dv.c

arm平台下的H264编码和解码源代码

            if (j < 4) {
                if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
                    /* NOTE: at end of line, the macroblock is handled as 420 */
                    idct_put(y_ptr + (j * 8), s->picture.linesize[0], block);
                } else {
                    idct_put(y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]),
                             s->picture.linesize[0], block);
                }
            } else {
                if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
                    uint64_t aligned_pixels[64/8];
                    uint8_t *pixels= (uint8_t*)aligned_pixels;
		    uint8_t *c_ptr, *c_ptr1, *ptr;
                    int y, linesize;
                    /* NOTE: at end of line, the macroblock is handled as 420 */
                    idct_put(pixels, 8, block);
                    linesize = s->picture.linesize[6 - j];
                    c_ptr = s->picture.data[6 - j] + c_offset;
                    ptr = pixels;
                    for(y = 0;y < 8; y++) {
                        /* convert to 411P */
                        c_ptr1 = c_ptr + 8*linesize;
                        c_ptr[0]= ptr[0]; c_ptr1[0]= ptr[4];
                        c_ptr[1]= ptr[1]; c_ptr1[1]= ptr[5];
                        c_ptr[2]= ptr[2]; c_ptr1[2]= ptr[6];
                        c_ptr[3]= ptr[3]; c_ptr1[3]= ptr[7];
                        c_ptr += linesize;
                        ptr += 8;
                    }
                } else {
                    /* don't ask me why they inverted Cb and Cr ! */
                    idct_put(s->picture.data[6 - j] + c_offset, 
                             s->picture.linesize[6 - j], block);
                }
            }
            block += 64;
            mb++;
        }
    }
}

#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 { 
        if (level < DV_VLC_MAP_LEV_SIZE) {
	    *vlc = dv_vlc_map[0][level].vlc | sign;
	    size = dv_vlc_map[0][level].size;
	} else {
            *vlc = 0xfe00 | (level << 1) | sign;
	    size = 16;
	}
	if (run) {
	    *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc : 
	                          (0x1f80 | (run - 1))) << size;
	    size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
	}
    }
    
    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; 
    }
    else { 
	size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;
	if (run) {
	    size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
	}
    }
    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;
    int i;
    
    /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */
    s = blk;
    for(i=0; i<7; i++) {
        score88 += SC(0,  8) + SC(1, 9) + SC(2, 10) + SC(3, 11) + 
	           SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);
        s += 8;
    }
    /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */
    s = blk;
    for(i=0; i<6; i++) {
        score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +
	            SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);
        s += 8;
    }

    return (score88 - score248 > -10);
}

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];
	     }
	  }
       }
    } while ((vs_total_ac_bits < size[0] + size[1] + size[2] + size[3] + size[4]) && 
             (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]));
}

/*
 * This is a very rough initial implementaion. The performance is
 * horrible and the weighting is missing. But it's missing from the 
 * decoding step also -- so at least we're on the same page with decoder ;-)
 */
static inline void dv_encode_video_segment(DVVideoContext *s, 
                                           uint8_t *dif, 
                                           const uint16_t *mb_pos_ptr)
{
    int mb_index, i, j, v;
    int mb_x, mb_y, c_offset, linesize; 
    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++;
        mb_x = v & 0xff;
        mb_y = v >> 8;
        y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
	c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ?
	           ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) :
		   (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
	do_edge_wrap = 0;
	qnos[mb_index] = 15; /* No quantization */
        ptr = dif + mb_index*80 + 4;
        for(j = 0;j < 6; j++) {
            if (j < 4) {  /* Four Y blocks */
		/* NOTE: at end of line, the macroblock is handled as 420 */
		if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
                    data = y_ptr + (j * 8);
                } else {
                    data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);
                }
		linesize = s->picture.linesize[0];
            } else {      /* Cr and Cb blocks */
	        /* don't ask Fabrice why they inverted Cb and Cr ! */
	        data = s->picture.data[6 - j] + c_offset;
		linesize = s->picture.linesize[6 - j];
		if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))
		    do_edge_wrap = 1;
	    }	
            
	    /* Everything is set up -- now just copy data -> DCT block */
	    if (do_edge_wrap) {  /* Edge wrap copy: 4x16 -> 8x8 */
		uint8_t* d;
		DCTELEM *b = block;
	        for (i=0;i<8;i++) {
		   d = data + 8 * linesize;
		   b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];
                   b[4] =    d[0]; b[5] =    d[1]; b[6] =    d[2]; b[7] =    d[3];
		   data += linesize;
		   b += 8;
		}
	    } 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];
	    ++enc_blk;
	    ++pb;
	    ptr += block_sizes[j]/8;
        }
    }

    if (vs_total_ac_bits < vs_bit_size)
        dv_guess_qnos(&enc_blks[0], &qnos[0]);

    for (i=0; i<5; i++) {
       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]);
}

static int dv_decode_mt(AVCodecContext *avctx, void* sl)
{
    DVVideoContext *s = avctx->priv_data;
    int slice = (size_t)sl;
    dv_decode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
	                    &s->sys->video_place[slice*5]);
    return 0;
}

static int dv_encode_mt(AVCodecContext *avctx, void* sl)
{
    DVVideoContext *s = avctx->priv_data;
    int slice = (size_t)sl;
    dv_encode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
	                    &s->sys->video_place[slice*5]);
    return 0;
}

/* NOTE: exactly one frame must be given (120000 bytes for NTSC,
   144000 bytes for PAL) */
static int dvvideo_decode_frame(AVCodecContext *avctx, 
                                 void *data, int *data_size,
                                 uint8_t *buf, int buf_size)
{
    DVVideoContext *s = avctx->priv_data;
  
    *data_size=0;
    /* special case for last picture */
    if(buf_size==0)
        return 0;
    
    s->sys = dv_frame_profile(buf);
    if (!s->sys || buf_size < s->sys->frame_size)
        return -1; /* NOTE: we only accept several full frames */

    if(s->picture.data[0])
        avctx->release_buffer(avctx, &s->picture);
    
    s->picture.reference = 0;
    avctx->pix_fmt = s->sys->pix_fmt;
    avctx->width = s->sys->width;
    avctx->height = s->sys->height;
    if(avctx->get_buffer(avctx, &s->picture) < 0) {
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return -1;
    }
    s->picture.interlaced_frame = 1;
    s->picture.top_field_first = 0;

    s->buf = buf;
    avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL, 
	           s->sys->difseg_size * 27);
    
    emms_c();

    /* return image */
    *data_size = sizeof(AVFrame);
    *(AVFrame*)data= s->picture;
    
    return s->sys->frame_size;
}

static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size, 
                                void *data)
{
    DVVideoContext *s = c->priv_data;

    s->sys = dv_codec_profile(c);
    if (!s->sys)
	return -1;
    
    c->pix_fmt = s->sys->pix_fmt;
    s->picture = *((AVFrame *)data);

    s->buf = buf;
    c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL, 
	       s->sys->difseg_size * 27);

    emms_c();
    return s->sys->frame_size;
}

AVCodec dvvideo_encoder = {
    "dvvideo",
    CODEC_TYPE_VIDEO,
    CODEC_ID_DVVIDEO,
    sizeof(DVVideoContext),
    dvvideo_init,
    dvvideo_encode_frame,
    dvvideo_end,
    NULL,
    CODEC_CAP_DR1,
    NULL
};

AVCodec dvvideo_decoder = {
    "dvvideo",
    CODEC_TYPE_VIDEO,
    CODEC_ID_DVVIDEO,
    sizeof(DVVideoContext),
    dvvideo_init,
    NULL,
    dvvideo_end,
    dvvideo_decode_frame,
    CODEC_CAP_DR1,
    NULL
};