Qt Canvas Painter blog post series

[Kaj Gronholm]

@JKSH Thanks! So Canvas2D it might be then 🙂

[beeka]

@Kaj-Gronholm There are many cases where GPL is problematic but LGPL would work. While I understand the commercial imperative, it is depressing that nothing new is being contributed to the core Qt (I appreciate there are bug fixes). It doesn't feel like a "difficult decision" but instead what is now normal. It is also frustrating that the commercial license is too expensive for me to justify to management (together with the clause preventing switching part-way through development). I still had hope that small features or performance-focused improvements might still find a home under the LGPL, but that does not look likely now.

[Kaj Gronholm]

@beeka I'm confident the licensing feedback will be monitored and adjusted if needed. Also, I'm sure there are ways to avoid the part-way through development issue. If in doubt, please get your management in touch with our sales representatives.

[Kaj Gronholm]

The third post of the trilogy is out now: https://www.qt.io/blog/accelerated-2d-canvas-benchmarks

TL;DR: Canvas Painter can be up to 10 times faster than QPainter with the OpenGL backend. Please read the blog post for details and comment here with your own results!

[kkoehne]

While I understand the commercial imperative, it is depressing that nothing new is being contributed to the core Qt (I appreciate
there are bug fixes). It doesn't feel like a "difficult decision" but instead what is now normal.

I can understand the frustration, but it's simply not true that everything 'new' in Qt is released only under Qt Commercial + GPL3 licenses. Actually the other three new modules in Qt 6.11 - Qt TaskTree, Qt OpenAPI and Qt Labs StyleKit - are made available under Qt Commercial + LGPLv3! And there are also new features in a lot of established Qt modules ...

This obviously doesn't help you though if you are interested primarily on Qt CanvasPainter.

[Dyami Caliri]

This is really cool, and something I've been waiting for to be able to transition our software from OpenGL to QRhi.

However, in my first attempts to use it, I can't seem to do what I want to do: use QCanvasPainter to draw OVER other content I've already drawn within a QRhi render pass. It seems to insist on clearing the contents. I want to draw grids, masking marks, and text over 2D composited images I've already drawn.

Is this possible?

[agocs]

@Dyami-Caliri said in Qt Canvas Painter blog post series:

This is really cool, and something I've been waiting for to be able to transition our software from OpenGL to QRhi.

However, in my first attempts to use it, I can't seem to do what I want to do: use QCanvasPainter to draw OVER other content I've already drawn within a QRhi render pass. It seems to insist on clearing the contents. I want to draw grids, masking marks, and text over 2D composited images I've already drawn.

Is this possible?

Yes. See https://doc-snapshots.qt.io/qt6-dev/qcanvasrhipaintdriver.html#endPaint (esp. the second snippet with the DoNotRecordRenderPass flag)

This can be seen in action in the hellorhi2 manual test that renders a triangle with QRhi and then some QCanvasPainter-based rendering within the same pass within a QRhiWidget: https://code.qt.io/cgit/qt/qtcanvaspainter.git/tree/tests/manual/hellorhi2/hellorhi2.cpp?h=6.11#n97
(there is no public example for this kind of low-level usage currently, perhaps it is something we should consider, since this is definitely a valid and important use case for QCanvasPainter)

[Dyami Caliri]

@agocs That's brilliant. Thank you. I'm sorry I didn't read the documentation more thoroughly.

[Dyami Caliri]

I'm making great progress with using this in our application. The antialiasing is really excellent.

One thing I've noticed is that QCanvasImage does not seem to have a devicePixelRatio. So, if I create one from a QImage that is a retina @2x resource, I have to then know when I paint to pass in the calculated width and height. Is my understanding correct, or am I missing something again?

[Kaj Gronholm]

@Dyami-Caliri Thanks!

You are correct, QCanvasImage currently works on actual pixels and doesn't take into account devicePixelRatio or "@nx" images. So it matches with "lower-level" APIs behavior mentioned here https://doc.qt.io/qt-6/highdpi.html#drawing

Please tell what would be the expected behavior in your opinion? What would QCanvasImage width&height return, how would different drawImage() methods behave, should there be additional API etc. Thanks!

[Dyami Caliri]

@Kaj-Gronholm Thank you for clarifying.

It seems to me that the rest of the QCanvasPainter does take devicePixelRatio into account. If you call the QCanvasRhiPaintDriver::beginPaint method that takes logicalSize and dpr parameters, the painter will automatically scale points and line widths, so that the drawing looks the same across different dpr screens. I have confirmed this with testing.

The main difference in behavior I would expect is for QCanvasPainter::drawImage(const QCanvasImage &image, float x, float y) to draw the image based on the logical dimensions (dimensions / image_dpr). So, if I've created a QCanvasImage from a QImage with dpr=2, these should be equivalent:
painter->drawImage(retinaImg, x, y);
painter->drawImage(retinaImg, x, y, retinaImg.width() / 2, retinaImg.height() / 2);

Or, put another way, if I had a @1px image and painted it with
painter->drawImage(img, x, y),
and then replaced it with a @2px or @3px image, I would expect it to take the same amount of logical space on the screen.