, ymax, ref_picture);
+ 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, xmin, ymin, xmax, ymax, ref_picture);
+ 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->block_wrap[0];
+ const int mot_stride = s->b8_stride;
const int mot_xy = s->block_index[0];
- P_LEFT[0] = s->motion_val[mot_xy - 1][0];
- P_LEFT[1] = s->motion_val[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] > (rel_xmax<<shift)) P_LEFT[0] = (rel_xmax<<shift);
+ if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
+
+ if(!s->first_slice_line) {
+ P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
+ P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
+ P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
+ P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][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(mb_y) {
- P_TOP[0] = s->motion_val[mot_xy - mot_stride ][0];
- P_TOP[1] = s->motion_val[mot_xy - mot_stride ][1];
- P_TOPRIGHT[0] = s->motion_val[mot_xy - mot_stride + 2][0];
- P_TOPRIGHT[1] = s->motion_val[mot_xy - mot_stride + 2][1];
- if(P_TOP[1] > (rel_ymax<<shift)) P_TOP[1] = (rel_ymax<<shift);
- if(P_TOPRIGHT[0] < (rel_xmin<<shift)) P_TOPRIGHT[0]= (rel_xmin<<shift);
- if(P_TOPRIGHT[1] > (rel_ymax<<shift)) P_TOPRIGHT[1]= (rel_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]);
if(s->out_format == FMT_H263){
- pred_x = P_MEDIAN[0];
- pred_y = P_MEDIAN[1];
+ c->pred_x = P_MEDIAN[0];
+ c->pred_y = P_MEDIAN[1];
}else { /* mpeg1 at least */
- pred_x= P_LEFT[0];
- pred_y= P_LEFT[1];
+ c->pred_x= P_LEFT[0];
+ c->pred_y= P_LEFT[1];
}
}else{
- pred_x= P_LEFT[0];
- pred_y= P_LEFT[1];
+ c->pred_x= P_LEFT[0];
+ c->pred_y= P_LEFT[1];
}
}
- dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
- &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);
-
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
+
break;
}
- /* intra / predictive decision */
- xx = mb_x * 16;
- yy = mb_y * 16;
-
- pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
/* At this point (mx,my) are full-pell and the relative displacement */
- ppix = ref_picture + ((yy+my) * s->linesize) + (xx+mx);
-
- sum = s->dsp.pix_sum(pix, s->linesize);
-
- varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
- vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize)+128)>>8;
-
-//printf("%d %d %d %X %X %X\n", s->mb_width, mb_x, mb_y,(int)s, (int)s->mb_var, (int)s->mc_mb_var); fflush(stdout);
- pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
- pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
- pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
-// pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
- pic->mb_var_sum += varc;
- pic->mc_mb_var_sum += vard;
-//printf("E%d %d %d %X %X %X\n", s->mb_width, mb_x, mb_y,(int)s, (int)s->mb_var, (int)s->mc_mb_var); fflush(stdout);
-
+ ppix = c->ref[0][0] + (my * s->linesize) + mx;
+
+ vard = s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16);
+
+ pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
+// pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
+ 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);
+ varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);
#endif
- if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
- if (vard <= 64 || vard < varc)
- s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
- else
- s->scene_change_score+= s->qscale;
-
- if (vard*2 + 200 > varc)
- mb_type|= MB_TYPE_INTRA;
- if (varc*2 + 200 > vard){
- mb_type|= MB_TYPE_INTER;
- s->me.sub_motion_search(s, &mx, &my, dmin, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
- pred_x, pred_y, &s->last_picture, 0, 0, mv_penalty);
+ if(mb_type){
+ 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);
+ set_p_mv_tables(s, mx, my, 1);
+ }else{
+ mx <<=shift;
+ my <<=shift;
+ }
+ if(mb_type == CANDIDATE_MB_TYPE_INTER4V){
+ h263_mv4_search(s, mx, my, shift);
+
+ set_p_mv_tables(s, mx, my, 0);
+ }
+ if(mb_type == CANDIDATE_MB_TYPE_INTER_I){
+ 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){
+ 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*256 > varc)
+ mb_type|= CANDIDATE_MB_TYPE_INTRA;
+ 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)
if(mx || my)
- mb_type |= MB_TYPE_SKIPED; //FIXME check difference
+ mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference
}else{
mx <<=shift;
my <<=shift;
}
if((s->flags&CODEC_FLAG_4MV)
- && !s->me.skip && varc>50 && vard>10){
- if(h263_mv4_search(s, rel_xmin, rel_ymin, rel_xmax, rel_ymax, mx, my, shift) < INT_MAX)
- mb_type|=MB_TYPE_INTER4V;
+ && !c->skip && varc>50<<8 && vard>10<<8){
+ if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
+ mb_type|=CANDIDATE_MB_TYPE_INTER4V;
set_p_mv_tables(s, mx, my, 0);
}else
set_p_mv_tables(s, mx, my, 1);
+ if((s->flags&CODEC_FLAG_INTERLACED_ME)
+ && !c->skip){ //FIXME varc/d checks
+ if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
+ mb_type |= CANDIDATE_MB_TYPE_INTER_I;
+ }
}else{
- mb_type= MB_TYPE_INTER;
+ int intra_score, i;
+ mb_type= CANDIDATE_MB_TYPE_INTER;
- dmin= s->me.sub_motion_search(s, &mx, &my, dmin, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
- pred_x, pred_y, &s->last_picture, 0, 0, mv_penalty);
-
- if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
- dmin= s->me.get_mb_score(s, mx, my, pred_x, pred_y, &s->last_picture, mv_penalty);
+ 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);
if((s->flags&CODEC_FLAG_4MV)
- && !s->me.skip && varc>50 && vard>10){
- int dmin4= h263_mv4_search(s, rel_xmin, rel_ymin, rel_xmax, rel_ymax, mx, my, shift);
+ && !c->skip && varc>50<<8 && vard>10<<8){
+ int dmin4= h263_mv4_search(s, mx, my, shift);
if(dmin4 < dmin){
- mb_type= MB_TYPE_INTER4V;
+ mb_type= CANDIDATE_MB_TYPE_INTER4V;
dmin=dmin4;
}
}
- pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
- set_p_mv_tables(s, mx, my, mb_type!=MB_TYPE_INTER4V);
-
- if (vard <= 64 || vard < varc) {
- s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ if((s->flags&CODEC_FLAG_INTERLACED_ME)
+ && !c->skip){ //FIXME varc/d checks
+ int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
+ if(dmin_i < dmin){
+ mb_type = CANDIDATE_MB_TYPE_INTER_I;
+ dmin= dmin_i;
+ }
+ }
+
+// 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 - 500;
}else{
- s->scene_change_score+= s->qscale;
+ int mean= (sum+128)>>8;
+ mean*= 0x01010101;
+
+ for(i=0; i<16; i++){
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
+ *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
+ }
+
+ 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
+ }else
+ s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;
+
+ {
+ 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 ff_pre_estimate_p_frame_motion(MpegEncContext * s,
int mb_x, int mb_y)
{
- int mx, my, range, dmin;
- int xmin, ymin, xmax, ymax;
- int rel_xmin, rel_ymin, rel_xmax, rel_ymax;
- int pred_x=0, pred_y=0;
+ MotionEstContext * const c= &s->me;
+ int mx, my, dmin;
int P[10][2];
const int shift= 1+s->quarter_sample;
- uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
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);
+
assert(s->quarter_sample==0 || s->quarter_sample==1);
- s->me.pre_penalty_factor = get_penalty_factor(s, s->avctx->me_pre_cmp);
+ c->pre_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
- get_limits(s, &range, &xmin, &ymin, &xmax, &ymax);
- rel_xmin= xmin - mb_x*16;
- rel_xmax= xmax - mb_x*16;
- rel_ymin= ymin - mb_y*16;
- rel_ymax= ymax - mb_y*16;
- s->me.skip=0;
+ get_limits(s, 16*mb_x, 16*mb_y);
+ c->skip=0;
P_LEFT[0] = s->p_mv_table[xy + 1][0];
P_LEFT[1] = s->p_mv_table[xy + 1][1];
- if(P_LEFT[0] < (rel_xmin<<shift)) P_LEFT[0] = (rel_xmin<<shift);
+ if(P_LEFT[0] < (c->xmin<<shift)) P_LEFT[0] = (c->xmin<<shift);
/* special case for first line */
- if (mb_y == s->mb_height-1) {
- pred_x= P_LEFT[0];
- pred_y= P_LEFT[1];
+ if (s->first_slice_line) {
+ c->pred_x= P_LEFT[0];
+ c->pred_y= P_LEFT[1];
P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
- P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
+ P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
} else {
P_TOP[0] = s->p_mv_table[xy + s->mb_stride ][0];
P_TOP[1] = s->p_mv_table[xy + s->mb_stride ][1];
P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
- if(P_TOP[1] < (rel_ymin<<shift)) P_TOP[1] = (rel_ymin<<shift);
- if(P_TOPRIGHT[0] > (rel_xmax<<shift)) P_TOPRIGHT[0]= (rel_xmax<<shift);
- if(P_TOPRIGHT[1] < (rel_ymin<<shift)) P_TOPRIGHT[1]= (rel_ymin<<shift);
-
+ if(P_TOP[1] < (c->ymin<<shift)) P_TOP[1] = (c->ymin<<shift);
+ if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
+ if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<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]);
- pred_x = P_MEDIAN[0];
- pred_y = P_MEDIAN[1];
+ c->pred_x = P_MEDIAN[0];
+ c->pred_y = P_MEDIAN[1];
}
- dmin = s->me.pre_motion_search(s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
- &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);
+
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
s->p_mv_table[xy][0] = mx<<shift;
s->p_mv_table[xy][1] = my<<shift;
-
+
return dmin;
}
static int ff_estimate_motion_b(MpegEncContext * s,
- int mb_x, int mb_y, int16_t (*mv_table)[2], Picture *picture, int f_code)
+ int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)
{
- int mx, my, range, dmin;
- int xmin, ymin, xmax, ymax;
- int rel_xmin, rel_ymin, rel_xmax, rel_ymax;
- int pred_x=0, pred_y=0;
+ MotionEstContext * const c= &s->me;
+ int mx, my, dmin;
int P[10][2];
const int shift= 1+s->quarter_sample;
const int mot_stride = s->mb_stride;
const int mot_xy = mb_y*mot_stride + mb_x;
- uint8_t * const ref_picture= picture->data[0];
- uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;
+ uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;
int mv_scale;
-
- s->me.penalty_factor = get_penalty_factor(s, s->avctx->me_cmp);
- s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);
- s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);
- get_limits(s, &range, &xmin, &ymin, &xmax, &ymax);
- rel_xmin= xmin - mb_x*16;
- rel_xmax= xmax - mb_x*16;
- rel_ymin= ymin - mb_y*16;
- rel_ymax= ymax - mb_y*16;
+ c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
+ c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
+ c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
+ c->current_mv_penalty= mv_penalty;
+
+ get_limits(s, 16*mb_x, 16*mb_y);
switch(s->me_method) {
case ME_ZERO:
default:
- no_motion_search(s, &mx, &my);
+ no_motion_search(s, &mx, &my);
dmin = 0;
mx-= mb_x*16;
my-= mb_y*16;
break;
+#if 0
case ME_FULL:
- dmin = full_motion_search(s, &mx, &my, range, xmin, ymin, xmax, ymax, ref_picture);
+ 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, xmin, ymin, xmax, ymax, ref_picture);
+ 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, xmin, ymin, xmax, ymax, ref_picture);
+ 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];
- if(P_LEFT[0] > (rel_xmax<<shift)) P_LEFT[0] = (rel_xmax<<shift);
+ if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
/* special case for first line */
- if (mb_y) {
+ 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] > (rel_ymax<<shift)) P_TOP[1]= (rel_ymax<<shift);
- if(P_TOPRIGHT[0] < (rel_xmin<<shift)) P_TOPRIGHT[0]= (rel_xmin<<shift);
- if(P_TOPRIGHT[1] > (rel_ymax<<shift)) P_TOPRIGHT[1]= (rel_ymax<<shift);
-
+ 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]);
}
- pred_x= P_LEFT[0];
- pred_y= P_LEFT[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);
}else{
mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
}
-
- dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
- picture, s->p_mv_table, mv_scale, mv_penalty);
-
+
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
+
break;
}
-
- dmin= s->me.sub_motion_search(s, &mx, &my, dmin, rel_xmin, rel_ymin, rel_xmax, rel_ymax,
- pred_x, pred_y, picture, 0, 0, mv_penalty);
-
- if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
- dmin= s->me.get_mb_score(s, mx, my, pred_x, pred_y, picture, mv_penalty);
+
+ 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);
//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;
}
static inline int check_bidir_mv(MpegEncContext * s,
- int mb_x, int mb_y,
int motion_fx, int motion_fy,
int motion_bx, int motion_by,
int pred_fx, int pred_fy,
- int pred_bx, int pred_by)
+ int pred_bx, int pred_by,
+ int size, int h)
{
//FIXME optimize?
- //FIXME move into template?
//FIXME better f_code prediction (max mv & distance)
- uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
- uint8_t *dest_y = s->me.scratchpad;
+ //FIXME pointers
+ MotionEstContext * const c= &s->me;
+ uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
+ uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame
+ int stride= c->stride;
+ uint8_t *dest_y = c->scratchpad;
uint8_t *ptr;
int dxy;
int src_x, src_y;
int fbmin;
+ uint8_t **src_data= c->src[0];
+ uint8_t **ref_data= c->ref[0];
+ uint8_t **ref2_data= c->ref[2];
if(s->quarter_sample){
dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
- src_x = mb_x * 16 + (motion_fx >> 2);
- src_y = mb_y * 16 + (motion_fy >> 2);
- assert(src_x >=-16 && src_x<=s->h_edge_pos);
- assert(src_y >=-16 && src_y<=s->v_edge_pos);
+ src_x = motion_fx >> 2;
+ src_y = motion_fy >> 2;
- ptr = s->last_picture.data[0] + (src_y * s->linesize) + src_x;
- s->dsp.put_qpel_pixels_tab[0][dxy](dest_y , ptr , s->linesize);
+ ptr = ref_data[0] + (src_y * stride) + src_x;
+ s->dsp.put_qpel_pixels_tab[0][dxy](dest_y , ptr , stride);
dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
- src_x = mb_x * 16 + (motion_bx >> 2);
- src_y = mb_y * 16 + (motion_by >> 2);
- assert(src_x >=-16 && src_x<=s->h_edge_pos);
- assert(src_y >=-16 && src_y<=s->v_edge_pos);
-
- ptr = s->next_picture.data[0] + (src_y * s->linesize) + src_x;
- s->dsp.avg_qpel_pixels_tab[0][dxy](dest_y , ptr , s->linesize);
+ src_x = motion_bx >> 2;
+ src_y = motion_by >> 2;
+
+ ptr = ref2_data[0] + (src_y * stride) + src_x;
+ s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y , ptr , stride);
}else{
dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
- src_x = mb_x * 16 + (motion_fx >> 1);
- src_y = mb_y * 16 + (motion_fy >> 1);
- assert(src_x >=-16 && src_x<=s->h_edge_pos);
- assert(src_y >=-16 && src_y<=s->v_edge_pos);
+ src_x = motion_fx >> 1;
+ src_y = motion_fy >> 1;
- ptr = s->last_picture.data[0] + (src_y * s->linesize) + src_x;
- s->dsp.put_pixels_tab[0][dxy](dest_y , ptr , s->linesize, 16);
+ ptr = ref_data[0] + (src_y * stride) + src_x;
+ s->dsp.put_pixels_tab[size][dxy](dest_y , ptr , stride, h);
dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
- src_x = mb_x * 16 + (motion_bx >> 1);
- src_y = mb_y * 16 + (motion_by >> 1);
- assert(src_x >=-16 && src_x<=s->h_edge_pos);
- assert(src_y >=-16 && src_y<=s->v_edge_pos);
-
- ptr = s->next_picture.data[0] + (src_y * s->linesize) + src_x;
- s->dsp.avg_pixels_tab[0][dxy](dest_y , ptr , s->linesize, 16);
+ 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);
}
- fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*s->me.mb_penalty_factor
- +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*s->me.mb_penalty_factor
- + s->dsp.mb_cmp[0](s, s->new_picture.data[0] + mb_x*16 + mb_y*16*s->linesize, dest_y, s->linesize);
-
- if(s->avctx->mb_cmp&FF_CMP_CHROMA){
+ fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
+ +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
+ + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic
+
+ if(c->avctx->mb_cmp&FF_CMP_CHROMA){
}
//FIXME CHROMA !!!
-
+
return fbmin;
}
/* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
-static inline int bidir_refine(MpegEncContext * s,
- int mb_x, int mb_y)
+static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
{
+ MotionEstContext * const c= &s->me;
const int mot_stride = s->mb_stride;
const int xy = mb_y *mot_stride + mb_x;
int fbmin;
int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];
int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];
int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];
-
- //FIXME do refinement and add flag
-
- fbmin= check_bidir_mv(s, mb_x, mb_y,
- motion_fx, motion_fy,
+ const int flags= c->sub_flags;
+ const int qpel= flags&FLAG_QPEL;
+ const int shift= 1+qpel;
+ const int xmin= c->xmin<<shift;
+ const int ymin= c->ymin<<shift;
+ const int xmax= c->xmax<<shift;
+ const int ymax= c->ymax<<shift;
+ uint8_t map[8][8][8][8];
+
+ memset(map,0,sizeof(map));
+#define BIDIR_MAP(fx,fy,bx,by) \
+ map[(motion_fx+fx)&7][(motion_fy+fy)&7][(motion_bx+bx)&7][(motion_by+by)&7]
+ BIDIR_MAP(0,0,0,0) = 1;
+
+ fbmin= check_bidir_mv(s, motion_fx, motion_fy,
motion_bx, motion_by,
pred_fx, pred_fy,
- pred_bx, pred_by);
+ pred_bx, pred_by,
+ 0, 16);
+
+ if(s->avctx->bidir_refine){
+ int score, end;
+#define CHECK_BIDIR(fx,fy,bx,by)\
+ if( !BIDIR_MAP(fx,fy,bx,by)\
+ &&(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)){\
+ BIDIR_MAP(fx,fy,bx,by) = 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){\
+ fbmin= score;\
+ motion_fx+=fx;\
+ motion_fy+=fy;\
+ motion_bx+=bx;\
+ motion_by+=by;\
+ end=0;\
+ }\
+ }
+#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)
+
+ do{
+ 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)
+ }
+ }
+ }
+ }while(!end);
+ }
- return fbmin;
+ 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;
}
-static inline int direct_search(MpegEncContext * s,
- int mb_x, int mb_y)
+static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
{
+ MotionEstContext * const c= &s->me;
int P[10][2];
const int mot_stride = s->mb_stride;
const int mot_xy = mb_y*mot_stride + mb_x;
const int time_pb= s->pb_time;
int mx, my, xmin, xmax, ymin, ymax;
int16_t (*mv_table)[2]= s->b_direct_mv_table;
- uint8_t * const mv_penalty= s->me.mv_penalty[1] + MAX_MV;
-
+
+ c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;
ymin= xmin=(-32)>>shift;
ymax= xmax= 31>>shift;
for(i=0; i<4; i++){
int index= s->block_index[i];
int min, max;
-
- s->me.co_located_mv[i][0]= s->motion_val[index][0];
- s->me.co_located_mv[i][1]= s->motion_val[index][1];
- s->me.direct_basis_mv[i][0]= s->me.co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
- s->me.direct_basis_mv[i][1]= s->me.co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
-// s->me.direct_basis_mv[1][i][0]= s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
-// s->me.direct_basis_mv[1][i][1]= s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
-
- max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;
- min= FFMIN(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;
- max+= (2*mb_x + (i& 1))*8 + 1; // +-1 is for the simpler rounding
- min+= (2*mb_x + (i& 1))*8 - 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->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
+
+ max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
+ min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
+ max+= 16*mb_x + 1; // +-1 is for the simpler rounding
+ min+= 16*mb_x - 1;
xmax= FFMIN(xmax, s->width - max);
xmin= FFMAX(xmin, - 16 - min);
- max= FFMAX(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;
- min= FFMIN(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;
- max+= (2*mb_y + (i>>1))*8 + 1; // +-1 is for the simpler rounding
- min+= (2*mb_y + (i>>1))*8 - 1;
+ max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
+ min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
+ max+= 16*mb_y + 1; // +-1 is for the simpler rounding
+ min+= 16*mb_y - 1;
ymax= FFMIN(ymax, s->height - max);
ymin= FFMAX(ymin, - 16 - min);
-
+
if(s->mv_type == MV_TYPE_16X16) break;
}
-
+
assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
-
+
if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
s->b_direct_mv_table[mot_xy][0]= 0;
s->b_direct_mv_table[mot_xy][1]= 0;
return 256*256*256*64;
}
- 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);
+ c->xmin= xmin;
+ c->ymin= ymin;
+ c->xmax= xmax;
+ c->ymax= ymax;
+ c->flags |= FLAG_DIRECT;
+ c->sub_flags |= FLAG_DIRECT;
+ c->pred_x=0;
+ c->pred_y=0;
+
+ 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 (mb_y) {
- 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]);
}
-
- //FIXME direct_search ptr in context!!! (needed for chroma anyway or this will get messy)
- if(s->flags&CODEC_FLAG_QPEL){
- dmin = simple_direct_qpel_epzs_motion_search(s, 0, &mx, &my, P, 0, 0, xmin, ymin, xmax, ymax,
- &s->last_picture, mv_table, 1<<14, mv_penalty);
- dmin = simple_direct_qpel_qpel_motion_search(s, &mx, &my, dmin, xmin, ymin, xmax, ymax,
- 0, 0, &s->last_picture, 0, 0, mv_penalty);
-
- if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
- dmin= simple_direct_qpel_qpel_get_mb_score(s, mx, my, 0, 0, &s->last_picture, mv_penalty);
- }else{
- dmin = simple_direct_hpel_epzs_motion_search(s, 0, &mx, &my, P, 0, 0, xmin, ymin, xmax, ymax,
- &s->last_picture, mv_table, 1<<15, mv_penalty);
- dmin = simple_direct_hpel_hpel_motion_search(s, &mx, &my, dmin, xmin, ymin, xmax, ymax,
- 0, 0, &s->last_picture, 0, 0, mv_penalty);
- if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)
- dmin= simple_direct_hpel_hpel_get_mb_score(s, mx, my, 0, 0, &s->last_picture, mv_penalty);
- }
+ dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
+ if(c->sub_flags&FLAG_QPEL)
+ dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
+ else
+ 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);
+
+ get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
+
+ mv_table[mot_xy][0]= mx;
+ mv_table[mot_xy][1]= my;
+ c->flags &= ~FLAG_DIRECT;
+ c->sub_flags &= ~FLAG_DIRECT;
- s->b_direct_mv_table[mot_xy][0]= mx;
- s->b_direct_mv_table[mot_xy][1]= my;
return dmin;
}
void ff_estimate_b_frame_motion(MpegEncContext * s,
int mb_x, int mb_y)
{
- const int penalty_factor= s->me.mb_penalty_factor;
- int fmin, bmin, dmin, fbmin;
+ MotionEstContext * const c= &s->me;
+ const int penalty_factor= c->mb_penalty_factor;
+ int fmin, bmin, dmin, fbmin, bimin, fimin;
int type=0;
-
- s->me.skip=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);
+
+ get_limits(s, 16*mb_x, 16*mb_y);
+
+ c->skip=0;
+
+ if(s->codec_id == 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(c->avctx->me_threshold){
+ int vard= check_input_motion(s, mb_x, mb_y, 0);
+
+ if((vard+128)>>8 < 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;
+
+// pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
+ s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
+/* pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
+ c->mb_var_sum_temp += (varc+128)>>8;*/
+ c->mc_mb_var_sum_temp += (vard+128)>>8;
+/* if (vard <= 64<<8 || vard < varc) {
+ c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
+ }else{
+ c->scene_change_score+= s->qscale * s->avctx->scenechange_factor;
+ }*/
+ return;
+ }
+ if((vard+128)>>8 < 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 == 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;
- s->me.skip=0;
- fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, &s->last_picture, s->f_code) + 3*penalty_factor;
-
- s->me.skip=0;
- bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, &s->next_picture, s->b_code) + 2*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]);
- s->me.skip=0;
+ c->skip=0;
fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
//printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);
+
+ if(s->flags & CODEC_FLAG_INTERLACED_ME){
+//FIXME mb type penalty
+ c->skip=0;
+ c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
+ fimin= 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], 0);
+ c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
+ bimin= 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], 0);
+ }else
+ fimin= bimin= INT_MAX;
+
{
int score= fmin;
- type = MB_TYPE_FORWARD;
-
+ type = CANDIDATE_MB_TYPE_FORWARD;
+
if (dmin <= score){
score = dmin;
- type = MB_TYPE_DIRECT;
+ type = CANDIDATE_MB_TYPE_DIRECT;
}
if(bmin<score){
score=bmin;
- type= MB_TYPE_BACKWARD;
+ type= CANDIDATE_MB_TYPE_BACKWARD;
}
if(fbmin<score){
score=fbmin;
- type= MB_TYPE_BIDIR;
+ type= CANDIDATE_MB_TYPE_BIDIR;
+ }
+ if(fimin<score){
+ score=fimin;
+ type= CANDIDATE_MB_TYPE_FORWARD_I;
+ }
+ if(bimin<score){
+ score=bimin;
+ type= CANDIDATE_MB_TYPE_BACKWARD_I;
}
-
+
score= ((unsigned)(score*score + 128*256))>>16;
- s->current_picture.mc_mb_var_sum += score;
+ c->mc_mb_var_sum_temp += score;
s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
}
- if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
- type= MB_TYPE_FORWARD | MB_TYPE_BACKWARD | MB_TYPE_BIDIR | MB_TYPE_DIRECT; //FIXME something smarter
- if(dmin>256*256*16) type&= ~MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB
+ if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
+ type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
+ if(fimin < INT_MAX)
+ type |= CANDIDATE_MB_TYPE_FORWARD_I;
+ if(bimin < INT_MAX)
+ type |= CANDIDATE_MB_TYPE_BACKWARD_I;
+ if(fimin < INT_MAX && bimin < INT_MAX){
+ type |= CANDIDATE_MB_TYPE_BIDIR_I;
+ }
+ //FIXME something smarter
+ 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 == 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;
+#if 0
+ if(s->out_format == FMT_MPEG1)
+ type |= CANDIDATE_MB_TYPE_INTRA;
+#endif
}
s->mb_type[mb_y*s->mb_stride + mb_x]= type;
{
if(s->me_method>=ME_EPZS){
int score[8];
- int i, y;
+ int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);
uint8_t * fcode_tab= s->fcode_tab;
int best_fcode=-1;
int best_score=-10000000;
+ if(s->msmpeg4_version)
+ range= FFMIN(range, 16);
+ else if(s->codec_id == 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);
for(y=0; y<s->mb_height; y++){
int xy= y*s->mb_stride;
for(x=0; x<s->mb_width; x++){
if(s->mb_type[xy] & type){
- int fcode= FFMAX(fcode_tab[mv_table[xy][0] + MAX_MV],
- fcode_tab[mv_table[xy][1] + MAX_MV]);
+ int mx= mv_table[xy][0];
+ int my= mv_table[xy][1];
+ int fcode= FFMAX(fcode_tab[mx + MAX_MV],
+ fcode_tab[my + MAX_MV]);
int j;
-
+
+ if(mx >= range || mx < -range ||
+ my >= range || my < -range)
+ 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])
score[j]-= 170;
xy++;
}
}
-
+
for(i=1; i<8; i++){
if(score[i] > best_score){
best_score= score[i];
void ff_fix_long_p_mvs(MpegEncContext * s)
{
+ MotionEstContext * const c= &s->me;
const int f_code= s->f_code;
int y, range;
assert(s->pict_type==P_TYPE);
- range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);
-
- if(s->msmpeg4_version) range= 16;
-
- if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;
-
- /* clip / convert to intra 16x16 type MVs */
- for(y=0; y<s->mb_height; y++){
- int x;
- int xy= y*s->mb_stride;
- for(x=0; x<s->mb_width; x++){
- if(s->mb_type[xy]&MB_TYPE_INTER){
- if( s->p_mv_table[xy][0] >=range || s->p_mv_table[xy][0] <-range
- || s->p_mv_table[xy][1] >=range || s->p_mv_table[xy][1] <-range){
- s->mb_type[xy] &= ~MB_TYPE_INTER;
- s->mb_type[xy] |= MB_TYPE_INTRA;
- s->p_mv_table[xy][0] = 0;
- s->p_mv_table[xy][1] = 0;
- }
- }
- xy++;
- }
- }
+ 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));
+
+ 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= 2+ s->mb_width*2;
+ const int wrap= s->b8_stride;
/* clip / convert to intra 8x8 type MVs */
for(y=0; y<s->mb_height; y++){
- int xy= (y*2 + 1)*wrap + 1;
+ int xy= y*2*wrap;
int i= y*s->mb_stride;
int x;
for(x=0; x<s->mb_width; x++){
- if(s->mb_type[i]&MB_TYPE_INTER4V){
+ if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){
int block;
for(block=0; block<4; block++){
int off= (block& 1) + (block>>1)*wrap;
- int mx= s->motion_val[ xy + off ][0];
- int my= s->motion_val[ 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] &= ~MB_TYPE_INTER4V;
- s->mb_type[i] |= MB_TYPE_INTRA;
+ 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;
}
}
}
}
}
-void ff_fix_long_b_mvs(MpegEncContext * s, int16_t (*mv_table)[2], int f_code, int type)
+/**
+ *
+ * @param truncate 1 for truncation, 0 for using intra
+ */
+void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
+ int16_t (*mv_table)[2], int f_code, int type, int truncate)
{
- int y;
+ MotionEstContext * const c= &s->me;
+ int y, h_range, v_range;
// RAL: 8 in MPEG-1, 16 in MPEG-4
- int range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);
-
- if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;
+ int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
+
+ if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
+
+ h_range= range;
+ v_range= field_select_table ? range>>1 : range;
/* clip / convert to intra 16x16 type MVs */
for(y=0; y<s->mb_height; y++){
int xy= y*s->mb_stride;
for(x=0; x<s->mb_width; x++){
if (s->mb_type[xy] & type){ // RAL: "type" test added...
- if( mv_table[xy][0] >=range || mv_table[xy][0] <-range
- || mv_table[xy][1] >=range || mv_table[xy][1] <-range){
-
- if(s->codec_id == CODEC_ID_MPEG1VIDEO && 0){
- }else{
- if (mv_table[xy][0] > range-1) mv_table[xy][0]= range-1;
- else if(mv_table[xy][0] < -range ) mv_table[xy][0]= -range;
- if (mv_table[xy][1] > range-1) mv_table[xy][1]= range-1;
- else if(mv_table[xy][1] < -range ) mv_table[xy][1]= -range;
+ if(field_select_table==NULL || field_select_table[xy] == field_select){
+ if( mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
+ || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
+
+ if(truncate){
+ if (mv_table[xy][0] > h_range-1) mv_table[xy][0]= h_range-1;
+ else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
+ if (mv_table[xy][1] > v_range-1) mv_table[xy][1]= v_range-1;
+ else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
+ }else{
+ s->mb_type[xy] &= ~type;
+ s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;
+ mv_table[xy][0]=
+ mv_table[xy][1]= 0;
+ }
}
}
}