| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
There is no need for other classes to directly access the shader cache,
and if we ever want a way to invalidate pieces of it, we better prevent
such access.
Change-Id: I79949b375854445c46b9d8bf6072bb53030b8dac
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
|
|
|
|
|
|
| |
Task-number: QTBUG-48594
Change-Id: Ifc207938de7f0c8995fc712df92665f222612647
Reviewed-by: Alan Alpert <aalpert@blackberry.com>
|
|
|
|
|
| |
Change-Id: I6e1f03bad9e957a60e1248e423f6a0e237d43f55
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/qml/debugger/qv4debugservice.cpp
src/qml/jsruntime/qv4value_inl_p.h
src/qml/jsruntime/qv4value_p.h
src/qml/memory/qv4mm.cpp
src/qml/memory/qv4mm_p.h
src/qml/qml/qqmlnotifier_p.h
src/qml/qml/qqmlproperty.cpp
src/quick/items/qquickflickable.cpp
src/quick/items/qquicktextedit.cpp
tests/auto/quick/qquickwindow/BLACKLIST
The extra changes in qqmlbinding.cpp are ported from changes to
qqmlproperty.cpp that occurred in parallel with writeBinding() being
moved to qqmlbinding.cpp.
Change-Id: I16d1920abf448c29a01822256f52153651a56356
|
| |
| |
| |
| |
| |
| | |
Change-Id: Ifc664b9c718744b9549953e42ac3450a88403dea
Task-number: QTBUG-47090
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
No reason to iterate through all items to create a temporary
QList we don't need.
Change-Id: I3c5388f026ded0290afb9dd35fc7934011c4998d
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Performing a full scan of the pages each time a new node is to be allocated is
wasteful. Instead, store the last known free page, and search from that position
forwards. When an item is freed, reset the last known free page to search
forward from that position. If no free page is found, don't fall back to a full
scan - just allocate a new page.
The creation of lots of new items is a pretty common pattern (PageStack on
mobile and similar), thus it makes sense to optimize for it and avoid scanning
where possible.
This takes the time to instantiate 142k Rectangles in a single
component from 2433ms on the render thread to 495ms on the render thread (down
slightly from 511ms when using no paging at all, and down from 638ms on 5.3.1).
Separately to this, it may make sense to increase the size of pages, but
ideally, this will wait until there is a shared pool between renderers.
Change-Id: I2b335e230bc28fffcbb38b3fb43f3d441789c256
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
|
|
|
|
|
|
|
|
| |
Found with GCC's -Wcast-qual.
Change-Id: Ia0aac2f09e9245339951ffff13c9589afabc7ade
Reviewed-by: Alan Alpert (Personal) <416365416c@gmail.com>
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
| |
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.
Change-Id: I61120571787870c0ed17066afb31779b1e6e30e9
Reviewed-by: Iikka Eklund <iikka.eklund@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Amongst other things, QSGRenderer::preprocess() triggers the rendering of
QQuickShaderEffectSource items into framebuffer objects. The renderers
responsible for these QQuickShaderEffectSource items bind and unbind
their own VAOs, which messes up the VAO state set in QSGRenderer::preprocess().
Scenegraph nodes must not depend on the vao state set by the renderer anyway as
the binding has to be considered as an implementation detail.
Change-Id: Idec34c65397394e12e7f8419a1c9633d5f70a22e
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Every batch would have its own buffer with its own memory
and the original idea was that it would stabilize so that
the same batch would be used over and over, so the memory
allocatd for each batch's buffer would be reused. In practice
batches get discarded and recycled in somewhat arbitrary
order so we always end up reallocating.
Instead, use a single buffer for all uploads. This saves us several
reallocations per frame, and also prevents pooling of several
potentially large allocated blocks inside the renderer.
The exception is when using visualization or for drivers
which have broken index buffer support (nouveau only currently),
where we keep the existing behavior as the buffer data is
needed during rendering and visualization.
Change-Id: Id29095e00128ec1ee6d4ac3caa8f2c17cc7bbd27
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>
|
|
|
|
|
|
|
|
| |
Cuts down quite a bit on the time spent on the render thread during
the sync phase.
Change-Id: Ide8c4348141c84ed8abd9a869607a022652c2828
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>
|
|
|
|
|
|
|
|
|
| |
- Renamed LICENSE.LGPL to LICENSE.LGPLv21
- Added LICENSE.LGPLv3 & LICENSE.GPLv2
- Removed LICENSE.GPL
Change-Id: I84a565e2e0caa3b76bf291a7d188a57a4b00e1b0
Reviewed-by: Jani Heikkinen <jani.heikkinen@digia.com>
|
|
|
|
|
|
|
|
|
| |
This convenience method is currently only used by the batch renderer.
Moving it allows removing the QOpenGLFunction inheritance of
QSGRenderer and unbinding it slightly from the rendering implementation.
Change-Id: I4322952f843de8d950ced32885feee8d6c4a2730
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
.qmake.conf
src/plugins/accessible/quick/quick.pro
src/quick/items/qquickpincharea.cpp
src/quick/scenegraph/coreapi/qsgbatchrenderer.cpp
src/quick/scenegraph/qsgthreadedrenderloop.cpp
Manually adjusted for TestHTTPServer constructor change:
tests/auto/quick/qquickimage/tst_qquickimage.cpp
Change-Id: I5e58a7c08ea92d6fc5e3bce98571c54f7b2ce08f
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When two separate changes have occurred, we need to not
only invalidate the batches directly overlapping those, but
also any batch which has render orders between the two. So,
keep track of the range of invalidated render orders and
invalidate everything in between.
Task-number: QTBUG-37422
Change-Id: Ie5a289d5c569b84917ec9ac52671173c566e69b3
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If a node sent DirtyGeometry and DirtyMaterial in the same round and
DirtyGeometry triggered its batch to be invalidated, we would take the
no-batch code path which did set the rebuild state to BuildBatches. This
looks ok on screen, but if the node changed from alpha to opaque, it
would not be put into an opaque batch. For things like full screen
rectangles, this can potentially hurt performance.
To prevent doing a full rebuild on any material change on batchless
items (aka all Rectangle nodes), we store the blend state of the material
in the element and do a full rebuild only when blend state changes.
Change-Id: Ifdf06fb72ef02ca47a135c821ddbcbe0d164ca29
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
When I introduced the invalidation of z ranges, I forgot that we need
to rebuild render lists when nodes move from opaque to alpha batches
or vice versa as that sorting happens in buildRenderLists(). To remedy
this, make Batch::isMaterialCompatible report blending changes
separately and trigger a full rebuild in this case.
Task-number: QTBUG-37422
Change-Id: I020813cb531ed58353f8340fcad58dec8d7856dd
Reviewed-by: Michael Brasser <michael.brasser@live.com>
|
|\
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/qml/jsruntime/qv4functionobject.cpp
src/quick/scenegraph/coreapi/qsgbatchrenderer_p.h
Change-Id: Id164f6c3b45501aa466908659ec4e3b957323753
|
| |
| |
| |
| |
| | |
Change-Id: I4eecff3c8a2c727d38d394305d248eddeef87e8e
Reviewed-by: Michael Brasser <michael.brasser@live.com>
|
|\|
| |
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/qml/compiler/qv4ssa.cpp
src/qml/jsruntime/qv4arrayobject.cpp
src/qml/jsruntime/qv4context.cpp
Change-Id: Ied5b23bec4dc14abe51127c507aed668f855c1e1
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Transform, material, opacity and geometry changes all used to trigger
full rebuilds, which meant that very little geometry could be retained
between frames. For instance, a rotating spinner on top of a gridview
this would cause a full rebuild where nothing is retained.
This change introduces new concept to the renderer: partial rebuilding
based on render order ranges. Since the render order of nodes is
strictly defined by their position in the tree and nothing else, we
should use that for the majority of rebuilds. When a change comes in
for a node, we invalidate its batch and all batches which it has
overlapping render orders with.
Render order rebuilds only happen when nodes are added and removed.
Change-Id: Ib4cb284164892b409e3fff5c492a54d60a5de2d7
Reviewed-by: Michael Brasser <michael.brasser@live.com>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There are two ways of setting this right now. One is to set it on
startup using an environment varible.
QSG_VISUALIZE=
"batches" - Visualize batchtes in the renderer. Merged batches are
drawn with solid color and Unmerged batches are drawn with a diagonal
line pattern. Few unique colors means good batching. Unmerged batches
are bad if they contain many individual nodes.
"clip" - Visualize clipping as red areas on top of the scene.
"overdraw" - Visualize all items in 3D to highlight overdraws. This mode
can also be used to detect geometry outside the viewport to some
extent. Opaque items are rendered with a green tint while translucent
items are rendered with a red tint. The bounding box for the viewport
is rendered in blue. Opaque content is easier for the scenegraph to
process and it can also be faster to render on some hardware.
"changes" - Changes in the scenegraph are visualized with a flashing
overlay with a random color. Changes on a primitive is visualized with
a solid color while changes in an ancestor, such as a matrix or
opacity changes is visualized with a pattern.
The second way to set the visualization mode is to set it at runtime
through QString QQuickWindowPrivate::customRenderMode. This "API" is
string based so it is not tied to the batch renderer and in theory can
support other custom renderers.
The visualized elements do not respect clipping and rendering order
is arbitrary.
Change-Id: I31efbe53fc905145bf48080ede3e36945cb60dcf
Reviewed-by: Michael Brasser <michael.brasser@live.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There is already a lot of code out there that makes use
of the rendernode hook and some of it is not capable
of dealing with the render target being something other
than 0. (Mozilla in this case).
By forcing everything into the "alpha" code path, the
hit on the renderer should be fairly minimal as long as
render nodes are kept to a minimum, so reverting back
to the ugly old behavior is acceptable.
The API is still internal and highly discouraged.
Change-Id: Ie35484f52da6b0420257c95710fdda07a2be2c23
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Some GL implementations (especially on embedded) will give us an
OpenGL context without a depth buffer. In low memory scenarios, it
might also be feasible to request a non-depth buffer context to save
the depth buffer memory. The renderer deals with this by treating all
nodes as translucent, by not adjusting the shaders and by not creating
the extra z-order vertex attribute for merged nodes.
Change-Id: I8edc92d530daa3e2628df2ba52901b47d87eaf26
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
|
|
|
|
|
|
|
|
|
|
| |
The renderer binds same buffer both for vertex and index data.
This is allowed on desktop & ES but is forbidden in WebGL.
Add a compile-time flag to disable this optimization(?) and
force using a separate buffer for the index data.
Change-Id: I57c17c883a55e02513a8e4427efb202cafaaf37e
Reviewed-by: Gunnar Sletta <gunnar.sletta@jollamobile.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
NaN does not compare well with other floats. The result is that the
bounding box is left at its initial values, FLT_MAX for top-left and
FLT_MIN for bottom-right. If so, treat geometry as invalid, aka
infinite.
Task-number: QTBUG-35192
Change-Id: I1df6987d56a0ce1f500b0eba344a5dcbc55f80a4
Reviewed-by: Mitch Curtis <mitch.curtis@digia.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The basic approach is to have the batched renderer create and bind a
vertex array object if it detects we are using an OpenGL Core profile
context. The VAO is bound for the duration of the QQ2 renderer's
work cycle and unbound at the end so as to not interfere with any
other VAO's a user may wish to use.
All shaders have been copied and ported to be compliant with the
GLSL 150 core specification which is the minimum for a Core profile
context (OpenGL 3.2 Core). We are not using any newer features as
yet so this will work anywhere we can get a Core profile context.
The QSGShaderSourceBuilder class has been extended to resolve any
requests for shaders to the same basefilename with "_core"
appended prior to any file extension. This could be extended in
the future to allow version, or GPU or platform specific shaders.
The QSGShaderSourceBuilder has also been extended to allow it to
insert #define definitions in the prologue of a shader. Any such
definition is inserted:
* After the last #extension directive (if any are found)
* Otherwise after the #version directive (if found)
* Otherwise at the start of the shader source
This is required by the custom particle shaders which make
extensive use of such #defines.
In addition the mechanism used by the distance field glyph cache to
extend the cache with new glyphs has been modified to work (and
work more efficiently) when using a Core profile context.
Rather than using a shader program and a buffer filling quad to
blit the old texture into the new cache texture, we instead use
the technique of framebuffer blitting. The existing fallback
implementation using glTexSubImage2D() is still available if
needed.
The DECLARATIVE_EXAMPLE_MAIN macro has been extended to allow easy
testing of any of the QtDeclarative examples with a core profile
context. Just run the example with
QT_QUICK_CORE_PROFILE=1 ./text
for e.g. The only ones that may not work out of the box are those
that provide GLSL shader source e.g. the customparticles or
shader effect examples. These work fine if the shader source is
adapted to GLSL 150 core.
In the future it may be a good idea to expose some context property
to QML that the user can use to determine what shader source
variation to provide to Qt Quick. Along these lines it would also
be very nice to allow the provision of shader source to
ShaderEffect or CustomParticle from a separate source file just as
we now do within Qt Quick.
Task-number: QTBUG-32050
Change-Id: Ia6e9f06dbb8508af9ae03c6b60fb418b4cc9e41f
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
|
|
|
|
|
|
|
| |
Task-number: QTBUG-34328
Change-Id: If0202a67d95500333a0fb6f4ca3eb19ecb027770
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
We pretransform vertices relative to their batch root and upload these
using single-precision floats. If the offsets are huge then the
floating point numbers start to get unstable and we get rendering
artifacts as a result. This typically happens for lists/tables with
huge models.
Task-number: QTBUG-34312
Change-Id: I2516f2b4fa93f44a1288659d05458fb1af0df943
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
When one item in a batch of other translucent items would
change opacity, we need to rebuild batches under the current
root as all items in a batch are expected to have the same
opacity.
Task-number: QTBUG-34311
Change-Id: I2b9db19f05bd3a82be65cfa8a91e9398e8d58d0f
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
See the task for the full reasoning behind this patch.
The threaded renderloop has been refactored to have one window per
thread. This is mostly a simplification of the current code path where
for loops over multiple windows are turned into if (window).
The QSGContext has been split into two classes, QSGRenderContext
for which there is one per OpenGLContext.
The rest of the patch is name changes and a couple of cleanups
in the hopes of simplifying this change.
Task-number: QTBUG-33993
Change-Id: I31c81f9694d7da7474a72333169be38de62613c4
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
If we merge geometry nodes that make actual use of the
z-coordinate, this information becomes lost when merging
and the result is that we end up with an arbitrary
wrongfully applied transformation to the merged element.
Task-number: QTBUG-33897
Change-Id: I6129243e9bb890949023c35dc6b7bce30d31709a
Reviewed-by: Andrew Knight <andrew.knight@digia.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
This is necessary for the QtWebEngine module.
This also adds an empty nokeywords test using the same mechanism
as qtbase/tests/auto/tools/moc/no-keywords.h to find conflicts
at compile time.
Change-Id: I9df541720797dd61f078178c2af68ead18ff8bfe
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When we do partial rebuilds, we need to update the render order
for the entire subtree unconditionally. The previous logic
was to reuse existing render orders beneath batch roots, which
caused overlapping orders to be used.
When geometry of an alpha batch changes, we need to rebuild
everything under the current root. Otherwise, we might end
up with previously batched content which incorrectly
overlaps the changed geometry.
Change-Id: I7b6de5ce34a02434294ac5ae29ae1b87eb3c4464
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
First, when doing a partial rebuild, we need to mark
all elements for orphaning, not just those without a batch,
as the renderlist should at all times contain everything
that is rendered on screen.
Second, we had a small bug with material changes where we
would overwrite the rebuild state instead of adding to it.
This led to the render lists not being rebuilt from scratch
in some scenarios.
Third, after the shadow nodes rewrite, we didn't mark
for a full rebuild when clips were added.
Change-Id: I6707ebfa2908fcedd7fb636fb25b844be15fdea1
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
|
|
The renderer tries to batch primitives together where possible, isolate
non-changing subparts of the scene from changing subparts and retain
vertexdata on the GPU as much as possible. Atlas textures are crucial
in enabling batching.
The renderer and atlas texture are described in detail in the doc page
"Qt Quick Scene Graph Renderer".
Change-Id: Ia476c7f0f42e1fc57a2cef528e93ee88cf8f7055
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
|