List of nodes to be rendered to an output. More...

Detailed Description

List of nodes to be rendered to an output.

This references all the score::gfx::Node that have an effect on a given output, and manages all the matching renderers, as well as a few shared data, such as output-specific UBOs, shared textures and buffers, etc.

The score::gfx::Graph creates one RenderList per OutputNode in the graph.

Public Types
using	Buffers = std::pair< const Mesh *const, MeshBuffers >

Public Member Functions
	RenderList (OutputNode &output, const std::shared_ptr< RenderState > &state)

void	init ()
	Initialize data for this renderer.

QRhiResourceUpdateBatch *	initialBatch () const noexcept
	Initial resource update batch.

void	setInitialBatch (QRhiResourceUpdateBatch *batch) noexcept
	Store a resource update batch to be submitted on the next render frame.

void	flushInitialBatch ()
	Submit the pending initial batch before destroying resources it may name.

MeshBuffers	initMeshBuffer (const Mesh &mesh, QRhiResourceUpdateBatch &res)
	Create buffers for a mesh and mark them for upload.

void	update (QRhiResourceUpdateBatch &res)
	Update / upload this RenderList's shared data.

void	render (QRhiCommandBuffer &commands, bool force=false)
	Render every node in order.

QRhiCommandBuffer *	currentCommandBuffer () const noexcept

void	release ()
	Release GPU resources owned by this render list.

void	releaseBuffer (QRhiBuffer *buf)

bool	maybeRebuild (bool force=false)
	Check if the render size has changed in order to rebuild the pipelines.

bool	resizeSwapchainSizedTargets (QSize newOutputSize, QSize newRenderSize)
	Fast-path for pure viewport resize.

TextureRenderTarget	renderTargetForOutput (const Edge &edge) const noexcept
	Obtain the texture corresponding to an output port.

TextureRenderTarget	renderTargetForInputPort (const Port &p) const noexcept
	Look up the render target for a given input port.

BufferView	bufferForInput (const Edge &edge) const noexcept

BufferView	bufferForOutput (const Edge &edge) const noexcept

QImage	adaptImage (const QImage &in)
	Adapts an image to the GPU limits / format.

Buffers	acquireMesh (const ossia::geometry_spec &, QRhiResourceUpdateBatch &res, const Mesh *current, MeshBuffers currentbufs) noexcept

void	clearRenderers ()
	Clear the renderers so that they get reinitialized on the next frame.

QRhiTexture &	emptyTexture () const noexcept
	Texture to use when a texture is missing (2D)

QRhiTexture &	emptyTexture3D () const noexcept
	Texture to use when a 3D (sampler3D) texture is missing.

QRhiTexture &	emptyTextureCube () const noexcept
	Texture to use when a cubemap (samplerCube) is missing.

QRhiTexture &	emptyTextureArray () const noexcept
	Texture to use when a 2D array (sampler2DArray) is missing.

QRhiBuffer &	outputUBO () const noexcept
	UBO corresponding to the output parameters:

GpuResourceRegistry &	registry () noexcept
	Per-output GPU arena store for scene-graph source nodes.

const GpuResourceRegistry &	registry () const noexcept

VertexFallbackPool &	vertexFallbackPool () noexcept
	Per-RenderList pool of neutral fallback vertex buffers for "REQUIRED: false" VERTEX_INPUTS whose upstream geometry does not provide a matching attribute.

GpuTimings &	gpuTimings () noexcept
	Per-RenderList GPU-timing collector.

const GpuTimings &	gpuTimings () const noexcept

Gfx::AssetTable *	assetTable () const noexcept
	Session-wide asset decode cache.

void	setAssetTable (Gfx::AssetTable *t) noexcept

const score::gfx::Mesh &	defaultQuad () const noexcept
	A quad mesh correct for this API.

const score::gfx::Mesh &	defaultTriangle () const noexcept
	A triangle mesh correct for this API.

bool	requiresDepth (const score::gfx::Port &p) const noexcept
	Whether this list of rendering actions requires depth testing at all.

bool	anyNodeRequiresDepth () const noexcept

int	samples () const noexcept

bool	canRender () const noexcept

QSize	renderSize (const Edge *e) const noexcept

void	createAllInputRenderTargets ()

void	markBuilt () noexcept
	Mark this render list as fully built.

bool	isBuilt () const noexcept
	Is this render list currently coherent with its output's GPU objects?

void	markRequiresDepth (bool value) noexcept

void	onEdgeRemoved (Edge &edge, const ossia::hash_set< const Port * > *preserveSinks=nullptr)

void	removeInputRenderTarget (const Port *port)
	Remove the render target for a specific input port.

QSize	resolveDownstreamSize (const Node node, const ossia::small_flat_map< const Port , RenderTargetSpecs, 16 > &resolvedSpecs) const noexcept
	Resolve the downstream render target size for a node.

Public Attributes
OutputNode &	output
	Output node to which this RenderList is rendering to.

RenderState &	state
	RenderState corresponding to this RenderList.

std::vector< score::gfx::Node * >	nodes
	Nodes present in this RenderList, in order.

std::vector< score::gfx::NodeRenderer * >	renderers
	Renderers - one per node.

int64_t	frame = 0

float	currentDate [4] {}

Friends
struct	Graph

Member Function Documentation

◆ adaptImage()

QImage score::gfx::RenderList::adaptImage ( const QImage & in )

Adapts an image to the GPU limits / format.

e.g. Y direction, texture size limits...

◆ assetTable()

Gfx::AssetTable * score::gfx::RenderList::assetTable ( ) const

inlinenoexcept

Session-wide asset decode cache.

Set by Graph::createRenderList from GfxContext's AssetTable. May be null on test RenderLists or after teardown. Consumers must guard.

◆ currentCommandBuffer()

QRhiCommandBuffer * score::gfx::RenderList::currentCommandBuffer ( ) const

inlinenoexcept

The command buffer render() is currently recording into, or null outside a frame. Strategies that must record raw-API commands (the Vulkan host-import upload) need it at update() time, where only the resource batch is passed.

◆ flushInitialBatch()

void score::gfx::RenderList::flushInitialBatch ( )

Submit the pending initial batch before destroying resources it may name.

A QRhiResourceUpdateBatch stores raw QRhiBuffer / QRhiTexture pointers until it is committed, and QRhiResource::deleteLater() outside a frame deletes immediately. Graph mutations run between frames, so any path that tears down renderer resources while this batch is outstanding must call this first; otherwise the next render() submits updates naming freed resources.

◆ gpuTimings()

GpuTimings & score::gfx::RenderList::gpuTimings ( )

inlinenoexcept

Per-RenderList GPU-timing collector.

Renderers wrap their begin/endPass regions in ScopedGpuTimer to attribute the CB-wide lastCompletedGpuTime to the named pass. The result is one frame stale — see GpuTiming.hpp for details.

◆ initialBatch()

QRhiResourceUpdateBatch * score::gfx::RenderList::initialBatch ( ) const

noexcept

Initial resource update batch.

See QRhiResourceUpdateBatch::merge for documentation on this

◆ initMeshBuffer()

MeshBuffers score::gfx::RenderList::initMeshBuffer	(	const Mesh &	mesh,
		QRhiResourceUpdateBatch &	res
	)

Create buffers for a mesh and mark them for upload.

The meshes used by the nodes are cached (as most are just rendering on a full-screen triangle, which we can reuse).

◆ isBuilt()

bool score::gfx::RenderList::isBuilt ( ) const

inlinenoexcept

Is this render list currently coherent with its output's GPU objects?

False between the moment something invalidates the list (a fast-path viewport resize: resizeSwapchainSizedTargets) and the maybeRebuild() that runs on the next render frame. In that window the output node has ALREADY destroyed and replaced its QRhiTextureRenderTarget / QRhiRenderPassDescriptor while the renderers still hold the pre-resize snapshot, so nothing outside the rebuild may ask a renderer for a render target.

◆ markBuilt()

void score::gfx::RenderList::markBuilt ( )

inlinenoexcept

Mark this render list as fully built.

Prevents maybeRebuild() from unnecessarily tearing down and recreating all resources on the first render frame after createRenderList() has already fully initialized everything.

◆ markRequiresDepth()

void score::gfx::RenderList::markRequiresDepth ( bool value )

inlinenoexcept

Set the "any node requires depth" flag computed from the node graph. Mirrors what maybeRebuild() recomputes; called from Graph::createRenderList so the freshly-built RL doesn't need a first-frame maybeRebuild to populate it.

◆ onEdgeRemoved()

void score::gfx::RenderList::onEdgeRemoved	(	Edge &	edge,
		const ossia::hash_set< const Port * > *	preserveSinks = `nullptr`
	)

Notify that an edge was removed. Notifies renderers, releases RT if unused.

Parameters

preserveSinks Optional set of sink Ports that should keep their input render target even if this edge was their only feed. Used by batched edge updates (see GfxContext::incrementalEdgeUpdate) so that inserting a filter between two nodes doesn't destroy and immediately re-allocate the same RT when the old and new edges share a sink port.

◆ outputUBO()

QRhiBuffer & score::gfx::RenderList::outputUBO ( ) const

inlinenoexcept

UBO corresponding to the output parameters:

Render size
Per-API adjustments and globals

◆ registry()

GpuResourceRegistry & score::gfx::RenderList::registry ( )

inlinenoexcept

Per-output GPU arena store for scene-graph source nodes.

Returns a reference to the registry that owns the Camera / Light / Material / PerDraw arena buffers. Source nodes (Camera, Light, PBRMesh, …) allocate a slot from this registry at construction and write their packed bytes into it at their own update().

Persist-across-rebuild contract: the registry is owned by the OutputNode (OutputNode::m_registry) and survives RenderList rebuilds — the same registry pointer is observed by both the pre- and post-rebuild RenderList for a given OutputNode. Consumers that cache the registry pointer (e.g. ScenePreprocessor's m_registry) can compare it against the new RL's registry on init() and skip the cache wipe when it is unchanged.

Valid between init() and release().

◆ renderTargetForInputPort()

TextureRenderTarget score::gfx::RenderList::renderTargetForInputPort ( const Port & p ) const

noexcept

Look up the render target for a given input port.

All input render targets are created centrally before any node init() so that they are available regardless of initialization order.

◆ renderTargetForOutput()

TextureRenderTarget score::gfx::RenderList::renderTargetForOutput ( const Edge & edge ) const

noexcept

Obtain the texture corresponding to an output port.

This is done by looking for the render target which corresponds to a given port.

◆ requiresDepth()

bool score::gfx::RenderList::requiresDepth ( const score::gfx::Port & p ) const

noexcept

Whether this list of rendering actions requires depth testing at all.

e.g. it's not needed if we're just doing some generative shaders.

◆ resizeSwapchainSizedTargets()

bool score::gfx::RenderList::resizeSwapchainSizedTargets	(	QSize	newOutputSize,
		QSize	newRenderSize
	)

Fast-path for pure viewport resize.

Update state.outputSize to newOutputSize and state.renderSize to newRenderSize, and mark every renderer's renderTargetSpecsChanged so the rt_changed block in renderInternal does the RT recreation and sampler rebinding on the next render frame.

The two sizes are distinct: the output node owns the render size (the /rendersize override, ScreenNode::setRenderSize), the platform owns the output size (the swapchain). Callers pass both from the RenderState the output node has already updated.

Skips the full recreateOutputRenderList teardown + rebuild (pipeline compiles, ScenePreprocessor rebuild, mesh slab uploads, texture array reallocation): the persistent registry and ScenePreprocessor caches make that work unnecessary for a pure size change. Cost is O(N renderers), with no GPU drain and no allocation until the next frame's rt_changed block recreates the RTs.

Returns true on success. Returns false, and the caller must fall back to recreateOutputRenderList, when:

either size is invalid
renderers vector is empty (RL not yet initialised)

◆ resolveDownstreamSize()

QSize score::gfx::RenderList::resolveDownstreamSize	(	const Node *	node,
		const ossia::small_flat_map< const Port *, RenderTargetSpecs, 16 > &	resolvedSpecs
	)		const

noexcept

Resolve the downstream render target size for a node.

Returns the maximum size across all downstream render targets that this node renders to. Used as fallback when a node's input port has no explicit render target size.

◆ setInitialBatch()

void score::gfx::RenderList::setInitialBatch ( QRhiResourceUpdateBatch * batch )

inlinenoexcept

Store a resource update batch to be submitted on the next render frame.

Used by incremental edge additions that happen after the first render frame (when the original m_initialBatch has already been consumed).

◆ vertexFallbackPool()

VertexFallbackPool & score::gfx::RenderList::vertexFallbackPool ( )

inlinenoexcept

Per-RenderList pool of neutral fallback vertex buffers for "REQUIRED: false" VERTEX_INPUTS whose upstream geometry does not provide a matching attribute.

Valid between init() and release(). See VertexFallbackPool.hpp.

The documentation for this class was generated from the following files:

RenderList.hpp
RenderList.cpp

Detailed Description

Public Types

Public Member Functions

Public Attributes

Friends

Member Function Documentation

◆ adaptImage()

◆ assetTable()

◆ currentCommandBuffer()

◆ flushInitialBatch()

◆ gpuTimings()

◆ initialBatch()

◆ initMeshBuffer()

◆ isBuilt()

◆ markBuilt()

◆ markRequiresDepth()

◆ onEdgeRemoved()

◆ outputUBO()

◆ registry()

◆ renderTargetForInputPort()

◆ renderTargetForOutput()

◆ requiresDepth()

◆ resizeSwapchainSizedTargets()

◆ resolveDownstreamSize()

◆ setInitialBatch()

◆ vertexFallbackPool()