cairo-xcb was deciding which type to cast a surface to based on its "type"
member. This is wrong, it should use "backend->type".
This bug was hit via xlib-xcb. This was painting a subsurface of a xlib-xcb
surface to an xcb surface. Because surface->type said "xlib", the code was
trying to check if the xcb surface had a fallback. However, this was done on the
subsurface. The end result was dereferencing a pointer to 0x28.
This was noticed while looking into
https://bugs.freedesktop.org/show_bug.cgi?id=42889
No test for this bug since I didn't manage to come up with one.
Signed-off-by: Uli Schlachter <psychon@znc.in>
I lost the '&& 0' I put in to disable the glyph rendering until I had
the glyph cache integration working again.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
And only upload the parts of the image that are modified during the
fallback. I have to keep reminding myself that the goal is always to
reduce the amount of fallbacks required...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This doesn't just need a clip without any path, it also needs pixel aligned
boxes.
This improves the result for unaligned boxes in tighten-bounds.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This function changed its behavior and no longer does what we want. Instead,
this now uses its own function which uses _cairo_clip_combine_with_surface().
This fixes crashes in the tighten-bounds and random-clip tests. These happened
because cairo-xcb was trying to be clever. ;-)
Since _cairo_clip_get_surface() did less, the resulting surface had
deferred_clear == true and picture == XCB_NONE. The code then tried using this
evil picture and either ran into an assert() or caused a BadPicture error.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This makes sure that errors from _cairo_clip_get_surface() aren't lost and that
we really got an xcb surface.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Having spent the last dev cycle looking at how we could specialize the
compositors for various backends, we once again look for the
commonalities in order to reduce the duplication. In part this is
motivated by the idea that spans is a good interface for both the
existent GL backend and pixman, and so they deserve a dedicated
compositor. xcb/xlib target an identical rendering system and so they
should be using the same compositor, and it should be possible to run
that same compositor locally against pixman to generate reference tests.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
P.S. This brings massive upheaval (read breakage) I've tried delaying in
order to fix as many things as possible but now this one patch does far,
far, far too much. Apologies in advance for breaking your favourite
backend, but trust me in that the end result will be much better. :)
This code was casting a pixman_box32_t* to cairo_box_t*. However, a box uses
fixed point numbers while the pixman box uses integers which means the result
was off by factor 256.
The fix is to replace the use of _cairo_boxes_limit() with
_cairo_boxes_init_with_clip(). However, this means this function no needs to be
passed a clip instead of a clip region which causes some minor changes to
_composite_boxes().
This improves the result for tighten-bounds again. Out of the tested
combinations, 10 are fixed by this. This bug was hit by code similar to this
(Repeating here since that test has so many different cases):
cairo_set_operator (cr, CAIRO_OPERATOR_IN);
cairo_set_fill_rule (cr, CAIRO_FILL_RULE_EVEN_ODD);
cairo_rectangle (cr, 0, 0, SIZE, SIZE);
cairo_rectangle (cr, 0, 0, SIZE, SIZE);
cairo_rectangle (cr, SIZE / 4, SIZE / 4, SIZE / 2, SIZE / 2);
cairo_clip_preserve (cr);
cairo_fill (cr);
Signed-off-by: Uli Schlachter <psychon@znc.in>
This only wraps the whole code block in an if which checks if the bounded
extents are empty. No other changes are done in here.
This fixes the failed assertion from the tighten-bounds test.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This code casts the result of _cairo_xcb_surface_create_similar() to
cairo_xcb_surface_t*. However, the tighten-bounds test makes this run with
bounded extents of size 0x0 and thus _cairo_xcb_surface_create_similar() falls
back to an image surface. Ugly.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The existing API only described the method to be used for performing
rasterisation and unlike other API provided no opportunity for the user
to give a hint as to how to trade off performance against speed. So in
order to no be overly prescriptive, we extend the NONE/GRAY/SUBPIXEL
methods with FAST/GOOD/BEST hints and leave the backend to decide how
best to achieve those goals.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This patch has been generated by the following Coccinelle semantic patch:
// Remove useless checks for NULL before freeing
//
// free (NULL) is a no-op, so there is no need to avoid it
@@
expression E;
@@
+ free (E);
+ E = NULL;
- if (unlikely (E != NULL)) {
- free(E);
(
- E = NULL;
|
- E = 0;
)
...
- }
@@
expression E;
@@
+ free (E);
- if (unlikely (E != NULL)) {
- free (E);
- }
Recording surfaces can be unbounded which causes
_cairo_surface_acquire_source_image to return a 0x0 image surface for them.
Since X11 doesn't like anything with a size of 0x0, we should reject such source
images. Users might still try to mess with 0x0 surfaces, so we will eventually
need a better idea for handling this.
Instead of failing the assertion that was added in the previous commit, this
commit makes cairo-xcb return an error.
This makes the recording-* tests fail instead of crash.
Signed-off-by: Uli Schlachter <psychon@znc.in>
For an unbounded surface we cannot assume (0, 0, surface_width,
surface_height) as that is wrong and causes the operation to appear
clipped.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
A common requirement is the fast upload of pixel data. In order to
allocate the most appropriate image buffer, we need knowledge of the
destination. The most obvious example is that we could use a
shared-memory region for the image to avoid the transfer cost of
uploading the pixels to the X server. Similarly, gl, win32, quartz...
The other side of the equation is that for manual modification of a
remote surface, it would be more efficient if we can create a similar
image to reduce the transfer costs. This strategy is already followed
for the destination fallbacks and this merely exposes the same
capability for the application fallbacks.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Being called with no clip, might be unexpected, but it means to fill the
whole extents with the opacity. So do so.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This is basically the same fix as e6c3efdd65. However, this was lost in
b132fae5e8 and thus had to be fixed again.
Fixes: clip-fill-eo-unbounded clip-fill-nz-unbounded
Signed-off-by: Uli Schlachter <psychon@znc.in>
As we do not control the geometry used for the individual glyphs, we
must always send a clip-region so that X can trim the glyph
appropriately. However, in order to avoid sending unnecessary data we
only do so if the clip extents is less than the ink extents of the
glyphs.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Every xcb surface had its own copy of the flags from the time that it was
created. This means that, if you want to make use of
cairo_xcb_device_debug_cap_xrender_version() and
cairo_xcb_device_debug_cap_xshm_version(), you first had to create a dummy xcb
surface, use that to get access to the cairo_device_t so that you can use these
functions and only then create your real surface, because the change only
affected new surfaces.
This commit changes everything to use the connection's flag and removes the
per-surface flags. This avoids the dummy surfaces completely.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Another 10% off fishbowl for both snb and pnv.
[Note this exposes the bugs in the polygon clipper; naive *and* broken.]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
A demonstration of step 2, improves performance for selected benchmarks
on selected GPUs by up to 30%.
firefox-fishbowl on snb {i5-2520m): 42s -> 29s.
firefox-talos-gfx on snb: 7.6 -> 5.2s.
firefox-fishbowl on pnv (n450): 380 -> 360s.
Whist this looks like it is getting close to as good as we can achieve,
we are constrained by both our API and Xrender and fishbowl is about 50%
slower than peak performance (on snb).
And it fixes the older performance regression in firefox-planet-gnome.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Step 1, fix the failings sighted recently by tracking clip-boxes as an
explicit property of the clipping and of composition.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Allow a backend to completely reimplement the Cairo API as it wants. The
goal is to pass operations to the native backends such as Quartz,
Direct2D, Qt, Skia, OpenVG with no overhead. And to permit complete
logging contexts, and whatever else the imagination holds. Perhaps to
experiment with double-paths?
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The experiment was at best a pyrrhic victory. Whilst it did show that
you could successfully subvert cairo_xcb_surface_t and provide the
rendering locally faster (than the xlib driver at that time), any
performance benefits were lost in the synchronisation overheads and
server-side buffer allocation.
Once cairo-gl is mature, we need to look at how we can overcome these to
improve client-side rendering
In the meantime, cairo-xcb is no longer my playground for
experimentation and is shaping up to become a stable backend...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Eventually someone might try to paint to the xcb surface again. However,
_cairo_surface_begin_modification doesn't like that:
cairo-surface.c:385: _cairo_surface_begin_modification: Assertion
`surface->snapshot_of == ((void *)0)' failed.
There was only a single place in the xcb backend where a cairo_xcb_surface_t
could be used as a snapshot, so the _cairo_surface_has_snapshot that checked for
such a surface can be removed, too.
This does *not* remove all snapshots from the xcb backend, but all the remaining
snapshots are instances of cairo_xcb_picture_t. These surfaces are only ever
created internally and thus can't be modified by users directly.
Fixes: xcb-snapshot-assert
Signed-off-by: Uli Schlachter <psychon@znc.in>
When e.g. using an image surface as the source for a xcb surface, a
cairo_xcb_picture_t is created and attached to that image surface as a snapshot.
This contains the Picture that was created on the X11 server.
However, as soon as the cairo_xcb_picture_t's cairo_xcb_screen_t is finished and
destroyed, this picture can't be used anymore. This commit now makes sure all
these Pictures are freed when the screen is finished.
This was found because my X server's memory usage grew quite large. Every time
the app was done drawing, it destroyed its last surface which also destroyed the
last reference to the cairo_xcb_screen_t. This meant that the existing Picture
snapshots couldn't be used anymore, but they were still kept around and used up
memory until there wasn't any free memory left.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The next commit will make cairo-xcb-screen.c use this struct and add new
members. Splitting off the move into its own commits makes that easier to
understand.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Since commit f1d313e0, the 'force' argument to _copy_to_picture() isn't used
anymore. Said commit should have removed it. Whoops.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The PDF blend operators, as offered by cairo, where added in RENDER 0.11. This
commit makes the XCB backend use them, if they are available.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Introduce cairo_xlib_device_debug_set_precision() to override the
automatic selection of rendering precision and force the Xorg/DDX to
strictly adhere to the precise rendering mode of the Render
specification. This allows us to test drivers without worrying, too
much, about minor discrepancies in antialiasing.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The RENDER specification requires radial gradients to have the first
circle completely inside the second one, but the error is not actually
generated.
The implementation produces the expected results if either circle
contains the other one, so only fall back in these cases.
The RENDER specification does not specify the constraints on the
gradient stops, but its implementation returns an error if less than 2
stops are used.
Xlib and XCB can work around this because gradients with just one stop
are by-definition the same as gradients with that stop repeated twice.
Fixes radial-gradient-one-stop.
