avfilter/vf_mix: additional speedup

This commit is contained in:
Paul B Mahol 2022-02-18 11:25:11 +01:00
parent 656cb0450a
commit 33417198aa
1 changed files with 31 additions and 6 deletions

View File

@ -35,6 +35,7 @@ typedef struct MixContext {
const AVPixFmtDescriptor *desc;
char *weights_str;
int nb_inputs;
int nb_threads;
int duration;
float *weights;
float scale;
@ -47,9 +48,12 @@ typedef struct MixContext {
int max;
int planes;
int nb_planes;
int linesize[4];
int linesizes[4];
int height[4];
uint8_t **data;
int *linesize;
AVFrame **frames;
FFFrameSync fs;
} MixContext;
@ -143,7 +147,7 @@ typedef struct ThreadData {
for (int p = 0; p < s->nb_planes; p++) { \
const int slice_start = (s->height[p] * jobnr) / nb_jobs; \
const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs; \
const int width = s->linesize[p] / sizeof(type); \
const int width = s->linesizes[p] / sizeof(type); \
type *dst = (type *)(out->data[p] + slice_start * out->linesize[p]); \
ptrdiff_t dst_linesize = out->linesize[p] / sizeof(type); \
\
@ -151,16 +155,22 @@ typedef struct ThreadData {
av_image_copy_plane((uint8_t *)dst, out->linesize[p], \
in[0]->data[p] + slice_start * in[0]->linesize[p], \
in[0]->linesize[p], \
s->linesize[p], slice_end - slice_start); \
s->linesizes[p], slice_end - slice_start); \
continue; \
} \
\
for (int i = 0; i < s->nb_inputs; i++) \
linesize[i] = in[i]->linesize[p]; \
\
for (int i = 0; i < s->nb_inputs; i++) \
srcf[i] = in[i]->data[p] + slice_start * linesize[i]; \
\
for (int y = slice_start; y < slice_end; y++) { \
for (int x = 0; x < width; x++) { \
float val = 0.f; \
\
for (int i = 0; i < s->nb_inputs; i++) { \
float src = *(type *)(in[i]->data[p] + y * in[i]->linesize[p] + x * sizeof(type));\
float src = *(type *)(srcf[i] + x * sizeof(type)); \
\
val += src * weights[i]; \
} \
@ -169,6 +179,8 @@ typedef struct ThreadData {
} \
\
dst += dst_linesize; \
for (int i = 0; i < s->nb_inputs; i++) \
srcf[i] += linesize[i]; \
} \
}
@ -184,6 +196,8 @@ static int mix_frames(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
AVFrame **in = td->in;
AVFrame *out = td->out;
const float *weights = s->weights;
uint8_t **srcf = s->data + jobnr * s->nb_inputs;
int *linesize = s->linesize + jobnr * s->nb_inputs;
if (s->depth <= 8) {
MIX_SLICE(uint8_t, lrintf, CLIP8)
@ -227,7 +241,7 @@ static int process_frame(FFFrameSync *fs)
td.in = in;
td.out = out;
ff_filter_execute(ctx, mix_frames, &td, NULL,
FFMIN(s->height[0], ff_filter_get_nb_threads(ctx)));
FFMIN(s->height[0], s->nb_threads));
return ff_filter_frame(outlink, out);
}
@ -253,6 +267,7 @@ static int config_output(AVFilterLink *outlink)
}
}
s->nb_threads = ff_filter_get_nb_threads(ctx);
s->desc = av_pix_fmt_desc_get(outlink->format);
if (!s->desc)
return AVERROR_BUG;
@ -260,12 +275,20 @@ static int config_output(AVFilterLink *outlink)
s->depth = s->desc->comp[0].depth;
s->max = (1 << s->depth) - 1;
if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
if ((ret = av_image_fill_linesizes(s->linesizes, inlink->format, inlink->w)) < 0)
return ret;
s->height[1] = s->height[2] = AV_CEIL_RSHIFT(inlink->h, s->desc->log2_chroma_h);
s->height[0] = s->height[3] = inlink->h;
s->data = av_calloc(s->nb_threads * s->nb_inputs, sizeof(*s->data));
if (!s->data)
return AVERROR(ENOMEM);
s->linesize = av_calloc(s->nb_threads * s->nb_inputs, sizeof(*s->linesize));
if (!s->linesize)
return AVERROR(ENOMEM);
if (s->tmix)
return 0;
@ -303,6 +326,8 @@ static av_cold void uninit(AVFilterContext *ctx)
ff_framesync_uninit(&s->fs);
av_freep(&s->weights);
av_freep(&s->data);
av_freep(&s->linesize);
if (s->tmix) {
for (i = 0; i < s->nb_frames && s->frames; i++)