From: Niklas Haas <g...@haasn.dev>
This gives vastly improved blending results than when blending directly in
the desired output colorspace. Overridable by the existing "disable_linear"
option.
This is functionally similar to combining multiple "libplacebo" filters,
but does not rely on the existence of a Vulkan filter link, so it can be used
without performance penalty in all circumstances. It's also enabled by
default, without requiring special action from the user.
---
libavfilter/vf_libplacebo.c | 51 +++++++++++++++++++++++++++++++++++++
1 file changed, 51 insertions(+)
diff --git a/libavfilter/vf_libplacebo.c b/libavfilter/vf_libplacebo.c
index 475030c80d..347151acd3 100644
--- a/libavfilter/vf_libplacebo.c
+++ b/libavfilter/vf_libplacebo.c
@@ -160,6 +160,10 @@ typedef struct LibplaceboContext {
pl_gpu gpu;
pl_tex tex[4];
+ /* dedicated renderer for linear output composition */
+ pl_renderer linear_rr;
+ pl_tex linear_tex;
+
/* input state */
LibplaceboInput *inputs;
int nb_inputs;
@@ -735,6 +739,7 @@ static int init_vulkan(AVFilterContext *avctx, const
AVVulkanDeviceContext *hwct
return AVERROR(ENOMEM);
for (int i = 0; i < s->nb_inputs; i++)
RET(input_init(avctx, &s->inputs[i], i));
+ s->linear_rr = pl_renderer_create(s->log, s->gpu);
/* fall through */
fail:
@@ -760,6 +765,8 @@ static void libplacebo_uninit(AVFilterContext *avctx)
#if PL_API_VER >= 351
pl_cache_destroy(&s->cache);
#endif
+ pl_renderer_destroy(&s->linear_rr);
+ pl_tex_destroy(s->gpu, &s->linear_tex);
pl_options_free(&s->opts);
pl_vulkan_destroy(&s->vulkan);
pl_log_destroy(&s->log);
@@ -953,6 +960,20 @@ static int output_frame(AVFilterContext *ctx, int64_t pts)
goto fail;
}
+ 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) {
+ use_linear_compositor = true;
+ target.color.transfer = PL_COLOR_TRC_LINEAR;
+ target.repr = pl_color_repr_rgb;
+ target.num_planes = 1;
+ target.planes[0] = (struct pl_plane) {
+ .components = 4,
+ .component_mapping = {0, 1, 2, 3},
+ .texture = s->linear_tex,
+ };
+ }
+
/* Draw first frame opaque, others with blending */
opts->params.blend_params = NULL;
#if PL_API_VER >= 346
@@ -981,6 +1002,13 @@ static int output_frame(AVFilterContext *ctx, int64_t pts)
#endif
}
+ if (use_linear_compositor) {
+ /* Blit the linear intermediate image to the output frame */
+ target.crop = orig_target.crop = (struct pl_rect2df) {0};
+ pl_render_image(s->linear_rr, &target, &orig_target, &opts->params);
+ target = orig_target;
+ }
+
if (outdesc->flags & AV_PIX_FMT_FLAG_HWACCEL) {
pl_unmap_avframe(s->gpu, &target);
} else if (!pl_download_avframe(s->gpu, &target, out)) {
@@ -1307,6 +1335,29 @@ static int libplacebo_config_output(AVFilterLink
*outlink)
s->force_divisible_by,
s->reset_sar ? sar_in : 1.0);
+
+ if (s->nb_inputs > 1 && !s->disable_fbos) {
+ /* Create a separate renderer and composition texture */
+ pl_fmt fmt = pl_find_fmt(s->gpu, PL_FMT_FLOAT, 4, 16, 0,
PL_FMT_CAP_BLENDABLE);
+ bool ok = !!fmt;
+ if (ok) {
+ ok = pl_tex_recreate(s->gpu, &s->linear_tex, pl_tex_params(
+ .format = fmt,
+ .w = outlink->w,
+ .h = outlink->h,
+ .renderable = true,
+ .sampleable = true,
+ .storable = fmt->caps & PL_FMT_CAP_STORABLE,
+ ));
+ }
+
+ if (!ok) {
+ av_log(s, AV_LOG_WARNING, "Failed to create a linear texture for "
+ "compositing multiple inputs, falling back to non-linear "
+ "blending.\n");
+ }
+ }
+
if (s->reset_sar) {
/* SAR is normalized, or we have multiple inputs, set out to 1:1 */
outlink->sample_aspect_ratio = (AVRational){ 1, 1 };
--
2.50.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
