aiie/teensy/teensy-speaker.cpp

219 lines
6.4 KiB
C++

#include <Arduino.h>
#include "teensy-speaker.h"
#include "teensy-println.h"
#include <Audio.h>
#include <SPI.h>
TeensyAudio audioDriver;
AudioMixer4 mixer2; //xy=280,253
AudioMixer4 mixer1; //xy=280,175
//AudioOutputI2S i2s; //xy=452,189
AudioOutputMQS i2s; //xy=452,189
AudioConnection patchCord1(audioDriver, 0, mixer1, 0);
AudioConnection patchCord2(audioDriver, 0, mixer2, 0);
AudioConnection patchCord3(mixer2, 0, i2s, 1);
AudioConnection patchCord4(mixer1, 0, i2s, 0);
#include "globals.h"
#define HIGHVAL (0x4FFF)
#define LOWVAL (-0x4FFF)
// Ring buffer that we fill with 44.1kHz data
#define BUFSIZE (4096)
#define CACHEMULTIPLIER 2
static volatile uint32_t bufIdx; // 0 .. BUFSIZE-1
static volatile uint64_t skippedSamples; // Who knows where this will
// wind up
static volatile uint8_t audioRunning = 0; // FIXME: needs constants abstracted
static volatile int64_t lastFilledTime = 0;
// how full do we want the audio buffer before we start it playing?
#define AUDIO_WATERLEVEL 4096
#define SAMPLEBYTES sizeof(short)
EXTMEM short soundBuf[BUFSIZE*CACHEMULTIPLIER];
static bool toggleState = false;
TeensySpeaker::TeensySpeaker(uint8_t sda, uint8_t scl) : PhysicalSpeaker()
{
toggleState = false;
mixerValue = 0x80;
AudioMemory(8);
}
TeensySpeaker::~TeensySpeaker()
{
}
void TeensySpeaker::begin()
{
mixer1.gain(0, 0.1f); // left channel (off for now)
mixer1.gain(1, 0.1f); // right channel (full volume for now)
memset(soundBuf, 0, sizeof(soundBuf));
toggleState = false;
bufIdx = 0;
skippedSamples = 0;
audioRunning = 0;
}
void TeensySpeaker::toggle(int64_t c)
{
// Figure out when the last time was that we put data in the audio buffer;
// then figure out how many audio buffer cycles we have to fill from that
// CPU time to this one.
__disable_irq();
// We expect to have filled to this cycle number...
int64_t expectedCycleNumber = (float)c * (float)AUDIO_SAMPLE_RATE_EXACT / (float)g_speed;
// Dynamically initialize the lastFilledTime based on the start time of the
// audio channel.
if (lastFilledTime == 0)
lastFilledTime = expectedCycleNumber;
// and we have filled to cycle number lastFilledTime. So how many do
// we need? This subtracts skippedSamples because those were filled
// automatically by the audioCallback when we had no data.
int64_t audioBufferSamples = expectedCycleNumber - lastFilledTime - skippedSamples;
// If audioBufferSamples < 0, then we need to keep some
// skippedSamples for later; otherwise we can keep moving forward.
if (audioBufferSamples < 0) {
skippedSamples = -audioBufferSamples;
audioBufferSamples = 0;
} else {
// Otherwise we consumed them and can forget about it.
skippedSamples = 0;
}
int32_t newIdx = bufIdx + audioBufferSamples;
if (audioBufferSamples == 0) {
// If the toggle wouldn't result in at least 1 buffer sample
// change, then we'll blatantly skip it here. If this turns out to
// be a problem, we could try setting audioBufferSamples++ and
// then twiddle the lastFilledTime so it looks like it's more in
// the future, but I suspect that would mean missing more future
// events, just like we would have missed this one.
//
// But I think this is probably okay - because something that's
// toggling the speaker fast enough that our 44k audio can't keep
// up with the individual changes is likely to toggle again in a
// moment without significant distortion?
__enable_irq();
return;
}
if (newIdx >= sizeof(soundBuf)/SAMPLEBYTES) {
// Buffer overrun error. Shouldn't happen?
println("OVERRUN");
newIdx = (sizeof(soundBuf)/SAMPLEBYTES) - 1;
}
lastFilledTime = expectedCycleNumber;
// Flip the toggle state
toggleState = !toggleState;
// Fill from bufIdx .. newIdx and set bufIdx to newIdx when done.
if (newIdx > bufIdx) {
long count = (long)newIdx - bufIdx;
for (long i=0; i<count; i++) {
soundBuf[bufIdx+i] = toggleState ? HIGHVAL : LOWVAL;
}
bufIdx = newIdx;
}
__enable_irq();
}
void TeensySpeaker::maintainSpeaker(int64_t c, uint64_t microseconds)
{
begin(); // flush! Hack. FIXME.
}
void TeensySpeaker::maintainSpeaker()
{
}
void TeensySpeaker::beginMixing()
{
// unused
}
void TeensySpeaker::mixOutput(uint8_t v)
{
// unused
}
void TeensyAudio::update(void)
{
audio_block_t *block;
short *stream;
if (audioRunning == 0)
audioRunning = 1;
block = allocate();
if (!block) {
return;
}
stream = block->data;
if (g_biosInterrupt) {
// While the BIOS is running, we don't put samples in the audio queue.
audioRunning = 0;
memset(stream, 0, AUDIO_BLOCK_SAMPLES * SAMPLEBYTES);
goto done;
}
if (audioRunning == 1 && bufIdx >= AUDIO_WATERLEVEL) {
// We have enough samples in the buffer to fill it, so we're fully
// up and running.
audioRunning = 2;
} else if (audioRunning == 1) {
// Still waiting for the first fill; return an empty buffer.
memset(stream, 0, AUDIO_BLOCK_SAMPLES * SAMPLEBYTES);
goto done;
}
static short lastKnownSample = 0; // saved for when the apple is quiescent
if (bufIdx >= AUDIO_BLOCK_SAMPLES) {
memcpy(stream, (void *)soundBuf, AUDIO_BLOCK_SAMPLES * SAMPLEBYTES);
lastKnownSample = stream[AUDIO_BLOCK_SAMPLES-1];
if (bufIdx > AUDIO_BLOCK_SAMPLES) {
// move the remaining data down
memcpy((void *)soundBuf, (void *)&soundBuf[AUDIO_BLOCK_SAMPLES], (bufIdx - AUDIO_BLOCK_SAMPLES + 1)*SAMPLEBYTES);
bufIdx -= AUDIO_BLOCK_SAMPLES;
}
} else {
if (bufIdx) {
// partial buffer exists
memcpy(stream, (void *)soundBuf, bufIdx * SAMPLEBYTES);
// and it's a partial underrun. Track the number of samples we skipped
// so we can keep the audio buffer in sync.
skippedSamples += AUDIO_BLOCK_SAMPLES - bufIdx;
for (int32_t i=0; i<AUDIO_BLOCK_SAMPLES-bufIdx; i++) {
stream[i+bufIdx] = lastKnownSample;
}
bufIdx = 0;
} else {
// No big deal - buffer underrun might just mean nothing is
// trying to play audio right now.
skippedSamples += AUDIO_BLOCK_SAMPLES;
for (int32_t i=0; i<AUDIO_BLOCK_SAMPLES; i++) {
stream[i] = 0;
}
// memset(stream, 0, AUDIO_BLOCK_SAMPLES * SAMPLEBYTES);
}
}
done:
transmit(block, 0);
release(block);
}