virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-10 06:10:52 +02:00
Code Issues Releases Activity

avfilter/vf_libplacebo: skip rendering fully invisible planes

Browse Source 
Sometimes, one input fully obscures another. In this case, we can skip
actually rendering any input below the obscuring one.

The reason I don't simply start the main render loop at `idx_start` will
become apparent in the following commit.

We can't use pl_frame_is_cropped() on this dummy frame, but we need to
determine the reference frame before we can map the real output, so to
resolve this conflict, we just reimplement the crop detection logic using
the output link dimensions.
This commit is contained in:
Niklas Haas
2025-08-07 22:04:27 +02:00
committed by Niklas Haas
parent 6627c8ea4b
commit 3091bca3ed
1 changed files with 35 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
libavfilter/vf_libplacebo.c
View File

@@ -16,6 +16,8 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <math.h>
#include "libavutil/avassert.h"
#include "libavutil/eval.h"
#include "libavutil/fifo.h"
@@ -913,10 +915,39 @@ static int output_frame(AVFilterContext *ctx, int64_t pts)
pl_options opts = s->opts;
AVFilterLink *outlink = ctx->outputs[0];
const AVPixFmtDescriptor *outdesc = av_pix_fmt_desc_get(outlink->format);
const double target_pts = pts * av_q2d(outlink->time_base);
struct pl_frame target;
const AVFrame *ref = NULL;
AVFrame *out;
/* Count the number of visible inputs, by excluding frames which are fully
* obscured or which have no frames in the mix */
int idx_start = 0, nb_visible = 0;
for (int i = 0; i < s->nb_inputs; i++) {
LibplaceboInput *in = &s->inputs[i];
struct pl_frame dummy;
if (in->qstatus != PL_QUEUE_OK || !in->mix.num_frames)
continue;
const struct pl_frame *cur = pl_frame_mix_current(&in->mix);
update_crops(ctx, in, &dummy, target_pts);
const int x0 = roundf(FFMIN(dummy.crop.x0, dummy.crop.x1)),
y0 = roundf(FFMIN(dummy.crop.y0, dummy.crop.y1)),
x1 = roundf(FFMAX(dummy.crop.x0, dummy.crop.x1)),
y1 = roundf(FFMAX(dummy.crop.y0, dummy.crop.y1));
/* If an opaque frame covers entire the output, disregard all lower layers */
const bool cropped = x0 > 0 || y0 > 0 || x1 < outlink->w || y1 < outlink->h;
if (!cropped && cur->repr.alpha == PL_ALPHA_NONE) {
idx_start = i;
nb_visible = 0;
}
nb_visible++;
}
/* It should be impossible to call output_frame() without at least one
* valid nonempty frame mix */
av_assert1(nb_visible > 0);
/* Use the first active input as metadata reference */
for (int i = 0; i < s->nb_inputs; i++) {
const LibplaceboInput *in = &s->inputs[i];
@@ -989,7 +1020,8 @@ static int output_frame(AVFilterContext *ctx, int64_t pts)
struct pl_frame orig_target = target;
bool use_linear_compositor = false;
if (s->linear_tex && target.color.transfer != PL_COLOR_TRC_LINEAR && !s->disable_linear) {
if (s->linear_tex && target.color.transfer != PL_COLOR_TRC_LINEAR &&
!s->disable_linear && nb_visible > 1) {
target = (struct pl_frame) {
.num_planes = 1,
.planes[0] = {
@@ -1017,10 +1049,10 @@ static int output_frame(AVFilterContext *ctx, int64_t pts)
FilterLink *il = ff_filter_link(ctx->inputs[i]);
FilterLink *ol = ff_filter_link(outlink);
int high_fps = av_cmp_q(il->frame_rate, ol->frame_rate) >= 0;
if (in->qstatus != PL_QUEUE_OK || !in->mix.num_frames)
if (in->qstatus != PL_QUEUE_OK || !in->mix.num_frames || i < idx_start)
continue;
opts->params.skip_caching_single_frame = high_fps;
update_crops(ctx, in, &target, out->pts * av_q2d(outlink->time_base));
update_crops(ctx, in, &target, target_pts);
pl_render_image_mix(in->renderer, &in->mix, &target, &opts->params);
/* Force straight output and set correct blend mode */