GCC uses a peculiar name for a real 128-bit integer on x86-64.
Speedups, xlib on a gm45
========================
poppler 41246.56 -> 35102.82: 1.18x speedup
swfdec-youtube 12623.01 -> 11936.79: 1.06x speedup
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The autoconf shipping with OpenBSD 4.5 converts sizeof(void*)
to SIZEOF_VOID__ rather than SIZEOF_VOID_P. Work around that
by defining SIZEOF_VOID_P if required.
This appears to be the simplest mechanism to build libglew at the moment -
should a system copy be unavailable. Fortunately libglew is now distributed
under a permissive licence.
If you want to pass 'make -C src check' you have to use the system copy,
or spend quite a bit of time cairo-fying libglew.
There you go Joonas, I don't always ignore your suggestions! This is
simple patch to allow the user to disable symbol loops in case the
auto-detection fails on some obscure (perhaps OpenBSD) platform. Or in
case the user really wants to trim a few bytes from a library only used
during tracing!
Project lines that exceed the 16.16 limits onto the XFixedLine, as we know
that top/bottom must fit within the confines of the surface and so will be
less than 16 bits.
Sadly this is a run-on patch that also does:
1. Make FillTiled the default for new GCs.
2. Stores extend mode as opposed to repeat, and thereby cleaning up some
inconsistent code.
3. Remove the special casing for unbounded trapezoids, as it is redundant
with the polygon clipping.
4. Tidy the logic for deciding when to use the core protocol
(_categorize_composite_operation)
Implement cheap calculation of glyph extents to see whether we can discard
the clip region. This is effective around 50% of the time for firefox (and
makes the xtrace so much neater).
Fill a region using IN and a white source is a frequent reduction of a
complex clip (viz a path that covers the whole operation extents will be
reduced to a single rectangle), and we can simply discard the fill (as it
always has bounded semantics).
If the clip path is a simple box, then for many operations it is entirely
representation by the alteration of extents and so we can skip a few steps
during reconstruction.
Jeff Muizelaar reported a bug from his firefox builds that was causing a
double free during XCloseDisplay, and suggested it was related to
c0e01d9cd. Reviewing the cleanup, suggested that the cause may just be a
pair of missing parenthesis.
Some environments may be broken beyond our capabilities to detect, or
maybe the user is just insane and doesn't want to build my nice shiny
cairo-trace. Whatever, give them the option to choose:
$ ./configure --disable-trace
If the XServer claims to support the required Render extension then send
it the operations. However for the cases where we know it does not work,
i.e. the current and previous generations (<=1.6) of Xorg servers, enable
the buggy_pad_reflect fallbacks.
The win32 backend handles surface sources directly and never calls
_cairo_pattern_acquire_surface() which is the only other possible user of
clone_similar().
Check for native mutex types before hooking into pthread, as this
workarounds broken builds on mingw that confuse us by including the
pthread header file.
FillRectangle is most frequently used to fill an entire imagee with the
background colour, i.e. with just a single, or few, rectangle. Avoid
heap allocation for this common case by allocating enough space for 4
rectangles (vertices+colors) on the stack.
The image surface code doesn't reliably work on images larger than
32767 in width or height. This patch makes the image surface
constructors fail by returning a surface in the CAIRO_STATUS_INVALID_SIZE
state when given negative or too large dimensions so that client code
gets a prompt and correct error rather than flaky rendering on large
images.
Creating an widthxheight solid picture for using with
RenderCompositeTrapezoids defeats the optimization with the xserver that
checks for a solid alpha pattern. The checks it performs are for
CONTENT_ALPHA, Repeat, 1x1 and value == 0xff.
The early discard checked if the line was below the last clip-box, or if
above the first. However, the clip-boxes are only sorted on by the bottom
(not the strict XY-banded sort of the regions) and so this was erroneously
discarding lines.
The issue Joonas was trying to solve was the unwanted inclusion of
the inlines via cairo-freelist-private.h. Unwittingly he included
cairoint.h from cairo-xlib-private.h instead, a far more heinous crime as
that causes the boilerplate to try to use the hidden, private symbols.
Instead we resolve this issue by making the cairo_xlib_display_t structure
private to cairo-xlib-display.c and provide functions to manipulate the
abstract data type. Whilst in the vicinity, we rename
cairo_xlib_screen_info_t to cairo_xlib_screen_t for consistency and
cleanliness.
The xlib boilerplate includes a cairo private header to be able
to disable usage of the render extension. This indirectly includes
cairo-freelist-private.h which contains a bunch of static inline
functions which use the private _cairo_freepool_alloc_from_new_pool
function, but since that is not inline also, it causes an undefined
extern which cannot be resolved. The binutils linker doesn't care
since the freelist function aren't actually used in the boilerplate
but the Solaris linker does. By making the .._alloc_from_new_pool
function inline no dangling references are created and linking
succeeds.
Ensure that the meta surface does not extend beyond the operation by
forcing a clip to the extents of the operation.
Fixes test/device-offset and test/device-offset-positive
If we have to rasterise a pattern for use by PS/PDF (for example, to
satisfy CAIRO_EXTENT_PAD) then only generate an image large enough to
cover the operation extents. We ensure tight coverage by computing the
extents afresh - we could do this lazily in the future, but we can not
rely on the bounds as computed by the analysis surface as for native
operations they may not be tight.
