Instead of maintaining an index and comparing it to the count, just mark
the last startstop event with NULL and stop dequeuing events upon seeing
that sentinel value. (Removes an unreadable line, and cachegrind hints
that it may be a tiny bit faster.)
First check if we can reject the intersection without having to perform
the full divides, based on analysing:
t * (ady*bdx - bdy*adx) = bdy * (ax - bx) - bdx * (ay - by)
s * (ady*bdx - bdy*adx) = ady * (ax - bx) - adx * (ay - by)
and excluding the result if by inspection we know that
(s, t) <= 0 || (s, t) => 1.
Doing so virtually eliminates all division and speeds up the strokes (when
performing self-intersection elimination using the tessellator) perf cases
by about 5%.
Use our prior knowledge of the inputs and trivial conditions to simplify
the edge equations and in many common conditions, such as vertical edges
and common points, reduce the operations down to a just returning the
non-degenerate 32 bit value. This adds an overhead of a few conditionals,
but on the fast paths we actually generate fewer branches and many fewer
arithmetic operations such that it improves performance of the fill
performance tests by around 10%.
The matrix is quite often just a simple scale and translate (or even
identity!). For this class of matrix, we can skip the full adjoint
rearrangement and determinant calculation and just compute the inverse
directly.
libXrender amalgamates sequences of XRenderFillRectangle() into a single
XRenderFillRectangles request (when permissible). Since it is common for a
cairo application to draw rectangles individually in order to exploit fast
paths within cairo [rectilinear fills], it is a reasonably common pattern.
No need to copy font options if the similar surface is from the same
backend and no special options have been applied by the user. Doing so
breaks lazy initialisation of backend specific options.
Constructing the font options cause the initialisation of Xlc and invoke
several round-trips to the X server, significantly delaying the creation
of the first surface. By deferring that operation until the first use of
fonts then we avoid that overhead for very simple applications (like the
test suite) and should improve start-up latency for larger application.
Exploit the auxiliary offset vector in the attributes to reduce
likelihood of range overflow in the translation components when converting
the pattern matrix to fixed-point pixman_matrix_t.
An example of this is bug 9148
Bug 9148 - invalid rendering when painting large scaled-down surfaces
(https://bugs.freedesktop.org/show_bug.cgi?id=9148)
but the issue is perhaps even more likely with high resolution fallback
images.
In order to workaound a directfb bug, tweak the reflect->repeat pattern so
that it covers the destination rectangle. Although the number of paint()
increases, the number of read/written pixels remain the same so that
performance should not deteriorate, but instead be improved by using a
cloned source. The early return of the REFLECT surface is discarded so
that the latter optimisations for surface sources can be applied. One side
effect of this is that acquire_source_image() is removed due to its lax
reference counting which thereby exposes the ROI optimisations for image
destinations as well.
We frequently need to find the bounds of a pattern under an identity
matrix, or a simple scale+translation. For these cases we do not need to
transform each corner and search for the bounds as the matrix is x/y
separable and so allows us to inspect the results for the extreme x/y
points independently.
Avoid allocating a default source pattern by using the static black pattern
object. The one complication is that we need to ensure that the static
pattern does leak to the application, so we replace it with an allocated
solid pattern within _cairo_gstate_get_source().
Only copy the pattern if we need to modify it, e.g. preserve a copy in a
snapshot or a soft-mask, or to modify the matrix. Otherwise we can
continue to use the original pattern and mark it as const in order to
generate compiler warnings if we do attempt to write to it.
We can only correct rounding errors between cairo and pixman matrices for
scaled matrices - so skip the inversion and point transformation overhead
for simple translation matrices.
Adrian Johnson discovered cases where we mistakenly compared the result
of unsigned arithmetic where we need signed quantities. Look for similar
cases in the users of cairo_rectangle_int_t.
_cairo_gstate_backend_to_user_rectangle() requires that its input
arguments are non-NULL and describe the input rectangle to be transformed.
However, we were passing through output parameters from the public API
which were allowed to be NULL. So we need to allocate temporary variables
in which to compute the output rectangle, but only copy them as required.
That is, the vesion that appears as:
http://cairographics.org/manual
and:
http://cairographics.org/cairo-manual.tar.gz
It was silly that we previously required a manual step to
upload the documentation (which we regularly forgot to do)
and that it uploaded with a date in the name rather than a
version. So we just drop the old doc-publish Makefile target
now as it's just not useful anymore.
If the image was opaque with no alpha channel, we filled the output alpha
with 0. Typically, the destination surface for dithering is an RGB window,
so this bug went unnoticed. However, test/xlib-expose-event is an example
where we generate an intermediate alpha-only pixmap for use as a stencil
and this was failing as the mask was left completely transparent. The
simple solution is to ensure that for opaque images, the output alpha is
set to the maximum permissible value.
