[FFmpeg-devel] [PATCH] avfilter/vf_psnr: add support for slice threading
Paul B Mahol
onemda at gmail.com
Tue Feb 16 22:54:26 EET 2021
Signed-off-by: Paul B Mahol <onemda at gmail.com>
---
libavfilter/vf_psnr.c | 126 +++++++++++++++++++++++++++++-------------
1 file changed, 89 insertions(+), 37 deletions(-)
diff --git a/libavfilter/vf_psnr.c b/libavfilter/vf_psnr.c
index 0d6bbad42f..0d0c0f35aa 100644
--- a/libavfilter/vf_psnr.c
+++ b/libavfilter/vf_psnr.c
@@ -51,9 +51,11 @@ typedef struct PSNRContext {
uint8_t rgba_map[4];
char comps[4];
int nb_components;
+ int nb_threads;
int planewidth[4];
int planeheight[4];
double planeweight[4];
+ uint64_t **score;
PSNRDSPContext dsp;
} PSNRContext;
@@ -104,29 +106,44 @@ static uint64_t sse_line_16bit(const uint8_t *_main_line, const uint8_t *_ref_li
return m2;
}
-static inline
-void compute_images_mse(PSNRContext *s,
- const uint8_t *main_data[4], const int main_linesizes[4],
- const uint8_t *ref_data[4], const int ref_linesizes[4],
- int w, int h, double mse[4])
+typedef struct ThreadData {
+ const uint8_t *main_data[4];
+ const uint8_t *ref_data[4];
+ int main_linesize[4];
+ int ref_linesize[4];
+ int planewidth[4];
+ int planeheight[4];
+ uint64_t **score;
+ int nb_components;
+ PSNRDSPContext *dsp;
+} ThreadData;
+
+static
+int compute_images_mse(AVFilterContext *ctx, void *arg,
+ int jobnr, int nb_jobs)
{
- int i, c;
-
- for (c = 0; c < s->nb_components; c++) {
- const int outw = s->planewidth[c];
- const int outh = s->planeheight[c];
- const uint8_t *main_line = main_data[c];
- const uint8_t *ref_line = ref_data[c];
- const int ref_linesize = ref_linesizes[c];
- const int main_linesize = main_linesizes[c];
+ ThreadData *td = arg;
+ uint64_t *score = td->score[jobnr];
+
+ for (int c = 0; c < td->nb_components; c++) {
+ const int outw = td->planewidth[c];
+ const int outh = td->planeheight[c];
+ const int slice_start = (outh * jobnr) / nb_jobs;
+ const int slice_end = (outh * (jobnr+1)) / nb_jobs;
+ const int ref_linesize = td->ref_linesize[c];
+ const int main_linesize = td->main_linesize[c];
+ const uint8_t *main_line = td->main_data[c] + main_linesize * slice_start;
+ const uint8_t *ref_line = td->ref_data[c] + ref_linesize * slice_start;
uint64_t m = 0;
- for (i = 0; i < outh; i++) {
- m += s->dsp.sse_line(main_line, ref_line, outw);
+ for (int i = slice_start; i < slice_end; i++) {
+ m += td->dsp->sse_line(main_line, ref_line, outw);
ref_line += ref_linesize;
main_line += main_linesize;
}
- mse[c] = m / (double)(outw * outh);
+ score[c] = m;
}
+
+ return 0;
}
static void set_meta(AVDictionary **metadata, const char *key, char comp, float d)
@@ -147,9 +164,11 @@ static int do_psnr(FFFrameSync *fs)
AVFilterContext *ctx = fs->parent;
PSNRContext *s = ctx->priv;
AVFrame *master, *ref;
- double comp_mse[4], mse = 0;
- int ret, j, c;
+ double comp_mse[4], mse = 0.;
+ uint64_t comp_sum[4] = { 0 };
AVDictionary **metadata;
+ ThreadData td;
+ int ret;
ret = ff_framesync_dualinput_get(fs, &master, &ref);
if (ret < 0)
@@ -158,23 +177,42 @@ static int do_psnr(FFFrameSync *fs)
return ff_filter_frame(ctx->outputs[0], master);
metadata = &master->metadata;
- compute_images_mse(s, (const uint8_t **)master->data, master->linesize,
- (const uint8_t **)ref->data, ref->linesize,
- master->width, master->height, comp_mse);
+ td.nb_components = s->nb_components;
+ td.dsp = &s->dsp;
+ td.score = s->score;
+ for (int c = 0; c < s->nb_components; c++) {
+ td.main_data[c] = master->data[c];
+ td.ref_data[c] = ref->data[c];
+ td.main_linesize[c] = master->linesize[c];
+ td.ref_linesize[c] = ref->linesize[c];
+ td.planewidth[c] = s->planewidth[c];
+ td.planeheight[c] = s->planeheight[c];
+ }
- for (j = 0; j < s->nb_components; j++)
- mse += comp_mse[j] * s->planeweight[j];
+ ctx->internal->execute(ctx, compute_images_mse, &td, NULL, FFMIN(s->planeheight[1], s->nb_threads));
+
+ for (int j = 0; j < s->nb_threads; j++) {
+ for (int c = 0; c < s->nb_components; c++)
+ comp_sum[c] += s->score[j][c];
+ }
+
+ for (int c = 0; c < s->nb_components; c++)
+ comp_mse[c] = comp_sum[c] / ((double)s->planewidth[c] * s->planeheight[c]);
+
+ for (int c = 0; c < s->nb_components; c++)
+ mse += comp_mse[c] * s->planeweight[c];
s->min_mse = FFMIN(s->min_mse, mse);
s->max_mse = FFMAX(s->max_mse, mse);
s->mse += mse;
- for (j = 0; j < s->nb_components; j++)
+
+ for (int j = 0; j < s->nb_components; j++)
s->mse_comp[j] += comp_mse[j];
s->nb_frames++;
- for (j = 0; j < s->nb_components; j++) {
- c = s->is_rgb ? s->rgba_map[j] : j;
+ for (int j = 0; j < s->nb_components; j++) {
+ int c = s->is_rgb ? s->rgba_map[j] : j;
set_meta(metadata, "lavfi.psnr.mse.", s->comps[j], comp_mse[c]);
set_meta(metadata, "lavfi.psnr.psnr.", s->comps[j], get_psnr(comp_mse[c], 1, s->max[c]));
}
@@ -185,16 +223,16 @@ static int do_psnr(FFFrameSync *fs)
if (s->stats_version == 2 && !s->stats_header_written) {
fprintf(s->stats_file, "psnr_log_version:2 fields:n");
fprintf(s->stats_file, ",mse_avg");
- for (j = 0; j < s->nb_components; j++) {
+ for (int j = 0; j < s->nb_components; j++) {
fprintf(s->stats_file, ",mse_%c", s->comps[j]);
}
fprintf(s->stats_file, ",psnr_avg");
- for (j = 0; j < s->nb_components; j++) {
+ for (int j = 0; j < s->nb_components; j++) {
fprintf(s->stats_file, ",psnr_%c", s->comps[j]);
}
if (s->stats_add_max) {
fprintf(s->stats_file, ",max_avg");
- for (j = 0; j < s->nb_components; j++) {
+ for (int j = 0; j < s->nb_components; j++) {
fprintf(s->stats_file, ",max_%c", s->comps[j]);
}
}
@@ -202,20 +240,20 @@ static int do_psnr(FFFrameSync *fs)
s->stats_header_written = 1;
}
fprintf(s->stats_file, "n:%"PRId64" mse_avg:%0.2f ", s->nb_frames, mse);
- for (j = 0; j < s->nb_components; j++) {
- c = s->is_rgb ? s->rgba_map[j] : j;
+ for (int j = 0; j < s->nb_components; j++) {
+ int c = s->is_rgb ? s->rgba_map[j] : j;
fprintf(s->stats_file, "mse_%c:%0.2f ", s->comps[j], comp_mse[c]);
}
fprintf(s->stats_file, "psnr_avg:%0.2f ", get_psnr(mse, 1, s->average_max));
- for (j = 0; j < s->nb_components; j++) {
- c = s->is_rgb ? s->rgba_map[j] : j;
+ for (int j = 0; j < s->nb_components; j++) {
+ int c = s->is_rgb ? s->rgba_map[j] : j;
fprintf(s->stats_file, "psnr_%c:%0.2f ", s->comps[j],
get_psnr(comp_mse[c], 1, s->max[c]));
}
if (s->stats_version == 2 && s->stats_add_max) {
fprintf(s->stats_file, "max_avg:%d ", s->average_max);
- for (j = 0; j < s->nb_components; j++) {
- c = s->is_rgb ? s->rgba_map[j] : j;
+ for (int j = 0; j < s->nb_components; j++) {
+ int c = s->is_rgb ? s->rgba_map[j] : j;
fprintf(s->stats_file, "max_%c:%d ", s->comps[j], s->max[c]);
}
}
@@ -289,6 +327,7 @@ static int config_input_ref(AVFilterLink *inlink)
unsigned sum;
int j;
+ s->nb_threads = ff_filter_get_nb_threads(ctx);
s->nb_components = desc->nb_components;
if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
ctx->inputs[0]->h != ctx->inputs[1]->h) {
@@ -329,6 +368,16 @@ static int config_input_ref(AVFilterLink *inlink)
if (ARCH_X86)
ff_psnr_init_x86(&s->dsp, desc->comp[0].depth);
+ s->score = av_calloc(s->nb_threads, sizeof(*s->score));
+ if (!s->score)
+ return AVERROR(ENOMEM);
+
+ for (int t = 0; t < s->nb_threads && s->score; t++) {
+ s->score[t] = av_calloc(s->nb_components, sizeof(*s->score[0]));
+ if (!s->score[t])
+ return AVERROR(ENOMEM);
+ }
+
return 0;
}
@@ -389,6 +438,9 @@ static av_cold void uninit(AVFilterContext *ctx)
}
ff_framesync_uninit(&s->fs);
+ for (int t = 0; t < s->nb_threads && s->score; t++)
+ av_freep(&s->score[t]);
+ av_freep(&s->score);
if (s->stats_file && s->stats_file != stdout)
fclose(s->stats_file);
@@ -427,5 +479,5 @@ AVFilter ff_vf_psnr = {
.priv_class = &psnr_class,
.inputs = psnr_inputs,
.outputs = psnr_outputs,
- .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
+ .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
};
--
2.17.1
More information about the ffmpeg-devel
mailing list