comparison snow.c @ 8214:c5276ad92ff8 libavcodec

snow: move scratch buffer from stack to malloced buffer in context This is required to ensure 16-byte alignment.

8213:93423820d82f

    int me_cache[ME_CACHE_SIZE];
    int me_cache_generation;
    slice_buffer sb;

    MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
}SnowContext;

typedef struct {
    IDWTELEM *b0;
    IDWTELEM *b1;

    BlockNode *rt= lt+1;
    BlockNode *lb= lt+b_stride;
    BlockNode *rb= lb+1;
    uint8_t *block[4];
    int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
    uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align
    uint8_t *ptmp;
    int x,y;

    if(b_x<0){
        lt= rt;

    const int obmc_stride= plane_index ? block_size : 2*block_size;
    const int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst= s->current_picture.data[plane_index];
    uint8_t *src= s->  input_picture.data[plane_index];
    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
    uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment
    uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS_MAX-1)];
    const int b_stride = s->b_width << s->block_max_depth;
    const int b_height = s->b_height<< s->block_max_depth;
    const int w= p->width;
    const int h= p->height;

    for(i=0; i<MAX_REF_FRAMES; i++)
        for(j=0; j<MAX_REF_FRAMES; j++)
            scale_mv_ref[i][j] = 256*(i+1)/(j+1);

    s->avctx->get_buffer(s->avctx, &s->mconly_picture);

    return 0;
}

static int common_init_after_header(AVCodecContext *avctx){

    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->m.obmc_scratchpad);

    av_freep(&s->block);

    for(i=0; i<MAX_REF_FRAMES; i++){
        av_freep(&s->ref_mvs[i]);
        av_freep(&s->ref_scores[i]);
        if(s->last_picture[i].data[0])

8214:c5276ad92ff8

    int me_cache[ME_CACHE_SIZE];
    int me_cache_generation;
    slice_buffer sb;

    MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)

    uint8_t *scratchbuf;
}SnowContext;

typedef struct {
    IDWTELEM *b0;
    IDWTELEM *b1;

    BlockNode *rt= lt+1;
    BlockNode *lb= lt+b_stride;
    BlockNode *rb= lb+1;
    uint8_t *block[4];
    int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
    uint8_t *tmp = s->scratchbuf;
    uint8_t *ptmp;
    int x,y;

    if(b_x<0){
        lt= rt;

    const int obmc_stride= plane_index ? block_size : 2*block_size;
    const int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst= s->current_picture.data[plane_index];
    uint8_t *src= s->  input_picture.data[plane_index];
    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
    uint8_t *cur = s->scratchbuf;
    uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS_MAX-1)];
    const int b_stride = s->b_width << s->block_max_depth;
    const int b_height = s->b_height<< s->block_max_depth;
    const int w= p->width;
    const int h= p->height;

    for(i=0; i<MAX_REF_FRAMES; i++)
        for(j=0; j<MAX_REF_FRAMES; j++)
            scale_mv_ref[i][j] = 256*(i+1)/(j+1);

    s->avctx->get_buffer(s->avctx, &s->mconly_picture);
    s->scratchbuf = av_malloc(s->mconly_picture.linesize[0]*7*MB_SIZE);

    return 0;
}

static int common_init_after_header(AVCodecContext *avctx){

    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->m.obmc_scratchpad);

    av_freep(&s->block);
    av_freep(&s->scratchbuf);

    for(i=0; i<MAX_REF_FRAMES; i++){
        av_freep(&s->ref_mvs[i]);
        av_freep(&s->ref_scores[i]);
        if(s->last_picture[i].data[0])