... when more than a return would be executed, to
prompt the compiler to move it out of the way of
the normal execution path.
Unexpectedly costs ~200b in text size on optimized
GCC 5.3 Linux AMD64 builds.
Change-Id: I0ebfb56af7c2262f64271a1b0ec46533e6000bc9
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Also print the failing type.
Sanity Bot complained about the spelling and Friedemann about
the grammar and missing type information in the output.
Change-Id: I7f38c56e569312e082e7b6baf9d556f5e7e40f80
Reviewed-by: Kai Koehne <kai.koehne@theqtcompany.com>
The printf-style version of QDebug expands to a lot less code than the
std::ostream-style version. Of course, you pay in type safety (but
compilers warn about it these days), you cannot stream complex Qt
types and streaming QStrings is awkward, but in many cases you
actually improve on readability.
But the main reason is that something that's not supposed to be
executed under normal operation has no business bloating executable
code size.
This is not an attempt at converting all qWarnings() to printf-style,
only the low-hanging fruit.
In this first part, replace
qWarning() << ""
with
qWarning("...").
Had to fix broken qImDebug() definition. Instead of defining it as
a nullary macro in the QT_NO_DEBUG case and as a variadic macro in
the other, define it in both cases, as is customary, as a non-function
macro so that overload selection works without requiring variadic
macro support of the compiler.
Saves e.g. ~250b in text size in QtPrintSupport on optimized GCC 5.3
AMD64 builds.
Change-Id: Ie30fe2f7942115d5dbf99fff1750ae0d477c379f
Reviewed-by: Kai Koehne <kai.koehne@theqtcompany.com>
Introduce template helper function qSelectionIndexes().
Template argument is container. Now we have the same code for
QVector and QList.
Also it's needed for a follow-up change in this file:
add method QModelIndex QItemSelection::index().
Change-Id: I7f86a9b96e5feac9873cf0df7a1cbca74f9191ec
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
QReadWriteLock is supposed to be a better alternative to QMutex when there
are only a few writers but potentially lots of reads. However, in practice
the previous implementation was much slower, unless you really do a lot of
work with the lock for read and you have lots of contention.
Indeed, the previous implementation was locking a QMutex both for lock, and
unlock (making it already at least twice as slow as QMutex).
This new implementation brings QReadWriteLock back to the same level as QMutex:
- No memory allocations in the uncontended case (almost no overhead allowing to
create many of them in classes)
- Lock-free if there is no contention
Should support up to 2^31 concurrent readers on 64 bit platforms, and 2^28
on 32 bit platforms
Change-Id: Ifa2fc999075cbb971088f4ee8e6fde78ce262da3
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>
I was surprised this compiled at all, since I thought
QT_NO_CAST_FROM_ASCII was in effect in Qt, but
apparently it isn't.
Change-Id: Id77743a2ca1b7f865960dc78d169584741f18d43
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
The printf-style version of QDebug expands to a lot less code than the
std::ostream-style version. Of course, you pay in type safety (but
compilers warn about it these days), you cannot stream complex Qt
types and streaming QStrings is awkward, but in many cases you
actually improve on readability.
But the main reason is that something that's not supposed to be
executed under normal operation has no business bloating executable
code size.
This is not an attempt at converting all qWarnings() to printf-style,
only the low-hanging fruit.
In this first part, replace
qWarning() << ""
with
qWarning("...").
Saves ~750b in text size on optimized GCC 5.3 AMD64 builds.
Change-Id: I8bf3e46cd5a6b2cae0ceb3e355a50f61925c63d3
Reviewed-by: Kai Koehne <kai.koehne@theqtcompany.com>
These types are held in QVariant, and QIBusAttribute is also
held in QVector. Now that they are no longer polymorphic,
they can be marked as movable.
Remove user-defined dtors to unlock the implicit move special
member functions, which I enforce in my local tree for all
Q_MOVABLE_TYPEs. Add std::move() when appending QIBusAttribute.
QVector has rvalue-push_back().
Change-Id: Ibb359939d5c11b5ef1f8ceced9a051cdde452dd5
Reviewed-by: Takao Fujiwara <takao.fujiwara1@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
These types don't inherit to be reused, they inherit to reuse.
Consequently, change the inheritance to private, remove the
virtual ~QIBusSerializable and rewrite the streaming operators
as member functions.
Remove the now-unused QIBusSerializable streaming operators and
meta-type registration.
Change-Id: Icf7a89174592ba62b39f73f0f016c8296cab5993
Reviewed-by: Takao Fujiwara <takao.fujiwara1@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
The fact that we override the big endian MAKE_TAG macro
(from qfontengine_p.h) with a little endian version on Windows
caused some confusion and was a bug waiting to happen. This patch
renames it instead to avoid future confusion.
Change-Id: I6224a4bfbd80eafc849ecd82e7fe5f83ee1953af
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
Despite the very similar name and hardware, the software stack is
completely different compared to the Jetson TK1 Pro we already have
a mkspec for.
Change-Id: I45353ece195035e961ff47df55d6361569aabb04
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
... with STL-style iterators or with algorithms.
Java-style iterators have overhead.
Change-Id: Ibeace7357c205a39dff3ca3fc0c835a026a15cac
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
To quote http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20160222/151168.html :
> AArch64: fix Cyclone CPU features list.
> It turns out we don't have CRC after all. Who knew?
So clang did define __ARM_FEATURE_CRC32, while the CPU didn't support
the crc32 instructions, resulting in EXC_BAD_INSTRUCTION.
Change-Id: I4b0123ac5e7fd04696c05bfe7dacce205cffac8f
Task-number: QTBUG-51168
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
ARMv8 defines the crc32 instructions as optional features. Even though
the target might be ARMv8, the compiler might have been told that the
target CPU doesn't support it, in which case __ARM_FEATURE_CRC32 is not
defined. Subsequently, the arm_acle.h header might only define the
intrinsics when __ARM_FEATURE_CRC32 is defined.
Change-Id: I85efcf9efdd2e152e3f3e72310122eebf543ca3b
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
The platform has been removed in Qt 5.7.
Task-number: QTBUG-51673
Change-Id: Id0b26e94bd8a673a35bfa5e02a5ba1c30891764a
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
Some consumers of wheel events need to know if the
scrolling direction is inverted in order to provide
consistent behavior.
For example, the scrolling direction of a horizontal
slider should not change when the user toggles the
"natural scrolling" setting on OS X. In this case
the inverted bit will change state and the slider
can compensate.
This change adds a bit to QWheelEvent and sets it
on OS X.
Task-number: QTBUG-35972
Change-Id: I221435dea45d436d570b113bd0e24ee6f6832211
Reviewed-by: Shawn Rutledge <shawn.rutledge@theqtcompany.com>
QUtf8::convertToUnicode() contains a SIMD-enabled
ASCII fast-path already which is likely faster
than what the compiler will emit for the old code
here.
Change-Id: I6afae9689424eb53a9f7c01359cc4f57ffcead26
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
iOS cannot do runtime feature detection, and querying the CPU is only
allowed in kernel mode (or beyond), so we have to decide the features
at compile time, in which case we might as well use the fallback code
path that uses the built in __ARM_* macros to point out which features
are supported, instead of hard-coding the features for iOS.
Change-Id: Ie507c0d8e962a7bdab16508c8b8122645276512e
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
Wrap the pairs of (void *ptr, void (*dtor)(void*)) in essentially
a std::unique_ptr. This simplifies code and provides the correct
implicit destruction, so we can drop the explicit glyph-cache
clear()ing in ~QFontEngine(), leaving that job to ~QLinkedList.
A subsequent change will turn the QLinkedList into a C array, the
clearing of which would otherwise cause excessive code bloat.
Since we can't use std::unique_ptr, yet, provide a hand-rolled
replacement for now, marking it for replacement with unique_ptr
once we can use it. Make that a local type instead of providing
a Qt-wide unique_ptr so we don't accidentally lock ourselves into
a half-baked std clone we can't get rid of anymore.
To prepare unique_ptr use with the same type-erased deleter
(function pointer) as now, replace a nullptr destroy_function
with a no-op function, so ~unique_ptr doesn't crash when we
port to it later.
Because QFreetypeFace contains the same construct and shares
payloads with QFontEngine, use the Holder there, too.
Even saves 150b in text size on optimized GCC 5.3 AMD64 builds.
Change-Id: I5ca11a3e6e1ff9e06199124403d96e1b280f3eb2
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
The printf-style version of QDebug expands to a lot less code than the
std::ostream-style version. Of course, you pay in type safety (but
compilers warn about it these days), you cannot stream complex Qt
types and streaming QStrings is awkward, but in many cases you
actually improve on readability.
But the main reason is that something that's not supposed to be
executed under normal operation has no business bloating executable
code size.
This is not an attempt at converting all qWarnings() to printf-style,
only the low-hanging fruit.
In this first part, replace
qWarning() << ""
with
qWarning("...").
Saves ~850b in text size on optimized GCC 5.3 AMD64 builds.
Change-Id: Ib1a087795a03b2a6b432e2c499968df779aaea37
Reviewed-by: Kai Koehne <kai.koehne@theqtcompany.com>
The printf-style version of QDebug expands to a lot less code than the
std::ostream-style version. Of course, you pay in type safety (but
compilers warn about it these days), you cannot stream complex Qt
types and streaming QStrings is awkward, but in many cases you
actually improve on readability.
But the main reason is that something that's not supposed to be
executed under normal operation has no business bloating executable
code size.
This is not an attempt at converting all qWarnings() to printf-style,
only the low-hanging fruit.
In this first part, replace
qWarning() << "...";
with
qWarning("...");
In QTransform shared warning strings.
Saves 3KiB in text size on optimized GCC 5.3 AMD64 builds.
Change-Id: I142a8020eaab043d78465178192f2c8c6d1cc4f9
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Reviewed-by: Kai Koehne <kai.koehne@theqtcompany.com>
Reuses the optimized routines from qimage to make painting RGB888
images faster on SSSE3 and NEON.
Change-Id: I99116b318322ba4cb0ddc2cb90bcf17a0350ef99
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
Just iterate over the container instead, saving one
iteration and the creation of a temporary QList.
Change-Id: I564e3e83cb247a12c413fc5a9dc17299ae089e30
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
Don't perform lookup twice. Just cache iterator
or position.
Change-Id: I454fd292614dee62167ff248fc3ddec0f79435b0
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Three hashes are handled similarly; so extract the loops over them as
methods hasUsedConfiguration() and cleanUpConfigurations() to avoid
duplicate loop code.
Change-Id: I1040724c4fc98caa48913fac339c03e60b04bae2
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
These are already held in QVectors.
Public API types need to wait until Qt 6, for BC reasons. Even
though Q_RELOCATABLE_TYPE deals with most of them, we lack a way
to mark a type as primitive, but still isStatic - for QList.
Change-Id: I91392b01ae6f94cc847007636e12d4e64c43b2bc
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@theqtcompany.com>
Change-Id: I5f8d72742cc4199bfa73df6037b851c58632ff86
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
For the remote X case the backing store previously always reuploaded
image data for every expose event. Instead of doing that create a remote
X pixmap and only flush repainted regions. For regular expose just copy
from the pixmap.
Additionally, atomically update the window by setting a clip mask and
flushing the entire region at once instead of doing it rect by rect.
Change-Id: I26bb1834b159e309c7ad93287dd297769f7e2633
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
It has a variant accepting QL1S directly, so
no need to go via a QString.
Change-Id: Ia8f1198ef2af7027bc9f7c2e1dad3a5f78a12eb4
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Many DirectFB types have constructors in C++, so we can't initialize
them with = {...}, like we would be able to if they had been regular POD
types.
Change-Id: Ic747cc2ab45e4dc6bb70ffff143840e5780ac2bc
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
Reviewed-by: Andy Nichols <andy.nichols@theqtcompany.com>
Marking them shared (which implies movable) now
would make QLists of these BiC.
Change-Id: If5638e8d9f43e0ad549aedf08934de31e1e189f1
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@theqtcompany.com>
Extract a loop repeated four times in the code into a
separate function. Port to use C++11 range-for loops
and not create temporary QLists.
Saves 2.4KiB in text size on optimized GCC 5.3 Linux
AMD64 builds.
Change-Id: I7eb78ffaac33627b595e35cafb6ce0769fb760a8
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
Use new QUtf8::convertToUnicode(QChar*, const char*, int)
overload instead of QString::fromUtf8(). The QUtf8 overload
allocates no memory, and is therefore marked as nothrow.
Using this function in convertFromAscii() allows to mark
this function nothrow, too.
All functions of QAbstractConcatenable can now be marked as
nothrow.
Since QUtf8::convertToUnicode() does not deal with lengths
of -1, insert a strlen() call when the len == -1 ASCII fast
path fails due to non-ASCII chars.
Saves 1.1KiB in text size on optimized GCC 5.3 Linux AMD64
builds of QtCore (other libraries are compiled without
exceptions, so this change should not have an effect on
those).
Change-Id: I7333e35844033831eae2a04203d13d9792c5d460
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Change-Id: I6864e227fceb133903979ac8f7a7434fc3e280bf
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>
The function resolving changes remove the special treatment for ES 3.0+
contexts, meaning that now all functions get resolved in the same way
irrespective of the current context.
For blitFramebuffer and renderbufferStorageMultisample this presented an
issue with ANGLE. There these functions are available both as an ANGLE
extension and as standard ES 3.0 functions. The latter are not functional
however in 2.0 contexts. We expect multisampled FBOs to work in 2.0
contexts too by prefering the ANGLE extension with 2.0 contexts.
Change-Id: I0a4b70e6d39c84d4b1f61f8fd0655d7326419a2a
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
Always try both e/wglGetProcAddress and ::GetProcAddress to
resolve the methods. Like this QOpengGLContext::getProcAddress is
able to return any OpenGL entry point, and we can both simplify
the code we have in the QPA backend as well as get rid of windows
specific code paths in Qt Gui.
Task-number: QTBUG-39531
Change-Id: I1ddf1b0974f69b56b219a619655b723eb0134b14
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
