nengel@2: #include "config.h"
nengel@2: #include "h264.h"
nengel@2: #include "h264_misc.h"
nengel@2: #include <math.h>
nengel@2: 
nengel@2: H264Context *get_h264dec_context(const char *file_name, int ifile, int ofile, int width, int height, h264_options *opts){
nengel@2:     int i;
nengel@2:     const int mb_height = (height + 15) / 16;
nengel@2:     const int mb_width  = (width  + 15) / 16;
nengel@2:     const int mb_stride = ((mb_width+1)/16 + 1) *16; //align mb_stride to 16
nengel@2: 
nengel@2:     ff_init_cabac_states();
nengel@2: 
nengel@2:     H264Context *h= av_mallocz(sizeof(H264Context));
nengel@2: 
nengel@2:     start_timer(h, TOTAL);
nengel@2:     h->file_name = file_name;
nengel@2:     h->profile = opts->profile;
nengel@2:     for (i=0; i<PROFILE_STAGES; i++)
nengel@2:         h->total_time[i]=0;
nengel@2: 
nengel@2:     h->ifile=ifile;
nengel@2:     h->ofile =ofile;
nengel@2: 
nengel@2:     h->verbose =opts->verbose;
nengel@2:     h->no_mbd =opts->no_mbd;
nengel@2:     h->static_3d =opts->static_3d;
nengel@2:     h->pipe_bufs = opts->pipe_bufs;
nengel@2:     h->slice_bufs = opts->slice_bufs;
nengel@2: 
nengel@2:     h->ed_ppe_threads =0;
nengel@2:     if (opts->ppe_ed){
nengel@2:         h->ed_ppe_threads = (opts->threads >opts->ppe_ed)? opts->ppe_ed :opts->threads;
nengel@2:     }
nengel@2: 
nengel@2:     h->threads = opts->threads - h->ed_ppe_threads;
nengel@2:     h->smt = opts->smt;
nengel@2:     if (h->smt){
nengel@2:         h->threads *= 2;
nengel@2:     }
nengel@2: 
nengel@2:     h->num_frames = opts->numframes;
nengel@2: 
nengel@2:     h->frame_width = width;
nengel@2:     h->frame_height = height;
nengel@2: 
nengel@2:     while ((width/2) %STRIDE_ALIGN)
nengel@2:         width+=STRIDE_ALIGN;
nengel@2:     h->width = width;
nengel@2:     h->height = mb_height*16;
nengel@2: 
nengel@2:     h->mb_height = mb_height;
nengel@2:     h->mb_width = mb_width;
nengel@2:     h->mb_stride = mb_stride;
nengel@2:     h->b4_stride = mb_width*4 + 1;
nengel@2:     h->b_stride = mb_width*4;
nengel@2: 
nengel@2:     h->smb_width = opts->smb_size[0];
nengel@2:     h->smb_height = opts->smb_size[1] < h->smb_width ?  opts->smb_size[1]  : h->smb_width;
nengel@2:     h->smbc = getSuperMBContext(h, h->smb_width, h->smb_height);    
nengel@2: 
nengel@2:     h->wave_order = opts->wave_order;
nengel@2: 
nengel@2:     h->pipe_bufs = opts->pipe_bufs;
nengel@2: 
nengel@2:     h->max_dpb_cnt = DPB_SIZE + opts->pipe_bufs;
nengel@2:     h->free_dpb_cnt = h->max_dpb_cnt;
nengel@2:     h->dpb = av_mallocz (h->max_dpb_cnt* sizeof (DecodedPicture));
nengel@2:     
nengel@2: 
nengel@2:     h->free_sb_cnt = h->threads*opts->slice_bufs + (h->no_mbd != 0) ;  //one extra to overlap some latency of signaling/freeing slicebuffers in entropy only mode
nengel@2:     h->sb_size = h->free_sb_cnt;
nengel@2:     h->sb = av_mallocz(h->sb_size* sizeof(SliceBufferEntry));
nengel@2: 
nengel@2:     h->rl_q.size = FFMAX(1, FFMIN( (h->height-3 - 512)/16, h->mb_width/2)) +1;
nengel@2:     h->rl_q.free = h->rl_q.size -1;
nengel@2:     h->rl_q.ready=0;
nengel@2:     h->rl_q.fi = h->rl_q.fo= 0;
nengel@2:     h->rl_q.queue = av_malloc(h->rl_q.size* sizeof(RingLineEntry*));
nengel@2:     for (i=0; i<h->rl_q.size; i++){
nengel@2:         if( posix_memalign((void**)&h->rl_q.queue[i],64,sizeof(RingLineEntry)))
nengel@2:             h->rl_q.queue[i]=NULL;
nengel@2:         h->rl_q.queue[i]->top = av_malloc(h->mb_width*sizeof(TopBorder));
nengel@2:     }
nengel@2: 
nengel@2:     h->rl_q.queue[0]->prev_line = h->rl_q.queue[h->rl_q.size-1];
nengel@2:     for (i=1; i<h->rl_q.size; i++){
nengel@2:         h->rl_q.queue[i]->prev_line = h->rl_q.queue[i-1];
nengel@2:     }
nengel@2: 
nengel@2:     if( HAVE_MMX | HAVE_ALTIVEC| HAVE_NEON ){
nengel@2:         for(i=0; i<16; i++){
nengel@2:             #define T(x) (x>>2) | ((x<<2) & 0xF)
nengel@2:             h->zigzag_scan[i] = T(zigzag_scan[i]);
nengel@2:             #undef T
nengel@2:         }
nengel@2:         for(i=0; i<64; i++){
nengel@2:             #define T(x) (x>>3) | ((x&7)<<3)
nengel@2:             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
nengel@2:             #undef T
nengel@2:         }
nengel@2:     }else{
nengel@2:         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
nengel@2:         memcpy(h->zigzag_scan8x8, ff_zigzag_direct, 64*sizeof(uint8_t));
nengel@2:     }
nengel@2: 
nengel@2:     pthread_mutex_init(&h->smb_lock, NULL);
nengel@2:     pthread_mutex_init(&h->sdl_lock, NULL);
nengel@2:     pthread_cond_init(&h->sdl_cond, NULL);
nengel@2: 
nengel@2:     ///pthread initialization
nengel@2:     pthread_mutex_init(&h->ilock, NULL);
nengel@2:     pthread_cond_init(&h->icond, NULL);
nengel@2:     pthread_mutex_init(&h->slock, NULL);
nengel@2:     pthread_cond_init(&h->scond, NULL);
nengel@2:     pthread_mutex_init(&h->tlock, NULL);
nengel@2:     pthread_cond_init(&h->tcond, NULL);
nengel@2:     pthread_mutex_init(&h->tdlock, NULL);
nengel@2:     pthread_cond_init(&h->tdcond, NULL);
nengel@2:     h->start =!opts->numamap; //default dont wait for start signal
nengel@2:     h->statmbd = opts->statmbd;
nengel@2:     h->rl_side_touch= opts->numamap;
nengel@2:     h->touch_start=0;
nengel@2:     h->setaff =opts->statsched;
nengel@2:     h->init_threads=0;
nengel@2: 
nengel@2:     pthread_mutex_init(&h->task_lock, NULL);
nengel@2:     pthread_cond_init(&h->task_cond, NULL);
nengel@2:     for (i=0; i<STAGES; i++){
nengel@2:         pthread_mutex_init (&h->lock[i], NULL);
nengel@2:         pthread_cond_init (&h->cond[i], NULL);
nengel@2: 
nengel@2:         pthread_mutex_init (&h->sb_q[i].lock, NULL);
nengel@2:         pthread_cond_init (&h->sb_q[i].cond, NULL);
nengel@2:         h->sb_q[i].size = h->free_sb_cnt; //change to num threads later
nengel@2:         h->sb_q[i].queue = av_malloc(h->free_sb_cnt* sizeof(SliceBufferEntry*));
nengel@2:         h->sb_q[i].cnt = h->sb_q[i].fi = h->sb_q[i].fo =0;
nengel@2:     }
nengel@2: 
nengel@2: #if HAVE_LIBSDL2
nengel@2:     h->sdlq.size=2;
nengel@2:     h->sdlq.ready=2;
nengel@2:     h->sdlq.queue = av_malloc(2* sizeof(SDL_Texture*));
nengel@2:     pthread_mutex_init (&h->sdlq.sdl_lock, NULL);
nengel@2:     pthread_cond_init (&h->sdlq.sdl_cond, NULL);
nengel@2: #endif
nengel@2: 
nengel@2:     h->display=opts->display;
nengel@2:     h->fullscreen=opts->fullscreen;
nengel@2: 
nengel@2:     return h;
nengel@2: }
nengel@2: 
nengel@2: 
nengel@2: void free_h264dec_context(H264Context *h) {
nengel@2:     int i;
nengel@2: 
nengel@2:     for(i=0; i<h->max_dpb_cnt; i++)
nengel@2:         free_dp(&h->dpb[i]);
nengel@2:     av_free (h->dpb);
nengel@2: 
nengel@2:     for(i=0; i<h->sb_size; i++){
nengel@2:         if (h->sb[i].initialized){
nengel@2:             free_sb_entry(&h->sb[i]);
nengel@2:         }
nengel@2:     }
nengel@2:     av_freep(&h->sb);
nengel@2: 
nengel@2:     for (i=0; i<h->rl_q.size; i++){
nengel@2:         av_freep(&h->rl_q.queue[i]->top);
nengel@2:         av_freep(&h->rl_q.queue[i]);
nengel@2:     }
nengel@2:     av_freep(&h->rl_q.queue);
nengel@2: 
nengel@2:     ///pthread cleanup
nengel@2:     pthread_mutex_destroy (&h->task_lock);
nengel@2:     pthread_cond_destroy (&h->task_cond);
nengel@2:     for (i=0; i<STAGES; i++){
nengel@2:         pthread_mutex_destroy (&h->lock[i]);
nengel@2:         pthread_cond_destroy (&h->cond[i]);
nengel@2: 
nengel@2:         pthread_mutex_destroy (&h->sb_q[i].lock);
nengel@2:         pthread_cond_destroy (&h->sb_q[i].cond);
nengel@2:         av_freep( &h->sb_q[i].queue);
nengel@2:     }
nengel@2:     pthread_mutex_destroy (&h->slock);
nengel@2:     pthread_cond_destroy (&h->scond);
nengel@2:     pthread_mutex_destroy (&h->ilock);
nengel@2:     pthread_cond_destroy (&h->icond);
nengel@2: 
nengel@2:     pthread_mutex_destroy(&h->smb_lock);
nengel@2:     pthread_mutex_destroy (&h->sdl_lock);
nengel@2:     pthread_cond_destroy (&h->sdl_cond);
nengel@2: #if HAVE_LIBSDL2
nengel@2:     av_free(h->sdlq.queue);
nengel@2:     pthread_mutex_destroy (&h->sdlq.sdl_lock);
nengel@2:     pthread_cond_destroy (&h->sdlq.sdl_cond);
nengel@2: #endif
nengel@2: 
nengel@2:     stop_timer(h, TOTAL);
nengel@2:     if (h->threads==0){
nengel@2:         for (i=0; i<PROFILE_STAGES; i++)
nengel@2:             h->total_time[i] /= h->num_frames;
nengel@2:         double others = h->total_time[TOTAL];
nengel@2:         for (i=1; i<PROFILE_STAGES; i++)
nengel@2:             others-=h->total_time[i];
nengel@2:         if (h->profile == 1){
nengel@2:             printf("\n[FRAME %.3fms] [FRONT %.3fms] [ENTROPY %.3fms] [MBREC %.3fms] [OTHERS %.3fms]\n", h->total_time[TOTAL], h->total_time[FRONT], h->total_time[ED], h->total_time[REC], others);
nengel@2:         }else if (h->profile ==2){
nengel@2:             printf("\n[FRAME %.3fms] [FRONT %.3fms] [ENTROPY %.3fms] [PRED  %.3fms] [OTHERS %.3fms]\n", h->total_time[TOTAL], h->total_time[FRONT], h->total_time[ED],h->total_time[REC], others);
nengel@2:         }
nengel@2:     }
nengel@2: 
nengel@2:     av_free(h);
nengel@2: }