| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
| |
Reviewed-by: Trust Me
|
|
|
|
|
| |
Reviewed-by: Trust Me
(cherry picked from commit ac5c099cc3c5b8c7eec7a49fdeb8a21037230350)
|
|
|
|
|
|
| |
Not sure why this worked before.
Reviewed-by: Benjamin Poulain <benjamin.poulain@nokia.com>
|
|
|
|
|
|
|
|
|
|
| |
By default only drawhelpers and image loaders will now use neon.
If -mfpu=neon has been explicitly enabled in the mkspec,
QT_ALWAYS_HAVE_NEON will be defined, allowing the use of neon
intrinsics elsewhere.
Task-number: QTBUG-15163
Reviewed-by: Benjamin Poulain <benjamin.poulain@nokia.com>
|
| |
|
|\ |
|
| |
| |
| |
| | |
Task-number: QTBUG-9248
|
| |
| |
| |
| | |
Correct link is resize()
|
|/
|
|
|
| |
Task-number: QTBUG-13414
Reviewed-by: David Boddie
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before Qt 4.7, QString::isRightToLeft was an inline function that
called QString::updateProperties(). In Qt 4.7, QString::isRightToLeft
was de-inlined and is now called from QString::updateProperties().
According to the Binary Compatibility Guidelines, it's ok to
de-inline a function provided that it's ok the old method is called.
Under some rare circumstances nowadays, the old method could be called
from updateProperties(), which would result in an infinite loop
(updateProperties -> isRightToLeft -> updateProperties -> ...)
This is usually prevented by -fvisibility-inlines-hidden in GCC
(automatic in Qt) and also by -Wl,-Bsymbolic-functions (not automatic,
must pass -reduced-relocations to configure).
Reviewed-by: Bradley T. Hughes
|
|\ |
|
| |
| |
| |
| | |
Reviewed-By: Bradley T. Hughes
|
| |
| |
| |
| |
| |
| |
| | |
constructor.
Reviewer: David Boddie
Task: QTBUG-13244
|
|\ \
| |/
| |
| |
| |
| |
| |
| | |
doc-4.7
Conflicts:
doc/src/getting-started/examples.qdoc
doc/src/index.qdoc
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This commit makes no code change at all.
Write down the conclusions from experimenting with the x86 SIMD
instructions for faster string comparison routines. Long story short:
we're already pretty good.
The SSE4.2 string instructions are probably more useful for
"implicit-length mode" (that is, the end of the string is marked by a
NUL), rather than QString's "explicit-length mode".
|
|\ \
| |/
|/| |
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Task-number: QTBUG-9239
Reviewed-by: joao
Conflicts:
src/corelib/tools/qstring.cpp
|
| |
| |
| |
| |
| |
| |
| | |
even if the local specify QLocale::OmitGroupSeparator
Task-number: QTBUG-9281
Reviewed-by: Denis
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| | |
This is (supposedly) more efficient as the compiler can optimise it to a
builtin, per Thiago.
Merge-request: 2430
Reviewed-by: Andreas Kling <andreas.kling@nokia.com>
|
| | |
|
| |
| |
| |
| |
| |
| |
| | |
Fix to use the native CompareC function instead.
Task-number: QTBUG-11350
Reviewed-by: Miikka Heikkinen
|
| |
| |
| |
| |
| |
| |
| | |
Rules P2 and P3 demand us to ignore the LRE/LRO/RLE/RLO
characters for determining the paragraph direction.
Task-Number: Part of QT-3292
|
|\ \
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
4.7-integration
* '4.7' of scm.dev.nokia.troll.no:qt/oslo-staging-2:
Fix compilation: disable -no-feature-* for bootstrapped
QString: Fix severals bugs when comparing with QStringRef
QProgressBar: make accessors const.
Changes: add patch for artificial emboldening
Added static version of QGLFramebufferObject::release().
Fix compilation on WinXP MinGW32;
Add a new qconfig feature GESTURES
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
the internal ucstricmp and ucstrcmp contains different bugs if the
strings are not 0-terminated, as it is with QStringRef.
- in ucstricmp, even if the pointer are the same, the lenght could
be different
- we used to deference the 'end' pointer, that would be 0 if the
string ends with 0, but we cannot do that in the general case
Task-number: QTBUG-10404
Reviewed-by: Denis
|
|/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
setRawData() itself (consistently with fromRawData()) sets alloc to the
passed raw size, so the check at the start would always find that we
cannot re-use the qstring object, thus completely defeating the purpose
of the function.
so also check that the string data pointer actually points into the
internal array before giving up.
Reviewed-by: olivier
|
| |
| |
| |
| |
| |
| |
| |
| | |
if normalization was requested for QChar::Unicode_Unassigned version;
treat it like latest supported version instead
Merge-request: 2393
Reviewed-by: Olivier Goffart <olivier.goffart@nokia.com>
|
| |
| |
| |
| |
| | |
Merge-request: 2393
Reviewed-by: Olivier Goffart <olivier.goffart@nokia.com>
|
| | |
|
|/
|
|
|
| |
this is solely for optimization purposes - saves a free()/malloc()
pair each time an existing object is re-used.
|
|
|
|
|
|
|
|
|
| |
QString::vsprintf was not decoding the format string according to QTextCodec::codecForCString,
thus making qDebug("ééé") display improperly, even if using a utf8 terminal, utf8 source files
and a codec for cstrings configured as utf8. Added a unit test.
Merge-request: 530
Reviewed-by: Benjamin Poulain <benjamin.poulain@nokia.com>
|
|
|
|
|
|
| |
absorb qt_winQString2MB() code directly into the QTextCodec code
Acked-by: thiago
|
| |
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
4.7-integration
* '4.7' of scm.dev.nokia.troll.no:qt/oslo-staging-1: (57 commits)
fix corewlan detection error when building for 10.5 when 10.6 is also
fixed treatment of zlib on Mac when crossbuilding
Partial overloading support for qdbus cli tool.
Allow empty authority in QUrl::setAuthority as per docs.
Added test for QTBUG-6962: Empty authority ignored by QUrl::setAuthority.
fixed case of GL include directory
check in MAC_APPLICATION_MENU translations
Re-generate the Unicode tables after updates to the program that generates them
Fix the code after merge: DerivedNormalizationProps has two or more columns
add some usefull definitions to qunicodetables_p.h
qchar.cpp: fix identation
finish last commit
prefer DerivedNormalizationProps.txt over CompositionExclusions.txt
improve error reporting a bit more
improve error reporting
fix incorect condition
check if string to int conversions were done w/o errors
improve error reporting for unassigned grapheme/word/sentence break classes
avoid using of qunicodetables_p.h in generator
use QHash for line break map
...
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* CURRENT_VERSION macro in qunicodetables.cpp was renamed
to UNICODE_DATA_VERSION and it's definition was moved to qunicodetables_p.h
* LAST_UNICODE_CHAR macro in qchar.cpp was renamed
to UNICODE_LAST_CODEPOINT and it's definition was moved to qunicodetables_p.h
Merge-request: 480
Reviewed-by: Thiago Macieira <thiago.macieira@nokia.com>
|
| |
| |
| |
| | |
Reviewed-By: Denis Dzyubenko
|
| | |
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Despite the names, the QString::toAscii and QString::fromAscii
functions are not about ASCII. One can set an ASCII-incompatible codec
using QTextCodec::setCodecForCStrings, which may lead to loss of
information.
Also update the docs about how the functions that take QByteArray take
embedded NULs in the QByteArray into account. The result is:
stop conversion at NUL: operator=, operator==, operator!=
include NUL in conversion: operator<, operator<=, operator>, operator>=
Reviewed-by: ossi
|
|\ \
| |/
|/|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
configure.exe
examples/multimedia/audioinput/audioinput.cpp
src/corelib/io/qfsfileengine.cpp
src/gui/egl/qegl_wince.cpp
src/gui/egl/qeglproperties.cpp
src/gui/egl/qeglproperties_p.h
src/gui/embedded/directfb.pri
src/gui/kernel/qapplication_win.cpp
src/gui/painting/qdrawutil.cpp
src/opengl/qgl_p.h
src/sql/drivers/odbc/qsql_odbc.cpp
src/sql/drivers/odbc/qsql_odbc.h
tests/auto/auto.pro
tests/auto/qgl/tst_qgl.cpp
translations/assistant_adp_ru.ts
|
| |
| |
| |
| |
| |
| |
| |
| | |
And start is out of bounds.
Reviewed-by: Thiago
Reviewed-by: Joao
Task-number: QTBUG-4306
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The SSE implementation used signed integers. This was failing for
characters with high value.
The Neon implementation was using >= instead of > for creating the mask
Reviewed-by: Samuel Rødal
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Implement toLatin1 with neon to process 8 characters at a time.
Using Neon improve the speed it there is at least 16 characters,
there is no improvement when using it for 8 characters.
Speed difference on a N900:
-8 characters: no change 100%
-16 characters: 126%
-1000 characters: 361%
-10000 characters: 423%
Reviewed-by: Samuel Rødal
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Encoding to latin1 can be done 8 characters at a time with SSE2.
The speed difference on a current CPU :
-<16 characters: no change: 100%
-16 characters string: 120%
-1000 characters: 230%
-10000 characters: 322%
Reviewed-by: Samuel Rødal
|
| |
| |
| |
| |
| |
| | |
Add comments to explain the intrinsics.
Comment on the general method. Share the information that it is not
worth it to do the same on with Neon.
|
|\ \ |
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
- avoid detaching if the string is already simplified
- avoid calling isSpace() multiple times on the same character
Reviewed-by: joao
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
detach() followed by resize() is suboptimal, as it first creates an
identically sized copy, and then a differently sized one.
Reviewed-by: joao
Reviewed-by: denis
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
to complement the QString(const QChar *, int) one, which unfortunately
clamps sizes below zero to zero instead of computing the length itself
instead.
Reviewed-by: denis
Reviewed-by: joao
|
|/ /
| |
| |
| |
| |
| |
| |
| | |
When there is at least 16 characters, use SSE2 to convert
from 8bits to 16 bits.
Reviewed-by: Samuel Rødal
Reveiwed-by: Thiago Macieira
|