[PATCH 1/2] compositor: Clear only the the damage that was actually repainted
Kristian Høgsberg
hoegsberg at gmail.com
Thu Aug 16 07:41:02 PDT 2012
On Thu, Aug 16, 2012 at 11:22:37AM +0300, Ander Conselvan de Oliveira wrote:
> Em 15-08-2012 23:02, Kristian Høgsberg escreveu:
> >On Tue, Aug 14, 2012 at 03:26:49PM +0300, Ander Conselvan de Oliveira wrote:
> >>From: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira at intel.com>
> >>
> >>Instead of clearing the whole output region after a repaint, clear
> >>only the regions that were actually painted. This way, the damage
> >>added when a surface moves from the primary plane to another one is
> >>kept while this region is obscured by the opaque region. This allows
> >>the contents below an overlaid surface to be culled, but to make this
> >>work properly, it is also necessary to change the way previous damage
> >>is drawn.
> >>
> >>Consider the following scenario: a surface is moved to an overlay plane
> >>leaving some damage in the primary plane. On the following frame, the
> >>surface on the overlay moves, revealing part of the damaged region on
> >>the primary plane. On the frame after that, the overlaid surface moves
> >>back to its previous position obscuring the region of the primary plane
> >>repainted before. At this point, the repainted region was added to the
> >>output's previous damage so that it is draw to both buffers. But since
> >>this region is now obscured, the redrawing is skipped. If the overlaid
> >>surface moves again revealing this region, one of the buffers actually
> >>contains the wrong content.
> >>
> >>To fix this problem, this patch ensures that any previous damage that
> >>would be lost is actually drawn, by subtracting it from the opaque
> >>region.
> >
> >Ah, hm, ok... that's really subtle. Good analysis. Now with your
> >fix, we still repaint when we move the opaque overlay back over a
> >region with "old damage" even though it's not visible. How about
> >unioning output->previous_damage into the primary plane damage just
> >before repainting? That way, if any old damage doesn't get repainted,
> >we add it back to the primary plane damage. The only drawback here is
> >that we move the overlay away again, we could end up repainting that
> >area again. But once all buffers has repainted the area, the residual
> >damage will disappear.
> >
> >I've tweaked your patch to do that and attached it. I think it's a
> >slight improvement, in that we probably typically end up painting less
> >and it's a little simpler.
>
> Yeah, that looks better. You did, however, remove one too many lines
> from surface_accumulate_damage() so the opaque region is always
> empty.
I did, didn't I... that kinda defeats the purpose... I put that line
back and committed the patch. Don't know if you noticed, but without
this patch, we got a one-frame flicker, as the sprite doesn't enable a
frame later. So when we repaint the primary fb to not contain the
opaque overlay, we get one frame where the overlay is not enabled and
yet also not in the primary plane. This patch works around that by
optimizing out the repaint, but of course, the right fix here is the
kms nuclear pageflip.
> >However, either approach is a workaround
> >how we should track the damage per buffer. When we repaint an output,
> >we add the new damage (from primary_plane.damage) to all buffers
> >damage, then render the damage (clipped by opaque regions) in current
> >back buffers damage region, subtracting what we render. This also
> >scales to triple buffering.
>
> And with some refactoring to track this on a per plane basis, it can
> also scale to the case where there are multiple surfaces in an ARGB
> overlay.
Yup, something that we should keep in mind.
thanks,
Kristian
>
> Ander
>
>
> >
> >Kristian
> >
> >>---
> >> src/compositor.c | 26 ++++++++++++++++++--------
> >> 1 file changed, 18 insertions(+), 8 deletions(-)
> >>
> >>diff --git a/src/compositor.c b/src/compositor.c
> >>index 7370435..26c57b9 100644
> >>--- a/src/compositor.c
> >>+++ b/src/compositor.c
> >>@@ -928,6 +928,9 @@ weston_surface_draw(struct weston_surface *es, struct weston_output *output,
> >> if (!pixman_region32_not_empty(&repaint))
> >> goto out;
> >>
> >>+ pixman_region32_subtract(&ec->primary_plane.damage,
> >>+ &ec->primary_plane.damage, &repaint);
> >>+
> >> glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
> >> if (es->blend || es->alpha < 1.0)
> >> glEnable(GL_BLEND);
> >>@@ -1095,7 +1098,8 @@ update_shm_texture(struct weston_surface *surface)
> >>
> >> static void
> >> surface_accumulate_damage(struct weston_surface *surface,
> >>- pixman_region32_t *opaque)
> >>+ pixman_region32_t *opaque,
> >>+ pixman_region32_t *exclude)
> >> {
> >> if (surface->buffer && wl_buffer_is_shm(surface->buffer))
> >> update_shm_texture(surface);
> >>@@ -1121,9 +1125,8 @@ surface_accumulate_damage(struct weston_surface *surface,
> >> pixman_region32_union(&surface->plane->damage,
> >> &surface->plane->damage, &surface->damage);
> >> empty_region(&surface->damage);
> >>- pixman_region32_copy(&surface->clip, opaque);
> >>- if (surface->plane == &surface->compositor->primary_plane)
> >>- pixman_region32_union(opaque, opaque, &surface->transform.opaque);
> >>+ pixman_region32_subtract(&surface->clip, opaque, exclude);
> >>+ pixman_region32_union(opaque, opaque, &surface->transform.opaque);
> >> }
> >>
> >> static void
> >>@@ -1135,7 +1138,7 @@ weston_output_repaint(struct weston_output *output, uint32_t msecs)
> >> struct weston_animation *animation, *next;
> >> struct weston_frame_callback *cb, *cnext;
> >> struct wl_list frame_callback_list;
> >>- pixman_region32_t opaque, output_damage;
> >>+ pixman_region32_t opaque, output_damage, previous;
> >> int32_t width, height;
> >>
> >> weston_compositor_update_drag_surfaces(ec);
> >>@@ -1170,8 +1173,16 @@ weston_output_repaint(struct weston_output *output, uint32_t msecs)
> >>
> >> pixman_region32_init(&opaque);
> >>
> >>+ /* Once we draw something on one of the back buffers it needs to be
> >>+ * drawn on the other one, otherwise we might endup with artifacts.
> >>+ * Make sure we draw any previous damage that would be lost otherwise
> >>+ */
> >>+ pixman_region32_init(&previous);
> >>+ pixman_region32_subtract(&previous, &output->previous_damage,
> >>+ &ec->primary_plane.damage);
> >>+
> >> wl_list_for_each(es, &ec->surface_list, link)
> >>- surface_accumulate_damage(es, &opaque);
> >>+ surface_accumulate_damage(es, &opaque, &previous);
> >>
> >> pixman_region32_init(&output_damage);
> >> pixman_region32_union(&output_damage, &ec->primary_plane.damage,
> >>@@ -1180,10 +1191,9 @@ weston_output_repaint(struct weston_output *output, uint32_t msecs)
> >> &ec->primary_plane.damage);
> >> pixman_region32_intersect(&output_damage,
> >> &output_damage, &output->region);
> >>- pixman_region32_subtract(&ec->primary_plane.damage,
> >>- &ec->primary_plane.damage, &output->region);
> >>
> >> pixman_region32_fini(&opaque);
> >>+ pixman_region32_fini(&previous);
> >>
> >> if (output->dirty)
> >> weston_output_update_matrix(output);
> >>--
> >>1.7.9.5
> >>
> >>_______________________________________________
> >>wayland-devel mailing list
> >>wayland-devel at lists.freedesktop.org
> >>http://lists.freedesktop.org/mailman/listinfo/wayland-devel
>
More information about the wayland-devel
mailing list