mirror of
https://github.com/RetroPortingToolKit/psxrecomp
synced 2026-10-26 02:11:36 -04:00
Matthew Stanley
3f3cfb7571
Import the supplied developer handoff against 03e635e8. Include regional capability controls, texture validation, SHA acceleration with portable fallback, SPU CD/XA correction, and overlay export/staging fixes. Validated with rebuilt EU runtime, guarded AOT audit and focused regressions.
78 lines
3.3 KiB
C
78 lines
3.3 KiB
C
/* spu_shadow.h — verified-enhancement float re-render of the SPU voice mix.
|
|
*
|
|
* A higher-fidelity reimplementation of the PS1 SPU's per-voice ADPCM playback
|
|
* that runs ALONGSIDE the faithful spu.c hardware model. Where the hardware
|
|
* resamples each voice with a 4-point Gaussian table (deliberately muffled) and
|
|
* truncates the mix to int16 at every step, the shadow keeps everything in
|
|
* float and resamples with a 4-point cubic (Catmull-Rom) kernel — no muffling,
|
|
* no intermediate requantization.
|
|
*
|
|
* It is NOT ground truth. The canon spu.c mix stays the authoritative output
|
|
* AND the verify oracle. The shadow is fed, sample-for-sample, into a
|
|
* ShadowVerifier against the canon stream; it only substitutes after a proven
|
|
* window and reverts LOUDLY (logs DEGRADED) after failed comparison windows.
|
|
* CD/XA input bypasses the voice-only shadow immediately. Opt-in via PSX_AUDIO_SHADOW=1, default OFF — with it off, spu_render
|
|
* output is byte-identical to upstream (this module is never entered).
|
|
*
|
|
* This file consumes the live per-voice state that spu.c already maintains via
|
|
* a small read-only accessor (spu_shadow_voice_snapshot); it does not duplicate
|
|
* the SPU register decode. ADSR envelope, key-on/off, and SPU RAM contents all
|
|
* come from the canon model, so the shadow can never diverge in WHICH note
|
|
* plays — only in the interpolation/headroom of HOW it is rendered.
|
|
*
|
|
* See docs/internal/PRINCIPLES.md "Verified-Enhancement HLE Is Allowed; Load-Bearing HLE Is
|
|
* Not" and docs/SHADOW_ENHANCEMENTS.md. ShadowVerifier attribution: see
|
|
* audio_shadow.h.
|
|
*/
|
|
|
|
#ifndef PSXRECOMP_SPU_SHADOW_H
|
|
#define PSXRECOMP_SPU_SHADOW_H
|
|
|
|
#include <stdbool.h>
|
|
#include <stdint.h>
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
/* Config/launcher gate. Sets the enable state the feature uses when the
|
|
* PSX_AUDIO_SHADOW env var is NOT set (env still wins for debug). Call before
|
|
* spu_init() so spu_shadow_reset() records the right state. Default OFF. */
|
|
void spu_shadow_set_enabled(int on);
|
|
|
|
/* Returns true if the shadow is enabled. Precedence: PSX_AUDIO_SHADOW env
|
|
* (if set) overrides the config value set via spu_shadow_set_enabled(). */
|
|
bool spu_shadow_enabled(void);
|
|
|
|
/* Called once per spu_init() to (re)create verifier + per-voice resampler
|
|
* state. Cheap; safe to call when disabled (no-op). */
|
|
void spu_shadow_reset(void);
|
|
|
|
/* Drive the shadow for one render block. `canon` is the int16 stereo mix that
|
|
* spu.c just produced (frames*2 samples, the authoritative output). The shadow
|
|
* re-renders the same block in float from live voice state, feeds (canon,
|
|
* shadow) to the verifier per frame, and — only while proven — overwrites
|
|
* `canon` in place with the float mix (clamped to int16). If the verifier
|
|
* reverts, it logs a DEGRADED line and leaves `canon` untouched.
|
|
*
|
|
* No-op (canon untouched, byte-identical) when the feature is disabled. */
|
|
void spu_shadow_process(int16_t* canon, int frames);
|
|
|
|
/* Diagnostics for the debug server / tests. */
|
|
typedef struct {
|
|
int enabled;
|
|
int proven;
|
|
uint64_t pauses;
|
|
float last_r; /* last window correlation */
|
|
float last_ratio; /* last window level ratio */
|
|
float gain; /* calibrated shadow gain */
|
|
char last_revert[160];
|
|
} SpuShadowInfo;
|
|
void spu_shadow_get_info(SpuShadowInfo* out);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|
|
|
|
#endif /* PSXRECOMP_SPU_SHADOW_H */
|