comparison dv.c @ 2847:712c84b9d8ce libavcodec

optimizations dc coeff rounding fix class=3 num of bits fix do interlaced check & idct only if CODEC_FLAG_INTERLACED_DCT

2846:40765c51a7a9

#include "dsputil.h"
#include "mpegvideo.h"
#include "simple_idct.h"
#include "dvdata.h"

typedef struct DVVideoContext {
    const DVprofile* sys;
    AVFrame picture;
    uint8_t *buf;
    
    uint8_t dv_zigzag[2][64];
    uint8_t dv_idct_shift[2][2][22][64];
  

#ifdef DV_CODEC_TINY_TARGET
#define DV_VLC_MAP_RUN_SIZE 15
#define DV_VLC_MAP_LEV_SIZE 23
#else
#define DV_VLC_MAP_RUN_SIZE  64 
#define DV_VLC_MAP_LEV_SIZE 512
#endif

/* MultiThreading */
static uint8_t** dv_anchor;

    /* FIXME: I really don't think this should be here */
    if (dv_codec_profile(avctx))
	avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt; 
    avctx->coded_frame = &s->picture;
    
    return 0;
}

// #define VLC_DEBUG

#endif
	pos += run; 	
	if (pos >= 64)
	    break;
        
	if (level) {
            pos1 = scan_table[pos];
            block[pos1] = level << shift_table[pos1];
        } 

        UPDATE_CACHE(re, gb);
    }
    CLOSE_READER(re, gb);
    mb->pos = pos;
}

static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
{
    int bits_left = get_bits_left(gb);
    while (bits_left >= 16) {
        put_bits(pb, 16, get_bits(gb, 16));
        bits_left -= 16;
    }
    if (bits_left > 0) {
        put_bits(pb, bits_left, get_bits(gb, bits_left));
    }
}

    uint8_t *buf_ptr;
    PutBitContext pb, vs_pb;
    GetBitContext gb;
    BlockInfo mb_data[5 * 6], *mb, *mb1;
    DCTELEM sblock[5*6][64] __align8;
    uint8_t mb_bit_buffer[80 + 4]; /* allow some slack */
    uint8_t vs_bit_buffer[5 * 80 + 4]; /* allow some slack */
	    
    memset(sblock, 0, sizeof(sblock));

    /* pass 1 : read DC and AC coefficients in blocks */
    buf_ptr = buf_ptr1;
    block1 = &sblock[0][0];

    }
}

#ifdef DV_CODEC_TINY_TARGET
/* Converts run and level (where level != 0) pair into vlc, returning bit size */
static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc)
{
    int sign = l >> 8;
    int level = (l ^ sign) - sign;
    int size;
    
    sign = (sign & 1);

    if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
        *vlc = dv_vlc_map[run][level].vlc | sign;
	size = dv_vlc_map[run][level].size;
    }
    else { 

    }
    
    return size;
}

static always_inline int dv_rl2vlc_size(int run, int l)
{
    int level = (l ^ (l >> 8)) - (l >> 8);
    int size;
    
    if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
	size = dv_vlc_map[run][level].size; 
    }

	}
    }
    return size;
}
#else
static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc)
{
    *vlc = dv_vlc_map[run][((uint16_t)l)&0x1ff].vlc;
    return dv_vlc_map[run][((uint16_t)l)&0x1ff].size;
}

static always_inline int dv_rl2vlc_size(int run, int l)
{
    return dv_vlc_map[run][((uint16_t)l)&0x1ff].size;
}
#endif

typedef struct EncBlockInfo {
    int area_q[4];
    int bit_size[4];
    int prev_run[4];
    int cur_ac;
    int cno;
    int dct_mode;
    DCTELEM *mb;
    uint8_t partial_bit_count;
    uint32_t partial_bit_buffer; /* we can't use uint16_t here */
} EncBlockInfo;

static always_inline void dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool, 
                                       int pb_size)
{
    int run;
    int bits_left;
    PutBitContext* pb = pb_pool;
    int size = bi->partial_bit_count;
    uint32_t vlc = bi->partial_bit_buffer;
    
    bi->partial_bit_count = bi->partial_bit_buffer = 0;
vlc_loop:
       /* Find suitable storage space */
       for (; size > (bits_left = put_bits_left(pb)); pb++) {
          if (bits_left) {
              size -= bits_left;
	      put_bits(pb, bits_left, vlc >> size);
	      vlc = vlc & ((1<<size)-1);
	  }
	  if (pb_size == 1) {
	      bi->partial_bit_count = size;
	      bi->partial_bit_buffer = vlc;
	      return;
	  }
	  --pb_size;
       }
       
       /* Store VLC */
       put_bits(pb, size, vlc);
       
       /* Construct the next VLC */
       run = 0;
       for (; bi->cur_ac < 64; bi->cur_ac++, run++) {
           if (bi->mb[bi->cur_ac]) {
	       size = dv_rl2vlc(run, bi->mb[bi->cur_ac], &vlc);
	       bi->cur_ac++;
	       goto vlc_loop;
	   }
       }
   
       if (bi->cur_ac == 64) {
           size = 4; vlc = 6; /* End Of Block stamp */
	   bi->cur_ac++;
	   goto vlc_loop;
       }
}

static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi, 
                                              const uint8_t* zigzag_scan, int bias)
{
    int i, area;
    int run;
    int classes[] = {12, 24, 36, 0xffff};

    run = 0;
    bi->mb[0] = blk[0]; 
    bi->cno = 0;
    for (area = 0; area < 4; area++) {
       bi->prev_run[area] = run;
       bi->bit_size[area] = 0;
       for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {
          bi->mb[i] = (blk[zigzag_scan[i]] / 16);
          while ((bi->mb[i] ^ (bi->mb[i] >> 8)) > classes[bi->cno])
              bi->cno++;
       
          if (bi->mb[i]) {
              bi->bit_size[area] += dv_rl2vlc_size(run, bi->mb[i]);
	      run = 0;
          } else
              ++run;
       }
    }
    bi->bit_size[3] += 4; /* EOB marker */
    bi->cno += bias;
    
    if (bi->cno >= 3) { /* FIXME: we have to recreate bit_size[], prev_run[] */
        bi->cno = 3;
	for (i=1; i<64; i++)
	   bi->mb[i] /= 2;
    }
}

#define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
static always_inline int dv_guess_dct_mode(DCTELEM *blk) {
    DCTELEM *s;
    int score88 = 0;
    int score248 = 0;

}

static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)
{
    int size[5];
    int i, j, k, a, run;
    EncBlockInfo* b;
    
    do {
       b = blks;
       for (i=0; i<5; i++) {
          if (!qnos[i])
	      continue;

	  qnos[i]--;
	  size[i] = 0;
          for (j=0; j<6; j++, b++) {
	     for (a=0; a<4; a++) {
	        if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
		    b->bit_size[a] = (a==3)?4:0;
		    b->area_q[a]++;
		    run = b->prev_run[a];
		    for (k=mb_area_start[a]; k<mb_area_start[a+1]; k++) {
		       b->mb[k] /= 2;
		       if (b->mb[k]) {
                           b->bit_size[a] += dv_rl2vlc_size(run, b->mb[k]);
	                   run = 0;
                       } else
                           ++run;
		    }
		}
		size[i] += b->bit_size[a];
	     }
	  }
       }

    uint8_t*  y_ptr;
    uint8_t*  data;
    uint8_t*  ptr;
    int       do_edge_wrap;
    DCTELEM   block[64] __align8;
    DCTELEM   sblock[5*6][64] __align8;
    EncBlockInfo  enc_blks[5*6];
    PutBitContext pbs[5*6];
    PutBitContext* pb; 
    EncBlockInfo* enc_blk;
    int       vs_bit_size = 0;
    int       qnos[5];
   
    enc_blk = &enc_blks[0];
    pb = &pbs[0];
    for(mb_index = 0; mb_index < 5; mb_index++) {
        v = *mb_pos_ptr++;

		}
	    } else {             /* Simple copy: 8x8 -> 8x8 */
	        s->get_pixels(block, data, linesize);
	    }
	  
            enc_blk->dct_mode = dv_guess_dct_mode(block);
	    enc_blk->mb = &sblock[mb_index*6+j][0];
	    enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;
	    enc_blk->partial_bit_count = 0;
	    enc_blk->partial_bit_buffer = 0;
	    enc_blk->cur_ac = 1;
	    
	    s->fdct[enc_blk->dct_mode](block);
	    
	    dv_set_class_number(block, enc_blk, 
	                        enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct,
				j/4*(j%2));
           
            init_put_bits(pb, ptr, block_sizes[j]/8);
	    put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024) >> 2));
	    put_bits(pb, 1, enc_blk->dct_mode);
	    put_bits(pb, 2, enc_blk->cno);
	    
	    vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +
	                   enc_blk->bit_size[2] + enc_blk->bit_size[3];

       dif[i*80 + 3] = qnos[i];
    }

    /* First pass over individual cells only */
    for (j=0; j<5*6; j++)
       dv_encode_ac(&enc_blks[j], &pbs[j], 1);

    /* Second pass over each MB space */
    for (j=0; j<5*6; j++) {
       if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count)
           dv_encode_ac(&enc_blks[j], &pbs[(j/6)*6], 6);
    }

    /* Third and final pass over the whole vides segment space */
    for (j=0; j<5*6; j++) {
       if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count)
           dv_encode_ac(&enc_blks[j], &pbs[0], 6*5);
    }

    for (j=0; j<5*6; j++)
       flush_put_bits(&pbs[j]);
}

2847:712c84b9d8ce

#include "dsputil.h"
#include "mpegvideo.h"
#include "simple_idct.h"
#include "dvdata.h"

//#undef NDEBUG
//#include <assert.h>

typedef struct DVVideoContext {
    const DVprofile* sys;
    AVFrame picture;
    AVCodecContext *avctx;
    uint8_t *buf;
    
    uint8_t dv_zigzag[2][64];
    uint8_t dv_idct_shift[2][2][22][64];
  

#ifdef DV_CODEC_TINY_TARGET
#define DV_VLC_MAP_RUN_SIZE 15
#define DV_VLC_MAP_LEV_SIZE 23
#else
#define DV_VLC_MAP_RUN_SIZE  64 
#define DV_VLC_MAP_LEV_SIZE 512 //FIXME sign was removed so this should be /2 but needs check
#endif

/* MultiThreading */
static uint8_t** dv_anchor;

    /* FIXME: I really don't think this should be here */
    if (dv_codec_profile(avctx))
	avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt; 
    avctx->coded_frame = &s->picture;
    s->avctx= avctx;
    
    return 0;
}

// #define VLC_DEBUG

#endif
	pos += run; 	
	if (pos >= 64)
	    break;
        
        assert(level);
        pos1 = scan_table[pos];
        block[pos1] = level << shift_table[pos1];

        UPDATE_CACHE(re, gb);
    }
    CLOSE_READER(re, gb);
    mb->pos = pos;
}

static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
{
    int bits_left = get_bits_left(gb);
    while (bits_left >= MIN_CACHE_BITS) {
        put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
        bits_left -= MIN_CACHE_BITS;
    }
    if (bits_left > 0) {
        put_bits(pb, bits_left, get_bits(gb, bits_left));
    }
}

    uint8_t *buf_ptr;
    PutBitContext pb, vs_pb;
    GetBitContext gb;
    BlockInfo mb_data[5 * 6], *mb, *mb1;
    DCTELEM sblock[5*6][64] __align8;
    uint8_t mb_bit_buffer[80 + 4] __align8; /* allow some slack */
    uint8_t vs_bit_buffer[5 * 80 + 4] __align8; /* allow some slack */
	    
    assert((((int)mb_bit_buffer)&7)==0);
    assert((((int)vs_bit_buffer)&7)==0);
    
    memset(sblock, 0, sizeof(sblock));

    /* pass 1 : read DC and AC coefficients in blocks */
    buf_ptr = buf_ptr1;
    block1 = &sblock[0][0];

    }
}

#ifdef DV_CODEC_TINY_TARGET
/* Converts run and level (where level != 0) pair into vlc, returning bit size */
static always_inline int dv_rl2vlc(int run, int level, int sign, uint32_t* vlc)
{
    int size;
    if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
        *vlc = dv_vlc_map[run][level].vlc | sign;
	size = dv_vlc_map[run][level].size;
    }
    else { 

    }
    
    return size;
}

static always_inline int dv_rl2vlc_size(int run, int level)
{
    int size;
    
    if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
	size = dv_vlc_map[run][level].size; 
    }

	}
    }
    return size;
}
#else
static always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t* vlc)
{
    *vlc = dv_vlc_map[run][l].vlc | sign;
    return dv_vlc_map[run][l].size;
}

static always_inline int dv_rl2vlc_size(int run, int l)
{
    return dv_vlc_map[run][l].size;
}
#endif

typedef struct EncBlockInfo {
    int area_q[4];
    int bit_size[4];
    int prev[5];
    int cur_ac;
    int cno;
    int dct_mode;
    DCTELEM mb[64];
    uint8_t next[64];
    uint8_t sign[64];
    uint8_t partial_bit_count;
    uint32_t partial_bit_buffer; /* we can't use uint16_t here */
} EncBlockInfo;

static always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool, 
                                       PutBitContext* pb_end)
{
    int prev;
    int bits_left;
    PutBitContext* pb = pb_pool;
    int size = bi->partial_bit_count;
    uint32_t vlc = bi->partial_bit_buffer;

    bi->partial_bit_count = bi->partial_bit_buffer = 0;
    for(;;){
       /* Find suitable storage space */
       for (; size > (bits_left = put_bits_left(pb)); pb++) {
          if (bits_left) {
              size -= bits_left;
	      put_bits(pb, bits_left, vlc >> size);
	      vlc = vlc & ((1<<size)-1);
	  }
	  if (pb + 1 >= pb_end) {
	      bi->partial_bit_count = size;
	      bi->partial_bit_buffer = vlc;
	      return pb;
	  }
       }
       
       /* Store VLC */
       put_bits(pb, size, vlc);
       
       if(bi->cur_ac>=64)
           break;
       
       /* Construct the next VLC */
       prev= bi->cur_ac;
       bi->cur_ac = bi->next[prev];
       if(bi->cur_ac < 64){
           size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], bi->sign[bi->cur_ac], &vlc);
       } else {
           size = 4; vlc = 6; /* End Of Block stamp */
       }
    }
    return pb;
}

static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi, 
                                              const uint8_t* zigzag_scan, int bias)
{
    int i, area;
    static const int classes[] = {12, 24, 36, 0xffff};
    int max=12;
    int prev=0;

    bi->mb[0] = blk[0]; 
    
    for (area = 0; area < 4; area++) {
       bi->prev[area] = prev;
       bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
       for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {
          int level = blk[zigzag_scan[i]];
       
          if (level+15 > 30U) {
              bi->sign[i] = (level>>31)&1;
              bi->mb[i] = level= ABS(level)>>4;
              if(level>max) max= level;
              bi->bit_size[area] += dv_rl2vlc_size(i - prev  - 1, level);
              bi->next[prev]= i;
              prev= i;
          }
       }
    }
    bi->next[prev]= i;
    for(bi->cno = 0; max > classes[bi->cno]; bi->cno++);

    bi->cno += bias;
    
    if (bi->cno >= 3) {
        bi->cno = 3;
        prev=0;
        i= bi->next[prev];
        for (area = 0; area < 4; area++) {
            bi->prev[area] = prev;
            bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
            for (; i<mb_area_start[area+1]; i= bi->next[i]) {
                bi->mb[i] >>=1;
            
                if (bi->mb[i]) {
                    bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]);
                    bi->next[prev]= i;
                    prev= i;
                }
            }
        }
        bi->next[prev]= i;
    }
}

//FIXME replace this by dsputil
#define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
static always_inline int dv_guess_dct_mode(DCTELEM *blk) {
    DCTELEM *s;
    int score88 = 0;
    int score248 = 0;

}

static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)
{
    int size[5];
    int i, j, k, a, prev;
    EncBlockInfo* b;

    do {
       b = blks;
       for (i=0; i<5; i++) {
          if (!qnos[i])
	      continue;

	  qnos[i]--;
	  size[i] = 0;
          for (j=0; j<6; j++, b++) {
	     for (a=0; a<4; a++) {
	        if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
		    b->bit_size[a] = 1; // 4 areas 4 bits for EOB :)
		    b->area_q[a]++;
                    prev= b->prev[a];
                    for (k= b->next[prev] ; k<mb_area_start[a+1]; k= b->next[k]) {
		       b->mb[k] >>= 1;
		       if (b->mb[k]) {
                           b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
	                   prev= k;
                       } else {
                           b->next[prev] = b->next[k];
                       }
		    }
                    b->prev[a+1]= prev;
		}
		size[i] += b->bit_size[a];
	     }
	  }
       }

    uint8_t*  y_ptr;
    uint8_t*  data;
    uint8_t*  ptr;
    int       do_edge_wrap;
    DCTELEM   block[64] __align8;
    EncBlockInfo  enc_blks[5*6];
    PutBitContext pbs[5*6];
    PutBitContext* pb; 
    EncBlockInfo* enc_blk;
    int       vs_bit_size = 0;
    int       qnos[5];
    
    assert((((int)block) & 7) == 0);
   
    enc_blk = &enc_blks[0];
    pb = &pbs[0];
    for(mb_index = 0; mb_index < 5; mb_index++) {
        v = *mb_pos_ptr++;

		}
	    } else {             /* Simple copy: 8x8 -> 8x8 */
	        s->get_pixels(block, data, linesize);
	    }
	  
            if(s->avctx->flags & CODEC_FLAG_INTERLACED_DCT)
                enc_blk->dct_mode = dv_guess_dct_mode(block);
            else
                enc_blk->dct_mode = 0;
	    enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;
	    enc_blk->partial_bit_count = 0;
	    enc_blk->partial_bit_buffer = 0;
	    enc_blk->cur_ac = 0;
	    
	    s->fdct[enc_blk->dct_mode](block);
	    
	    dv_set_class_number(block, enc_blk, 
	                        enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct, j/4);
           
            init_put_bits(pb, ptr, block_sizes[j]/8);
	    put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024 + 2) >> 2));
	    put_bits(pb, 1, enc_blk->dct_mode);
	    put_bits(pb, 2, enc_blk->cno);
	    
	    vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +
	                   enc_blk->bit_size[2] + enc_blk->bit_size[3];

       dif[i*80 + 3] = qnos[i];
    }

    /* First pass over individual cells only */
    for (j=0; j<5*6; j++)
       dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j+1]);

    /* Second pass over each MB space */
    for (j=0; j<5*6; j+=6) {
        pb= &pbs[j];
        for (i=0; i<6; i++) {
            if (enc_blks[i+j].partial_bit_count)
                pb=dv_encode_ac(&enc_blks[i+j], pb, &pbs[j+6]);
        }
    }

    /* Third and final pass over the whole vides segment space */
    pb= &pbs[0];
    for (j=0; j<5*6; j++) {
       if (enc_blks[j].partial_bit_count)
           pb=dv_encode_ac(&enc_blks[j], pb, &pbs[6*5]);
    }

    for (j=0; j<5*6; j++)
       flush_put_bits(&pbs[j]);
}