pulp::gpu_audio::GpuStft Class Reference

GPU short-time Fourier transform primitive (spectral toolkit). More...

#include <gpu_stft.hpp>

Public Member Functions
bool	prepare (uint32_t fft_size, signal::WindowFunction::Type window=signal::WindowFunction::Type::hann, float window_param=0.0f, render::GpuCompute *shared_device=nullptr)
	shared_device (optional) reuses an existing GpuCompute so this STFT and a sibling primitive don't each spin up a device (the prerequisite for cross-effect batching); it must already be initialized and must outlive this GpuStft.
bool	gpu_available () const
bool	device_usable () const
	True while the prepared device can still run this STFT's work: present, not lost, and on the generation recorded at prepare().
uint32_t	fft_size () const
const std::vector< float > &	window () const
render::GpuCompute *	compute ()
	The underlying GPU compute device (or null if none initialized).
bool	analyze (const float *frame_in, float *spectrum_out)
	frame_in: fft_size real samples.
bool	synthesize (const float *spectrum_in, float *frame_out)
	spectrum_in: 2*fft_size interleaved complex.

Detailed Description

GPU short-time Fourier transform primitive (spectral toolkit).

analyze(): apply the analysis window to one frame, then a GPU forward FFT → an interleaved-complex spectral frame. synthesize(): GPU inverse FFT → the real time-domain frame (no synthesis window; the caller overlap-adds). With a COLA window/hop (e.g. Hann at 50% overlap), analyze→synthesize→overlap-add reconstructs the input. fft_size must be a power of two. GPU work runs via render::GpuCompute; not real-time-safe (blocks on readback) — for the worker / offline analysis, the foundation for spectral freeze / morph / paint.

Member Function Documentation

analyze()

bool pulp::gpu_audio::GpuStft::analyze	(	const float *	frame_in,
		float *	spectrum_out )

frame_in: fft_size real samples.

spectrum_out: 2*fft_size interleaved complex (windowed forward FFT). Returns false if not GPU-ready.

compute()

render::GpuCompute * pulp::gpu_audio::GpuStft::compute ( )

inline

The underlying GPU compute device (or null if none initialized).

Exposed so a sibling primitive (e.g. the multi-layer spectral stack) can run on the SAME device and keep spectral frames GPU-resident without a second device. Worker / offline use only — not real-time-safe.

device_usable()

bool pulp::gpu_audio::GpuStft::device_usable

(

)

const

True while the prepared device can still run this STFT's work: present, not lost, and on the generation recorded at prepare().

False is the refusal state analyze()/synthesize() report.

fft_size()

uint32_t pulp::gpu_audio::GpuStft::fft_size ( ) const

inline

gpu_available()

bool pulp::gpu_audio::GpuStft::gpu_available ( ) const

inline

prepare()

bool pulp::gpu_audio::GpuStft::prepare	(	uint32_t	fft_size,
		signal::WindowFunction::Type	window = signal::WindowFunction::Type::hann,
		float	window_param = 0.0f,
		render::GpuCompute *	shared_device = nullptr )

shared_device (optional) reuses an existing GpuCompute so this STFT and a sibling primitive don't each spin up a device (the prerequisite for cross-effect batching); it must already be initialized and must outlive this GpuStft.

null = create + own a private device.

prepare() records the device's device_generation(). From then on analyze() and synthesize() refuse (return false, output zeroed) once the device reports device_lost() or its generation differs from the one recorded, because work prepared on the old device cannot run on a new one. Prepare again against the re-acquired device to resume.

synthesize()

bool pulp::gpu_audio::GpuStft::synthesize	(	const float *	spectrum_in,
		float *	frame_out )

spectrum_in: 2*fft_size interleaved complex.

frame_out: fft_size real samples (inverse FFT, no synthesis window). Returns false if not ready.

window()

const std::vector< float > & pulp::gpu_audio::GpuStft::window ( ) const

inline

---

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

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