Re-work how the shaders and emitted vertices work. Rather than always
rendering clip-rect sized quads and doing transformation in tex coords
(and requiring the corresponding clipping in frag shader), instead
emit transformed vertices, clipped wrt. dirty region, and use simpler
frag shaders. Also, split the rendering, so blended surfaces with an
opaque region have the opaque region drawn with blend disabled. The
result is considerably fewer pixels drawn with blend enabled, and much
fewer cycles in the frag shader.
This requires having some more complex logic to figure out the vertices
of the shape which forms the intersection of the clip rect and the
transformed surface. Which has perhaps got a few bugs or missing cases,
still (visual glitches in some cases) but at this point more or less is
starting to work. I think it is at least far enough along to get some
initial review.
The result, on small SoC GPU (omap4/pandaboard) on 1920x1080 display,
for simple stuff like moving windows around, I get 60fps (before 30fps
or less), and pushing YUV buffers for hw decoded 1080p video goes from
~6fps to 30fps, with no drop in framerate for transformed/rotated video
surface.
v1: original
v2: check that perpendicular intersect vertex falls within bounds of
transformed surface
v3: update w/ comments and fixes from Pekka Paalanen
v4: fix for full surface alpha from Pekka Paalanen, fix compositor-
wayland build
Signed-off-by: Rob Clark <rob@ti.com>
In cases where the GPU can natively handle certain YUV formats,
eglQueryWaylandBufferWL() can return the value EGL_TEXTURE_EXTERNAL_WL
and the compositor will treat the buffer as a single egl-image-external.
See:
http://www.khronos.org/registry/gles/extensions/OES/OES_EGL_image_external.txt
v1: original
v2: rename EGL_TEXTURE_EXTERNAL_OES -> EGL_TEXTURE_EXTERNAL_WL and query
for the extension
v3: fix build without updated mesa headers, if EGL_TEXTURE_EXTERNAL_WL
#define is missing from older mesa headers.
v4: resend without missing parts
Signed-off-by: Rob Clark <rob@ti.com>
The correspondence between cursor functions and names of cursors has
never been standardized. As a consequence, each cursor function can be
represented as a cursor with one of several names. Be more robust when
loading cursor by trying all known names that correspond to a cursor.
This should fix https://bugs.freedesktop.org/show_bug.cgi?id=50487
and https://bugs.freedesktop.org/show_bug.cgi?id=52609 a bit more
thoroughly.
This is a workaround for screenshots with transformed outputs. It reorders
the output positions so the correct buffer size is determined for the final
image. This assumes the outputs are in succession on the x axis. The outputs
are rendered in their transformed state.
This patch allows rotation and mirroring outputs for x11 and drm backends.
A new 'transform' key can be set in the [output] section. From the protocol:
"The flipped values correspond to an initial flip around a vertical axis
followed by rotation."
The transform key can be one of the following 8 strings:
normal
90
180
270
flipped
flipped-90
flipped-180
flipped-270
E.g. this can happen when you grab the lower right corner of a window
and move over the top of the window when resizing. In this case, the
changed width is still important and should be acted upon.
https://bugs.freedesktop.org/show_bug.cgi?id=53560
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 preserved by folding it back into the
primary plane damage just before repainting.
We don't want to send enter/leave events to the compositor when the pointer
enters the surface, only when the pointer enters the content area. This
avoids hiding the cursor when entering the frame and sending out-of-bounds
coordinates to notify_pointer_focus().
This way we map the surface if it currently isn't mapped and avoid
duplicating some of the code already in pointer_cursor_surface_configure().
Without this, the cursor code relied on a wl_surface.attach() to show the
new pointer surface. If we're not changing the cursor buffer, we don't
get that, but we still need to map the cursor.
Replace all occurrences of workspace_damage_all_surfaces() with
weston_compositor_schedule_repaint(). This fixes a bug exposed by
982387011f where trying to switch workspaces while there are no
surfaces on the current workspace fails to start the animation.
This patch, along with the wayland patch, adds the ability to specify
a cursor theme in the weston.ini file:
[cursors]
theme=THEME_NAME
If specified, than Weston can use a specific X cursor theme for the
pointer. This relies on the 0001-Add-support-for-X-cursor-themes.patch
for wayland.
[krh: edited to use shell section and key name cursor-theme]
Place the window in a random position on the output where the first seat with
a pointer is. When calculating the random position limit the range to the
area that would ensure that the whole surface is visible. If the surface is
larger than the output then the surface is placed at the origin of the
output.
This change is based on the good work of Scott Moreau <oreaus@gmail.com>
If a client sends a pong message out of the blue, we deref ping_timer
which is NULL and then crash. Unsolicited pong requests indicate that
something is wrong on the client, but the compositor should survide that.
Add a wl_seat argument to the activate and deactivate requests of
text_method.
On activation a text_model gets assigned to the input_method of the
wl_seat specified in the activate request.
Rob Clark
Pekka Paalanen
Philipp Brüschweiler
Jan Arne Petersen
Scott Moreau
Ander Conselvan de Oliveira
Kristian Høgsberg
Christopher Michael
Rob Bradford