/*
 * Apple // emulator for *ix
 *
 * This software package is subject to the GNU General Public License
 * version 3 or later (your choice) as published by the Free Software
 * Foundation.
 *
 * Copyright 2013-2015 Aaron Culliney
 *
 */

/* Apple //e speaker support. Source inspired/derived from AppleWin.
 *
 *  - ~23 //e cycles per PC sample (played back at 44.100kHz)
 *
 * The soundcard output drives how much 6502 emulation is done in real-time.  If the soundcard buffer is running out of
 * sample-data, then more 6502 cycles need to be executed to top-up the buffer, and vice-versa.
 *
 * This is in contrast to the AY8910 voices (mockingboard/phasor), which can simply generate more data if their buffers
 * are running low.
 */

#include "common.h"

#define DEBUG_SPEAKER 0
#if DEBUG_SPEAKER
#   define SPEAKER_LOG(...) LOG(__VA_ARGS__)
#else
#   define SPEAKER_LOG(...)
#endif

#define SPKR_SILENT_STEP 1

// TODO FIXME : still need to investigate better way to fix audio glitches when fast-loading (auto-adjusting speed) ...
#define HACKISHLY_REDUCE_AUDIO_GLITCHES_FOR_FAST_LOADING 1

static unsigned long bufferTotalSize = 0;
static unsigned long bufferSizeIdealMin = 0;
static unsigned long bufferSizeIdealMax = 0;
static unsigned long channelsSampleRateHz = 0;

static bool speaker_isAvailable = false;

static int16_t *samples_buffer = NULL; // holds max 1 second of samples
static int16_t *remainder_buffer = NULL; // holds enough to create one sample (averaged)
static unsigned long samples_buffer_idx = 0;
static unsigned int remainder_buffer_size = 0;
static unsigned long remainder_buffer_size_max = 0;
static unsigned int remainder_buffer_idx = 0;

static long speaker_volume = 0;
static int16_t speaker_amplitude = SPKR_DATA_INIT;
static int16_t speaker_data = 0;

static double cycles_per_sample = 0.0;
static unsigned long cycles_last_update = 0;
static unsigned long cycles_quiet_time = 0;

static bool speaker_accessed_since_last_flush = false;
static bool speaker_recently_active = false;

static bool speaker_going_silent = false;

static int samples_adjustment_counter = 0;

static AudioBuffer_s *speakerBuffer = NULL;

#if SPEAKER_TRACING
static FILE *speaker_trace_fp = NULL;
static unsigned long cycles_trace_toggled = 0;
#endif

// --------------------------------------------------------------------------------------------------------------------

static void speaker_prefsChanged(const char *domain) {
    long lVal = 0;
    speaker_volume = prefs_parseLongValue(domain, PREF_SPEAKER_VOLUME, &lVal, /*base:*/10) ? lVal : 5; // expected range 0-10
    if (speaker_volume < 0) {
        speaker_volume = 0;
    }
    if (speaker_volume > 10) {
        speaker_volume = 10;
    }
    float samplesScale = speaker_volume/10.f;
    speaker_amplitude = (int16_t)(SPKR_DATA_INIT * samplesScale);
}

static __attribute__((constructor)) void _init_speaker(void) {
    prefs_registerListener(PREF_DOMAIN_AUDIO, &speaker_prefsChanged);
}

/*
 * Because disk image loading is slow (AKA close-to-original-//e-speed), we may auto-switch to "fullspeed" for faster
 * loading when all the following heuristics hold true:
 *      - Disk motor is on
 *      - Speaker has not been toggled in some time (is not "active")
 *      - The graphics state is not "dirty"
 *
 * In fullspeed mode we output only quiet samples (zero-amplitude) at such a rate as to prevent the streaming audio from
 * either underflowing or overflowing.
 *
 * We will also auto-switch back to the last configured "scaled" speed when the speaker is toggled.
 */
static void _speaker_init_timing(void) {
    // 46.28 //e cycles for 22.05kHz sample rate

    // AppleWin NOTE : use integer value: Better for MJ Mahon's RT.SYNTH.DSK (integer multiples of 1.023MHz Clk)
    cycles_per_sample = (unsigned int)(cycles_persec_target / (double)audio_getCurrentBackend()->systemSettings.sampleRateHz);

    unsigned int last_remainder_buffer_size = remainder_buffer_size;
    remainder_buffer_size = (unsigned int)cycles_per_sample;
    if ((double)remainder_buffer_size != cycles_per_sample) {
        ++remainder_buffer_size;
    }
    assert(remainder_buffer_size <= remainder_buffer_size_max);

    if (last_remainder_buffer_size == remainder_buffer_size) {
        // no change ... insure seamless remainder_buffer
    } else {
        SPEAKER_LOG("changing remainder buffer size");
        remainder_buffer_idx = 0;
    }
    if (cycles_last_update > cycles_count_total) {
        SPEAKER_LOG("resetting speaker cycles counter");
        cycles_last_update = 0;
    }

    LOG("Speaker initialize timing ... cycles_persec_target:%f cycles_per_sample:%f speaker sampleRateHz:%lu", cycles_persec_target, cycles_per_sample, audio_getCurrentBackend()->systemSettings.sampleRateHz);

    if (is_fullspeed) {
        remainder_buffer_idx = 0;
        samples_buffer_idx = 0;
    }
}

/*
 * Adds to the samples_buffer the number of samples since the last invocation of this function.
 *
 * Speaker output square wave example:
 *        _______             ____      _____________________!        . +speaker_amplitude
 *                                                    silence _       .
 *                                                  threshold  _      .
 *               _ remainder                                    _     .
 *                 average                                       _    .
 * _______        ____________    ______                          ___ . 0
 *
 *      - When the speaker is accessed by the emulated program, the output (speaker_data) is toggled between the
 *        positive amplitude or zero
 *      - Evenly-divisible samples since last function call (cycles_diff/cycles_per_sample) are put directly into the
 *        samples_buffer for output to audio system backend
 *      - (+) Fractional samples are put into a remainder_buffer to be averaged and then transfered to the sample_buffer
 *        when there is enough data for 1 whole sample (possibly on a subsequent invocation of this function)
 *      - (+) If the speaker has not been toggled with output at +amplitude for a certain number of machine cycles, we
 *        gradually step the samples down to the zero bound of true quiet.  (This is done to avoid glitching when
 *        pausing/unpausing emulation for GUI/menus and auto-switching between full and configured speeds)
 */
static void _speaker_update(/*bool toggled*/) {

    do {
        if (is_fullspeed) {
            break;
        }

        if (UNLIKELY(cycles_last_update > cycles_count_total)) {
            LOG("ignoring cycles_count_total overflow ...");
            break; // ignore cycles_count_total overflow ...
        }

        unsigned long cycles_diff = cycles_count_total - cycles_last_update;

        if (remainder_buffer_idx) {
            // top-off previous previous fractional remainder cycles
            while (cycles_diff && (remainder_buffer_idx < remainder_buffer_size)) {
                remainder_buffer[remainder_buffer_idx] = speaker_data;
                ++remainder_buffer_idx;
                --cycles_diff;
            }

            if (remainder_buffer_idx == remainder_buffer_size) {
                // now have a complete extra sample from (previous + now) ... calculate mean value
                remainder_buffer_idx = 0;
                int sample_mean = 0;
                for (unsigned int i=0; i<remainder_buffer_size; i++) {
                    sample_mean += (int)remainder_buffer[i];
                }

                assert(remainder_buffer_size > 0);
                sample_mean /= (int)remainder_buffer_size;

                if (samples_buffer_idx < channelsSampleRateHz) {
                    samples_buffer[samples_buffer_idx++] = (int16_t)sample_mean;
                    if (NUM_CHANNELS == 2) {
                        samples_buffer[samples_buffer_idx++] = (int16_t)sample_mean;
                    }
                }
            }
        }

        const unsigned long samples_count = (unsigned long)((double)cycles_diff / cycles_per_sample);
        unsigned long num_samples = samples_count;
        const unsigned long cycles_remainder = (unsigned long)((double)cycles_diff - (double)num_samples * cycles_per_sample);

        // populate samples_buffer with whole samples
        while (num_samples && (samples_buffer_idx < channelsSampleRateHz)) {
            samples_buffer[samples_buffer_idx++] = speaker_data;
            if (NUM_CHANNELS == 2) {
                samples_buffer[samples_buffer_idx++] = speaker_data;
            }
#if HACKISHLY_REDUCE_AUDIO_GLITCHES_FOR_FAST_LOADING
            if (speaker_going_silent && speaker_data) {
                if (speaker_data < 0) {
                    speaker_data += SPKR_SILENT_STEP;
                } else {
                    speaker_data -= SPKR_SILENT_STEP;
                }
            }
#endif
            --num_samples;
        }

        if (cycles_remainder > 0) {
            // populate remainder_buffer with fractional samples
            assert(remainder_buffer_idx == 0 && "should have already dealt with remainder buffer");
            if (UNLIKELY(cycles_remainder >= remainder_buffer_size)) {
                LOG("OOPS, overflow in cycles_remainder:%lu", cycles_remainder);
            } else {
                while (remainder_buffer_idx<cycles_remainder) {
                    remainder_buffer[remainder_buffer_idx] = speaker_data;
                    ++remainder_buffer_idx;
                }
            }
        }

        if (UNLIKELY(samples_buffer_idx > channelsSampleRateHz)) {
            ////assert(samples_buffer_idx == channelsSampleRateHz && "should be at exactly the end, no further");
            if (UNLIKELY(samples_buffer_idx > channelsSampleRateHz)) {
                LOG("OOPS, possible overflow in speaker samples buffer ... samples_buffer_idx:%lu channelsSampleRateHz:%lu", (unsigned long)samples_buffer_idx, channelsSampleRateHz);
            }
        }
    } while (0);

    cycles_last_update = cycles_count_total;
}

/*
 * Submits "quiet" samples to the audio system backend when CPU thread is running fullspeed, to keep the audio streaming
 * topped up.
 *
 * 20150131 NOTE : it seems that there are still cases where we glitch OpenAL (seemingly on transitioning back to
 * regular speed sample submissions).  I have not been able to fully isolate the cause, but this has been minimized by
 * always trending the samples to the zero value when there has been a sufficient amount of speaker silence.
 */
static void _submit_samples_buffer_fullspeed(void) {
    samples_adjustment_counter = 0;

    unsigned long bytes_queued = 0;
    long err = speakerBuffer->GetCurrentPosition(speakerBuffer, &bytes_queued);
    if (err) {
        return;
    }
    ////assert(bytes_queued <= bufferTotalSize); -- wtf failing on Desktop on launch

    if (bytes_queued >= bufferSizeIdealMax) {
        return;
    }

    unsigned long num_samples_pad = (bufferSizeIdealMax - bytes_queued) / sizeof(int16_t);
    if (num_samples_pad == 0) {
        return;
    }

    unsigned long system_buffer_size = 0;
    int16_t *system_samples_buffer = NULL;
    if (speakerBuffer->Lock(speakerBuffer, num_samples_pad*sizeof(int16_t), &system_samples_buffer, &system_buffer_size)) {
        return;
    }
    if (num_samples_pad > system_buffer_size/sizeof(int16_t)) {
        num_samples_pad = system_buffer_size/sizeof(int16_t);
    }

    //SPEAKER_LOG("bytes_queued:%d enqueueing %d quiet samples", bytes_queued, num_samples_pad);
    for (unsigned int i=0; i<num_samples_pad; i++) {
        system_samples_buffer[i] = speaker_data;
    }

    speakerBuffer->Unlock(speakerBuffer, system_buffer_size);
}

// Submits samples from the samples_buffer to the audio system backend when running at a normal scaled-speed.  This also
// generates cycles feedback to the main CPU timing routine depending on the needs of the streaming audio (more or less
// data).
static unsigned long _submit_samples_buffer(const unsigned long num_channel_samples) {

    assert(num_channel_samples);

    unsigned long bytes_queued = 0;
    long err = speakerBuffer->GetCurrentPosition(speakerBuffer, &bytes_queued);
    if (err) {
        return num_channel_samples;
    }
    ////assert(bytes_queued <= bufferTotalSize);  -- this is failing on desktop FIXME TODO ...

    //
    // calculate CPU cycles feedback adjustment to prevent system audio buffer under/overflow
    //

    if (bytes_queued < bufferSizeIdealMin) {
        samples_adjustment_counter += SOUNDCORE_ERROR_INC; // need moar data
    } else if (bytes_queued > bufferSizeIdealMax) {
        samples_adjustment_counter -= SOUNDCORE_ERROR_INC; // need less data
    } else {
        samples_adjustment_counter = 0; // Acceptable amount of data in buffer
    }

    if (samples_adjustment_counter < -SOUNDCORE_ERROR_MAX) {
        samples_adjustment_counter = -SOUNDCORE_ERROR_MAX;
    } else if (samples_adjustment_counter > SOUNDCORE_ERROR_MAX) {
        samples_adjustment_counter =  SOUNDCORE_ERROR_MAX;
    }

    cycles_speaker_feedback = (int)(samples_adjustment_counter * cycles_per_sample);

    //SPEAKER_LOG("feedback:%d samples_adjustment_counter:%d bytes_queued:%lu", cycles_speaker_feedback, samples_adjustment_counter, bytes_queued);

    //
    // copy samples to audio system backend
    //

    const unsigned long bytes_free = bufferTotalSize - bytes_queued;
    unsigned long requested_samples = num_channel_samples;
    unsigned long requested_buffer_size = num_channel_samples * sizeof(int16_t);

    if (requested_buffer_size > bytes_free) {
        requested_samples = bytes_free / sizeof(int16_t);
        requested_buffer_size = bytes_free;
    }

    if (requested_buffer_size) {
        unsigned long system_buffer_size = 0;
        int16_t *system_samples_buffer = NULL;

        const unsigned long maxSpeakerBytes = channelsSampleRateHz * sizeof(int16_t);
        unsigned long curr_buffer_size = requested_buffer_size;
        unsigned long samples_idx = 0;
        unsigned long counter = 0;
        do {
            if (speakerBuffer->Lock(speakerBuffer, curr_buffer_size, &system_samples_buffer, &system_buffer_size)) {
                LOG("Problem locking speaker buffer");
                break;
            }

            if (system_buffer_size > maxSpeakerBytes) {
                LOG("AVOIDING BUFOVER...");
                system_buffer_size = maxSpeakerBytes;
                requested_buffer_size = maxSpeakerBytes;
            }

            memcpy(system_samples_buffer, &samples_buffer[samples_idx], system_buffer_size);

            err = speakerBuffer->Unlock(speakerBuffer, system_buffer_size);
            if (err) {
                LOG("Problem unlocking speaker buffer");
                break;
            }

            curr_buffer_size -= system_buffer_size;
            samples_idx += system_buffer_size;
            ++counter;
        } while (samples_idx < requested_buffer_size && counter < 2);
    }

    return requested_samples;
}

// --------------------------------------------------------------------------------------------------------------------
// speaker public API functions

void speaker_destroy(void) {
    ASSERT_ON_CPU_THREAD();
    speaker_isAvailable = false;
    audio_destroySoundBuffer(&speakerBuffer);
    FREE(samples_buffer);
    FREE(remainder_buffer);
}

void speaker_init(void) {
    ASSERT_ON_CPU_THREAD();

    long err = 0;
    speaker_isAvailable = false;
    do {
        err = audio_createSoundBuffer(&speakerBuffer);
        if (err) {
            break;
        }

        assert(audio_getCurrentBackend()->systemSettings.bytesPerSample == sizeof(int16_t));
        assert(NUM_CHANNELS == 2 || NUM_CHANNELS == 1);

        if (NUM_CHANNELS == 2) {
            bufferTotalSize = audio_getCurrentBackend()->systemSettings.stereoBufferSizeSamples * audio_getCurrentBackend()->systemSettings.bytesPerSample * NUM_CHANNELS;
        } else {
            bufferTotalSize = audio_getCurrentBackend()->systemSettings.monoBufferSizeSamples * audio_getCurrentBackend()->systemSettings.bytesPerSample;
        }
        bufferSizeIdealMin = bufferTotalSize/4;
        bufferSizeIdealMax = bufferTotalSize/2;
        channelsSampleRateHz = audio_getCurrentBackend()->systemSettings.sampleRateHz * NUM_CHANNELS;
        LOG("Speaker initializing with %lu buffer size (bytes), sample rate of %lu", bufferTotalSize, audio_getCurrentBackend()->systemSettings.sampleRateHz);

        remainder_buffer_size_max = ((CLK_6502_INT*(unsigned long)CPU_SCALE_FASTEST)/audio_getCurrentBackend()->systemSettings.sampleRateHz)+1;

        samples_buffer = CALLOC(1, channelsSampleRateHz * sizeof(int16_t));
        if (!samples_buffer) {
            err = -1;
            break;
        }
        samples_buffer_idx = bufferSizeIdealMax;

        remainder_buffer = MALLOC(remainder_buffer_size_max * sizeof(int16_t));
        if (!remainder_buffer) {
            err = -1;
            break;
        }

        _speaker_init_timing();

        speaker_isAvailable = true;
    } while (0);

    if (err) {
        LOG("Error creating speaker subsystem, regular sound will be disabled!");
        if (samples_buffer) {
            FREE(samples_buffer);
        }
        if (remainder_buffer) {
            FREE(remainder_buffer);
        }
    }
}

void speaker_reset(void) {
    if (speaker_isAvailable) {
        _speaker_init_timing();
    }
}

void speaker_flush(void) {
    SCOPE_TRACE_AUDIO("speaker_flush");
    if (!speaker_isAvailable) {
        return;
    }

    ASSERT_ON_CPU_THREAD();

    timing_checkpointCycles();

    if (is_fullspeed) {
        cycles_quiet_time = cycles_count_total;
        speaker_going_silent = false;
        speaker_accessed_since_last_flush = false;
    } else {
        if (speaker_accessed_since_last_flush) {
            cycles_quiet_time = 0;
            speaker_going_silent = false;
            speaker_accessed_since_last_flush = false;
        } else {
            if (!cycles_quiet_time) {
                cycles_quiet_time = cycles_count_total;
            }
            const unsigned int cycles_diff = (unsigned int)(cycles_persec_target/10); // 0.1sec of cycles
            if ((cycles_count_total - cycles_quiet_time) > cycles_diff*2) {
                // After 0.2sec of //e cycles time set inactive flag (allows auto-switch to full speed for fast disk access)
                speaker_recently_active = false;
            } else if ((speaker_data != 0) && (cycles_count_total - cycles_quiet_time > cycles_diff)) {
#if !HACKISHLY_REDUCE_AUDIO_GLITCHES_FOR_FAST_LOADING
                // OpenSLES seems to be able to pause output without the nasty pops that I hear with OpenAL on Linux
                // desktop.  So this speaker_going_silent hack is not needed.  There is also a noticeable glitch in
                // OpenSLES when this codepath is enabled.
                //
                // Furthermore, the simple mixer in soundcore-opensles.c now requires signed 16bit samples
#else
                // After 0.1sec of //e cycles time start reducing samples to zero (if they aren't there already).  This
                // process attempts to mask the extraneous clicks when freezing/restarting emulation (GUI access) and
                // glitching from the audio system backend
                speaker_going_silent = true;
                SPEAKER_LOG("speaker going silent");
#endif
            }
        }
    }
    _speaker_update(/*toggled:false*/);

    unsigned long samples_used = 0;
    if (is_fullspeed) {
        assert(!samples_buffer_idx && "should be all quiet samples");
        _submit_samples_buffer_fullspeed();
    } else if (samples_buffer_idx) {
        samples_used = _submit_samples_buffer(samples_buffer_idx);
    }
    assert(samples_used <= samples_buffer_idx);

    if (samples_used) {
        size_t unsubmitted_size = samples_buffer_idx-samples_used;
        if (unsubmitted_size) {
            memmove(samples_buffer, &samples_buffer[samples_used], unsubmitted_size);
        }
        samples_buffer_idx -= samples_used;
    }
}

bool speaker_isActive(void) {
    return speaker_recently_active;
}

double speaker_cyclesPerSample(void) {
    return cycles_per_sample;
}

// --------------------------------------------------------------------------------------------------------------------
// VM system entry point

uint8_t speaker_toggle(void) {
    ASSERT_ON_CPU_THREAD();

    timing_checkpointCycles();

#if SPEAKER_TRACING
    // output cycle count delta when toggled
    if (speaker_trace_fp) {
        unsigned long cycles_trace_delta = cycles_count_total - cycles_trace_toggled;
        fprintf(speaker_trace_fp, "%lu\n", cycles_trace_delta);
        cycles_trace_toggled = cycles_count_total;
    }
#endif

#if DIRECT_SPEAKER_ACCESS
#error this used to be implemented ...
    // AFAIK ... this requires an actual speaker device and ability to access it (usually requiring this program to be
    // running setuid-operator or (gasp) setuid-root ... maybe
#else
    speaker_accessed_since_last_flush = true;
    speaker_recently_active = true;

    if (timing_shouldAutoAdjustSpeed()) {
        is_fullspeed = false;
    }

    if (speaker_isAvailable) {
        _speaker_update(/*toggled:true*/);
    }

    if (!is_fullspeed) {
        if (speaker_data == speaker_amplitude) {
            speaker_data = -speaker_amplitude;
        } else {
            speaker_data = speaker_amplitude;
        }
    }
#endif

    return floating_bus();
}

#if SPEAKER_TRACING
// --------------------------------------------------------------------------------------------------------------------
// Speaker audio tracing (binary samples output)

void speaker_traceBegin(const char *trace_file) {
    if (trace_file) {
        speaker_trace_fp = fopen(trace_file, "w");
    }
}

void speaker_traceFlush(void) {
    if (speaker_trace_fp) {
        fflush(speaker_trace_fp);
    }
}

void speaker_traceEnd(void) {
    if (speaker_trace_fp) {
        fflush(speaker_trace_fp);
        fclose(speaker_trace_fp);
        speaker_trace_fp = NULL;
    }
}
#endif