mirror of
https://github.com/libretro/libretro-o2em.git
synced 2024-11-30 03:40:24 +00:00
1295 lines
35 KiB
C
1295 lines
35 KiB
C
/* Copyright (C) 2010-2020 The RetroArch team
|
|
*
|
|
* ---------------------------------------------------------------------------------------
|
|
* The following license statement only applies to this file (core_audio_mixer.c).
|
|
* ---------------------------------------------------------------------------------------
|
|
*
|
|
* Permission is hereby granted, free of charge,
|
|
* to any person obtaining a copy of this software and associated documentation files (the "Software"),
|
|
* to deal in the Software without restriction, including without limitation the rights to
|
|
* use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software,
|
|
* and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
|
|
* INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
|
|
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
|
|
* WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
|
*/
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
#include "../../config.h"
|
|
#endif
|
|
|
|
#include <core_audio_mixer.h>
|
|
#include <audio/audio_resampler.h>
|
|
|
|
#ifdef HAVE_RWAV
|
|
#include <formats/rwav.h>
|
|
#endif
|
|
#include <memalign.h>
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <math.h>
|
|
|
|
#ifdef HAVE_STB_VORBIS
|
|
#define STB_VORBIS_NO_PUSHDATA_API
|
|
#define STB_VORBIS_NO_STDIO
|
|
#define STB_VORBIS_NO_CRT
|
|
|
|
#include <stb/stb_vorbis.h>
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_FLAC
|
|
#define DR_FLAC_IMPLEMENTATION
|
|
#include <dr/dr_flac.h>
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_MP3
|
|
#define DR_MP3_IMPLEMENTATION
|
|
#include <retro_assert.h>
|
|
#define DRMP3_ASSERT(expression) retro_assert(expression)
|
|
#include <dr/dr_mp3.h>
|
|
#endif
|
|
|
|
#ifdef HAVE_IBXM
|
|
#include <ibxm/ibxm.h>
|
|
#endif
|
|
|
|
#define CORE_AUDIO_MIXER_MAX_VOICES 1
|
|
#define CORE_AUDIO_MIXER_TEMP_BUFFER 8192
|
|
|
|
struct core_audio_mixer_sound
|
|
{
|
|
enum core_audio_mixer_type type;
|
|
|
|
union
|
|
{
|
|
struct
|
|
{
|
|
/* wav */
|
|
const float* pcm;
|
|
unsigned frames;
|
|
} wav;
|
|
|
|
#ifdef HAVE_STB_VORBIS
|
|
struct
|
|
{
|
|
/* ogg */
|
|
const void* data;
|
|
unsigned size;
|
|
} ogg;
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_FLAC
|
|
struct
|
|
{
|
|
/* flac */
|
|
const void* data;
|
|
unsigned size;
|
|
} flac;
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_MP3
|
|
struct
|
|
{
|
|
/* mp */
|
|
const void* data;
|
|
unsigned size;
|
|
} mp3;
|
|
#endif
|
|
|
|
#ifdef HAVE_IBXM
|
|
struct
|
|
{
|
|
/* mod/s3m/xm */
|
|
const void* data;
|
|
unsigned size;
|
|
} mod;
|
|
#endif
|
|
} types;
|
|
};
|
|
|
|
struct core_audio_mixer_voice
|
|
{
|
|
union
|
|
{
|
|
struct
|
|
{
|
|
unsigned position;
|
|
} wav;
|
|
|
|
#ifdef HAVE_STB_VORBIS
|
|
struct
|
|
{
|
|
stb_vorbis *stream;
|
|
void *resampler_data;
|
|
const retro_resampler_t *resampler;
|
|
float *buffer;
|
|
unsigned position;
|
|
unsigned samples;
|
|
unsigned buf_samples;
|
|
float ratio;
|
|
} ogg;
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_FLAC
|
|
struct
|
|
{
|
|
float* buffer;
|
|
drflac *stream;
|
|
void *resampler_data;
|
|
const retro_resampler_t *resampler;
|
|
unsigned position;
|
|
unsigned samples;
|
|
unsigned buf_samples;
|
|
float ratio;
|
|
} flac;
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_MP3
|
|
struct
|
|
{
|
|
drmp3 stream;
|
|
void *resampler_data;
|
|
const retro_resampler_t *resampler;
|
|
float* buffer;
|
|
unsigned position;
|
|
unsigned samples;
|
|
unsigned buf_samples;
|
|
float ratio;
|
|
} mp3;
|
|
#endif
|
|
|
|
#ifdef HAVE_IBXM
|
|
struct
|
|
{
|
|
int* buffer;
|
|
struct replay* stream;
|
|
struct module* module;
|
|
unsigned position;
|
|
unsigned samples;
|
|
unsigned buf_samples;
|
|
} mod;
|
|
#endif
|
|
} types;
|
|
core_audio_mixer_sound_t *sound;
|
|
core_audio_mixer_stop_cb_t stop_cb;
|
|
unsigned type;
|
|
float volume;
|
|
bool repeat;
|
|
|
|
};
|
|
|
|
/* TODO/FIXME - static globals */
|
|
static struct core_audio_mixer_voice core_s_voices[CORE_AUDIO_MIXER_MAX_VOICES] = {0};
|
|
static unsigned core_s_rate = 0;
|
|
|
|
#ifdef HAVE_RWAV
|
|
static bool wav_to_float(const rwav_t* wav, float** pcm, size_t samples_out)
|
|
{
|
|
size_t i;
|
|
/* Allocate on a 16-byte boundary, and pad to a multiple of 16 bytes */
|
|
float *f = (float*)memalign_alloc(16,
|
|
((samples_out + 15) & ~15) * sizeof(float));
|
|
|
|
if (!f)
|
|
return false;
|
|
|
|
*pcm = f;
|
|
|
|
if (wav->bitspersample == 8)
|
|
{
|
|
float sample = 0.0f;
|
|
const uint8_t *u8 = (const uint8_t*)wav->samples;
|
|
|
|
if (wav->numchannels == 1)
|
|
{
|
|
for (i = wav->numsamples; i != 0; i--)
|
|
{
|
|
sample = (float)*u8++ / 255.0f;
|
|
sample = sample * 2.0f - 1.0f;
|
|
*f++ = sample;
|
|
*f++ = sample;
|
|
}
|
|
}
|
|
else if (wav->numchannels == 2)
|
|
{
|
|
for (i = wav->numsamples; i != 0; i--)
|
|
{
|
|
sample = (float)*u8++ / 255.0f;
|
|
sample = sample * 2.0f - 1.0f;
|
|
*f++ = sample;
|
|
sample = (float)*u8++ / 255.0f;
|
|
sample = sample * 2.0f - 1.0f;
|
|
*f++ = sample;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
/* TODO/FIXME note to leiradel - can we use audio/conversion/s16_to_float
|
|
* functions here? */
|
|
|
|
float sample = 0.0f;
|
|
const int16_t *s16 = (const int16_t*)wav->samples;
|
|
|
|
if (wav->numchannels == 1)
|
|
{
|
|
for (i = wav->numsamples; i != 0; i--)
|
|
{
|
|
sample = (float)((int)*s16++ + 32768) / 65535.0f;
|
|
sample = sample * 2.0f - 1.0f;
|
|
*f++ = sample;
|
|
*f++ = sample;
|
|
}
|
|
}
|
|
else if (wav->numchannels == 2)
|
|
{
|
|
for (i = wav->numsamples; i != 0; i--)
|
|
{
|
|
sample = (float)((int)*s16++ + 32768) / 65535.0f;
|
|
sample = sample * 2.0f - 1.0f;
|
|
*f++ = sample;
|
|
sample = (float)((int)*s16++ + 32768) / 65535.0f;
|
|
sample = sample * 2.0f - 1.0f;
|
|
*f++ = sample;
|
|
}
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool one_shot_resample(const float* in, size_t samples_in,
|
|
unsigned rate, const char *resampler_ident, enum resampler_quality quality,
|
|
float** out, size_t* samples_out)
|
|
{
|
|
struct resampler_data info;
|
|
void* data = NULL;
|
|
const retro_resampler_t* resampler = NULL;
|
|
float ratio = (double)core_s_rate / (double)rate;
|
|
|
|
if (!retro_resampler_realloc(&data, &resampler,
|
|
resampler_ident, quality, ratio))
|
|
return false;
|
|
|
|
/* Allocate on a 16-byte boundary, and pad to a multiple of 16 bytes. We
|
|
* add 16 more samples in the formula below just as safeguard, because
|
|
* resampler->process sometimes reports more output samples than the
|
|
* formula below calculates. Ideally, audio resamplers should have a
|
|
* function to return the number of samples they will output given a
|
|
* count of input samples. */
|
|
*samples_out = (size_t)(samples_in * ratio);
|
|
*out = (float*)memalign_alloc(16,
|
|
(((*samples_out + 16) + 15) & ~15) * sizeof(float));
|
|
|
|
if (*out == NULL)
|
|
return false;
|
|
|
|
info.data_in = in;
|
|
info.data_out = *out;
|
|
info.input_frames = samples_in / 2;
|
|
info.output_frames = 0;
|
|
info.ratio = ratio;
|
|
|
|
resampler->process(data, &info);
|
|
resampler->free(data);
|
|
return true;
|
|
}
|
|
#endif
|
|
|
|
void core_audio_mixer_init(unsigned rate)
|
|
{
|
|
unsigned i;
|
|
|
|
core_s_rate = rate;
|
|
|
|
for (i = 0; i < CORE_AUDIO_MIXER_MAX_VOICES; i++)
|
|
core_s_voices[i].type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
}
|
|
|
|
void core_audio_mixer_done(void)
|
|
{
|
|
unsigned i;
|
|
|
|
for (i = 0; i < CORE_AUDIO_MIXER_MAX_VOICES; i++)
|
|
core_s_voices[i].type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
}
|
|
|
|
core_audio_mixer_sound_t* core_audio_mixer_load_wav(void *buffer, int32_t size,
|
|
const char *resampler_ident, enum resampler_quality quality)
|
|
{
|
|
#ifdef HAVE_RWAV
|
|
/* WAV data */
|
|
rwav_t wav;
|
|
/* WAV samples converted to float */
|
|
float* pcm = NULL;
|
|
size_t samples = 0;
|
|
/* Result */
|
|
core_audio_mixer_sound_t* sound = NULL;
|
|
|
|
wav.bitspersample = 0;
|
|
wav.numchannels = 0;
|
|
wav.samplerate = 0;
|
|
wav.numsamples = 0;
|
|
wav.subchunk2size = 0;
|
|
wav.samples = NULL;
|
|
|
|
if ((rwav_load(&wav, buffer, size)) != RWAV_ITERATE_DONE)
|
|
return NULL;
|
|
|
|
samples = wav.numsamples * 2;
|
|
|
|
if (!wav_to_float(&wav, &pcm, samples))
|
|
return NULL;
|
|
|
|
if (wav.samplerate != core_s_rate)
|
|
{
|
|
float* resampled = NULL;
|
|
|
|
if (!one_shot_resample(pcm, samples, wav.samplerate,
|
|
resampler_ident, quality,
|
|
&resampled, &samples))
|
|
return NULL;
|
|
|
|
memalign_free((void*)pcm);
|
|
pcm = resampled;
|
|
}
|
|
|
|
sound = (core_audio_mixer_sound_t*)calloc(1, sizeof(*sound));
|
|
|
|
if (!sound)
|
|
{
|
|
memalign_free((void*)pcm);
|
|
return NULL;
|
|
}
|
|
|
|
sound->type = CORE_AUDIO_MIXER_TYPE_WAV;
|
|
sound->types.wav.frames = (unsigned)(samples / 2);
|
|
sound->types.wav.pcm = pcm;
|
|
|
|
rwav_free(&wav);
|
|
|
|
return sound;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
core_audio_mixer_sound_t* core_audio_mixer_load_ogg(void *buffer, int32_t size)
|
|
{
|
|
#ifdef HAVE_STB_VORBIS
|
|
core_audio_mixer_sound_t* sound = (core_audio_mixer_sound_t*)calloc(1, sizeof(*sound));
|
|
|
|
if (!sound)
|
|
return NULL;
|
|
|
|
sound->type = CORE_AUDIO_MIXER_TYPE_OGG;
|
|
sound->types.ogg.size = size;
|
|
sound->types.ogg.data = buffer;
|
|
|
|
return sound;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
core_audio_mixer_sound_t* core_audio_mixer_load_flac(void *buffer, int32_t size)
|
|
{
|
|
#ifdef HAVE_DR_FLAC
|
|
core_audio_mixer_sound_t* sound = (core_audio_mixer_sound_t*)calloc(1, sizeof(*sound));
|
|
|
|
if (!sound)
|
|
return NULL;
|
|
|
|
sound->type = CORE_AUDIO_MIXER_TYPE_FLAC;
|
|
sound->types.flac.size = size;
|
|
sound->types.flac.data = buffer;
|
|
|
|
return sound;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
core_audio_mixer_sound_t* core_audio_mixer_load_mp3(void *buffer, int32_t size)
|
|
{
|
|
#ifdef HAVE_DR_MP3
|
|
core_audio_mixer_sound_t* sound = (core_audio_mixer_sound_t*)calloc(1, sizeof(*sound));
|
|
|
|
if (!sound)
|
|
return NULL;
|
|
|
|
sound->type = CORE_AUDIO_MIXER_TYPE_MP3;
|
|
sound->types.mp3.size = size;
|
|
sound->types.mp3.data = buffer;
|
|
|
|
return sound;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
core_audio_mixer_sound_t* core_audio_mixer_load_mod(void *buffer, int32_t size)
|
|
{
|
|
#ifdef HAVE_IBXM
|
|
core_audio_mixer_sound_t* sound = (core_audio_mixer_sound_t*)calloc(1, sizeof(*sound));
|
|
|
|
if (!sound)
|
|
return NULL;
|
|
|
|
sound->type = CORE_AUDIO_MIXER_TYPE_MOD;
|
|
sound->types.mod.size = size;
|
|
sound->types.mod.data = buffer;
|
|
|
|
return sound;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
void core_audio_mixer_destroy(core_audio_mixer_sound_t* sound)
|
|
{
|
|
void *handle = NULL;
|
|
if (!sound)
|
|
return;
|
|
|
|
switch (sound->type)
|
|
{
|
|
case CORE_AUDIO_MIXER_TYPE_WAV:
|
|
handle = (void*)sound->types.wav.pcm;
|
|
if (handle)
|
|
memalign_free(handle);
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_OGG:
|
|
#ifdef HAVE_STB_VORBIS
|
|
handle = (void*)sound->types.ogg.data;
|
|
if (handle)
|
|
free(handle);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_MOD:
|
|
#ifdef HAVE_IBXM
|
|
handle = (void*)sound->types.mod.data;
|
|
if (handle)
|
|
free(handle);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_FLAC:
|
|
#ifdef HAVE_DR_FLAC
|
|
handle = (void*)sound->types.flac.data;
|
|
if (handle)
|
|
free(handle);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_MP3:
|
|
#ifdef HAVE_DR_MP3
|
|
handle = (void*)sound->types.mp3.data;
|
|
if (handle)
|
|
free(handle);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_NONE:
|
|
break;
|
|
}
|
|
|
|
free(sound);
|
|
}
|
|
|
|
static bool core_audio_mixer_play_wav(core_audio_mixer_sound_t* sound,
|
|
core_audio_mixer_voice_t* voice, bool repeat, float volume,
|
|
core_audio_mixer_stop_cb_t stop_cb)
|
|
{
|
|
voice->types.wav.position = 0;
|
|
return true;
|
|
}
|
|
|
|
#ifdef HAVE_STB_VORBIS
|
|
static bool core_audio_mixer_play_ogg(
|
|
core_audio_mixer_sound_t* sound,
|
|
core_audio_mixer_voice_t* voice,
|
|
bool repeat, float volume,
|
|
const char *resampler_ident,
|
|
enum resampler_quality quality,
|
|
core_audio_mixer_stop_cb_t stop_cb)
|
|
{
|
|
stb_vorbis_info info;
|
|
int res = 0;
|
|
float ratio = 1.0f;
|
|
unsigned samples = 0;
|
|
void *ogg_buffer = NULL;
|
|
void *resampler_data = NULL;
|
|
const retro_resampler_t* resamp = NULL;
|
|
stb_vorbis *stb_vorbis = stb_vorbis_open_memory(
|
|
(const unsigned char*)sound->types.ogg.data,
|
|
sound->types.ogg.size, &res, NULL);
|
|
|
|
if (!stb_vorbis)
|
|
return false;
|
|
|
|
info = stb_vorbis_get_info(stb_vorbis);
|
|
|
|
if (info.sample_rate != core_s_rate)
|
|
{
|
|
ratio = (double)core_s_rate / (double)info.sample_rate;
|
|
|
|
if (!retro_resampler_realloc(&resampler_data,
|
|
&resamp, resampler_ident, quality,
|
|
ratio))
|
|
goto error;
|
|
}
|
|
|
|
/* Allocate on a 16-byte boundary, and pad to a multiple of 16 bytes. We
|
|
* add 16 more samples in the formula below just as safeguard, because
|
|
* resampler->process sometimes reports more output samples than the
|
|
* formula below calculates. Ideally, audio resamplers should have a
|
|
* function to return the number of samples they will output given a
|
|
* count of input samples. */
|
|
samples = (unsigned)(CORE_AUDIO_MIXER_TEMP_BUFFER * ratio);
|
|
ogg_buffer = (float*)memalign_alloc(16,
|
|
(((samples + 16) + 15) & ~15) * sizeof(float));
|
|
|
|
if (!ogg_buffer)
|
|
{
|
|
if (resamp && resampler_data)
|
|
resamp->free(resampler_data);
|
|
goto error;
|
|
}
|
|
|
|
/* "system" menu sounds may reuse the same voice without freeing anything first, so do that here if needed */
|
|
if (voice->types.ogg.stream)
|
|
stb_vorbis_close(voice->types.ogg.stream);
|
|
if (voice->types.ogg.resampler && voice->types.ogg.resampler_data)
|
|
voice->types.ogg.resampler->free(voice->types.ogg.resampler_data);
|
|
if (voice->types.ogg.buffer)
|
|
memalign_free(voice->types.ogg.buffer);
|
|
|
|
voice->types.ogg.resampler = resamp;
|
|
voice->types.ogg.resampler_data = resampler_data;
|
|
voice->types.ogg.buffer = (float*)ogg_buffer;
|
|
voice->types.ogg.buf_samples = samples;
|
|
voice->types.ogg.ratio = ratio;
|
|
voice->types.ogg.stream = stb_vorbis;
|
|
voice->types.ogg.position = 0;
|
|
voice->types.ogg.samples = 0;
|
|
|
|
return true;
|
|
|
|
error:
|
|
stb_vorbis_close(stb_vorbis);
|
|
return false;
|
|
}
|
|
#endif
|
|
|
|
#ifdef HAVE_IBXM
|
|
static bool core_audio_mixer_play_mod(
|
|
core_audio_mixer_sound_t* sound,
|
|
core_audio_mixer_voice_t* voice,
|
|
bool repeat, float volume,
|
|
core_audio_mixer_stop_cb_t stop_cb)
|
|
{
|
|
struct data data;
|
|
char message[64];
|
|
int buf_samples = 0;
|
|
int samples = 0;
|
|
void *mod_buffer = NULL;
|
|
struct module* module = NULL;
|
|
struct replay* replay = NULL;
|
|
|
|
data.buffer = (char*)sound->types.mod.data;
|
|
data.length = sound->types.mod.size;
|
|
module = module_load(&data, message);
|
|
|
|
if (!module)
|
|
{
|
|
printf("core_audio_mixer_play_mod module_load() failed with error: %s\n", message);
|
|
goto error;
|
|
}
|
|
|
|
if (voice->types.mod.module)
|
|
dispose_module(voice->types.mod.module);
|
|
|
|
voice->types.mod.module = module;
|
|
|
|
replay = new_replay(module, core_s_rate, 1);
|
|
|
|
if (!replay)
|
|
{
|
|
printf("core_audio_mixer_play_mod new_replay() failed\n");
|
|
goto error;
|
|
}
|
|
|
|
buf_samples = calculate_mix_buf_len(core_s_rate);
|
|
mod_buffer = memalign_alloc(16, ((buf_samples + 15) & ~15) * sizeof(int));
|
|
|
|
if (!mod_buffer)
|
|
{
|
|
printf("core_audio_mixer_play_mod cannot allocate mod_buffer !\n");
|
|
goto error;
|
|
}
|
|
|
|
samples = replay_calculate_duration(replay);
|
|
|
|
if (!samples)
|
|
{
|
|
printf("core_audio_mixer_play_mod cannot retrieve duration !\n");
|
|
goto error;
|
|
}
|
|
|
|
/* FIXME: stopping and then starting a mod stream will crash here in dispose_replay (ASAN says struct replay is misaligned?) */
|
|
if (voice->types.mod.stream)
|
|
dispose_replay(voice->types.mod.stream);
|
|
if (voice->types.mod.buffer)
|
|
memalign_free(voice->types.mod.buffer);
|
|
|
|
voice->types.mod.buffer = (int*)mod_buffer;
|
|
voice->types.mod.buf_samples = buf_samples;
|
|
voice->types.mod.stream = replay;
|
|
voice->types.mod.position = 0;
|
|
voice->types.mod.samples = 0; /* samples; */
|
|
|
|
return true;
|
|
|
|
error:
|
|
if (mod_buffer)
|
|
memalign_free(mod_buffer);
|
|
if (module)
|
|
dispose_module(module);
|
|
return false;
|
|
|
|
}
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_FLAC
|
|
static bool core_audio_mixer_play_flac(
|
|
core_audio_mixer_sound_t* sound,
|
|
core_audio_mixer_voice_t* voice,
|
|
bool repeat, float volume,
|
|
const char *resampler_ident,
|
|
enum resampler_quality quality,
|
|
core_audio_mixer_stop_cb_t stop_cb)
|
|
{
|
|
float ratio = 1.0f;
|
|
unsigned samples = 0;
|
|
void *flac_buffer = NULL;
|
|
void *resampler_data = NULL;
|
|
const retro_resampler_t* resamp = NULL;
|
|
drflac *dr_flac = drflac_open_memory((const unsigned char*)sound->types.flac.data,sound->types.flac.size);
|
|
|
|
if (!dr_flac)
|
|
return false;
|
|
if (dr_flac->sampleRate != core_s_rate)
|
|
{
|
|
ratio = (double)core_s_rate / (double)(dr_flac->sampleRate);
|
|
|
|
if (!retro_resampler_realloc(&resampler_data,
|
|
&resamp, resampler_ident, quality,
|
|
ratio))
|
|
goto error;
|
|
}
|
|
|
|
/* Allocate on a 16-byte boundary, and pad to a multiple of 16 bytes. We
|
|
* add 16 more samples in the formula below just as safeguard, because
|
|
* resampler->process sometimes reports more output samples than the
|
|
* formula below calculates. Ideally, audio resamplers should have a
|
|
* function to return the number of samples they will output given a
|
|
* count of input samples. */
|
|
samples = (unsigned)(CORE_AUDIO_MIXER_TEMP_BUFFER * ratio);
|
|
flac_buffer = (float*)memalign_alloc(16,
|
|
(((samples + 16) + 15) & ~15) * sizeof(float));
|
|
|
|
if (!flac_buffer)
|
|
{
|
|
if (resamp && resamp->free)
|
|
resamp->free(resampler_data);
|
|
goto error;
|
|
}
|
|
|
|
if (voice->types.flac.stream)
|
|
drflac_close(voice->types.flac.stream);
|
|
if (voice->types.flac.resampler && voice->types.flac.resampler_data)
|
|
voice->types.flac.resampler->free(voice->types.flac.resampler_data);
|
|
if (voice->types.flac.buffer)
|
|
memalign_free(voice->types.flac.buffer);
|
|
|
|
voice->types.flac.resampler = resamp;
|
|
voice->types.flac.resampler_data = resampler_data;
|
|
voice->types.flac.buffer = (float*)flac_buffer;
|
|
voice->types.flac.buf_samples = samples;
|
|
voice->types.flac.ratio = ratio;
|
|
voice->types.flac.stream = dr_flac;
|
|
voice->types.flac.position = 0;
|
|
voice->types.flac.samples = 0;
|
|
|
|
return true;
|
|
|
|
error:
|
|
drflac_close(dr_flac);
|
|
return false;
|
|
}
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_MP3
|
|
static bool core_audio_mixer_play_mp3(
|
|
core_audio_mixer_sound_t* sound,
|
|
core_audio_mixer_voice_t* voice,
|
|
bool repeat, float volume,
|
|
const char *resampler_ident,
|
|
enum resampler_quality quality,
|
|
core_audio_mixer_stop_cb_t stop_cb)
|
|
{
|
|
float ratio = 1.0f;
|
|
unsigned samples = 0;
|
|
void *mp3_buffer = NULL;
|
|
void *resampler_data = NULL;
|
|
const retro_resampler_t* resamp = NULL;
|
|
bool res;
|
|
|
|
if (voice->types.mp3.stream.pData)
|
|
{
|
|
drmp3_uninit(&voice->types.mp3.stream);
|
|
memset(&voice->types.mp3.stream, 0, sizeof(voice->types.mp3.stream));
|
|
}
|
|
|
|
res = drmp3_init_memory(&voice->types.mp3.stream, (const unsigned char*)sound->types.mp3.data, sound->types.mp3.size, NULL);
|
|
|
|
if (!res)
|
|
return false;
|
|
|
|
if (voice->types.mp3.stream.sampleRate != core_s_rate)
|
|
{
|
|
ratio = (double)core_s_rate / (double)(voice->types.mp3.stream.sampleRate);
|
|
|
|
if (!retro_resampler_realloc(&resampler_data,
|
|
&resamp, resampler_ident, quality,
|
|
ratio))
|
|
goto error;
|
|
}
|
|
|
|
/* Allocate on a 16-byte boundary, and pad to a multiple of 16 bytes. We
|
|
* add 16 more samples in the formula below just as safeguard, because
|
|
* resampler->process sometimes reports more output samples than the
|
|
* formula below calculates. Ideally, audio resamplers should have a
|
|
* function to return the number of samples they will output given a
|
|
* count of input samples. */
|
|
samples = (unsigned)(CORE_AUDIO_MIXER_TEMP_BUFFER * ratio);
|
|
mp3_buffer = (float*)memalign_alloc(16,
|
|
(((samples + 16) + 15) & ~15) * sizeof(float));
|
|
|
|
if (!mp3_buffer)
|
|
{
|
|
if (resamp && resampler_data)
|
|
resamp->free(resampler_data);
|
|
goto error;
|
|
}
|
|
|
|
/* "system" menu sounds may reuse the same voice without freeing anything first, so do that here if needed */
|
|
if (voice->types.mp3.resampler && voice->types.mp3.resampler_data)
|
|
voice->types.mp3.resampler->free(voice->types.mp3.resampler_data);
|
|
if (voice->types.mp3.buffer)
|
|
memalign_free(voice->types.mp3.buffer);
|
|
|
|
voice->types.mp3.resampler = resamp;
|
|
voice->types.mp3.resampler_data = resampler_data;
|
|
voice->types.mp3.buffer = (float*)mp3_buffer;
|
|
voice->types.mp3.buf_samples = samples;
|
|
voice->types.mp3.ratio = ratio;
|
|
voice->types.mp3.position = 0;
|
|
voice->types.mp3.samples = 0;
|
|
|
|
return true;
|
|
|
|
error:
|
|
drmp3_uninit(&voice->types.mp3.stream);
|
|
return false;
|
|
}
|
|
#endif
|
|
|
|
core_audio_mixer_voice_t* core_audio_mixer_play(core_audio_mixer_sound_t* sound,
|
|
bool repeat, float volume,
|
|
const char *resampler_ident,
|
|
enum resampler_quality quality,
|
|
core_audio_mixer_stop_cb_t stop_cb)
|
|
{
|
|
unsigned i;
|
|
bool res = false;
|
|
core_audio_mixer_voice_t* voice = core_s_voices;
|
|
|
|
if (!sound)
|
|
return NULL;
|
|
|
|
for (i = 0; i < CORE_AUDIO_MIXER_MAX_VOICES; i++, voice++)
|
|
{
|
|
if (voice->type != CORE_AUDIO_MIXER_TYPE_NONE)
|
|
continue;
|
|
|
|
switch (sound->type)
|
|
{
|
|
case CORE_AUDIO_MIXER_TYPE_WAV:
|
|
res = core_audio_mixer_play_wav(sound, voice, repeat, volume, stop_cb);
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_OGG:
|
|
#ifdef HAVE_STB_VORBIS
|
|
res = core_audio_mixer_play_ogg(sound, voice, repeat, volume,
|
|
resampler_ident, quality, stop_cb);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_MOD:
|
|
#ifdef HAVE_IBXM
|
|
res = core_audio_mixer_play_mod(sound, voice, repeat, volume, stop_cb);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_FLAC:
|
|
#ifdef HAVE_DR_FLAC
|
|
res = core_audio_mixer_play_flac(sound, voice, repeat, volume,
|
|
resampler_ident, quality, stop_cb);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_MP3:
|
|
#ifdef HAVE_DR_MP3
|
|
res = core_audio_mixer_play_mp3(sound, voice, repeat, volume,
|
|
resampler_ident, quality, stop_cb);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_NONE:
|
|
break;
|
|
}
|
|
|
|
break;
|
|
}
|
|
|
|
if (res)
|
|
{
|
|
voice->type = sound->type;
|
|
voice->repeat = repeat;
|
|
voice->volume = volume;
|
|
voice->sound = sound;
|
|
voice->stop_cb = stop_cb;
|
|
}
|
|
else
|
|
voice = NULL;
|
|
|
|
return voice;
|
|
}
|
|
|
|
void core_audio_mixer_stop(core_audio_mixer_voice_t* voice)
|
|
{
|
|
core_audio_mixer_stop_cb_t stop_cb = NULL;
|
|
core_audio_mixer_sound_t* sound = NULL;
|
|
|
|
if (voice)
|
|
{
|
|
stop_cb = voice->stop_cb;
|
|
sound = voice->sound;
|
|
|
|
voice->type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
|
|
if (stop_cb)
|
|
stop_cb(sound, CORE_AUDIO_MIXER_SOUND_STOPPED);
|
|
}
|
|
}
|
|
|
|
static void core_audio_mixer_mix_wav(float* buffer, size_t num_frames,
|
|
core_audio_mixer_voice_t* voice,
|
|
float volume)
|
|
{
|
|
int i;
|
|
unsigned buf_free = (unsigned)(num_frames * 2);
|
|
const core_audio_mixer_sound_t* sound = voice->sound;
|
|
unsigned pcm_available = sound->types.wav.frames
|
|
* 2 - voice->types.wav.position;
|
|
const float* pcm = sound->types.wav.pcm +
|
|
voice->types.wav.position;
|
|
|
|
again:
|
|
if (pcm_available < buf_free)
|
|
{
|
|
for (i = pcm_available; i != 0; i--)
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
if (voice->repeat)
|
|
{
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_REPEATED);
|
|
|
|
buf_free -= pcm_available;
|
|
pcm_available = sound->types.wav.frames * 2;
|
|
pcm = sound->types.wav.pcm;
|
|
voice->types.wav.position = 0;
|
|
goto again;
|
|
}
|
|
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_FINISHED);
|
|
|
|
voice->type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
}
|
|
else
|
|
{
|
|
for (i = buf_free; i != 0; i--)
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
voice->types.wav.position += buf_free;
|
|
}
|
|
}
|
|
|
|
#ifdef HAVE_STB_VORBIS
|
|
static void core_audio_mixer_mix_ogg(float* buffer, size_t num_frames,
|
|
core_audio_mixer_voice_t* voice,
|
|
float volume)
|
|
{
|
|
int i;
|
|
float* temp_buffer = NULL;
|
|
unsigned buf_free = (unsigned)(num_frames * 2);
|
|
unsigned temp_samples = 0;
|
|
float* pcm = NULL;
|
|
|
|
if (voice->types.ogg.position == voice->types.ogg.samples)
|
|
{
|
|
again:
|
|
if (temp_buffer == NULL)
|
|
temp_buffer = (float*)malloc(CORE_AUDIO_MIXER_TEMP_BUFFER * sizeof(float));
|
|
|
|
temp_samples = stb_vorbis_get_samples_float_interleaved(
|
|
voice->types.ogg.stream, 2, temp_buffer,
|
|
CORE_AUDIO_MIXER_TEMP_BUFFER) * 2;
|
|
|
|
if (temp_samples == 0)
|
|
{
|
|
if (voice->repeat)
|
|
{
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_REPEATED);
|
|
|
|
stb_vorbis_seek_start(voice->types.ogg.stream);
|
|
goto again;
|
|
}
|
|
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_FINISHED);
|
|
|
|
voice->type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
goto cleanup;
|
|
}
|
|
|
|
if (voice->types.ogg.resampler)
|
|
{
|
|
struct resampler_data info;
|
|
info.data_in = temp_buffer;
|
|
info.data_out = voice->types.ogg.buffer;
|
|
info.input_frames = temp_samples / 2;
|
|
info.output_frames = 0;
|
|
info.ratio = voice->types.ogg.ratio;
|
|
|
|
voice->types.ogg.resampler->process(
|
|
voice->types.ogg.resampler_data, &info);
|
|
}
|
|
else
|
|
memcpy(voice->types.ogg.buffer, temp_buffer,
|
|
temp_samples * sizeof(float));
|
|
|
|
voice->types.ogg.position = 0;
|
|
voice->types.ogg.samples = voice->types.ogg.buf_samples;
|
|
}
|
|
|
|
pcm = voice->types.ogg.buffer + voice->types.ogg.position;
|
|
|
|
if (voice->types.ogg.samples < buf_free)
|
|
{
|
|
for (i = voice->types.ogg.samples; i != 0; i--)
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
buf_free -= voice->types.ogg.samples;
|
|
goto again;
|
|
}
|
|
|
|
for (i = buf_free; i != 0; --i )
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
voice->types.ogg.position += buf_free;
|
|
voice->types.ogg.samples -= buf_free;
|
|
|
|
cleanup:
|
|
if (temp_buffer != NULL)
|
|
free(temp_buffer);
|
|
}
|
|
#endif
|
|
|
|
#ifdef HAVE_IBXM
|
|
static void core_audio_mixer_mix_mod(float* buffer, size_t num_frames,
|
|
core_audio_mixer_voice_t* voice,
|
|
float volume)
|
|
{
|
|
int i;
|
|
float samplef = 0.0f;
|
|
unsigned temp_samples = 0;
|
|
unsigned buf_free = (unsigned)(num_frames * 2);
|
|
int* pcm = NULL;
|
|
|
|
if (voice->types.mod.samples == 0)
|
|
{
|
|
again:
|
|
temp_samples = replay_get_audio(
|
|
voice->types.mod.stream, voice->types.mod.buffer, 0 ) * 2;
|
|
|
|
if (temp_samples == 0)
|
|
{
|
|
if (voice->repeat)
|
|
{
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_REPEATED);
|
|
|
|
replay_seek( voice->types.mod.stream, 0);
|
|
goto again;
|
|
}
|
|
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_FINISHED);
|
|
|
|
voice->type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
return;
|
|
}
|
|
|
|
voice->types.mod.position = 0;
|
|
voice->types.mod.samples = temp_samples;
|
|
}
|
|
pcm = voice->types.mod.buffer + voice->types.mod.position;
|
|
|
|
if (voice->types.mod.samples < buf_free)
|
|
{
|
|
for (i = voice->types.mod.samples; i != 0; i--)
|
|
{
|
|
samplef = ((float)(*pcm++) + 32768.0f) / 65535.0f;
|
|
samplef = samplef * 2.0f - 1.0f;
|
|
*buffer++ += samplef * volume;
|
|
}
|
|
|
|
buf_free -= voice->types.mod.samples;
|
|
goto again;
|
|
}
|
|
|
|
for (i = buf_free; i != 0; --i )
|
|
{
|
|
samplef = ((float)(*pcm++) + 32768.0f) / 65535.0f;
|
|
samplef = samplef * 2.0f - 1.0f;
|
|
*buffer++ += samplef * volume;
|
|
}
|
|
|
|
voice->types.mod.position += buf_free;
|
|
voice->types.mod.samples -= buf_free;
|
|
}
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_FLAC
|
|
static void core_audio_mixer_mix_flac(float* buffer, size_t num_frames,
|
|
core_audio_mixer_voice_t* voice,
|
|
float volume)
|
|
{
|
|
int i;
|
|
struct resampler_data info;
|
|
float temp_buffer[CORE_AUDIO_MIXER_TEMP_BUFFER] = { 0 };
|
|
unsigned buf_free = (unsigned)(num_frames * 2);
|
|
unsigned temp_samples = 0;
|
|
float *pcm = NULL;
|
|
|
|
if (voice->types.flac.position == voice->types.flac.samples)
|
|
{
|
|
again:
|
|
temp_samples = (unsigned)drflac_read_f32( voice->types.flac.stream, CORE_AUDIO_MIXER_TEMP_BUFFER, temp_buffer);
|
|
if (temp_samples == 0)
|
|
{
|
|
if (voice->repeat)
|
|
{
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_REPEATED);
|
|
|
|
drflac_seek_to_sample(voice->types.flac.stream,0);
|
|
goto again;
|
|
}
|
|
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_FINISHED);
|
|
|
|
voice->type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
return;
|
|
}
|
|
|
|
info.data_in = temp_buffer;
|
|
info.data_out = voice->types.flac.buffer;
|
|
info.input_frames = temp_samples / 2;
|
|
info.output_frames = 0;
|
|
info.ratio = voice->types.flac.ratio;
|
|
|
|
if (voice->types.flac.resampler)
|
|
voice->types.flac.resampler->process(
|
|
voice->types.flac.resampler_data, &info);
|
|
else
|
|
memcpy(voice->types.flac.buffer, temp_buffer, temp_samples * sizeof(float));
|
|
voice->types.flac.position = 0;
|
|
voice->types.flac.samples = voice->types.flac.buf_samples;
|
|
}
|
|
|
|
pcm = voice->types.flac.buffer + voice->types.flac.position;
|
|
|
|
if (voice->types.flac.samples < buf_free)
|
|
{
|
|
for (i = voice->types.flac.samples; i != 0; i--)
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
buf_free -= voice->types.flac.samples;
|
|
goto again;
|
|
}
|
|
|
|
for (i = buf_free; i != 0; --i )
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
voice->types.flac.position += buf_free;
|
|
voice->types.flac.samples -= buf_free;
|
|
}
|
|
#endif
|
|
|
|
#ifdef HAVE_DR_MP3
|
|
static void core_audio_mixer_mix_mp3(float* buffer, size_t num_frames,
|
|
core_audio_mixer_voice_t* voice,
|
|
float volume)
|
|
{
|
|
int i;
|
|
struct resampler_data info;
|
|
float temp_buffer[CORE_AUDIO_MIXER_TEMP_BUFFER] = { 0 };
|
|
unsigned buf_free = (unsigned)(num_frames * 2);
|
|
unsigned temp_samples = 0;
|
|
float* pcm = NULL;
|
|
|
|
if (voice->types.mp3.position == voice->types.mp3.samples)
|
|
{
|
|
again:
|
|
temp_samples = (unsigned)drmp3_read_f32(
|
|
&voice->types.mp3.stream,
|
|
CORE_AUDIO_MIXER_TEMP_BUFFER / 2, temp_buffer) * 2;
|
|
|
|
if (temp_samples == 0)
|
|
{
|
|
if (voice->repeat)
|
|
{
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_REPEATED);
|
|
|
|
drmp3_seek_to_frame(&voice->types.mp3.stream,0);
|
|
goto again;
|
|
}
|
|
|
|
if (voice->stop_cb)
|
|
voice->stop_cb(voice->sound, CORE_AUDIO_MIXER_SOUND_FINISHED);
|
|
|
|
voice->type = CORE_AUDIO_MIXER_TYPE_NONE;
|
|
return;
|
|
}
|
|
|
|
info.data_in = temp_buffer;
|
|
info.data_out = voice->types.mp3.buffer;
|
|
info.input_frames = temp_samples / 2;
|
|
info.output_frames = 0;
|
|
info.ratio = voice->types.mp3.ratio;
|
|
|
|
if (voice->types.mp3.resampler)
|
|
voice->types.mp3.resampler->process(
|
|
voice->types.mp3.resampler_data, &info);
|
|
else
|
|
memcpy(voice->types.mp3.buffer, temp_buffer,
|
|
temp_samples * sizeof(float));
|
|
voice->types.mp3.position = 0;
|
|
voice->types.mp3.samples = voice->types.mp3.buf_samples;
|
|
}
|
|
|
|
pcm = voice->types.mp3.buffer + voice->types.mp3.position;
|
|
|
|
if (voice->types.mp3.samples < buf_free)
|
|
{
|
|
for (i = voice->types.mp3.samples; i != 0; i--)
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
buf_free -= voice->types.mp3.samples;
|
|
goto again;
|
|
}
|
|
|
|
for (i = buf_free; i != 0; --i )
|
|
*buffer++ += *pcm++ * volume;
|
|
|
|
voice->types.mp3.position += buf_free;
|
|
voice->types.mp3.samples -= buf_free;
|
|
}
|
|
#endif
|
|
|
|
void core_audio_mixer_mix(float* buffer, size_t num_frames,
|
|
float volume_override, bool override)
|
|
{
|
|
unsigned i;
|
|
size_t j = 0;
|
|
float* sample = NULL;
|
|
core_audio_mixer_voice_t* voice = core_s_voices;
|
|
|
|
for (i = 0; i < CORE_AUDIO_MIXER_MAX_VOICES; i++, voice++)
|
|
{
|
|
float volume = (override) ? volume_override : voice->volume;
|
|
|
|
switch (voice->type)
|
|
{
|
|
case CORE_AUDIO_MIXER_TYPE_WAV:
|
|
core_audio_mixer_mix_wav(buffer, num_frames, voice, volume);
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_OGG:
|
|
#ifdef HAVE_STB_VORBIS
|
|
core_audio_mixer_mix_ogg(buffer, num_frames, voice, volume);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_MOD:
|
|
#ifdef HAVE_IBXM
|
|
core_audio_mixer_mix_mod(buffer, num_frames, voice, volume);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_FLAC:
|
|
#ifdef HAVE_DR_FLAC
|
|
core_audio_mixer_mix_flac(buffer, num_frames, voice, volume);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_MP3:
|
|
#ifdef HAVE_DR_MP3
|
|
core_audio_mixer_mix_mp3(buffer, num_frames, voice, volume);
|
|
#endif
|
|
break;
|
|
case CORE_AUDIO_MIXER_TYPE_NONE:
|
|
break;
|
|
}
|
|
}
|
|
|
|
for (j = 0, sample = buffer; j < num_frames * 2; j++, sample++)
|
|
{
|
|
if (*sample < -1.0f)
|
|
*sample = -1.0f;
|
|
else if (*sample > 1.0f)
|
|
*sample = 1.0f;
|
|
}
|
|
}
|
|
|
|
float core_audio_mixer_voice_get_volume(core_audio_mixer_voice_t *voice)
|
|
{
|
|
if (!voice)
|
|
return 0.0f;
|
|
|
|
return voice->volume;
|
|
}
|
|
|
|
void core_audio_mixer_voice_set_volume(core_audio_mixer_voice_t *voice, float val)
|
|
{
|
|
if (!voice)
|
|
return;
|
|
|
|
voice->volume = val;
|
|
}
|