comparison svq1enc.c @ 10855:423410fc29b1 libavcodec

Reduce stack usage in svq1_encode_plane(). Reuse context scratch buffer instead. Avoid a crash on MinGW.

10854:f6fc6ace95e3

    int x, y;
    int i;
    int block_width, block_height;
    int level;
    int threshold[6];
    const int lambda= (s->picture.quality*s->picture.quality) >> (2*FF_LAMBDA_SHIFT);

    /* figure out the acceptable level thresholds in advance */
    threshold[5] = QUALITY_THRESHOLD;
    for (level = 4; level >= 0; level--)

        ff_init_me(&s->m);

        s->m.me.dia_size= s->avctx->dia_size;
        s->m.first_slice_line=1;
        for (y = 0; y < block_height; y++) {
            uint8_t src[stride*16];

            s->m.new_picture.data[0]= src - y*16*stride; //ugly
            s->m.mb_y= y;

            for(i=0; i<16 && i + 16*y<height; i++){
                memcpy(&src[i*stride], &src_plane[(i+16*y)*src_stride], width);

        ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code, CANDIDATE_MB_TYPE_INTER, 0);
    }

    s->m.first_slice_line=1;
    for (y = 0; y < block_height; y++) {
        uint8_t src[stride*16];

        for(i=0; i<16 && i + 16*y<height; i++){
            memcpy(&src[i*stride], &src_plane[(i+16*y)*src_stride], width);
            for(x=width; x<16*block_width; x++)
                src[i*stride+x]= src[i*stride+x-1];
        }

    }

    if(!s->current_picture.data[0]){
        avctx->get_buffer(avctx, &s->current_picture);
        avctx->get_buffer(avctx, &s->last_picture);
        s->scratchbuf = av_malloc(s->current_picture.linesize[0] * 16);
    }

    temp= s->current_picture;
    s->current_picture= s->last_picture;
    s->last_picture= temp;

10855:423410fc29b1

    int x, y;
    int i;
    int block_width, block_height;
    int level;
    int threshold[6];
    uint8_t *src = s->scratchbuf + stride * 16;
    const int lambda= (s->picture.quality*s->picture.quality) >> (2*FF_LAMBDA_SHIFT);

    /* figure out the acceptable level thresholds in advance */
    threshold[5] = QUALITY_THRESHOLD;
    for (level = 4; level >= 0; level--)

        ff_init_me(&s->m);

        s->m.me.dia_size= s->avctx->dia_size;
        s->m.first_slice_line=1;
        for (y = 0; y < block_height; y++) {
            s->m.new_picture.data[0]= src - y*16*stride; //ugly
            s->m.mb_y= y;

            for(i=0; i<16 && i + 16*y<height; i++){
                memcpy(&src[i*stride], &src_plane[(i+16*y)*src_stride], width);

        ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code, CANDIDATE_MB_TYPE_INTER, 0);
    }

    s->m.first_slice_line=1;
    for (y = 0; y < block_height; y++) {
        for(i=0; i<16 && i + 16*y<height; i++){
            memcpy(&src[i*stride], &src_plane[(i+16*y)*src_stride], width);
            for(x=width; x<16*block_width; x++)
                src[i*stride+x]= src[i*stride+x-1];
        }

    }

    if(!s->current_picture.data[0]){
        avctx->get_buffer(avctx, &s->current_picture);
        avctx->get_buffer(avctx, &s->last_picture);
        s->scratchbuf = av_malloc(s->current_picture.linesize[0] * 16 * 2);
    }

    temp= s->current_picture;
    s->current_picture= s->last_picture;
    s->last_picture= temp;