comparison vp8.c @ 12222:7acdbfd2a222 libavcodec

Calculate deblock strength per-MB instead of per-row Gives better cache locality, since the VP8Macroblock structs are still in cache. Inspired by the way x264 does it.

12221:45852dac8338

#include "vp8dsp.h"
#include "h264pred.h"
#include "rectangle.h"

typedef struct {
    uint8_t segment;
    uint8_t skip;
    // todo: make it possible to check for at least (i4x4 or split_mv)
    // in one op. are others needed?
    uint8_t mode;

    int num_coeff_partitions;
    VP56RangeCoder coeff_partition[8];

    VP8Macroblock *macroblocks;
    VP8Macroblock *macroblocks_base;
    int mb_stride;

    uint8_t *intra4x4_pred_mode;
    uint8_t *intra4x4_pred_mode_base;
    int b4_stride;

    // this is 4 blocks for intra4x4 to keep 4-byte alignment for fill_rectangle
    s->mb_stride = s->mb_width+1;
    s->b4_stride = 4*s->mb_stride;

    s->macroblocks_base        = av_mallocz(s->mb_stride*(s->mb_height+1)*sizeof(*s->macroblocks));
    s->intra4x4_pred_mode_base = av_mallocz(s->b4_stride*(4*s->mb_height+1));
    s->top_nnz                 = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
    s->top_border              = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));

    if (!s->macroblocks_base || !s->intra4x4_pred_mode_base || !s->top_nnz || !s->top_border)
        return AVERROR(ENOMEM);

    s->macroblocks        = s->macroblocks_base        + 1 + s->mb_stride;
    s->intra4x4_pred_mode = s->intra4x4_pred_mode_base + 4 + s->b4_stride;

        u_dst += 4*s->uvlinesize;
        v_dst += 4*s->uvlinesize;
    }
}

static void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb, int *level, int *inner, int *hev_thresh)
{
    int interior_limit, filter_level;

    if (s->segmentation.enabled) {
        filter_level = s->segmentation.filter_level[mb->segment];

        interior_limit >>= s->filter.sharpness > 4 ? 2 : 1;
        interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
    }
    interior_limit = FFMAX(interior_limit, 1);

    *level = filter_level;
    *inner = interior_limit;

    if (hev_thresh) {
        *hev_thresh = filter_level >= 15;

        if (s->keyframe) {
            if (filter_level >= 40)
                *hev_thresh = 2;
        } else {
            if (filter_level >= 40)
                *hev_thresh = 3;
            else if (filter_level >= 20)
                *hev_thresh = 2;
        }
    }
}

static void filter_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb, int mb_x, int mb_y)
{
    int filter_level, inner_limit, hev_thresh, mbedge_lim, bedge_lim;

    filter_level_for_mb(s, mb, &filter_level, &inner_limit, &hev_thresh);
    if (!filter_level)
        return;

    mbedge_lim = 2*(filter_level+2) + inner_limit;
     bedge_lim = 2* filter_level    + inner_limit;

    if (mb_x) {
        s->vp8dsp.vp8_h_loop_filter16y(dst[0],     s->linesize,
                                       mbedge_lim, inner_limit, hev_thresh);
        s->vp8dsp.vp8_h_loop_filter8uv(dst[1],     dst[2],      s->uvlinesize,

                                             s->uvlinesize, bedge_lim,
                                             inner_limit,   hev_thresh);
    }
}

static void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8Macroblock *mb, int mb_x, int mb_y)
{
    int filter_level, inner_limit, mbedge_lim, bedge_lim;

    filter_level_for_mb(s, mb, &filter_level, &inner_limit, NULL);
    if (!filter_level)
        return;

    mbedge_lim = 2*(filter_level+2) + inner_limit;
     bedge_lim = 2* filter_level    + inner_limit;

    }
}

static void filter_mb_row(VP8Context *s, int mb_y)
{
    VP8Macroblock *mb = s->macroblocks + mb_y*s->mb_stride;
    uint8_t *dst[3] = {
        s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize,
        s->framep[VP56_FRAME_CURRENT]->data[1] +  8*mb_y*s->uvlinesize,
        s->framep[VP56_FRAME_CURRENT]->data[2] +  8*mb_y*s->uvlinesize
    };
    int mb_x;

    for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
        backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
        filter_mb(s, dst, mb++, mb_x, mb_y);
        dst[0] += 16;
        dst[1] += 8;
        dst[2] += 8;
    }
}

static void filter_mb_row_simple(VP8Context *s, int mb_y)
{
    uint8_t *dst = s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize;
    VP8Macroblock *mb = s->macroblocks + mb_y*s->mb_stride;
    int mb_x;

    for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
        backup_mb_border(s->top_border[mb_x+1], dst, NULL, NULL, s->linesize, 0, 1);
        filter_mb_simple(s, dst, mb++, mb_x, mb_y);
        dst += 16;
    }
}

static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,

                if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
                    s->left_nnz[8]      = 0;
                    s->top_nnz[mb_x][8] = 0;
                }
            }

            dst[0] += 16;
            dst[1] += 8;
            dst[2] += 8;
            mb++;

12222:7acdbfd2a222

#include "vp8dsp.h"
#include "h264pred.h"
#include "rectangle.h"

typedef struct {
    uint8_t filter_level;
    uint8_t inner_limit;
} VP8FilterStrength;

typedef struct {
    uint8_t segment;
    uint8_t skip;
    // todo: make it possible to check for at least (i4x4 or split_mv)
    // in one op. are others needed?
    uint8_t mode;

    int num_coeff_partitions;
    VP56RangeCoder coeff_partition[8];

    VP8Macroblock *macroblocks;
    VP8Macroblock *macroblocks_base;
    VP8FilterStrength *filter_strength;
    int mb_stride;

    uint8_t *intra4x4_pred_mode;
    uint8_t *intra4x4_pred_mode_base;
    int b4_stride;

    // this is 4 blocks for intra4x4 to keep 4-byte alignment for fill_rectangle
    s->mb_stride = s->mb_width+1;
    s->b4_stride = 4*s->mb_stride;

    s->macroblocks_base        = av_mallocz(s->mb_stride*(s->mb_height+1)*sizeof(*s->macroblocks));
    s->filter_strength         = av_mallocz(s->mb_stride*sizeof(*s->filter_strength));
    s->intra4x4_pred_mode_base = av_mallocz(s->b4_stride*(4*s->mb_height+1));
    s->top_nnz                 = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
    s->top_border              = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));

    if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_base || !s->top_nnz || !s->top_border)
        return AVERROR(ENOMEM);

    s->macroblocks        = s->macroblocks_base        + 1 + s->mb_stride;
    s->intra4x4_pred_mode = s->intra4x4_pred_mode_base + 4 + s->b4_stride;

        u_dst += 4*s->uvlinesize;
        v_dst += 4*s->uvlinesize;
    }
}

static void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb, VP8FilterStrength *f )
{
    int interior_limit, filter_level;

    if (s->segmentation.enabled) {
        filter_level = s->segmentation.filter_level[mb->segment];

        interior_limit >>= s->filter.sharpness > 4 ? 2 : 1;
        interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
    }
    interior_limit = FFMAX(interior_limit, 1);

    f->filter_level = filter_level;
    f->inner_limit = interior_limit;
}

static void filter_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb, VP8FilterStrength *f, int mb_x, int mb_y)
{
    int mbedge_lim, bedge_lim, hev_thresh;
    int filter_level = f->filter_level;
    int inner_limit = f->inner_limit;

    if (!filter_level)
        return;

    mbedge_lim = 2*(filter_level+2) + inner_limit;
     bedge_lim = 2* filter_level    + inner_limit;
    hev_thresh = filter_level >= 15;

    if (s->keyframe) {
        if (filter_level >= 40)
            hev_thresh = 2;
    } else {
        if (filter_level >= 40)
            hev_thresh = 3;
        else if (filter_level >= 20)
            hev_thresh = 2;
    }

    if (mb_x) {
        s->vp8dsp.vp8_h_loop_filter16y(dst[0],     s->linesize,
                                       mbedge_lim, inner_limit, hev_thresh);
        s->vp8dsp.vp8_h_loop_filter8uv(dst[1],     dst[2],      s->uvlinesize,

                                             s->uvlinesize, bedge_lim,
                                             inner_limit,   hev_thresh);
    }
}

static void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8Macroblock *mb, VP8FilterStrength *f, int mb_x, int mb_y)
{
    int mbedge_lim, bedge_lim;
    int filter_level = f->filter_level;
    int inner_limit = f->inner_limit;

    if (!filter_level)
        return;

    mbedge_lim = 2*(filter_level+2) + inner_limit;
     bedge_lim = 2* filter_level    + inner_limit;

    }
}

static void filter_mb_row(VP8Context *s, int mb_y)
{
    VP8FilterStrength *f = s->filter_strength;
    VP8Macroblock *mb = s->macroblocks + mb_y*s->mb_stride;
    uint8_t *dst[3] = {
        s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize,
        s->framep[VP56_FRAME_CURRENT]->data[1] +  8*mb_y*s->uvlinesize,
        s->framep[VP56_FRAME_CURRENT]->data[2] +  8*mb_y*s->uvlinesize
    };
    int mb_x;

    for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
        backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
        filter_mb(s, dst, mb++, f++, mb_x, mb_y);
        dst[0] += 16;
        dst[1] += 8;
        dst[2] += 8;
    }
}

static void filter_mb_row_simple(VP8Context *s, int mb_y)
{
    VP8FilterStrength *f = s->filter_strength;
    VP8Macroblock *mb = s->macroblocks + mb_y*s->mb_stride;
    uint8_t *dst = s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize;
    int mb_x;

    for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
        backup_mb_border(s->top_border[mb_x+1], dst, NULL, NULL, s->linesize, 0, 1);
        filter_mb_simple(s, dst, mb++, f++, mb_x, mb_y);
        dst += 16;
    }
}

static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,

                if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
                    s->left_nnz[8]      = 0;
                    s->top_nnz[mb_x][8] = 0;
                }
            }

            if (s->deblock_filter)
                filter_level_for_mb(s, mb, &s->filter_strength[mb_x]);

            dst[0] += 16;
            dst[1] += 8;
            dst[2] += 8;
            mb++;