pulp::gpu_audio::GpuConvolutionReverb Class Reference

Optional, control-parity GPU route for the two-channel Forge convolution reverb. More...

#include <gpu_convolution_reverb.hpp>

Public Member Functions
	GpuConvolutionReverb (GpuConvolutionReverbConfig config)
	~GpuConvolutionReverb ()
	GpuConvolutionReverb (const GpuConvolutionReverb &)=delete
GpuConvolutionReverb &	operator= (const GpuConvolutionReverb &)=delete
bool	prepare () noexcept
	Prepare the opt-in route.
void	release () noexcept
bool	prepared () const noexcept
bool	gpu_enabled () const noexcept
std::uint32_t	block_size () const noexcept
std::uint32_t	latency_samples () const noexcept
void	set_ir_gain_db (double db) noexcept
void	set_predelay_ms (double ms) noexcept
void	set_wet_percent (double percent) noexcept
void	set_dry_percent (double percent) noexcept
void	set_width_percent (double percent) noexcept
void	set_lowcut_hz (double hz) noexcept
void	set_highcut_hz (double hz) noexcept
void	process (const audio::BufferView< const float > &input, audio::BufferView< float > &output, std::uint32_t n) noexcept
	Audio-thread entry point.
GpuConvolutionReverbReport	report () const noexcept
GpuAudioStatus	status_snapshot () const noexcept
	Allocation-free status of the least healthy lane (lane 0 on a tie), because one fenced or lost lane already changes what the route emits.
GpuAudioRecoveryReason	recovery_reason () const noexcept
	First latched recovery cause of that lane, or None.
bool	fenced () const noexcept
	True before preparation, or when any lane's GPU delivery is fenced.

Friends
struct	detail::GpuConvolutionReverbTestAccess

Detailed Description

Optional, control-parity GPU route for the two-channel Forge convolution reverb.

It uses two separately authenticated mono GpuConvolver instances, one per output lane, so a two-channel IR remains dual-mono and distinct L/R stimuli cannot accidentally collapse to one resident mono spectrum.

The wrapper owns only host-side control, filtering, predelay, dry alignment, and lane composition. GPU work and the continuously primed CPU fallbacks remain inside the existing GpuAudioTransport instances. Its fixed latency is exactly three internal host quanta: two transport lead quanta plus one staging quantum that makes the result independent of callback partitioning. It is reported through latency_samples().

Constructor & Destructor Documentation

GpuConvolutionReverb() [1/2]

pulp::gpu_audio::GpuConvolutionReverb::GpuConvolutionReverb ( GpuConvolutionReverbConfig config )

explicit

~GpuConvolutionReverb()

pulp::gpu_audio::GpuConvolutionReverb::~GpuConvolutionReverb

(

)

GpuConvolutionReverb() [2/2]

pulp::gpu_audio::GpuConvolutionReverb::GpuConvolutionReverb ( const GpuConvolutionReverb & )

delete

Member Function Documentation

block_size()

std::uint32_t pulp::gpu_audio::GpuConvolutionReverb::block_size ( ) const

inlinenoexcept

fenced()

bool pulp::gpu_audio::GpuConvolutionReverb::fenced ( ) const

noexcept

True before preparation, or when any lane's GPU delivery is fenced.

gpu_enabled()

bool pulp::gpu_audio::GpuConvolutionReverb::gpu_enabled ( ) const

inlinenoexcept

latency_samples()

std::uint32_t pulp::gpu_audio::GpuConvolutionReverb::latency_samples ( ) const

inlinenoexcept

operator=()

GpuConvolutionReverb & pulp::gpu_audio::GpuConvolutionReverb::operator= ( const GpuConvolutionReverb & )

delete

prepare()

bool pulp::gpu_audio::GpuConvolutionReverb::prepare ( )

noexcept

Prepare the opt-in route.

The default config is deliberately disabled and fails closed until gpu_enabled is set true by the selected mode.

prepared()

bool pulp::gpu_audio::GpuConvolutionReverb::prepared ( ) const

inlinenoexcept

process()

void pulp::gpu_audio::GpuConvolutionReverb::process ( const audio::BufferView< const float > & input, audio::BufferView< float > & output, std::uint32_t n )

noexcept

Audio-thread entry point.

It never allocates, locks, or calls a GPU API. Each lane transport receives the same quantum and sequence shape.

recovery_reason()

GpuAudioRecoveryReason pulp::gpu_audio::GpuConvolutionReverb::recovery_reason ( ) const

noexcept

First latched recovery cause of that lane, or None.

release()

void pulp::gpu_audio::GpuConvolutionReverb::release ( )

noexcept

report()

GpuConvolutionReverbReport pulp::gpu_audio::GpuConvolutionReverb::report ( ) const

noexcept

set_dry_percent()

void pulp::gpu_audio::GpuConvolutionReverb::set_dry_percent ( double percent )

noexcept

set_highcut_hz()

void pulp::gpu_audio::GpuConvolutionReverb::set_highcut_hz ( double hz )

noexcept

set_ir_gain_db()

void pulp::gpu_audio::GpuConvolutionReverb::set_ir_gain_db ( double db )

noexcept

set_lowcut_hz()

void pulp::gpu_audio::GpuConvolutionReverb::set_lowcut_hz ( double hz )

noexcept

set_predelay_ms()

void pulp::gpu_audio::GpuConvolutionReverb::set_predelay_ms ( double ms )

noexcept

set_wet_percent()

void pulp::gpu_audio::GpuConvolutionReverb::set_wet_percent ( double percent )

noexcept

set_width_percent()

void pulp::gpu_audio::GpuConvolutionReverb::set_width_percent ( double percent )

noexcept

status_snapshot()

GpuAudioStatus pulp::gpu_audio::GpuConvolutionReverb::status_snapshot ( ) const

noexcept

Allocation-free status of the least healthy lane (lane 0 on a tie), because one fenced or lost lane already changes what the route emits.

Per-lane statuses are in report().lanes[i].status. Uninitialized until prepare() succeeds. Host/UI thread; not concurrent with prepare().

Friends And Related Symbol Documentation

detail::GpuConvolutionReverbTestAccess

friend struct detail::GpuConvolutionReverbTestAccess

friend

---

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

• /home/runner/work/pulp/pulp/core/gpu_audio/include/pulp/gpu_audio/gpu_convolution_reverb.hpp