crossfade.hpp File Reference

The one shared old→new crossfade gain law for the SIGNAL side. More...

#include <cmath>

Go to the source code of this file.

Namespaces
namespace	pulp
namespace	pulp::signal

Enumerations
enum class	pulp::signal::CrossfadeGainLaw { pulp::signal::EqualGain , pulp::signal::EqualPower }
	The old→new gain split applied to a shaped progress u ∈ [0,1]. More...

Functions
template<typename T>
constexpr T	pulp::signal::crossfade_smoothstep (T t) noexcept
	Smoothstep progress shaping: cubic 3u²−2u³ with zero slope at both ends, clamped to [0,1].
template<typename T>
void	pulp::signal::crossfade_gains (T u, CrossfadeGainLaw law, T &old_gain, T &new_gain) noexcept
	Old/new blend gains for an ALREADY-SHAPED progress u (the caller applies any smoothstep/clamp first).

Detailed Description

The one shared old→new crossfade gain law for the SIGNAL side.

Every click-free/gap-free transition that blends an OLD render into a NEW one — the live plugin swap (TransitionMixer / crossfade_plugin_slot), the convolver IR swapper (PartitionedConvolver), the loop-wrap crossfade (LoopRenderer), and the live-kernel structural swap — computes its per-frame gain pair HERE instead of hand-rolling cos/sin. Pure math, no state, no allocation, no locks: safe to call per sample on the audio thread.

A crossfade is fully described by two orthogonal choices:

1. The PROGRESS SHAPING — how the raw fade fraction t = pos/len maps to a shaped progress u ∈ [0,1]:
   • identity (u = t) — a plain ramp; the loop-wrap law.
   • crossfade_smoothstep(t) — cubic, zero slope at both ends; the live-swap law (click-free at BOTH the swap instant and the fade end).
2. The GAIN LAW — how the shaped progress u splits into (old, new) gains:
   • EqualGain → (1 − u, u): amplitude sum == 1. No level bump for CORRELATED old/new (the usual hot-reload: same input, a small DSP tweak); a −6 dB mid dip when decorrelated.
   • EqualPower → (cos(u·π/2), sin(u·π/2)): power sum == 1. Avoids the mid-fade dip when old/new are DECORRELATED (a big DSP change); the −3 dB "constant power" cross-fade.

Callers compose the two: crossfade_gains(shaping(t), law). Keeping the shaping OUT of crossfade_gains is deliberate — it lets the loop-wrap path use the raw ramp and the live-swap path use smoothstep while sharing the one gain law, so all of them stay identical bit-for-bit to their prior hand-rolled math (the loop wrap and live swap only differed in shaping + a duplicated cos/sin, never in the underlying law).