mirror of
https://github.com/libretro/Mesen.git
synced 2024-11-23 09:09:45 +00:00
122 lines
3.0 KiB
C++
122 lines
3.0 KiB
C++
#pragma once
|
|
#include "stdafx.h"
|
|
#include "EmulationSettings.h"
|
|
#include "../Utilities/LowPassFilter.h"
|
|
#include "../Utilities/blip_buf.h"
|
|
#include "../Libretro/libretro.h"
|
|
#include "Snapshotable.h"
|
|
#include "StereoPanningFilter.h"
|
|
#include "StereoDelayFilter.h"
|
|
#include "StereoCombFilter.h"
|
|
#include "ReverbFilter.h"
|
|
#include "CrossFeedFilter.h"
|
|
|
|
class Console;
|
|
class OggMixer;
|
|
|
|
namespace orfanidis_eq {
|
|
class freq_grid;
|
|
class eq1;
|
|
}
|
|
|
|
class SoundMixer : public Snapshotable
|
|
{
|
|
public:
|
|
static constexpr uint32_t CycleLength = 10000;
|
|
static constexpr uint32_t BitsPerSample = 16;
|
|
|
|
private:
|
|
static constexpr uint32_t MaxSampleRate = 96000;
|
|
static constexpr uint32_t MaxSamplesPerFrame = MaxSampleRate / 60 * 4 * 2; //x4 to allow CPU overclocking up to 10x, x2 for panning stereo
|
|
static constexpr uint32_t MaxChannelCount = 11;
|
|
|
|
retro_audio_sample_batch_t _sendAudioSample = nullptr;
|
|
vector<int16_t> _audioSampleBuffer;
|
|
size_t _audioSampleBufferPos = 0;
|
|
|
|
bool _skipMode = false;
|
|
EmulationSettings* _settings;
|
|
double _fadeRatio;
|
|
uint32_t _muteFrameCount;
|
|
unique_ptr<OggMixer> _oggMixer;
|
|
|
|
unique_ptr<orfanidis_eq::freq_grid> _eqFrequencyGrid;
|
|
unique_ptr<orfanidis_eq::eq1> _equalizerLeft;
|
|
unique_ptr<orfanidis_eq::eq1> _equalizerRight;
|
|
shared_ptr<Console> _console;
|
|
|
|
CrossFeedFilter _crossFeedFilter;
|
|
LowPassFilter _lowPassFilter;
|
|
StereoPanningFilter _stereoPanning;
|
|
StereoDelayFilter _stereoDelay;
|
|
StereoCombFilter _stereoCombFilter;
|
|
ReverbFilter _reverbFilter;
|
|
|
|
int16_t _previousOutputLeft = 0;
|
|
int16_t _previousOutputRight = 0;
|
|
|
|
vector<uint32_t> _timestamps;
|
|
int16_t _channelOutput[MaxChannelCount][CycleLength];
|
|
int16_t _currentOutput[MaxChannelCount];
|
|
|
|
blip_t* _blipBufLeft;
|
|
blip_t* _blipBufRight;
|
|
int16_t *_outputBuffer;
|
|
double _volumes[MaxChannelCount];
|
|
double _panning[MaxChannelCount];
|
|
|
|
NesModel _model;
|
|
uint32_t _sampleRate;
|
|
uint32_t _clockRate;
|
|
|
|
bool _hasPanning;
|
|
|
|
double _previousTargetRate;
|
|
|
|
__forceinline double GetChannelOutput(AudioChannel channel, bool forRightChannel);
|
|
__forceinline int16_t GetOutputVolume(bool forRightChannel);
|
|
void EndFrame(uint32_t time);
|
|
|
|
void UpdateRates(bool forceUpdate);
|
|
|
|
void UpdateEqualizers(bool forceUpdate);
|
|
void ApplyEqualizer(orfanidis_eq::eq1* equalizer, size_t sampleCount);
|
|
|
|
void UpdateTargetSampleRate();
|
|
|
|
protected:
|
|
virtual void StreamState(bool saving) override;
|
|
|
|
public:
|
|
SoundMixer(shared_ptr<Console> console);
|
|
~SoundMixer();
|
|
|
|
void SetNesModel(NesModel model);
|
|
void Reset();
|
|
|
|
void PlayAudioBuffer(uint32_t cycle);
|
|
void UploadAudioSamples();
|
|
void AddDelta(AudioChannel channel, uint32_t time, int16_t delta);
|
|
|
|
void StartRecording(string filepath);
|
|
void StopRecording();
|
|
bool IsRecording();
|
|
|
|
//For NSF/NSFe
|
|
uint32_t GetMuteFrameCount();
|
|
void ResetMuteFrameCount();
|
|
void SetFadeRatio(double fadeRatio);
|
|
|
|
OggMixer* GetOggMixer();
|
|
|
|
void SetSendAudioSample(retro_audio_sample_batch_t sendAudioSample)
|
|
{
|
|
_sendAudioSample = sendAudioSample;
|
|
}
|
|
|
|
void SetSkipMode(bool skip)
|
|
{
|
|
_skipMode = skip;
|
|
}
|
|
};
|