summaryrefslogtreecommitdiff
path: root/libavfilter/vf_xmedian.c
diff options
context:
space:
mode:
authorPaul B Mahol <onemda@gmail.com>2022-02-18 16:13:16 +0100
committerPaul B Mahol <onemda@gmail.com>2022-02-18 17:12:46 +0100
commit51cd565208123f5e69053bb37b9ae7f1c6e9d9b2 (patch)
tree68d98214a25aca831f0399b319b78d9061570e54 /libavfilter/vf_xmedian.c
parentcca6fe7fe0482d06391fb34858aa9e3d59b590b2 (diff)
avfilter/vf_xmedian: small additional speedup
Diffstat (limited to 'libavfilter/vf_xmedian.c')
-rw-r--r--libavfilter/vf_xmedian.c32
1 files changed, 29 insertions, 3 deletions
diff --git a/libavfilter/vf_xmedian.c b/libavfilter/vf_xmedian.c
index 28d8e6d4e9..148ab555bb 100644
--- a/libavfilter/vf_xmedian.c
+++ b/libavfilter/vf_xmedian.c
@@ -36,6 +36,7 @@ typedef struct XMedianContext {
const AVPixFmtDescriptor *desc;
int nb_inputs;
int nb_frames;
+ int nb_threads;
int planes;
float percentile;
@@ -49,6 +50,9 @@ typedef struct XMedianContext {
int width[4];
int height[4];
+ uint8_t **data;
+ int *linesize;
+
AVFrame **frames;
FFFrameSync fs;
@@ -129,6 +133,8 @@ static int median_frames ## name(AVFilterContext *ctx, void *arg, int jobnr, int
AVFrame **in = td->in; \
AVFrame *out = td->out; \
const int nb_inputs = s->nb_inputs; \
+ uint8_t **srcf = s->data + jobnr * nb_inputs; \
+ int *linesize = s->linesize + jobnr * nb_inputs; \
const int radius = s->radius; \
const int index = s->index; \
type values[256]; \
@@ -136,6 +142,7 @@ static int median_frames ## name(AVFilterContext *ctx, void *arg, int jobnr, int
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->width[p]; \
type *dst = (type *)(out->data[p] + slice_start * out->linesize[p]); \
ptrdiff_t dst_linesize = out->linesize[p] / sizeof(type); \
\
@@ -147,10 +154,16 @@ static int median_frames ## name(AVFilterContext *ctx, void *arg, int jobnr, int
continue; \
} \
\
+ for (int i = 0; i < nb_inputs; i++) \
+ linesize[i] = in[i]->linesize[p]; \
+ \
+ for (int i = 0; i < 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 < s->width[p]; x++) { \
+ for (int x = 0; x < width; x++) { \
for (int i = 0; i < nb_inputs; i++) { \
- const type *src = (const type *)(in[i]->data[p] + y * in[i]->linesize[p]); \
+ const type *src = (const type *)srcf[i]; \
values[i] = src[x]; \
} \
\
@@ -162,6 +175,8 @@ static int median_frames ## name(AVFilterContext *ctx, void *arg, int jobnr, int
} \
\
dst += dst_linesize; \
+ for (int i = 0; i < nb_inputs; i++) \
+ srcf[i] += linesize[i]; \
} \
} \
\
@@ -199,7 +214,7 @@ static int process_frame(FFFrameSync *fs)
td.in = in;
td.out = out;
ff_filter_execute(ctx, s->median_frames, &td, NULL,
- FFMIN(s->height[1], ff_filter_get_nb_threads(ctx)));
+ FFMIN(s->height[1], s->nb_threads));
}
return ff_filter_frame(outlink, out);
@@ -230,6 +245,7 @@ static int config_output(AVFilterLink *outlink)
s->nb_planes = av_pix_fmt_count_planes(outlink->format);
s->depth = s->desc->comp[0].depth;
s->max = (1 << s->depth) - 1;
+ s->nb_threads = ff_filter_get_nb_threads(ctx);
if (s->depth <= 8)
s->median_frames = median_frames8;
@@ -244,6 +260,14 @@ static int config_output(AVFilterLink *outlink)
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->xmedian)
return 0;
@@ -283,6 +307,8 @@ static av_cold void uninit(AVFilterContext *ctx)
for (int i = 0; i < s->nb_frames && s->frames && !s->xmedian; i++)
av_frame_free(&s->frames[i]);
av_freep(&s->frames);
+ av_freep(&s->data);
+ av_freep(&s->linesize);
}
static int activate(AVFilterContext *ctx)