GpuSync node that owns a ShaderExecutionContext and dispatches it asynchronously, polling for completion on each subsequent compute_frame(). More...

#include <GpuComputeNode.hpp>

Inheritance diagram for MayaFlux::Nodes::GpuSync::GpuComputeNode:

Collaboration diagram for MayaFlux::Nodes::GpuSync::GpuComputeNode:

Public Member Functions
void	clear_gpu_update_flag () override
	Clear the "needs update" flag after GPU binding.

void	compute_frame () override
	Drive the async dispatch lifecycle.

NodeContext &	get_last_context () override
	Retrieves the last created context object.

	GpuComputeNode (std::shared_ptr< Yantra::ShaderExecutionContext<> > executor, bool continuous=false)
	Construct with a configured ShaderExecutionContext.

bool	is_pending () const
	Returns true if a dispatch has been submitted and not yet completed.

bool	is_ready () const
	Returns true if the last dispatch completed and callbacks have fired.

bool	needs_gpu_update () const override
	Mark if this node needs GPU binding update.

void	notify_tick (double value) override
	Notifies all registered callbacks with the current context.

void	on_complete (const TypedHook< GpuComputeContext > &callback)
	Register a callback fired once per completed dispatch.

bool	remove_complete_hook (const TypedHook< GpuComputeContext > &callback)
	Remove a previously registered on_complete callback.

void	restore_state () override
	Restores the node's state from the last save Recursively cascades through all connected modulator nodes Protected - only NodeSourceProcessor and NodeBuffer can call.

void	save_state () override
	Saves the node's current state for later restoration Recursively cascades through all connected modulator nodes Protected - only NodeSourceProcessor and NodeBuffer can call.

void	set_dirty ()
	Arm a dispatch on the next compute_frame() call.

void	update_context (double value) override
	Updates the context object with the current node state.

	~GpuComputeNode () override=default

Public Member Functions inherited from MayaFlux::Nodes::GpuSync::GpuSync
std::vector< double >	process_batch (unsigned int num_samples) override
	Batch processing for GPU nodes.

double	process_sample (double) override
	Single sample processing hook.

	~GpuSync () override=default

Public Member Functions inherited from MayaFlux::Nodes::Node
void	add_buffer_reference ()
	Increments the buffer reference count This method is called when a new buffer starts using this node to ensure proper lifecycle management.

uint64_t	get_active_snapshot_context () const
	Get the active snapshot context ID.

const std::atomic< uint32_t > &	get_channel_mask () const
	Retrieves the current bitmask of active channels using this node.

uint32_t	get_frame_rate () const

std::span< const float >	get_gpu_data_buffer () const
	Provides access to the GPU data buffer.

virtual double	get_last_output ()
	Retrieves the most recent output value produced by the node.

std::vector< ModulatorTree >	get_modulator_tree () const
	Returns the full modulator tree rooted at this node.

virtual std::vector< std::pair< ModulatorRole, std::shared_ptr< Node > > >	get_modulators () const
	Returns direct modulator nodes and their roles.

RoutingState &	get_routing_state ()
	Retrieves the current routing state of the network (non-const)

const RoutingState &	get_routing_state () const
	Retrieves the current routing state of the network.

uint32_t	get_sample_rate () const

bool	has_active_snapshot () const
	Check if node is currently being snapshotted by any context.

bool	has_capability (NodeCapability cap) const
	Query a single capability.

bool	is_buffer_processed () const
	Checks if the buffer has been processed.

bool	is_gpu_compatible () const
	Checks if the node supports GPU processing.

bool	is_in_network () const
	Sets whether the node is part of a NodeNetwork.

bool	is_in_snapshot_context (uint64_t context_id) const
	Check if currently in a snapshot context.

bool	is_used_by_channel (uint32_t channel_id) const
	Checks if the node is currently used by a specific channel.

bool	mark_buffer_processed ()
	Marks the node as having been processed by a buffer.

bool	needs_channel_routing () const
	Checks if the network is currently in a routing transition phase.

virtual uint8_t	node_capabilities () const
	Declare which data shapes this node's context can produce.

virtual void	on_tick (const NodeHook &callback)
	Registers a callback to be called on each tick.

virtual void	on_tick_if (const NodeCondition &condition, const NodeHook &callback)
	Registers a conditional callback.

void	register_channel_usage (uint32_t channel_id)
	Mark the specificed channel as a processor/user.

void	release_snapshot_context (uint64_t context_id)
	Release snapshot context.

virtual void	remove_all_hooks ()
	Removes all registered callbacks.

void	remove_buffer_reference ()
	Decrements the buffer reference count This method is called when a buffer stops using this node to ensure proper lifecycle management.

virtual bool	remove_conditional_hook (const NodeCondition &callback)
	Removes a previously registered conditional callback.

virtual bool	remove_hook (const NodeHook &callback)
	Removes a previously registered callback.

void	request_buffer_reset ()
	Requests a reset of the buffer state.

void	request_reset_from_channel (uint32_t channel_id)
	Requests a reset of the processed state from a specific channel.

virtual void	reset_processed_state ()
	Resets the processed state of the node and any attached input nodes.

void	set_frame_rate (uint32_t frame_rate)

virtual void	set_gpu_compatible (bool compatible)
	Sets whether the node is compatible with GPU processing.

void	set_in_network (bool networked)
	Marks the node as being part of a NodeNetwork.

void	set_sample_rate (uint32_t sample_rate)

bool	try_claim_snapshot_context (uint64_t context_id)
	Attempt to claim snapshot context for this processing cycle.

void	unregister_channel_usage (uint32_t channel_id)
	Removes the specified channel from the usage tracking.

virtual	~Node ()=default
	Virtual destructor for proper cleanup of derived classes.

Private Attributes
std::vector< TypedHook< GpuComputeContext > >	m_complete_callbacks

std::unique_ptr< GpuComputeContext >	m_context_storage

bool	m_continuous { false }

bool	m_dirty { false }

std::shared_ptr< Yantra::ShaderExecutionContext<> >	m_executor

Yantra::GpuChannelResult	m_last_result

Portal::Graphics::FenceID	m_pending_fence { Portal::Graphics::INVALID_FENCE }

bool	m_ready { false }

Additional Inherited Members
Public Attributes inherited from MayaFlux::Nodes::Node
bool	m_fire_events_during_snapshot = false
	Internal flag controlling whether notify_tick fires during state snapshots Default: false (events don't fire during isolated buffer processing) Can be exposed in future if needed via concrete implementation in parent.

std::atomic< uint32_t >	m_modulator_count { 0 }
	Counter tracking how many other nodes are using this node as a modulator.

std::atomic< NodeState >	m_state { NodeState::INACTIVE }
	Atomic state flag tracking the node's processing status.

Protected Member Functions inherited from MayaFlux::Nodes::GpuSync::GpuSync
void	notify_tick (double) override
	GPU sync nodes don't emit tick callbacks.

Protected Member Functions inherited from MayaFlux::Nodes::Node
virtual void	reset_processed_state_internal ()
	Resets the processed state of the node directly.

Protected Attributes inherited from MayaFlux::Nodes::Node
std::vector< NodeHook >	m_callbacks
	Collection of standard callback functions.

std::vector< std::pair< NodeHook, NodeCondition > >	m_conditional_callbacks
	Collection of conditional callback functions with their predicates.

uint32_t	m_frame_rate { 60 }
	Frame rate for time-based processing, used for normalization.

bool	m_gpu_compatible {}
	Flag indicating if the node supports GPU processing This flag is set by derived classes to indicate whether the node can be processed on the GPU.

std::vector< float >	m_gpu_data_buffer
	GPU data buffer for context objects.

double	m_last_output { 0 }
	The most recent sample value generated by this oscillator.

bool	m_networked_node {}
	Flag indicating if the node is part of a NodeNetwork This flag is used to disable event firing when the node is managed within a NodeNetwork, preventing redundant or conflicting event notifications.

uint8_t	m_node_capability { NodeCapability::SCALAR }
	Bitmask of capabilities declared by this node.

uint32_t	m_sample_rate { 48000 }
	Sample rate for audio processing, used for normalization.

bool	m_state_saved {}
	tracks if the node's state has been saved by a snapshot operation

uint32_t	m_timing_rate { m_sample_rate }
	Current timing rate for the node, used for timing calculations (can be sample rate or frame rate)

Detailed Description

GpuSync node that owns a ShaderExecutionContext and dispatches it asynchronously, polling for completion on each subsequent compute_frame().

Lifecycle per dispatch cycle:

set_dirty() is called (or continuous mode is active).
compute_frame() calls dispatch_core_async on the executor, stores the returned FenceID, and returns immediately. No GPU work blocks the caller.
On each subsequent compute_frame() while m_pending_fence is live, ShaderFoundry::is_fence_signaled is polled. While not signaled the call is a no-op.
Once signaled, readback_primary / readback_aux collect the result, on_complete callbacks are fired with a GpuComputeContext, and the fence is cleared. If continuous mode is active a new dispatch is armed immediately.

The node does not prescribe what consumers do with the result. Callbacks write into whatever NDData or external storage they own. The node itself stores nothing beyond the last fence.

Usage:

auto exec = std::make_shared<Yantra::ShaderExecutionContext<>>(
    Yantra::GpuShaderConfig { "my_shader.comp", { 256, 1, 1 }, sizeof(MyPC) });
exec->input(input_data).output(output_bytes).push(pc);
 
auto node = std::make_shared<GpuComputeNode>(std::move(exec));
node->on_complete([](GpuComputeContext& ctx) {
    auto result = Yantra::ShaderExecutionContext<>::read_output<float>(ctx.gpu_result, 1);
    // write into NDData, update state, etc.
});
node->set_dirty();

Definition at line 64 of file GpuComputeNode.hpp.

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

Public Member Functions

Private Attributes

Additional Inherited Members

Detailed Description