QSGRenderNode Class

Member Function Documentation

[override virtual] QSGRenderNode::~QSGRenderNode()

Destructs the render node. Derived classes are expected to perform cleanup similar to releaseResources() in here.

When a low-level graphics API is in use, the scenegraph will make sure there is a CPU-side wait for the GPU to complete all work submitted to the scenegraph's graphics command queue before the scenegraph's nodes are deleted. Therefore there is no need to issue additional waits here, unless the render() implementation is using additional command queues.

See also releaseResources().

[virtual] QSGRenderNode::StateFlags QSGRenderNode::changedStates() const

When the underlying rendering API is OpenGL, this function should return a mask where each bit represents graphics states changed by the render() function:

With APIs other than OpenGL, the only relevant values are the ones that correspond to dynamic state changes recorded on the command list/buffer. For example, RSSetViewports, RSSetScissorRects, OMSetBlendState, OMSetDepthStencilState in case of D3D11, or vkCmdSetViewport, vkCmdSetScissor, vkCmdSetBlendConstants, vkCmdSetStencilRef in case of Vulkan, and only when such commands were added to the scenegraph's command list queried via the QSGRendererInterface::CommandList resource enum. States set in pipeline state objects do not need to be reported here. Similarly, draw call related settings (pipeline states, descriptor sets, vertex or index buffer bindings, root signature, descriptor heaps, etc.) are always set again by the scenegraph so render() can freely change them.

The software backend exposes its QPainter and saves and restores before and after invoking render(). Therefore reporting any changed states from here is not necessary.

The function is called by the renderer so it can reset the states after rendering this node. This makes the implementation of render() simpler since it does not have to query and restore these states.

The default implementation returns 0, meaning no relevant state was changed in render().

const QSGClipNode *QSGRenderNode::clipList() const

Returns the current clip list.

[virtual] QSGRenderNode::RenderingFlags QSGRenderNode::flags() const

Returns flags describing the behavior of this render node.

The default implementation returns 0.

See also RenderingFlag and rect().

qreal QSGRenderNode::inheritedOpacity() const

Returns the current effective opacity.

const QMatrix4x4 *QSGRenderNode::matrix() const

Returns pointer to the current model-view matrix.

[virtual, since 6.0] void QSGRenderNode::prepare()

Called from the frame preparation phase. There is a call to this function before each invocation of render().

Unlike render(), this function is called before the scenegraph starts recording the render pass for the current frame on the underlying command buffer. This is useful when doing rendering with graphics APIs, such as Vulkan, where copy type of operations will need to be recorded before the render pass.

The default implementation is empty.

This function was introduced in Qt 6.0.

[since 6.5] const QMatrix4x4 *QSGRenderNode::projectionMatrix() const

Returns pointer to the current projection matrix.

In render() this is the same matrix that is returned from RenderState::projectionMatrix(). This getter exists so that prepare() also has a way to query the projection matrix.

When working with a modern graphics API, or Qt's own graphics abstraction layer, it is more than likely that one will want to load *projectionMatrix() * *matrix() into a uniform buffer. That is however something that needs to be done in prepare(), so outside the recording of a render pass. That is why both matrices are queriable directly from the QSGRenderNode, both in prepare() and render().

This function was introduced in Qt 6.5.

[virtual] QRectF QSGRenderNode::rect() const

Returns the bounding rectangle in item coordinates for the area render() touches. The value is only in use when flags() includes BoundedRectRendering, ignored otherwise.

Reporting the rectangle in combination with BoundedRectRendering is particularly important with the software backend because otherwise having a rendernode in the scene would trigger fullscreen updates, skipping all partial update optimizations.

For rendernodes covering the entire area of a corresponding QQuickItem the return value will be (0, 0, item->width(), item->height()).

See also flags().

[virtual] void QSGRenderNode::releaseResources()

This function is called when all custom graphics resources allocated by this node have to be freed immediately. In case the node does not directly allocate graphics resources (buffers, textures, render targets, fences, etc.) through the graphics API that is in use, there is nothing to do here.

Failing to release all custom resources can lead to incorrect behavior in graphics device loss scenarios on some systems since subsequent reinitialization of the graphics system may fail.

Unlike with the destructor, it is expected that render() can reinitialize all resources it needs when called after a call to releaseResources().

With OpenGL, the scenegraph's OpenGL context will be current both when calling the destructor and this function.

[pure virtual] void QSGRenderNode::render(const QSGRenderNode::RenderState *state)

This function is called by the renderer and should paint this node with directly invoking commands in the graphics API (OpenGL, Direct3D, etc.) currently in use.

The effective opacity can be retrieved with inheritedOpacity().

The projection matrix is available through state, while the model-view matrix can be fetched with matrix(). The combined matrix is then the projection matrix times the model-view matrix. The correct stacking of the items in the scene is ensured by the projection matrix.

When using the provided matrices, the coordinate system for vertex data follows the usual QQuickItem conventions: top-left is (0, 0), bottom-right is the corresponding QQuickItem's width() and height() minus one. For example, assuming a two float (x-y) per vertex coordinate layout, a triangle covering half of the item can be specified as (width - 1, height - 1), (0, 0), (0, height - 1) using counter-clockwise direction.

Clip information is calculated before the function is called. Implementations wishing to take clipping into account can set up scissoring or stencil based on the information in state. The stencil buffer is filled with the necessary clip shapes, but it is up to the implementation to enable stencil testing.

Some scenegraph backends, software in particular, use no scissor or stencil. There the clip region is provided as an ordinary QRegion.

With the legacy, direct OpenGL based renderer, the following states are set on the render thread's context before this function is called:

• glColorMask(true, true, true, true)
• glDepthMask(false)
• glDisable(GL_DEPTH_TEST)
• glStencilFunc(GL_EQUAL, state.stencilValue, 0xff); glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP) depending on clip
• glScissor(state.scissorRect.x(), state.scissorRect.y(), state.scissorRect.width(), state.scissorRect.height()) depending on clip
• glEnable(GL_BLEND)
• glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA)
• glDisable(GL_CULL_FACE)

States that are not listed above, but are covered by StateFlags, can have arbitrary values.

changedStates() should return which states this function changes. If a state is not covered by StateFlags, the state should be set to the default value according to the OpenGL specification. For other APIs, see the documentation for changedStates() for more information.

For APIs other than OpenGL, it will likely be necessary to query certain API-specific resources (for example, the graphics device or the command list/buffer to add the commands to). This is done via QSGRendererInterface.

Assume nothing about the pipelines and dynamic states bound on the command list/buffer when this function is called.

With some graphics APIs it can be necessary to reimplement prepare() in addition, or alternatively connect to the QQuickWindow::beforeRendering() signal. These are called/emitted before recording the beginning of a renderpass on the command buffer (vkCmdBeginRenderPass with Vulkan, or starting to encode via MTLRenderCommandEncoder in case of Metal. Recording copy operations cannot be done inside render() with such APIs. Rather, do such operations either in prepare() or the slot connected to beforeRendering (with DirectConnection).

See also QSGRendererInterface and QQuickWindow::rendererInterface().