Rework the way the QResourceManager allocates and manages
it's resources. Get rid of the huge pre-allocated freelist,
and instead merge it with the counter field (that is now
always a quintptr large).

Give QHandle a direct pointer to it's data in addition to
the counter. This makes QHandle 8 or 16 bytes large, but
speeds up access to it's data and gives more safety on the
counter as that one now always has at least 31 bits.

Counter and the linked freelist use the same memory location.
To avoid potential conflicts, make sure the counter is always
an odd number, so it can't be a valid pointer.

Do not use INDEXBITS anymore, the resource manager can now
always allocate as many resources as we have RAM available
and will grow with the required use (and not allocate fixed
memory upfront).

This change reduces the memory consumption of the qardboard
example from a bit above 40 to 10MB.

Change-Id: I514f2d3f957f8635098fb88342e42e3361456340
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

We ended up having corrupted meshes if the application which exported
the OBJ was adding trailing white spaces. Also make sure we got a test
case using crlf for end of lines.

Change-Id: Iace9dbc3d0d124fefe9e3350d396fdf26555cd17
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

Change-Id: Ib06d84088d93e00ffdcaa6baa9d34e03358943cb
Reviewed-by: Svenn-Arne Dragly <svenn-arne.dragly@qt.io>
Reviewed-by: Andy Nichols <andy.nichols@qt.io>

qanimationclip.h:56: warning: No documentation for Qt3DAnimation::QAnimationClip::clipData

Change-Id: Ic44c96d47d3dff095b6b299f5280f21345dba452
Reviewed-by: Venugopal Shivashankar <Venugopal.Shivashankar@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>

qtexturewrapmode.h:60: warning: No documentation for 'Qt3DRender::QTextureWrapMode::WrapMode'
qtexturewrapmode.h:55: warning: No documentation for 'Qt3DRender::QTextureWrapMode::x'
qtexturewrapmode.h:56: warning: No documentation for 'Qt3DRender::QTextureWrapMode::y'
qtexturewrapmode.h:57: warning: No documentation for 'Qt3DRender::QTextureWrapMode::z'
qtexturegenerator.h:59: warning: No documentation for 'Qt3DRender::QTextureGenerator::operator()()
qtexturegenerator.h:60: warning: No documentation for 'Qt3DRender::QTextureGenerator::operator==(const QTextureGenerator &other)
qtexturewrapmode.h:60: warning: No documentation for 'Qt3DRender::QTextureWrapMode::WrapMode'

Change-Id: I6a77f43dfb153de142d12ef1c3cdb1850669d49f
Reviewed-by: Venugopal Shivashankar <Venugopal.Shivashankar@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>

Change-Id: I342a74f6605d7e6a9ac6f07af21dc8f7fd80c3ce
Reviewed-by: Venugopal Shivashankar <Venugopal.Shivashankar@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>

After a fix for a crash, this test now FAILs, which requires
blacklisting until it is fixed

Task-number: QTBUG-63639
Change-Id: I918600f36053a5a3e1adb31da4f1d4ec3303b8b0
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>

After a fix for a crash, this test now FAILs, which requires
blacklisting until it is fixed

Task-number: QTBUG-63639
Change-Id: I10ce9b6b80aa5adf83f2113be7c6b2007463b73e
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>

Add a few pointer checks. This allows for the test to be blacklisted
instead of marked as insignificant.

Task-number: QTBUG-63639
Change-Id: Ide9216226d880909a600c3becc0d6ca7259e5a69
Reviewed-by: Tony Sarajärvi <tony.sarajarvi@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>

Use new version with commercial exception.

Change-Id: I09809579db277135edce1c6c1938ac3915600e11
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>

Texture data from generators may not be available in the current frame and
the next frame should be rendered immediately, even in OnDemand rendering.
The problem was that this information was not propagated back to the
renderer, it was just silently ignored inside
GraphicsContext::setParameters. Added a bool return type to
GraphicsContext::setParameters so that the renderer could check for
errors.

Task-number: QTBUG-63561
Task-number: QTBUG-59752
Change-Id: I18fb6f8236bf47d3eeab07d879dffaa599665e6b
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

Change-Id: I8d6bb938f7ad06fa97898f4242af76cb48efbfc5

Change-Id: Ib54e714e9152ac669441d0067ceb27f5b29ca436

All buffers are uploaded to GPU before the frame submission takes place.
If during frame submission we check for a CPU buffer whether it's dirty, we end up
(rightfully) with a race condition as we are preparing for frame n+1 already.
There must have been a left over when the above changes were made as the
UBO/SSBO code paths would, during rendering, check whether a CPU buffer was
dirty, and if so try to upload, causing a race.
Removed these code paths which aren't needed.

Change-Id: I713a92b881335ecfb30f24bd7485a7bac29be4b7
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

This now correctly handles both the old ChannelMapping and the
new SkeletonMapping use cases. There was some ambiguity about
which way around the indices were being generated. It worked in
previous cases for the limited data available to test with. This
has been tested to work with both the JSON animation format and
skeletal animations loaded from JSON and glTF 2 files.

Change-Id: Ie8558c764cd8dc57d6cdad19bbc6a66052bc68d0
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

Change-Id: I4a9044933b182a2e0eba745483f1b8bc214dcdba
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

This only loads the shader source once instead of 1000 times. Reduces
the time to first frame by a further 30% by massively reducing the
work the main thread has to do at startup. The Qt3DExtras material
classes are convenient but lead to bad practices like this. We should
provide a way to allow them to share the effect easily but still use
the provided materials for their API. Perhaps a new ctor overload that
takes a QEffect pointer would work nicely.

Also fixed some float vs double vs int precision warnings as a drive
by.

Change-Id: I181de8d95841eb88c91c29ef6f45d215da079b98
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

No point loading the same data 1k times. Shaves 35% off the time from
application start to first frame being complete.

Change-Id: I8cd7c8a5aa377c0663c03e35d7a4c03f50b052f6
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

Fixes clang warning.

Change-Id: I1831bd112c33cdd1fe9cc7ae169cfba05cb8a694
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

The backend doesn't need to know about the rotation angles used to
update the transform. Saves about 3% of time on the main thread
related to mallocing the property change notifications.

Change-Id: I784baf705cdd697ba860dc7a0c2e0e1c9ee467ec
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>

Everything goes through the other overload now.

Change-Id: Id0c196a6967a9f83e47092ca1fc7f4ab4ebe9309
Reviewed-by: Sérgio Martins <sergio.martins@kdab.com>

Will make it possible to reduce the amount of code to maintain and test.
There are still several ways to optimize this too so best to only do it
for one code path that all animators use.

Change-Id: I05ebd4f41d88aa0bea37d4d17130bc79c96ebd5c
Reviewed-by: Sérgio Martins <sergio.martins@kdab.com>

Should pass before and after the refactoring.

Change-Id: I9ef8f948f03a131546fc6485e9c66494ff383714
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

Change-Id: I87e8292ee606af33f729e2b492a616325a49c89a
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

The timer is parented to the QMouseHandler (introduced in
595b4add

) and was therefore deleted both
by the parent and by the QScopedPointer, leading to crash on exit.
This commit removes the QScopedPointer and uses a raw pointer instead.

Task-number: QTBUG-63462
Change-Id: I6b031caf7cb69ccbde74995661f4ce8c73f21d88
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

Caches the resolution of node ids to ChannelMappings. Updated test to
check this new feature.

Change-Id: I0c2f539c2bd91e7918b98c5b60c9ca486b63ec08
Reviewed-by: Kevin Ottens <kevin.ottens@kdab.com>

Change-Id: I9015546607b5663f1feada64bc03cb73d3538ef2
Reviewed-by: Kevin Ottens <kevin.ottens@kdab.com>

Rather than assuming they are in order. The blended clip animator
code path now behaves the same as the simple QClipAnimator in
that it can handle clips with channel components not in the
expected order. We can now look at making both share a common
code path.

Change-Id: I61848e07b4f418ba9f4c56f56f4f49c9a5fbd588
Reviewed-by: Kevin Ottens <kevin.ottens@kdab.com>

Change-Id: I5a86547ae669394c290b30427f98821bc7552a63
Reviewed-by: Kevin Ottens <kevin.ottens@kdab.com>

This correctly handles the case where the blend tree consists of a
single value node. Added test case to catch it.

Change-Id: I8982d2081748866d163f107a4513bd2d17feb83a
Reviewed-by: Kevin Ottens <kevin.ottens@kdab.com>

Copy externalRenderTargetSize and surfacePixelRatio from
old to new surfaceSelector when setting activeFrameGraph
on RenderSettings. Previously, when switching
activeFrameGraph, surface was copied from old to new
surfaceSelector, but these other properties were not. The
rendered image was then stretched until window is resized
again.

Change-Id: I62b78da05c764f981813ab96b9a769bae7d786b2
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

For < 3 vertices, array accesses would fail, and face creation not
make sense anyways

Change-Id: I5eb746dfdec92d1c836e9870d87e7900530ec81c
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

Or else we will continue uploading data indefinitely

Change-Id: If81d8a0a790964c699462fbcb166808ee7859f59
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

Change order of local variables affecting the destructions order
and making sure window outlives the surface selector

Change-Id: Ia04f110656f4cc563643fb056ae491ffadf3a8b4
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

In some cases, the list of pickable objects may contain entities that don’t have an object picker (or not parent has).
Now filter the pick results to only keep entities which either have an object picker or are children of an entity that does.

Added unit test which now passes

Change-Id: I930c3d60cf2d19e845fe6c0de904c53b93ebe8be
Reviewed-by: Kevin Ottens <kevin.ottens@kdab.com>

We always use the same policy anyway.

Change-Id: I3349b8c19ce0e6239b140dfac7ba66f8dda1b9be
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

Instead, direclty use the Allocator to acquire and release handles
without an additional indirection. This removes around 30% of the
overhead in run().

Change-Id: Ic4a9343dd52a900eb1c7eb6b4135bc7670076df1
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

It's not used anywhere

Change-Id: Id65ec9fc8e38697c0dbd23fe4f816e1637a22483
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

This should be removeAll instead of remove(). Remove takes an
integer (index into the array). It compiles by chance (but doesn't
do the right thing), as QHandle has a cast operator to int.

Change-Id: If8d59a535021bab3d2553f83bef7cf769c738dd6
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>

The mutex right now is used for controlling access to the m_dirty flags,
and for guarding access to m_gl and other data.

We don't need to use it for accessing the flags, as an atomic int will
do just fine and relieve us of potential deadlocks. The mutex must only
guard changes to the actual data.

Task-number: QTBUG-61130
Change-Id: Ia1f25af2233387f375c077965e901c67f972f1ec
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>