We test that inline conversion between same image depths always succeed
inline, but that requires that any direct conversions exists both in
non-inline and inline versions.
This patch adds a non-sse2 inline conversion from ARGB32 to ARGB32PM
which was missing.
Change-Id: I71937cd4b77fb41fe2064da937f6dcbf2a6534e6
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
DestFormat is not defined here but should be Format_RGBX8888
Change-Id: Ie33989a6fc31650434281a6b7b0f4806524268af
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Timo Jyrinki <timo.jyrinki@canonical.com>
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
Completes the inplace converters so that we can rely on inplace
conversions to succede as long as the image depth is the same.
Change-Id: Ia1ae34b5de1bc16e87ff5403bdacfcae44a22791
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
Some inplace conversions would not set the requested image format in
the returned image, due to the same conversion being used for several
destination formats.
This patch ensures all inplace conversions return the right format,
and adds testing infrastructure to easily test iterate over all formats
so that they all can be tested.
Change-Id: I68ab17a713ddb2aa8ed6aaf0a0a1ec8bbcb56090
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
The one converters from RGB30 was misplaced in the method table, and
the unpremultiplication from A2RGB30 to RGB30 had an underflow mistake
when alpha was 2.
Change-Id: I92c11ede28611a3dbdce72aca1898845c120c209
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
Could causedSSE4 instructions to be used on non SSE4 machines
in cases when qUnpremultiplywas not inlined.
This reverts commit 964ccc5853.
Change-Id: Ic676ade8f75129e8d37c4d96cbfb2bdb5b794919
Task-number: QTBUG-45741
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Merges the SSE4 specific unpremultiply with the normal version, and
adds a SSE2 fallback. There was no reason to split the two since
compile time options will ensure the right version is inlined.
Also adds short-cut for 0 and 255 values.
Change-Id: Ie5aa262f6964219fd3062d4a498f697cf79a4595
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Following up on using GCC's autovectorizing for faster SSE4.1
premultiply, this patch adds specialized autovectorized versions
of premultiply for AVX2, giving another almost doubling in speed.
To make the speed up for AVX2 and also SSE4_1 available to non-GCC
compilers, the target-specific methods have been moved to separate
files.
Change-Id: I97ce05be67f4adeeb9a096eef80fd5fb662099f3
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
This continues 68762151db
When doing format conversion, the optimized inplace codepath did not
check if the image data was external, i.e. if the QImage had been
created by the constructor taking an existing external buffer.
The previous commit fixed the readonly case. But in the case of the
QImage constructor taking non-const uchar*, data will be read-write,
but still external. This would of course crash if the converter tries
to realloc it.
Task-number: QTBUG-44610
Change-Id: I94d275d464e8af221682b538fc3e4897a59c061e
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
Improves the conversion from RGB888 to RGB32 on platforms without SIMD
versions. This includes the fallback used on non-neon ARM devices.
Besides image conversion the routine is also used for decoding JPEG.
On x86 this version is within 0.7x of the speed of the SSSE3 version.
Change-Id: Id131994d7c3c4f879d89e80f9d6c435bb5535ed7
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
There are many direct QImage conversions that doesn't need to be direct
but only are because they are faster than the generic conversion. This
patch optimizes the generic conversions and then removes all the direct
conversions that are now no faster than the generic.
Change-Id: I3dc5f44cc7f6358fd66420e9974eebaf2c7ca59c
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
Qt copyrights are now in The Qt Company, so we could update the source
code headers accordingly. In the same go we should also fix the links to
point to qt.io.
Outdated header.LGPL removed (use header.LGPL21 instead)
Old header.LGPL3 renamed to header.LGPL3-COMM to match actual licensing
combination. New header.LGPL-COMM taken in the use file which were
using old header.LGPL3 (src/plugins/platforms/android/extract.cpp)
Added new header.LGPL3 containing Commercial + LGPLv3 + GPLv2 license
combination
Change-Id: I6f49b819a8a20cc4f88b794a8f6726d975e8ffbe
Reviewed-by: Matti Paaso <matti.paaso@theqtcompany.com>
The autovectorized versions of premultiplying conversions are almost
twice as fast with SSE4.1 as with SSE2. Therefore this patch lets
compilers that can make those versions convenient without duplicating
code do that and lets us use them when available.
Change-Id: I699035963abe55a38b9ef8ba7b4a8c961c8dfcdd
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
The methods for inplace conversion from RGBA to ARGB was misplaced by
one step causing conversion to RGB16 to get an invalid method.
Change-Id: I3b2b4cffe993705c48613eec7d9b7c6213f57fc2
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
Coverity warns that we may access the gray array without initializing
it. The array is initialized if d (the source image depth) is 8.
However, when doing ordered dithering, we don't check that depth
any more and just use the array. There are instead checks in place
for the other dither modes; the one for ordered has been commented
out before public history.
So, for the love of kittens, put the check back.
Change-Id: I1cc5ced8edbb626777e54e9f4e58f152c6b28ddc
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Extend the QImage format with two 8-bit grayscale and alpha formats.
These formats have the advantage over Indexed8 that they have simpler
conversion and can be rendered to by the raster engine.
[ChangeLog][QtGui][QImage] Added support grayscale and alpha 8-bit
formats which can also be rendered to.
Change-Id: I4343c80a92a3dda196aa38d0c3ea251b094fc274
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
The convert_generic_inline method was not correctly handling the case
where both the conversion methods were passthrough and the last store
is therefore not needed and may trigger an invalid memcpy call.
Change-Id: Ic88780f50e1ff9dedc04b8ff1ab3527dd0c8150c
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
Adds basic support for 10-bit per color channel formats to QImage
and the XCB plugin. This will make it possible to paint to and from
these formats, but only at 8-bit per color channel accuracy.
This also fixes Qt5 applications on X11 with native 30bit depth.
[ChangeLog][QtGui][QImage] Added support for 10-bit per color channel image formats.
Task-number: QTBUG-25998
Change-Id: I93ccd3c74bfbb0bd94b352476e5fe58a94119e1f
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
Add detection of MIPS DSPr2 at run-time in qsimd.cpp. This makes it
possible to have generic Qt builds for MIPS that can enable the fast
code paths for processors with the DSP ASE at run-time. Also, this
makes it possible to manually disable them by setting the environment
variable "QT_NO_CPU_FEATURE=dspr2". Last, but not least, functions
requiring DSPr2 are not enabled when running in CPUs with version-1
DSP.
Change-Id: Ia5a01d84119553c22ab83386c74a6cb8ba5fee53
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Add a hand-coded MIPS assembler function to convert a stream of rgb888
bytes to an array of 32-bit ARGB values. This is used in QImage for
image conversion, and also in QJpegHandler for decoding JPEG image
scanlines.
Change-Id: Ia74a5ff799b29fad3d4acbfcafe790cd60586d72
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This patch optimizes the unpremultiply method further by using a
lookup table to avoid any divisions at all.
The opportunity is taken to export both premultiply and unpremultiply
since they are commonly used methods relevant to the exported QRgb type
that can be both premultiplied and unpremultipled ARGB.
[ChangeLog][QtGui][QColor] Exported highly optimized methods for
premultiply and unpremultiply of QRgb values.
Change-Id: I658bcf57b0bc73c34c1765b64617d43b63ae820b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
The generic converter can easily be modified to work inplace. Such
an inplace converter can ensure we can do direct conversion between
formats of similar pixel depth.
Inplace conversion is currently only used by QPixmaps, but will be used
by ref qualified versions of QImage::convertToFormat in a future patch.
Change-Id: Idbdaf087b53deecdee8258ba0b5d80edd37cfe79
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
Now the only way to enable Neon support is to change the mkspec.
[ChangeLog][Important Behavior Changes] Qt no longer checks for
support for the Neon FPU on ARM platforms at runtime. Code optimized
for Neon must be enabled unconditionally at compile time by ensuring
the compiler supports Neon. You may need to edit your mkspec for that.
Task-number: QTBUG-30440
Change-Id: I4df9b2bf3cd022f8ed70f02f16878cb2cb3fe6fb
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
If the compiler supports SSE2, we'll use our SSE2-optimised code
unconditionally. Runtime detection is left for SSSE3 code.
The SSE2 codebase is big and thus a timebomb if an inline function
gets leaked out and run without runtime check. In reality, it's
extremely unlikely people running CPUs without SSE2 support are
running Qt 5 at this moment (they're either too old or too new,
e.g. Intel Quark).
The SSSE3 codebase is a lot more manageable.
Task-number: QTBUG-30440
Change-Id: I3e586e4434e820365d5316b650ee3061d0acf767
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Currently if the converter_map does not have an entry for how to
do a direct conversion, the image is first converted to ARGB32 and
then to the final format. In most cases the conversion to ARGB32 is
the generic converter which convert over ARGB32PM, which mean the data
is converted to ARGB32PM then to ARGB32, then to ARBG32PM and then to
final format.
This patch uses the generic converter directly everywhere there isn't
an optimized converter declared and it is valid.
Change-Id: Ibad57432ef8c58025e63a032b09f92dad4a559ee
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
A large part of qimage.cpp is the implementation of the various
conversions methods from one format to another. These methods could all
be moved to a separate file to make qimage.cpp smaller, and possible
later enable more aggressive optimization on the conversions.
Change-Id: I355806710281b4222114cdd9c35a7c28cbc0f692
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
Allan Sandfeld Jensen
Liang Qi
Thiago Macieira
Eirik Aavitsland
Jani Heikkinen
Frederik Gladhorn
Dmitry Shachnev
Tor Arne Vestbø
Frederik Gladhorn
Oswald Buddenhagen
Gabriel de Dietrich
Matti Paaso
Giuseppe D'Angelo
Allan Sandfeld Jensen
Adrian Perez de Castro