X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fmotion_est.c;h=0afb697eb9c7b8738153b4048392fdb87f5f6621;hb=36ef5369ee9b336febc2c270f8718cec4476cb85;hp=82b5f766eca913c7b9b86b57ccfdabfb83a9d695;hpb=8437ff75f865cdc3849c6518409aa0ef85eb88cf;p=ffmpeg diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c index 82b5f766eca..0afb697eb9c 100644 --- a/libavcodec/motion_est.c +++ b/libavcodec/motion_est.c @@ -1,36 +1,39 @@ /* * Motion estimation - * Copyright (c) 2000,2001 Fabrice Bellard. + * Copyright (c) 2000,2001 Fabrice Bellard * Copyright (c) 2002-2004 Michael Niedermayer * + * new motion estimation (X1/EPZS) by Michael Niedermayer * - * This library is free software; you can redistribute it and/or + * This file is part of Libav. + * + * Libav is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. + * version 2.1 of the License, or (at your option) any later version. * - * This library is distributed in the hope that it will be useful, + * Libav is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software + * License along with Libav; 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 */ /** - * @file motion_est.c + * @file * Motion estimation. */ #include #include #include +#include "libavutil/intmath.h" #include "avcodec.h" #include "dsputil.h" +#include "mathops.h" #include "mpegvideo.h" #undef NDEBUG @@ -49,7 +52,7 @@ static inline int sad_hpel_motion_search(MpegEncContext * s, int src_index, int ref_index, int size, int h); -static inline int update_map_generation(MotionEstContext *c) +static inline unsigned update_map_generation(MotionEstContext *c) { c->map_generation+= 1<<(ME_MAP_MV_BITS*2); if(c->map_generation==0){ @@ -101,22 +104,18 @@ static int get_flags(MotionEstContext *c, int direct, int chroma){ + (chroma ? FLAG_CHROMA : 0); } -static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby, +static av_always_inline int cmp_direct_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby, const int size, const int h, int ref_index, int src_index, - me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){ + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel){ MotionEstContext * const c= &s->me; const int stride= c->stride; - const int uvstride= c->uvstride; - const int qpel= flags&FLAG_QPEL; - const int chroma= flags&FLAG_CHROMA; - const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel? const int hx= subx + (x<<(1+qpel)); const int hy= suby + (y<<(1+qpel)); uint8_t * const * const ref= c->ref[ref_index]; uint8_t * const * const src= c->src[src_index]; int d; //FIXME check chroma 4mv, (no crashes ...) - if(flags&FLAG_DIRECT){ + assert(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)); if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){ const int time_pp= s->pp_time; const int time_pb= s->pb_time; @@ -174,7 +173,22 @@ static always_inline int cmp(MpegEncContext *s, const int x, const int y, const d = cmp_func(s, c->temp, src[0], stride, 16); }else d= 256*256*256*32; - }else{ + return d; +} + +static av_always_inline int cmp_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby, + const int size, const int h, int ref_index, int src_index, + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel, int chroma){ + MotionEstContext * const c= &s->me; + const int stride= c->stride; + const int uvstride= c->uvstride; + const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel? + const int hx= subx + (x<<(1+qpel)); + const int hy= suby + (y<<(1+qpel)); + uint8_t * const * const ref= c->ref[ref_index]; + uint8_t * const * const src= c->src[src_index]; + int d; + //FIXME check chroma 4mv, (no crashes ...) int uvdxy; /* no, it might not be used uninitialized */ if(dxy){ if(qpel){ @@ -205,16 +219,71 @@ static always_inline int cmp(MpegEncContext *s, const int x, const int y, const d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1); d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1); } + return d; +} + +static int cmp_simple(MpegEncContext *s, const int x, const int y, + int ref_index, int src_index, + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){ + return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0); +} + +static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y, + const int size, const int h, int ref_index, int src_index, + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){ + if(flags&FLAG_DIRECT){ + return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL); + }else{ + return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA); } -#if 0 - if(full_pel){ - const int index= (((y)<mv_penalty[hx - c->pred_x] + c->mv_penalty[hy - c->pred_y])*c->penalty_factor; -#endif - return d; +/** @brief compares a block (either a full macroblock or a partition thereof) + against a proposed motion-compensated prediction of that block + */ +static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby, + const int size, const int h, int ref_index, int src_index, + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){ + if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size) + && av_builtin_constant_p(subx) && av_builtin_constant_p(suby) + && flags==0 && h==16 && size==0 && subx==0 && suby==0){ + return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func); + }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby) + && subx==0 && suby==0){ + return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags); + }else{ + return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags); + } +} + +static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby, + const int size, const int h, int ref_index, int src_index, + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){ + if(flags&FLAG_DIRECT){ + return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0); + }else{ + return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA); + } +} + +static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby, + const int size, const int h, int ref_index, int src_index, + me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){ + if(flags&FLAG_DIRECT){ + return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1); + }else{ + return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA); + } } #include "motion_est_template.c" @@ -226,10 +295,27 @@ static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){ } -void ff_init_me(MpegEncContext *s){ +int ff_init_me(MpegEncContext *s){ MotionEstContext * const c= &s->me; + int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255); + + if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -ME_MAP_SIZE){ + av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n"); + return -1; + } + //special case of snow is needed because snow uses its own iterative ME code + if(s->me_method!=ME_ZERO && s->me_method!=ME_EPZS && s->me_method!=ME_X1 && s->avctx->codec_id != AV_CODEC_ID_SNOW){ + av_log(s->avctx, AV_LOG_ERROR, "me_method is only allowed to be set to zero and epzs; for hex,umh,full and others see dia_size\n"); + return -1; + } + c->avctx= s->avctx; + if(cache_size < 2*dia_size && !c->stride){ + av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n"); + } + ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp); ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp); ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp); @@ -267,8 +353,10 @@ void ff_init_me(MpegEncContext *s){ c->uvstride= 8*s->mb_width + 16; } - // 8x8 fullpel search would need a 4x4 chroma compare, which we dont have yet, and even if we had the motion estimation code doesnt expect it - if(s->codec_id != CODEC_ID_SNOW){ + /* 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 != AV_CODEC_ID_SNOW){ if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){ s->dsp.me_cmp[2]= zero_cmp; } @@ -279,36 +367,12 @@ void ff_init_me(MpegEncContext *s){ c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel; } - if(s->codec_id == CODEC_ID_H261){ + if(s->codec_id == AV_CODEC_ID_H261){ c->sub_motion_search= no_sub_motion_search; } - c->temp= c->scratchpad; -} - -#if 0 -static int pix_dev(uint8_t * pix, int line_size, int mean) -{ - int s, i, j; - - s = 0; - for (i = 0; i < 16; i++) { - for (j = 0; j < 16; j += 8) { - s += ABS(pix[0]-mean); - s += ABS(pix[1]-mean); - s += ABS(pix[2]-mean); - s += ABS(pix[3]-mean); - s += ABS(pix[4]-mean); - s += ABS(pix[5]-mean); - s += ABS(pix[6]-mean); - s += ABS(pix[7]-mean); - pix += 8; - } - pix += line_size - 16; - } - return s; + return 0; } -#endif static inline void no_motion_search(MpegEncContext * s, int *mx_ptr, int *my_ptr) @@ -317,229 +381,6 @@ static inline void no_motion_search(MpegEncContext * s, *my_ptr = 16 * s->mb_y; } -#if 0 /* the use of these functions is inside #if 0 */ -static int full_motion_search(MpegEncContext * s, - int *mx_ptr, int *my_ptr, int range, - int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture) -{ - int x1, y1, x2, y2, xx, yy, x, y; - int mx, my, dmin, d; - uint8_t *pix; - - xx = 16 * s->mb_x; - yy = 16 * s->mb_y; - x1 = xx - range + 1; /* we loose one pixel to avoid boundary pb with half pixel pred */ - if (x1 < xmin) - x1 = xmin; - x2 = xx + range - 1; - if (x2 > xmax) - x2 = xmax; - y1 = yy - range + 1; - if (y1 < ymin) - y1 = ymin; - y2 = yy + range - 1; - if (y2 > ymax) - y2 = ymax; - pix = s->new_picture.data[0] + (yy * s->linesize) + xx; - dmin = 0x7fffffff; - mx = 0; - my = 0; - for (y = y1; y <= y2; y++) { - for (x = x1; x <= x2; x++) { - d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, - s->linesize, 16); - if (d < dmin || - (d == dmin && - (abs(x - xx) + abs(y - yy)) < - (abs(mx - xx) + abs(my - yy)))) { - dmin = d; - mx = x; - my = y; - } - } - } - - *mx_ptr = mx; - *my_ptr = my; - -#if 0 - if (*mx_ptr < -(2 * range) || *mx_ptr >= (2 * range) || - *my_ptr < -(2 * range) || *my_ptr >= (2 * range)) { - av_log(NULL, AV_LOG_ERROR, "error %d %d\n", *mx_ptr, *my_ptr); - } -#endif - return dmin; -} - - -static int log_motion_search(MpegEncContext * s, - int *mx_ptr, int *my_ptr, int range, - int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture) -{ - int x1, y1, x2, y2, xx, yy, x, y; - int mx, my, dmin, d; - uint8_t *pix; - - xx = s->mb_x << 4; - yy = s->mb_y << 4; - - /* Left limit */ - x1 = xx - range; - if (x1 < xmin) - x1 = xmin; - - /* Right limit */ - x2 = xx + range; - if (x2 > xmax) - x2 = xmax; - - /* Upper limit */ - y1 = yy - range; - if (y1 < ymin) - y1 = ymin; - - /* Lower limit */ - y2 = yy + range; - if (y2 > ymax) - y2 = ymax; - - pix = s->new_picture.data[0] + (yy * s->linesize) + xx; - dmin = 0x7fffffff; - mx = 0; - my = 0; - - do { - for (y = y1; y <= y2; y += range) { - for (x = x1; x <= x2; x += range) { - d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16); - if (d < dmin || (d == dmin && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) { - dmin = d; - mx = x; - my = y; - } - } - } - - range = range >> 1; - - x1 = mx - range; - if (x1 < xmin) - x1 = xmin; - - x2 = mx + range; - if (x2 > xmax) - x2 = xmax; - - y1 = my - range; - if (y1 < ymin) - y1 = ymin; - - y2 = my + range; - if (y2 > ymax) - y2 = ymax; - - } while (range >= 1); - -#ifdef DEBUG - av_log(s->avctx, AV_LOG_DEBUG, "log - MX: %d\tMY: %d\n", mx, my); -#endif - *mx_ptr = mx; - *my_ptr = my; - return dmin; -} - -static int phods_motion_search(MpegEncContext * s, - int *mx_ptr, int *my_ptr, int range, - int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture) -{ - int x1, y1, x2, y2, xx, yy, x, y, lastx, d; - int mx, my, dminx, dminy; - uint8_t *pix; - - xx = s->mb_x << 4; - yy = s->mb_y << 4; - - /* Left limit */ - x1 = xx - range; - if (x1 < xmin) - x1 = xmin; - - /* Right limit */ - x2 = xx + range; - if (x2 > xmax) - x2 = xmax; - - /* Upper limit */ - y1 = yy - range; - if (y1 < ymin) - y1 = ymin; - - /* Lower limit */ - y2 = yy + range; - if (y2 > ymax) - y2 = ymax; - - pix = s->new_picture.data[0] + (yy * s->linesize) + xx; - mx = 0; - my = 0; - - x = xx; - y = yy; - do { - dminx = 0x7fffffff; - dminy = 0x7fffffff; - - lastx = x; - for (x = x1; x <= x2; x += range) { - d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16); - if (d < dminx || (d == dminx && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) { - dminx = d; - mx = x; - } - } - - x = lastx; - for (y = y1; y <= y2; y += range) { - d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16); - if (d < dminy || (d == dminy && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) { - dminy = d; - my = y; - } - } - - range = range >> 1; - - x = mx; - y = my; - x1 = mx - range; - if (x1 < xmin) - x1 = xmin; - - x2 = mx + range; - if (x2 > xmax) - x2 = xmax; - - y1 = my - range; - if (y1 < ymin) - y1 = ymin; - - y2 = my + range; - if (y2 > ymax) - y2 = ymax; - - } while (range >= 1); - -#ifdef DEBUG - av_log(s->avctx, AV_LOG_DEBUG, "phods - MX: %d\tMY: %d\n", mx, my); -#endif - - /* half pixel search */ - *mx_ptr = mx; - *my_ptr = my; - return dminy; -} -#endif /* 0 */ - #define Z_THRESHOLD 256 #define CHECK_SAD_HALF_MV(suffix, x, y) \ @@ -668,16 +509,16 @@ static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4) if(mv4){ int mot_xy= s->block_index[0]; - s->current_picture.motion_val[0][mot_xy ][0]= mx; - s->current_picture.motion_val[0][mot_xy ][1]= my; - s->current_picture.motion_val[0][mot_xy+1][0]= mx; - s->current_picture.motion_val[0][mot_xy+1][1]= my; + s->current_picture.f.motion_val[0][mot_xy ][0] = mx; + s->current_picture.f.motion_val[0][mot_xy ][1] = my; + s->current_picture.f.motion_val[0][mot_xy + 1][0] = mx; + s->current_picture.f.motion_val[0][mot_xy + 1][1] = my; mot_xy += s->b8_stride; - s->current_picture.motion_val[0][mot_xy ][0]= mx; - s->current_picture.motion_val[0][mot_xy ][1]= my; - s->current_picture.motion_val[0][mot_xy+1][0]= mx; - s->current_picture.motion_val[0][mot_xy+1][1]= my; + s->current_picture.f.motion_val[0][mot_xy ][0] = mx; + s->current_picture.f.motion_val[0][mot_xy ][1] = my; + s->current_picture.f.motion_val[0][mot_xy + 1][0] = mx; + s->current_picture.f.motion_val[0][mot_xy + 1][1] = my; } } @@ -687,6 +528,7 @@ static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4) static inline void get_limits(MpegEncContext *s, int x, int y) { MotionEstContext * const c= &s->me; + int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL)); /* if(c->avctx->me_range) c->range= c->avctx->me_range >> 1; else c->range= 16; @@ -708,6 +550,12 @@ static inline void get_limits(MpegEncContext *s, int x, int 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); + } } static inline void init_mv4_ref(MotionEstContext *c){ @@ -743,8 +591,8 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) const int mot_stride = s->b8_stride; const int mot_xy = s->block_index[block]; - P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0]; - P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1]; + P_LEFT[0] = s->current_picture.f.motion_val[0][mot_xy - 1][0]; + P_LEFT[1] = s->current_picture.f.motion_val[0][mot_xy - 1][1]; if(P_LEFT[0] > (c->xmax<xmax<pred_x= pred_x4= P_LEFT[0]; c->pred_y= pred_y4= P_LEFT[1]; } else { - 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]; + P_TOP[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][0]; + P_TOP[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][1]; + P_TOPRIGHT[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + off[block]][0]; + P_TOPRIGHT[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + off[block]][1]; if(P_TOP[1] > (c->ymax<ymax<xmin<xmin< (c->xmax<xmax<current_picture.motion_val[0][ s->block_index[block] ][0]= mx4; - s->current_picture.motion_val[0][ s->block_index[block] ][1]= my4; + s->current_picture.f.motion_val[0][s->block_index[block]][0] = mx4; + s->current_picture.f.motion_val[0][s->block_index[block]][1] = my4; if(mx4 != mx || my4 != my) same=0; } @@ -818,7 +666,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) return INT_MAX; if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){ - dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16); + dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16); } if(c->avctx->mb_cmp&FF_CMP_CHROMA){ @@ -833,15 +681,15 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift) offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize; if(s->no_rounding){ - s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8); - s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8); + s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.f.data[1] + offset, s->uvlinesize, 8); + s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8); }else{ - s->dsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.data[1] + offset, s->uvlinesize, 8); - s->dsp.put_pixels_tab [1][dxy](c->scratchpad+8 , s->last_picture.data[2] + offset, s->uvlinesize, 8); + s->dsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.f.data[1] + offset, s->uvlinesize, 8); + s->dsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8); } - dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad , s->uvlinesize, 8); - dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8); + dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.f.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad , s->uvlinesize, 8); + dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.f.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8); } c->pred_x= mx; @@ -1007,7 +855,7 @@ static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int 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 mb_type= s->current_picture.f.mb_type[mb_xy]; int flags= c->flags; int shift= (flags&FLAG_QPEL) + 1; int mask= (1<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)); + clip_input_mv(s, p->f.motion_val[0][xy], !!IS_INTERLACED(mb_type)); + clip_input_mv(s, p->f.motion_val[1][xy], !!IS_INTERLACED(mb_type)); } if(IS_INTERLACED(mb_type)){ @@ -1040,8 +888,8 @@ static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int } if(USES_LIST(mb_type, 0)){ - int field_select0= p->ref_index[0][xy ]; - int field_select1= p->ref_index[0][xy2]; + int field_select0= p->f.ref_index[0][4*mb_xy ]; + int field_select1= p->f.ref_index[0][4*mb_xy+2]; assert(field_select0==0 ||field_select0==1); assert(field_select1==0 ||field_select1==1); init_interlaced_ref(s, 0); @@ -1049,46 +897,46 @@ static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int 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]; + *(uint32_t*)s->p_field_mv_table[0][field_select0][mb_xy] = *(uint32_t*)p->f.motion_val[0][xy ]; + *(uint32_t*)s->p_field_mv_table[1][field_select1][mb_xy] = *(uint32_t*)p->f.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]; + *(uint32_t*)s->b_field_mv_table[0][0][field_select0][mb_xy] = *(uint32_t*)p->f.motion_val[0][xy ]; + *(uint32_t*)s->b_field_mv_table[0][1][field_select1][mb_xy] = *(uint32_t*)p->f.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]; + x = p->f.motion_val[0][xy ][0]; + y = p->f.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]; + x = p->f.motion_val[0][xy2][0]; + y = p->f.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]; + int field_select0 = p->f.ref_index[1][4 * mb_xy ]; + int field_select1 = p->f.ref_index[1][4 * mb_xy + 2]; 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]; + *(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy] = *(uint32_t*)p->f.motion_val[1][xy ]; + *(uint32_t*)s->b_field_mv_table[1][1][field_select1][mb_xy] = *(uint32_t*)p->f.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]; + x = p->f.motion_val[1][xy ][0]; + y = p->f.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]; + x = p->f.motion_val[1][xy2][0]; + y = p->f.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 } @@ -1104,33 +952,33 @@ static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int 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]; + x= p->f.motion_val[0][xy][0]; + y= p->f.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]; + *(uint32_t*)s->p_mv_table[mb_xy] = *(uint32_t*)p->f.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]; + *(uint32_t*)s->b_bidir_forw_mv_table[mb_xy] = *(uint32_t*)p->f.motion_val[0][xy]; + *(uint32_t*)s->b_bidir_back_mv_table[mb_xy] = *(uint32_t*)p->f.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]; + *(uint32_t*)s->b_forw_mv_table[mb_xy] = *(uint32_t*)p->f.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]; + x = p->f.motion_val[0][xy][0]; + y = p->f.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]; + *(uint32_t*)s->b_back_mv_table[mb_xy] = *(uint32_t*)p->f.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]; + x = p->f.motion_val[1][xy][0]; + y = p->f.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; @@ -1143,13 +991,15 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, { MotionEstContext * const c= &s->me; uint8_t *pix, *ppix; - int sum, varc, vard, mx, my, dmin; + int sum, mx, my, dmin; + int varc; ///< the variance of the block (sum of squared (p[y][x]-average)) + int vard; ///< sum of squared differences with the estimated motion vector int P[10][2]; const int shift= 1+s->quarter_sample; int mb_type=0; Picture * const pic= &s->current_picture; - init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0); + init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0); assert(s->quarter_sample==0 || s->quarter_sample==1); assert(s->linesize == c->stride); @@ -1166,7 +1016,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, /* 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; + 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; @@ -1176,13 +1026,11 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, 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; - if (vard <= 64<<8 || vard < varc) { //FIXME - c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc); - }else{ - c->scene_change_score+= s->qscale * s->avctx->scenechange_factor; - } + c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score); return; } if((vard+128)>>8 < c->avctx->mb_threshold) @@ -1197,39 +1045,22 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, my-= mb_y*16; dmin = 0; break; -#if 0 - case ME_FULL: - dmin = full_motion_search(s, &mx, &my, range, ref_picture); - mx-= mb_x*16; - my-= mb_y*16; - break; - case ME_LOG: - dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture); - mx-= mb_x*16; - my-= mb_y*16; - break; - case ME_PHODS: - dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture); - mx-= mb_x*16; - my-= mb_y*16; - break; -#endif case ME_X1: case ME_EPZS: { const int mot_stride = s->b8_stride; const int mot_xy = s->block_index[0]; - P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0]; - P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1]; + P_LEFT[0] = s->current_picture.f.motion_val[0][mot_xy - 1][0]; + P_LEFT[1] = s->current_picture.f.motion_val[0][mot_xy - 1][1]; if(P_LEFT[0] > (c->xmax<xmax<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]; + P_TOP[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][0]; + P_TOP[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][1]; + P_TOPRIGHT[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + 2][0]; + P_TOPRIGHT[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + 2][1]; if(P_TOP[1] > (c->ymax<ymax<xmin<xmin< (c->ymax<ymax<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); -#endif if(mb_type){ - 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; + 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); @@ -1290,14 +1116,14 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1); } }else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){ - if (vard <= 64<<8 || vard < varc) - c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc); - else - c->scene_change_score+= s->qscale * s->avctx->scenechange_factor; + 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){ + 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) @@ -1352,7 +1178,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){ intra_score= varc - 500; }else{ - int mean= (sum+128)>>8; + unsigned mean = (sum+128)>>8; mean*= 0x01010101; for(i=0; i<16; i++){ @@ -1364,42 +1190,18 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, intra_score= s->dsp.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16); } -#if 0 //FIXME - /* get chroma score */ - if(c->avctx->mb_cmp&FF_CMP_CHROMA){ - for(i=1; i<3; i++){ - uint8_t *dest_c; - int mean; - - if(s->out_format == FMT_H263){ - mean= (s->dc_val[i][mb_x + mb_y*s->b8_stride] + 4)>>3; //FIXME not exact but simple ;) - }else{ - mean= (s->last_dc[i] + 4)>>3; - } - dest_c = s->new_picture.data[i] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; - - mean*= 0x01010101; - for(i=0; i<8; i++){ - *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 0]) = mean; - *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 4]) = mean; - } - - intra_score+= s->dsp.mb_cmp[1](s, c->scratchpad, dest_c, s->uvlinesize); - } - } -#endif intra_score += c->mb_penalty_factor*16; if(intra_score < dmin){ mb_type= CANDIDATE_MB_TYPE_INTRA; - s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup + s->current_picture.f.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup }else - s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0; + s->current_picture.f.mb_type[mb_y*s->mb_stride + mb_x] = 0; - if (vard <= 64<<8 || vard < varc) { //FIXME - c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc); - }else{ - c->scene_change_score+= s->qscale * s->avctx->scenechange_factor; + { + 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); } } @@ -1414,7 +1216,7 @@ int ff_pre_estimate_p_frame_motion(MpegEncContext * s, int P[10][2]; const int shift= 1+s->quarter_sample; const int xy= mb_x + mb_y*s->mb_stride; - init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0); + init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0); assert(s->quarter_sample==0 || s->quarter_sample==1); @@ -1486,47 +1288,28 @@ static int ff_estimate_motion_b(MpegEncContext * s, mx-= mb_x*16; my-= mb_y*16; break; -#if 0 - case ME_FULL: - dmin = full_motion_search(s, &mx, &my, range, ref_picture); - mx-= mb_x*16; - my-= mb_y*16; - break; - case ME_LOG: - dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture); - mx-= mb_x*16; - my-= mb_y*16; - break; - case ME_PHODS: - dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture); - mx-= mb_x*16; - my-= mb_y*16; - break; -#endif case ME_X1: case ME_EPZS: - { - P_LEFT[0] = mv_table[mot_xy - 1][0]; - P_LEFT[1] = mv_table[mot_xy - 1][1]; - - if(P_LEFT[0] > (c->xmax<xmax<first_slice_line) { - P_TOP[0] = mv_table[mot_xy - mot_stride ][0]; - P_TOP[1] = mv_table[mot_xy - mot_stride ][1]; - P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1 ][0]; - P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1 ][1]; - if(P_TOP[1] > (c->ymax<ymax<xmin<xmin< (c->ymax<ymax< (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift); - P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); - P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); - } - c->pred_x= P_LEFT[0]; - c->pred_y= P_LEFT[1]; + /* special case for first line */ + if (!s->first_slice_line) { + P_TOP[0] = mv_table[mot_xy - mot_stride ][0]; + P_TOP[1] = mv_table[mot_xy - mot_stride ][1]; + P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0]; + P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1]; + if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift); + if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift); + if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift); + + P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); + P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); } + c->pred_x = P_LEFT[0]; + c->pred_y = P_LEFT[1]; if(mv_table == s->b_forw_mv_table){ mv_scale= (s->pb_time<<16) / (s->pp_time<ymin<xmax<ymax<avctx->bidir_refine){ - int score, end; + int end; + static const uint8_t limittab[5]={0,8,32,64,80}; + const int limit= limittab[s->avctx->bidir_refine]; + static const int8_t vect[][4]={ +{ 0, 0, 0, 1}, { 0, 0, 0,-1}, { 0, 0, 1, 0}, { 0, 0,-1, 0}, { 0, 1, 0, 0}, { 0,-1, 0, 0}, { 1, 0, 0, 0}, {-1, 0, 0, 0}, + +{ 0, 0, 1, 1}, { 0, 0,-1,-1}, { 0, 1, 1, 0}, { 0,-1,-1, 0}, { 1, 1, 0, 0}, {-1,-1, 0, 0}, { 1, 0, 0, 1}, {-1, 0, 0,-1}, +{ 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0}, +{ 0, 0,-1, 1}, { 0, 0, 1,-1}, { 0,-1, 1, 0}, { 0, 1,-1, 0}, {-1, 1, 0, 0}, { 1,-1, 0, 0}, { 1, 0, 0,-1}, {-1, 0, 0, 1}, +{ 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0}, + +{ 0, 1, 1, 1}, { 0,-1,-1,-1}, { 1, 1, 1, 0}, {-1,-1,-1, 0}, { 1, 1, 0, 1}, {-1,-1, 0,-1}, { 1, 0, 1, 1}, {-1, 0,-1,-1}, +{ 0,-1, 1, 1}, { 0, 1,-1,-1}, {-1, 1, 1, 0}, { 1,-1,-1, 0}, { 1, 1, 0,-1}, {-1,-1, 0, 1}, { 1, 0,-1, 1}, {-1, 0, 1,-1}, +{ 0, 1,-1, 1}, { 0,-1, 1,-1}, { 1,-1, 1, 0}, {-1, 1,-1, 0}, {-1, 1, 0, 1}, { 1,-1, 0,-1}, { 1, 0, 1,-1}, {-1, 0,-1, 1}, +{ 0, 1, 1,-1}, { 0,-1,-1, 1}, { 1, 1,-1, 0}, {-1,-1, 1, 0}, { 1,-1, 0, 1}, {-1, 1, 0,-1}, {-1, 0, 1, 1}, { 1, 0,-1,-1}, + +{ 1, 1, 1, 1}, {-1,-1,-1,-1}, +{ 1, 1, 1,-1}, {-1,-1,-1, 1}, { 1, 1,-1, 1}, {-1,-1, 1,-1}, { 1,-1, 1, 1}, {-1, 1,-1,-1}, {-1, 1, 1, 1}, { 1,-1,-1,-1}, +{ 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1}, + }; + static const uint8_t hash[]={ +HASH8( 0, 0, 0, 1), HASH8( 0, 0, 0,-1), HASH8( 0, 0, 1, 0), HASH8( 0, 0,-1, 0), HASH8( 0, 1, 0, 0), HASH8( 0,-1, 0, 0), HASH8( 1, 0, 0, 0), HASH8(-1, 0, 0, 0), + +HASH8( 0, 0, 1, 1), HASH8( 0, 0,-1,-1), HASH8( 0, 1, 1, 0), HASH8( 0,-1,-1, 0), HASH8( 1, 1, 0, 0), HASH8(-1,-1, 0, 0), HASH8( 1, 0, 0, 1), HASH8(-1, 0, 0,-1), +HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0), +HASH8( 0, 0,-1, 1), HASH8( 0, 0, 1,-1), HASH8( 0,-1, 1, 0), HASH8( 0, 1,-1, 0), HASH8(-1, 1, 0, 0), HASH8( 1,-1, 0, 0), HASH8( 1, 0, 0,-1), HASH8(-1, 0, 0, 1), +HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0), + +HASH8( 0, 1, 1, 1), HASH8( 0,-1,-1,-1), HASH8( 1, 1, 1, 0), HASH8(-1,-1,-1, 0), HASH8( 1, 1, 0, 1), HASH8(-1,-1, 0,-1), HASH8( 1, 0, 1, 1), HASH8(-1, 0,-1,-1), +HASH8( 0,-1, 1, 1), HASH8( 0, 1,-1,-1), HASH8(-1, 1, 1, 0), HASH8( 1,-1,-1, 0), HASH8( 1, 1, 0,-1), HASH8(-1,-1, 0, 1), HASH8( 1, 0,-1, 1), HASH8(-1, 0, 1,-1), +HASH8( 0, 1,-1, 1), HASH8( 0,-1, 1,-1), HASH8( 1,-1, 1, 0), HASH8(-1, 1,-1, 0), HASH8(-1, 1, 0, 1), HASH8( 1,-1, 0,-1), HASH8( 1, 0, 1,-1), HASH8(-1, 0,-1, 1), +HASH8( 0, 1, 1,-1), HASH8( 0,-1,-1, 1), HASH8( 1, 1,-1, 0), HASH8(-1,-1, 1, 0), HASH8( 1,-1, 0, 1), HASH8(-1, 1, 0,-1), HASH8(-1, 0, 1, 1), HASH8( 1, 0,-1,-1), + +HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1), +HASH8( 1, 1, 1,-1), HASH8(-1,-1,-1, 1), HASH8( 1, 1,-1, 1), HASH8(-1,-1, 1,-1), HASH8( 1,-1, 1, 1), HASH8(-1, 1,-1,-1), HASH8(-1, 1, 1, 1), HASH8( 1,-1,-1,-1), +HASH8( 1, 1,-1,-1), HASH8(-1,-1, 1, 1), HASH8( 1,-1,-1, 1), HASH8(-1, 1, 1,-1), HASH8( 1,-1, 1,-1), HASH8(-1, 1,-1, 1), +}; + #define CHECK_BIDIR(fx,fy,bx,by)\ - if( !BIDIR_MAP(fx,fy,bx,by)\ + if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\ &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\ &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\ - BIDIR_MAP(fx,fy,bx,by) = 1;\ + int score;\ + map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\ score= check_bidir_mv(s, motion_fx+fx, motion_fy+fy, motion_bx+bx, motion_by+by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);\ if(score < fbmin){\ + hashidx += HASH(fx,fy,bx,by);\ fbmin= score;\ motion_fx+=fx;\ motion_fy+=fy;\ @@ -1670,36 +1492,47 @@ static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y) } #define CHECK_BIDIR2(a,b,c,d)\ CHECK_BIDIR(a,b,c,d)\ -CHECK_BIDIR(-a,-b,-c,-d) - -#define CHECK_BIDIRR(a,b,c,d)\ -CHECK_BIDIR2(a,b,c,d)\ -CHECK_BIDIR2(b,c,d,a)\ -CHECK_BIDIR2(c,d,a,b)\ -CHECK_BIDIR2(d,a,b,c) +CHECK_BIDIR(-(a),-(b),-(c),-(d)) do{ + int i; + int borderdist=0; end=1; - CHECK_BIDIRR( 0, 0, 0, 1) - if(s->avctx->bidir_refine > 1){ - CHECK_BIDIRR( 0, 0, 1, 1) - CHECK_BIDIR2( 0, 1, 0, 1) - CHECK_BIDIR2( 1, 0, 1, 0) - CHECK_BIDIRR( 0, 0,-1, 1) - CHECK_BIDIR2( 0,-1, 0, 1) - CHECK_BIDIR2(-1, 0, 1, 0) - if(s->avctx->bidir_refine > 2){ - CHECK_BIDIRR( 0, 1, 1, 1) - CHECK_BIDIRR( 0,-1, 1, 1) - CHECK_BIDIRR( 0, 1,-1, 1) - CHECK_BIDIRR( 0, 1, 1,-1) - if(s->avctx->bidir_refine > 3){ - CHECK_BIDIR2( 1, 1, 1, 1) - CHECK_BIDIRR( 1, 1, 1,-1) - CHECK_BIDIR2( 1, 1,-1,-1) - CHECK_BIDIR2( 1,-1,-1, 1) - CHECK_BIDIR2( 1,-1, 1,-1) + CHECK_BIDIR2(0,0,0,1) + CHECK_BIDIR2(0,0,1,0) + CHECK_BIDIR2(0,1,0,0) + CHECK_BIDIR2(1,0,0,0) + + for(i=8; i>shift; ymax= xmax= 31>>shift; - if(IS_8X8(s->next_picture.mb_type[mot_xy])){ + if (IS_8X8(s->next_picture.f.mb_type[mot_xy])) { s->mv_type= MV_TYPE_8X8; }else{ s->mv_type= MV_TYPE_16X16; @@ -1741,8 +1574,8 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) int index= s->block_index[i]; int min, max; - c->co_located_mv[i][0]= s->next_picture.motion_val[0][index][0]; - c->co_located_mv[i][1]= s->next_picture.motion_val[0][index][1]; + c->co_located_mv[i][0] = s->next_picture.f.motion_val[0][index][0]; + c->co_located_mv[i][1] = s->next_picture.f.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); @@ -1783,15 +1616,15 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y) c->pred_x=0; c->pred_y=0; - P_LEFT[0] = clip(mv_table[mot_xy - 1][0], xmin<first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped - P_TOP[0] = clip(mv_table[mot_xy - mot_stride ][0], xmin<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<?min/max, maybe not needed - s->b_direct_mv_table[mot_xy][0]= mx; - s->b_direct_mv_table[mot_xy][1]= my; + mv_table[mot_xy][0]= mx; + mv_table[mot_xy][1]= my; c->flags &= ~FLAG_DIRECT; c->sub_flags &= ~FLAG_DIRECT; @@ -1824,11 +1657,24 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, int fmin, bmin, dmin, fbmin, bimin, fimin; int type=0; const int xy = mb_y*s->mb_stride + mb_x; - init_ref(c, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2); + init_ref(c, s->new_picture.f.data, s->last_picture.f.data, + s->next_picture.f.data, 16 * mb_x, 16 * mb_y, 2); get_limits(s, 16*mb_x, 16*mb_y); c->skip=0; + + if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.f.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); @@ -1880,7 +1726,7 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, } } - if (s->codec_id == CODEC_ID_MPEG4) + if (s->codec_id == AV_CODEC_ID_MPEG4) dmin= direct_search(s, mb_x, mb_y); else dmin= INT_MAX; @@ -1950,11 +1796,9 @@ void ff_estimate_b_frame_motion(MpegEncContext * s, type |= CANDIDATE_MB_TYPE_BIDIR_I; } //FIXME something smarter - if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB -#if 0 - if(s->out_format == FMT_MPEG1) - type |= CANDIDATE_MB_TYPE_INTRA; -#endif + if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB + if(s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT && s->flags&CODEC_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy]) + type |= CANDIDATE_MB_TYPE_DIRECT0; } s->mb_type[mb_y*s->mb_stride + mb_x]= type; @@ -1972,7 +1816,7 @@ int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type) if(s->msmpeg4_version) range= FFMIN(range, 16); - else if(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL) + else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL) range= FFMIN(range, 256); for(i=0; i<8; i++) score[i]= s->mb_num*(8-i); @@ -1993,7 +1837,7 @@ int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type) continue; for(j=0; jpict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy]) + if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy]) score[j]-= 170; } } @@ -2025,12 +1869,12 @@ 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); + assert(s->pict_type==AV_PICTURE_TYPE_P); range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code); assert(range <= 16 || !s->msmpeg4_version); - assert(range <=256 || !(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)); + assert(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)); if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range; @@ -2049,14 +1893,14 @@ void ff_fix_long_p_mvs(MpegEncContext * s) int block; for(block=0; block<4; block++){ int off= (block& 1) + (block>>1)*wrap; - int mx= s->current_picture.motion_val[0][ xy + off ][0]; - int my= s->current_picture.motion_val[0][ xy + off ][1]; + int mx = s->current_picture.f.motion_val[0][ xy + off ][0]; + int my = s->current_picture.f.motion_val[0][ xy + off ][1]; if( mx >=range || mx <-range || my >=range || my <-range){ s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V; s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA; - s->current_picture.mb_type[i]= CANDIDATE_MB_TYPE_INTRA; + s->current_picture.f.mb_type[i] = CANDIDATE_MB_TYPE_INTRA; } } }