| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
| |
This cuts down quite some intructions in some use cases,
making esp. discardUpdateRequest() a bit cheaper.
|
| |
|
|
|
|
|
| |
Using qFuzzyCompare for checking whether two transformations
are equal doesn't give us too much and is inconsistent with
our other matrix classes. Using simple floating point
equality is a lot faster as well.
|
| |
|
|
|
|
| |
This cut's off some cycles in discardUpdateRequest()
which is called from most places when something in
the items changes.
|
| |
|
|
| |
This caused a crash in the contacts demo.
|
| |
|
|
|
|
|
|
|
|
| |
This function works much faster than the last one, but still it can be
much faster. The main expense right now it seems is the transform
calculations, and the item->collidesWithPath call. There's still much
to gain here.
This function does not make use of the BSP tree's fast lookup so it makes
lookups slower in (e.g.) the chip demo.
|
| |
|
|
|
| |
Also change the order of comparisons, as the transformable flags are
most likely to fail first.
|
| |
|
|
|
|
|
|
| |
The QStyleOption constructor is expensive, as it allocates a QFont,
a QPalette and a QFontMetrics. By simply reusing a temporary style
option object instead we carve away wasted cycles.
Reviewed-by: Ariya
|
| |
|
|
|
| |
The recursive approach is faster and fits better into the new
scene transform cache we'll do later.
|
| |
|
|
|
|
|
|
|
|
| |
Mark the children list for sorting when the Z value for items changes.
Change the signature of the recursive draw function slightly so that
the expose region is optional, and ensure we don't intersect with this
region if it's not available.
This change also flips the direct painting so that the default is to
use the recursive approach.
|
| |
|
|
|
|
|
|
|
| |
This change introduces two helper functions in QGraphicsItemPrivate,
that combine the item's transform into the current transform, effectively
merging the code that calculates any item's combined to-parent transform.
This makes the code more readable, and also makes it easier for us to
reintroduce the componentized transform API in QGraphicsItem (which
was previously reverted).
|
| |
|
|
|
|
| |
This change also changes the direct painting path to be opt-in as a
temporary testing measure to see what's broken when using the old code
path.
|
| |
|
|
|
|
|
|
| |
Convert some repaint() calls back to updates(). This ensures that any
updates triggered before this call are processed normally, while still
keeping the synchronous repaint() call intact.
Reviewed-by: bnilsen
|
| | |
|
| |
|
|
|
| |
This seems to be the only way to get a high frame rate, regardless
of the performance of painting and all.
|
| |
|
|
|
| |
This change is only to test how much of an impact the itemChange
mechanism has on performance. It's easy to revert later on.
|
| |
|
|
|
|
| |
just calculate it top-down.
Reviewed-by: bnilsen
|
| |
|
|
|
|
|
| |
To avoid sorting siblings in real-time, ensure we lazily sort the list
in place when needed.
Reviewed-by: bnilsen
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
This time with a recursive approach of processing dirty items. I've kept
the previous approach using a dirty list, but the recursive one is now
the default. Use QT_GV_USE_DIRTY_LIST=1 to swap.
I've also cached the item's device transform in both cases so that we
can re-use it later when drawing the item.
|
| |
|
|
|
|
|
|
|
|
|
| |
This flag helps optimize the case where an item is used only as a
transformation node in a scene graph, and where the item itself
doesn't paint anything. This is the default for FxItem (the
subclasses that do paint enable the HasContents flag). This lets
Graphics View know whether there's any point in setting up the
world transform, opacity and other things.
Reviewed-by: Lars
|
| |
|
|
| |
Regression introduced during refactoring earlier today.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Greatly simplify how we handle opacity and store it as a member in
QGraphicsItemPrivate. Remove the caching of effectiveOpacity. It's
faster to calculate it on the fly, and the recursive painting algorithm
will make even that need go away.
Reviewed-by: Andreas
|
| |
|
|
|
|
|
|
| |
Since the transform is now a pointer in QGraphicsItemPrivate, we can
use this pointer directly and there's no need for a separate bitflag
anymore.
Reviewed-by: Andreas
|
| |
|
|
|
|
|
|
|
| |
Now that we have a recursive painting algorithm these types of
optimizations are no longer necessary. In fact they only cause more
problems and clutter up the code unnecessarily. Removing this also
removes extra overhead from moving and transforming items.
Reviewed-by: Lars
|
| |
|
|
|
|
|
|
|
|
|
| |
This flag helps optimize the case where an item is used only as a
transformation node in a scene graph, and where the item itself
doesn't paint anything. This is the default for FxItem (the
subclasses that do paint enable the HasContents flag). This lets
Graphics View know whether there's any point in setting up the
world transform, opacity and other things.
Reviewed-by: Lars
|
| |
|
|
| |
Regression introduced during refactoring earlier today.
|
| |
|
|
|
|
|
|
|
| |
The code so far was converting the rect to a painterpath, mapping
that one and then taking the bounding rect. It is actually sufficient
to simply map the four corners of the rectangle and take the bounding
rect of these four points even in the projective case.
Reviewed-by: Andreas
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Greatly simplify how we handle opacity and store it as a member in
QGraphicsItemPrivate. Remove the caching of effectiveOpacity. It's
faster to calculate it on the fly, and the recursive painting algorithm
will make even that need go away.
Reviewed-by: Andreas
|
| |
|
|
|
|
|
|
| |
Since the transform is now a pointer in QGraphicsItemPrivate, we can
use this pointer directly and there's no need for a separate bitflag
anymore.
Reviewed-by: Andreas
|
| |
|
|
|
|
|
|
|
| |
Now that we have a recursive painting algorithm these types of
optimizations are no longer necessary. In fact they only cause more
problems and clutter up the code unnecessarily. Removing this also
removes extra overhead from moving and transforming items.
Reviewed-by: Lars
|
| |
|
|
|
|
|
|
|
|
|
| |
Removed some experimental code to handle scaling and rotating around
different axis. It cuased setTransform and transform not to behave
symmetrically and caused some performance regressions.
Additionally moved the QTransform out of the (relatively slow) extra
list and made it a pointer in QGraphicsItemPrivate.
Reviewed-by: Andreas
|
| |
|
|
|
|
|
|
|
|
|
| |
This is work-in-progress, so don't expect everything to work perfectly.
Most of the auto-test pass and examples and demos seem to run fine.
Unfortunately I'm too tired to write about the actual update mehanism
now, but it's faster than the old approach (if that helps:)). There's
more to optimize, but I'll come back to that later. I need some sleep now :)
To be continued.
|
| |
|
|
| |
We'll need this later when making a smarter update mechanism.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
the important thing is by effectively implementing Simple Canvas' approach
to drawing, we're in theory (and in practise measured on the desktop)
as fast as Simple Canvas when rendering.
|
| |
|
|
|
|
|
| |
This compiler doesn't seem to follow the same rule as others
for implicit conversions to/from wchar_t.
Add the necessary casts, keeping in mind that sizeof(wchar_t) == 2
on Windows.
|
| |
|
|
|
|
|
|
|
|
|
| |
Thunderbird sets only the user level shell association for mailto. This
is now being read before the default mail application registry.
The registry crawling could have been avoided by using the ShellExecute()
but it supports only around 2KBytes of data as parameter, so we will
continue using CreateProcess().
Task-number: 251554
Reviewed-by: Jens Bache-Wiig
|
| |
|
|
|
|
|
|
|
|
|
| |
When using Qt/E with VNC screen driver on top of the Linux framebuffer
(export QWS_DISPLAY="VNC:LinuxFb") the region to be exposed needs be clipped
to the real screen to avoid possible negative coordinates which in turn
cause access to invalid memory locations when comparing the content of the
VNC buffer with the Linux framebuffer.
Merge-request: 603
Reviewed-by: Tom Cooksey <thomas.cooksey@nokia.com>
|
| | |
|
| |
|
|
|
|
|
|
|
| |
still paint them.
The autotest is included.
Task-number: 254449
Reviewed-by: ogoffart
|
| | |
|
| |\ |
|
