i_sdlsound.c

//
// Copyright(C) 1993-1996 Id Software, Inc.
// Copyright(C) 2005-2014 Simon Howard
// Copyright(C) 2008 David Flater
// Copyright(C) 2021-2022 Graham Sanderson
//
// This program is free software; you can redistribute it and/or
// modify it under the terms of the GNU General Public License
// as published by the Free Software Foundation; either version 2
// of the License, or (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License for more details.
//
// DESCRIPTION:
//	System interface for sound.
//

#include "config.h"

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include "SDL.h"
#include "SDL_mixer.h"

#ifdef HAVE_LIBSAMPLERATE
#include <samplerate.h>
#endif

#include "deh_str.h"
#include "i_sound.h"
#include "i_system.h"
#include "i_swap.h"
#include "m_argv.h"
#include "m_misc.h"
#include "w_wad.h"
#include "z_zone.h"

#include "doomtype.h"

#define LOW_PASS_FILTER
//#define DEBUG_DUMP_WAVS
#define NUM_CHANNELS 16

typedef struct allocated_sound_s allocated_sound_t;

struct allocated_sound_s
{
    should_be_const sfxinfo_t *sfxinfo;
    Mix_Chunk chunk;
    int use_count;
    int pitch;
    allocated_sound_t *prev, *next;
};

static boolean sound_initialized = false;

static allocated_sound_t *channels_playing[NUM_CHANNELS];

static int mixer_freq;
static Uint16 mixer_format;
static int mixer_channels;
static boolean use_sfx_prefix;
static boolean (*ExpandSoundData)(should_be_const sfxinfo_t *sfxinfo,
                                  should_be_const byte *data,
                                  int samplerate,
                                  int length) = NULL;

// Doubly-linked list of allocated sounds.
// When a sound is played, it is moved to the head, so that the oldest
// sounds not used recently are at the tail.

#define MUNGE_AUDIO 0
#if MUNGE_AUDIO
static uint8_t *munge_audio(const uint8_t *data, int length, const char *filename);
#endif

static allocated_sound_t *allocated_sounds_head = NULL;
static allocated_sound_t *allocated_sounds_tail = NULL;
static int allocated_sounds_size = 0;

int use_libsamplerate = 0;

// Scale factor used when converting libsamplerate floating point numbers
// to integers. Too high means the sounds can clip; too low means they
// will be too quiet. This is an amount that should avoid clipping most
// of the time: with all the Doom IWAD sound effects, at least. If a PWAD
// is used, clipping might occur.

float libsamplerate_scale = 0.65f;

// Hook a sound into the linked list at the head.

static void AllocatedSoundLink(allocated_sound_t *snd)
{
    snd->prev = NULL;

    snd->next = allocated_sounds_head;
    allocated_sounds_head = snd;

    if (allocated_sounds_tail == NULL)
    {
        allocated_sounds_tail = snd;
    }
    else
    {
        snd->next->prev = snd;
    }
}

// Unlink a sound from the linked list.

static void AllocatedSoundUnlink(allocated_sound_t *snd)
{
    if (snd->prev == NULL)
    {
        allocated_sounds_head = snd->next;
    }
    else
    {
        snd->prev->next = snd->next;
    }

    if (snd->next == NULL)
    {
        allocated_sounds_tail = snd->prev;
    }
    else
    {
        snd->next->prev = snd->prev;
    }
}

static void FreeAllocatedSound(allocated_sound_t *snd)
{
    // Unlink from linked list.

    AllocatedSoundUnlink(snd);

    // Keep track of the amount of allocated sound data:

    allocated_sounds_size -= snd->chunk.alen;

    free(snd);
}

// Search from the tail backwards along the allocated sounds list, find
// and free a sound that is not in use, to free up memory.  Return true
// for success.

static boolean FindAndFreeSound(void)
{
    allocated_sound_t *snd;

    snd = allocated_sounds_tail;

    while (snd != NULL)
    {
        if (snd->use_count == 0)
        {
            FreeAllocatedSound(snd);
            return true;
        }

        snd = snd->prev;
    }

    // No available sounds to free...

    return false;
}

// Enforce SFX cache size limit.  We are just about to allocate "len"
// bytes on the heap for a new sound effect, so free up some space
// so that we keep allocated_sounds_size < snd_cachesize

static void ReserveCacheSpace(size_t len)
{
    if (snd_cachesize <= 0)
    {
        return;
    }

    // Keep freeing sound effects that aren't currently being played,
    // until there is enough space for the new sound.

    while (allocated_sounds_size + len > snd_cachesize)
    {
        // Free a sound.  If there is nothing more to free, stop.

        if (!FindAndFreeSound())
        {
            break;
        }
    }
}

// Allocate a block for a new sound effect.

static allocated_sound_t *AllocateSound(should_be_const sfxinfo_t *sfxinfo, size_t len)
{
    allocated_sound_t *snd;

    // Keep allocated sounds within the cache size.

    ReserveCacheSpace(len);

    // Allocate the sound structure and data.  The data will immediately
    // follow the structure, which acts as a header.

    do
    {
        snd = malloc(sizeof(allocated_sound_t) + len);

        // Out of memory?  Try to free an old sound, then loop round
        // and try again.

        if (snd == NULL && !FindAndFreeSound())
        {
            return NULL;
        }

    } while (snd == NULL);

    // Skip past the chunk structure for the audio buffer

    snd->chunk.abuf = (byte *) (snd + 1);
    snd->chunk.alen = len;
    snd->chunk.allocated = 1;
    snd->chunk.volume = MIX_MAX_VOLUME;
    snd->pitch = NORM_PITCH;

    snd->sfxinfo = sfxinfo;
    snd->use_count = 0;

    // Keep track of how much memory all these cached sounds are using...

    allocated_sounds_size += len;

    AllocatedSoundLink(snd);

    return snd;
}

// Lock a sound, to indicate that it may not be freed.

static void LockAllocatedSound(allocated_sound_t *snd)
{
    // Increase use count, to stop the sound being freed.

    ++snd->use_count;

    //printf("++ %s: Use count=%i\n", snd->sfxinfo->name, snd->use_count);

    // When we use a sound, re-link it into the list at the head, so
    // that the oldest sounds fall to the end of the list for freeing.

    AllocatedSoundUnlink(snd);
    AllocatedSoundLink(snd);
}

// Unlock a sound to indicate that it may now be freed.

static void UnlockAllocatedSound(allocated_sound_t *snd)
{
    if (snd->use_count <= 0)
    {
        I_Error("Sound effect released more times than it was locked...");
    }

    --snd->use_count;

    //printf("-- %s: Use count=%i\n", snd->sfxinfo->name, snd->use_count);
}

// Search through the list of allocated sounds and return the one that matches
// the supplied sfxinfo entry and pitch level.

static allocated_sound_t * GetAllocatedSoundBySfxInfoAndPitch(should_be_const sfxinfo_t *sfxinfo, int pitch)
{
    allocated_sound_t * p = allocated_sounds_head;

    while (p != NULL)
    {
        if (p->sfxinfo == sfxinfo && p->pitch == pitch)
        {
            return p;
        }
        p = p->next;
    }

    return NULL;
}

// Allocate a new sound chunk and pitch-shift an existing sound up-or-down
// into it.

static allocated_sound_t * PitchShift(allocated_sound_t *insnd, int pitch)
{
    allocated_sound_t * outsnd;
    Sint16 *inp, *outp;
    Sint16 *srcbuf, *dstbuf;
    Uint32 srclen, dstlen;

    srcbuf = (Sint16 *)insnd->chunk.abuf;
    srclen = insnd->chunk.alen;

    // determine ratio pitch:NORM_PITCH and apply to srclen, then invert.
    // This is an approximation of vanilla behaviour based on measurements
    dstlen = (int)((1 + (1 - (float)pitch / NORM_PITCH)) * srclen);

    // ensure that the new buffer is an even length
    if ((dstlen % 2) == 0)
    {
        dstlen++;
    }

    outsnd = AllocateSound(insnd->sfxinfo, dstlen);

    if (!outsnd)
    {
        return NULL;
    }

    outsnd->pitch = pitch;
    dstbuf = (Sint16 *)outsnd->chunk.abuf;

    // loop over output buffer. find corresponding input cell, copy over
    for (outp = dstbuf; outp < dstbuf + dstlen/2; ++outp)
    {
        inp = srcbuf + (int)((float)(outp - dstbuf) / dstlen * srclen);
        *outp = *inp;
    }

    return outsnd;
}

// When a sound stops, check if it is still playing.  If it is not,
// we can mark the sound data as CACHE to be freed back for other
// means.

static void ReleaseSoundOnChannel(int channel)
{
    allocated_sound_t *snd = channels_playing[channel];

    Mix_HaltChannel(channel);

    if (snd == NULL)
    {
        return;
    }

    channels_playing[channel] = NULL;

    UnlockAllocatedSound(snd);

    // if the sound is a pitch-shift and it's not in use, immediately
    // free it
    if (snd->pitch != NORM_PITCH && snd->use_count <= 0)
    {
        FreeAllocatedSound(snd);
    }
}

#ifdef HAVE_LIBSAMPLERATE

// Returns the conversion mode for libsamplerate to use.

static int SRC_ConversionMode(void)
{
    switch (use_libsamplerate)
    {
        // 0 = disabled

        default:
        case 0:
            return -1;

        // Ascending numbers give higher quality

        case 1:
            return SRC_LINEAR;
        case 2:
            return SRC_ZERO_ORDER_HOLD;
        case 3:
            return SRC_SINC_FASTEST;
        case 4:
            return SRC_SINC_MEDIUM_QUALITY;
        case 5:
            return SRC_SINC_BEST_QUALITY;
    }
}

// libsamplerate-based generic sound expansion function for any sample rate
//   unsigned 8 bits --> signed 16 bits
//   mono --> stereo
//   samplerate --> mixer_freq
// Returns number of clipped samples.
// DWF 2008-02-10 with cleanups by Simon Howard.

static boolean ExpandSoundData_SRC(sfxinfo_t *sfxinfo,
                                   byte *data,
                                   int samplerate,
                                   int length)
{
    SRC_DATA src_data;
    float *data_in;
    uint32_t i, abuf_index=0, clipped=0;
//    uint32_t alen;
    int retn;
    int16_t *expanded;
    allocated_sound_t *snd;
    Mix_Chunk *chunk;

    src_data.input_frames = length;
    data_in = malloc(length * sizeof(float));
    src_data.data_in = data_in;
    src_data.src_ratio = (double)mixer_freq / samplerate;

    // We include some extra space here in case of rounding-up.
    src_data.output_frames = src_data.src_ratio * length + (mixer_freq / 4);
    src_data.data_out = malloc(src_data.output_frames * sizeof(float));

    assert(src_data.data_in != NULL && src_data.data_out != NULL);

    // Convert input data to floats

    for (i=0; i<length; ++i)
    {
        // Unclear whether 128 should be interpreted as "zero" or whether a
        // symmetrical range should be assumed.  The following assumes a
        // symmetrical range.
        data_in[i] = data[i] / 127.5 - 1;
    }

    // Do the sound conversion

    retn = src_simple(&src_data, SRC_ConversionMode(), 1);
    assert(retn == 0);

    // Allocate the new chunk.

//    alen = src_data.output_frames_gen * 4;

    snd = AllocateSound(sfxinfo, src_data.output_frames_gen * 4);

    if (snd == NULL)
    {
        return false;
    }

    chunk = &snd->chunk;
    expanded = (int16_t *) chunk->abuf;

    // Convert the result back into 16-bit integers.

    for (i=0; i<src_data.output_frames_gen; ++i)
    {
        // libsamplerate does not limit itself to the -1.0 .. 1.0 range on
        // output, so a multiplier less than INT16_MAX (32767) is required
        // to avoid overflows or clipping.  However, the smaller the
        // multiplier, the quieter the sound effects get, and the more you
        // have to turn down the music to keep it in balance.

        // 22265 is the largest multiplier that can be used to resample all
        // of the Vanilla DOOM sound effects to 48 kHz without clipping
        // using SRC_SINC_BEST_QUALITY.  It is close enough (only slightly
        // too conservative) for SRC_SINC_MEDIUM_QUALITY and
        // SRC_SINC_FASTEST.  PWADs with interestingly different sound
        // effects or target rates other than 48 kHz might still result in
        // clipping--I don't know if there's a limit to it.

        // As the number of clipped samples increases, the signal is
        // gradually overtaken by noise, with the loudest parts going first.
        // However, a moderate amount of clipping is often tolerated in the
        // quest for the loudest possible sound overall.  The results of
        // using INT16_MAX as the multiplier are not all that bad, but
        // artifacts are noticeable during the loudest parts.

        float cvtval_f =
            src_data.data_out[i] * libsamplerate_scale * INT16_MAX;
        int32_t cvtval_i = cvtval_f + (cvtval_f < 0 ? -0.5 : 0.5);

        // Asymmetrical sound worries me, so we won't use -32768.
        if (cvtval_i < -INT16_MAX)
        {
            cvtval_i = -INT16_MAX;
            ++clipped;
        }
        else if (cvtval_i > INT16_MAX)
        {
            cvtval_i = INT16_MAX;
            ++clipped;
        }

        // Left and right channels

        expanded[abuf_index++] = cvtval_i;
        expanded[abuf_index++] = cvtval_i;
    }

    free(data_in);
    free(src_data.data_out);

    if (clipped > 0)
    {
        fprintf(stderr, "Sound '%s': clipped %u samples (%0.2f %%)\n", 
                        sfxinfo->name, clipped,
                        400.0 * clipped / chunk->alen);
    }

    return true;
}

#endif

static boolean ConvertibleRatio(int freq1, int freq2)
{
    int ratio;

    if (freq1 > freq2)
    {
        return ConvertibleRatio(freq2, freq1);
    }
    else if ((freq2 % freq1) != 0)
    {
        // Not in a direct ratio

        return false;
    }
    else
    {
        // Check the ratio is a power of 2

        ratio = freq2 / freq1;

        while ((ratio & 1) == 0)
        {
            ratio = ratio >> 1;
        }

        return ratio == 1;
    }
}

#ifdef DEBUG_DUMP_WAVS

// Debug code to dump resampled sound effects to WAV files for analysis.

static void WriteWAV(char *filename, byte *data,
                     uint32_t length, int samplerate)
{
    FILE *wav;
    unsigned int i;
    unsigned short s;

    wav = fopen(filename, "wb");

    // Header

    fwrite("RIFF", 1, 4, wav);
    i = LONG(36 + samplerate);
    fwrite(&i, 4, 1, wav);
    fwrite("WAVE", 1, 4, wav);

    // Subchunk 1

    fwrite("fmt ", 1, 4, wav);
    i = LONG(16);
    fwrite(&i, 4, 1, wav);           // Length
    s = SHORT(1);
    fwrite(&s, 2, 1, wav);           // Format (PCM)
    s = SHORT(2);
    fwrite(&s, 2, 1, wav);           // Channels (2=stereo)
    i = LONG(samplerate);
    fwrite(&i, 4, 1, wav);           // Sample rate
    i = LONG(samplerate * 2 * 2);
    fwrite(&i, 4, 1, wav);           // Byte rate (samplerate * stereo * 16 bit)
    s = SHORT(2 * 2);
    fwrite(&s, 2, 1, wav);           // Block align (stereo * 16 bit)
    s = SHORT(16);
    fwrite(&s, 2, 1, wav);           // Bits per sample (16 bit)

    // Data subchunk

    fwrite("data", 1, 4, wav);
    i = LONG(length);
    fwrite(&i, 4, 1, wav);           // Data length
    fwrite(data, 1, length, wav);    // Data

    fclose(wav);
}

#endif

// Generic sound expansion function for any sample rate.
// Returns number of clipped samples (always 0).

static boolean ExpandSoundData_SDL(should_be_const sfxinfo_t *sfxinfo,
                                   should_be_const byte *data,
                                   int samplerate,
                                   int length)
{
    SDL_AudioCVT convertor;
    allocated_sound_t *snd;
    Mix_Chunk *chunk;
    uint32_t expanded_length;

    // Calculate the length of the expanded version of the sample.

    expanded_length = (uint32_t) ((((uint64_t) length) * mixer_freq) / samplerate);

    // Double up twice: 8 -> 16 bit and mono -> stereo

    expanded_length *= 4;

    // Allocate a chunk in which to expand the sound

    snd = AllocateSound(sfxinfo, expanded_length);

    if (snd == NULL)
    {
        return false;
    }

    chunk = &snd->chunk;

    // If we can, use the standard / optimized SDL conversion routines.

    if (samplerate <= mixer_freq
     && ConvertibleRatio(samplerate, mixer_freq)
     && SDL_BuildAudioCVT(&convertor,
                          AUDIO_U8, 1, samplerate,
                          mixer_format, mixer_channels, mixer_freq))
    {
        convertor.len = length;
        convertor.buf = malloc(convertor.len * convertor.len_mult);
        assert(convertor.buf != NULL);
        memcpy(convertor.buf, data, length);

        SDL_ConvertAudio(&convertor);

        memcpy(chunk->abuf, convertor.buf, chunk->alen);
        free(convertor.buf);
    }
    else
    {
        Sint16 *expanded = (Sint16 *) chunk->abuf;
        int expanded_length;
        int expand_ratio;
        int i;

        // Generic expansion if conversion does not work:
        //
        // SDL's audio conversion only works for rate conversions that are
        // powers of 2; if the two formats are not in a direct power of 2
        // ratio, do this naive conversion instead.

        // number of samples in the converted sound

        expanded_length = ((uint64_t) length * mixer_freq) / samplerate;
        expand_ratio = (length << 8) / expanded_length;

        for (i=0; i<expanded_length; ++i)
        {
            Sint16 sample;
            int src;

            src = (i * expand_ratio) >> 8;

            sample = data[src] | (data[src] << 8);
            sample -= 32768;

            // expand 8->16 bits, mono->stereo

            expanded[i * 2] = expanded[i * 2 + 1] = sample;
        }

#ifdef LOW_PASS_FILTER
        // Perform a low-pass filter on the upscaled sound to filter
        // out high-frequency noise from the conversion process.

        {
            float rc, dt, alpha;

            // Low-pass filter for cutoff frequency f:
            //
            // For sampling rate r, dt = 1 / r
            // rc = 1 / 2*pi*f
            // alpha = dt / (rc + dt)

            // Filter to the half sample rate of the original sound effect
            // (maximum frequency, by nyquist)

            dt = 1.0f / mixer_freq;
            rc = 1.0f / (3.14f * samplerate);
            alpha = dt / (rc + dt);

            // Both channels are processed in parallel, hence [i-2]:

            for (i=2; i<expanded_length * 2; ++i)
            {
                expanded[i] = (Sint16) (alpha * expanded[i]
                                      + (1 - alpha) * expanded[i-2]);
            }
        }
#endif /* #ifdef LOW_PASS_FILTER */
    }

    return true;
}

// Load and convert a sound effect
// Returns true if successful

static boolean CacheSFX(should_be_const sfxinfo_t *sfxinfo)
{
    int lumpnum;
    unsigned int lumplen;
    int samplerate;
    unsigned int length;
    should_be_const byte *data;

    // need to load the sound

    lumpnum = sfx_mut(sfxinfo)->lumpnum;
    data = W_CacheLumpNum(lumpnum, PU_STATIC);
    lumplen = W_LumpLength(lumpnum);

    // Check the header, and ensure this is a valid sound

    if (lumplen < 8
     || data[0] != 0x03 || data[1] != 0x00)
    {
        // Invalid sound

        return false;
    }

    // 16 bit sample rate field, 32 bit length field

    samplerate = (data[3] << 8) | data[2];
    length = (data[7] << 24) | (data[6] << 16) | (data[5] << 8) | data[4];

    // If the header specifies that the length of the sound is greater than
    // the length of the lump itself, this is an invalid sound lump

    // We also discard sound lumps that are less than 49 samples long,
    // as this is how DMX behaves - although the actual cut-off length
    // seems to vary slightly depending on the sample rate.  This needs
    // further investigation to better understand the correct
    // behavior.

    if (length > lumplen - 8 || length <= 48)
    {
        return false;
    }

    // The DMX sound library seems to skip the first 16 and last 16
    // bytes of the lump - reason unknown.

    data += 16;
    length -= 32;

    // Sample rate conversion

#if MUNGE_AUDIO
    char filename[16];
    allocated_sound_t * snd;

    M_snprintf(filename, sizeof(filename), "%s.d2",
               DEH_String(sfxinfo->name));

    uint8_t *d2 = munge_audio(data, (int)length, filename);
    if (1)
    {
        char filename[16];
        allocated_sound_t * snd;

        M_snprintf(filename, sizeof(filename), "%s.dff",
                   DEH_String(sfxinfo->name));
        uint8_t *d = malloc(length);
        d[0] = data[0];
        for(int s=1;s<length;s++) {
            d[s] = data[s] - data[s-1];
        }
        FILE *out = fopen(filename, "wb");
        fwrite(d, 1, length, out);
        fclose(out);
        free(d);
    }
    data = d2;
#endif
    if (!ExpandSoundData(sfxinfo, data + 8, samplerate, (int)length))
    {
#if MUNGE_AUDIO
        free(d2);
#endif
        return false;
    }
#if MUNGE_AUDIO
    free(d2);
#endif

#ifdef DEBUG_DUMP_WAVS
    {
        char filename[16];
        allocated_sound_t * snd;

        M_snprintf(filename, sizeof(filename), "%s.wav",
                   DEH_String(sfxinfo->name));
        snd = GetAllocatedSoundBySfxInfoAndPitch(sfxinfo, NORM_PITCH);
        WriteWAV(filename, snd->chunk.abuf, snd->chunk.alen,mixer_freq);
    }
#endif

    // don't need the original lump any more
  
    W_ReleaseLumpNum(lumpnum);

    return true;
}

static void GetSfxLumpName(const sfxinfo_t *sfx, char *buf, size_t buf_len)
{
    // Linked sfx lumps? Get the lump number for the sound linked to.

    if (sfx->link != NULL)
    {
        sfx = sfx->link;
    }

    // Doom adds a DS* prefix to sound lumps; Heretic and Hexen don't
    // do this.

    if (use_sfx_prefix)
    {
        M_snprintf(buf, buf_len, "ds%s", DEH_String(sfx->name));
    }
    else
    {
        M_StringCopy(buf, DEH_String(sfx->name), buf_len);
    }
}

#ifdef HAVE_LIBSAMPLERATE

// Preload all the sound effects - stops nasty ingame freezes

#if DOOM_SMALL
#include "adpcm-lib.h"
#endif
static void I_SDL_PrecacheSounds(sfxinfo_t *sounds, int num_sounds)
{
    char namebuf[9];
    int i;

#if 1 && DOOM_SMALL
    int total = 0;
    static int buckets[512];
    for (i=0; i<num_sounds; ++i)
    {
        GetSfxLumpName(&sounds[i], namebuf, sizeof(namebuf));
        int lumpnum = W_CheckNumForName(namebuf);
        if (lumpnum != -1)
        {
            uint8_t *data = W_CacheLumpNum(lumpnum, PU_STATIC);
            int lumplen = W_LumpLength(lumpnum);

            // Check the header, and ensure this is a valid sound

            if (lumplen < 8
                || data[0] != 0x03 || data[1] != 0x00)
            {
                printf("Invalid %s\n", DEH_String(sounds[i].name));
                continue;
            }

            // 16 bit sample rate field, 32 bit length field

            int samplerate = (data[3] << 8) | data[2];
            int length = (data[7] << 24) | (data[6] << 16) | (data[5] << 8) | data[4];

            // If the header specifies that the length of the sound is greater than
            // the length of the lump itself, this is an invalid sound lump

            // We also discard sound lumps that are less than 49 samples long,
            // as this is how DMX behaves - although the actual cut-off length
            // seems to vary slightly depending on the sample rate.  This needs
            // further investigation to better understand the correct
            // behavior.

            if (length > lumplen - 8 || length <= 48)
            {
                printf("Invalid %s\n", DEH_String(sounds[i].name));
                continue;
            }

            // The DMX sound library seems to skip the first 16 and last 16
            // bytes of the lump - reason unknown.

            data += 16;
            length -= 32;
            printf("%s %d %d\n", DEH_String(sounds[i].name), samplerate, length);
#if 0
#if 0
            uint8_t predict = data[0];
            for(int s=1;s<length;s++) {
                int delta = data[s] - predict;
                buckets[delta + 256]++;
                predict = data[s];
            }
#else
            int predict2 = data[0];
            int predict1 = data[1];
            for(int s=2;s<length;s++) {
                int predict = predict1;// 2 * predict1 - predict2;
                int delta = data[s] - predict;
                buckets[delta + 256]++;
                predict2 = predict1;
                predict1 = data[s];
            }
#endif
#endif
#if MUNGE_AUDIO
            free(munge_audio(data, length, NULL));
#endif
            total += length;
        } else {
            printf("Missing %s\n", DEH_String(sounds[i].name));
        }

    }
    printf("TOTAL %d\n", total);
#if 0
    long thuf=0;
    for(int i=0;i<512;i++) {
        int huffy = (int)ceil(-log2(buckets[i]/(double)total));
        printf("%d %d %f %f %d\n", i-256, buckets[i], buckets[i]/(double)total, -log2(buckets[i]/(double)total), huffy);
        thuf += buckets[i] * huffy;
    }
    printf("HUFFY %ld\n", thuf/8);
#endif
#endif

    // Don't need to precache the sounds unless we are using libsamplerate.

    if (use_libsamplerate == 0)
    {
	return;
    }

    printf("I_SDL_PrecacheSounds: Precaching all sound effects..");

    for (i=0; i<num_sounds; ++i)
    {
        if ((i % 6) == 0)
        {
            printf(".");
            fflush(stdout);
        }

        GetSfxLumpName(&sounds[i], namebuf, sizeof(namebuf));

        sfx_mut(&sounds[i])->lumpnum = W_CheckNumForName(namebuf);

        if (sfx_mut(&sounds[i])->lumpnum != -1)
        {
            CacheSFX(&sounds[i]);
        }
    }

    printf("\n");
}

#if MUNGE_AUDIO
static uint8_t *munge_audio(const uint8_t *data, int length, const char *filename) {
    int16_t *raw = calloc(2, length);
    for (int s = 0; s < length; s++) {
        raw[s] = (int16_t) ((((int) data[s]) - 128)) << 8;
    }
    int32_t average_deltas[2];
    int i;

    average_deltas[0] = average_deltas[1] = 0;

    for (i = 255; i--;) {
        average_deltas[0] -= average_deltas[0] >> 3;
        average_deltas[0] += abs((int32_t) raw[i] - raw[i - 1]);
    }

    average_deltas[0] >>= 3;
    average_deltas[1] >>= 3;

    void *ctx = adpcm_create_context(1, 3, NOISE_SHAPING_DYNAMIC, average_deltas);
#define MUNGE_BLOCK_SIZE 256
#define MUNGE_ENCODED_BLOCK_SIZE (MUNGE_BLOCK_SIZE / 2)

    int block_size = (MUNGE_BLOCK_SIZE - 1) / (1 ^ 3) + (1 * 4);
    printf("block size should be %d\n", block_size);
    // todo not correct
    uint8_t *enc = malloc(((length + MUNGE_BLOCK_SIZE - 1) / 2) & ~(MUNGE_ENCODED_BLOCK_SIZE-1));
    uint8_t *encout = enc;
    for (int off = 0; off < length; off += MUNGE_BLOCK_SIZE) {
        int local_len = length - off;
        if (local_len > MUNGE_BLOCK_SIZE) local_len = MUNGE_BLOCK_SIZE;
        size_t outbufsize = 0;
        adpcm_encode_block(ctx, encout, &outbufsize, raw + off, local_len);
//        printf("ENCODE %p + %04x -> %p + %04x\n", raw + off, local_len * 2, encout, (int)outbufsize);
        encout += outbufsize;
            printf("OB SIZE %d\n", (int)outbufsize);
    }
    if (filename) {
        FILE *out = fopen(filename, "wb");
        fwrite(enc, 1, encout-enc, out);
        fclose(out);
    }

    const uint8_t *src = enc;
    int16_t *raw2 = calloc(2, length);
    for (int off = 0; off < length; off += MUNGE_BLOCK_SIZE) {
        int local_len = length - off;
        if (local_len > MUNGE_BLOCK_SIZE) local_len = MUNGE_BLOCK_SIZE;
        int samps = adpcm_decode_block(raw2 + off, src, local_len / 2, 1);
        printf("%d\n", samps);
//        printf("DDECODE %p + %04x <- %p + %04x\n", raw2 + off, local_len * 2, src, local_len/2);
        src += MUNGE_ENCODED_BLOCK_SIZE;
    }
    free(enc);
    uint8_t *data2 = malloc(length);
    for (int s = 0; s < length; s++) {
        data2[s] = (raw2[s] >> 8) ^ 0x80;
    }
    free(raw2);
    adpcm_free_context(ctx);
    free(raw);
    return data2;
}
#endif

#else

static void I_SDL_PrecacheSounds(should_be_const sfxinfo_t *sounds, int num_sounds)
{
    // no-op
}

#endif

// Load a SFX chunk into memory and ensure that it is locked.

static boolean LockSound(should_be_const sfxinfo_t *sfxinfo)
{
    // If the sound isn't loaded, load it now
    if (GetAllocatedSoundBySfxInfoAndPitch(sfxinfo, NORM_PITCH) == NULL)
    {
        if (!CacheSFX(sfxinfo))
        {
            return false;
        }
    }

    LockAllocatedSound(GetAllocatedSoundBySfxInfoAndPitch(sfxinfo, NORM_PITCH));

    return true;
}

//
// Retrieve the raw data lump index
//  for a given SFX name.
//

static int I_SDL_GetSfxLumpNum(should_be_const sfxinfo_t *sfx)
{
    char namebuf[9];

    GetSfxLumpName(sfx, namebuf, sizeof(namebuf));

    return W_GetNumForName(namebuf);
}

static void I_SDL_UpdateSoundParams(int handle, int vol, int sep)
{
    int left, right;

    if (!sound_initialized || handle < 0 || handle >= NUM_CHANNELS)
    {
        return;
    }

    left = ((254 - sep) * vol) / 127;
    right = ((sep) * vol) / 127;

    if (left < 0) left = 0;
    else if ( left > 255) left = 255;
    if (right < 0) right = 0;
    else if (right > 255) right = 255;

    Mix_SetPanning(handle, left, right);
}

//
// Starting a sound means adding it
//  to the current list of active sounds
//  in the internal channels.
// As the SFX info struct contains
//  e.g. a pointer to the raw data,
//  it is ignored.
// As our sound handling does not handle
//  priority, it is ignored.
// Pitching (that is, increased speed of playback)
//  is set, but currently not used by mixing.
//

static int I_SDL_StartSound(should_be_const sfxinfo_t *sfxinfo, int channel, int vol, int sep, int pitch)
{
    allocated_sound_t *snd;

    if (!sound_initialized || channel < 0 || channel >= NUM_CHANNELS)
    {
        return -1;
    }

    // Release a sound effect if there is already one playing
    // on this channel

    ReleaseSoundOnChannel(channel);

    // Get the sound data

    if (!LockSound(sfxinfo))
    {
        return -1;
    }

    snd = GetAllocatedSoundBySfxInfoAndPitch(sfxinfo, pitch);

    if (snd == NULL)
    {
        allocated_sound_t *newsnd;
        // fetch the base sound effect, un-pitch-shifted
        snd = GetAllocatedSoundBySfxInfoAndPitch(sfxinfo, NORM_PITCH);

        if (snd == NULL)
        {
            return -1;
        }

        if (snd_pitchshift)
        {
            newsnd = PitchShift(snd, pitch);

            if (newsnd)
            {
                LockAllocatedSound(newsnd);
                UnlockAllocatedSound(snd);
                snd = newsnd;
            }
        }
    }
    else
    {
        LockAllocatedSound(snd);
    }

    // play sound

    Mix_PlayChannel(channel, &snd->chunk, 0);

    channels_playing[channel] = snd;

    // set separation, etc.

    I_SDL_UpdateSoundParams(channel, vol, sep);

    return channel;
}

static void I_SDL_StopSound(int handle)
{
    if (!sound_initialized || handle < 0 || handle >= NUM_CHANNELS)
    {
        return;
    }

    // Sound data is no longer needed; release the
    // sound data being used for this channel

    ReleaseSoundOnChannel(handle);
}


static boolean I_SDL_SoundIsPlaying(int handle)
{
    if (!sound_initialized || handle < 0 || handle >= NUM_CHANNELS)
    {
        return false;
    }

    return Mix_Playing(handle);
}

//
// Periodically called to update the sound system
//

static void I_SDL_UpdateSound(void)
{
    int i;

    // Check all channels to see if a sound has finished

    for (i=0; i<NUM_CHANNELS; ++i)
    {
        if (channels_playing[i] && !I_SDL_SoundIsPlaying(i))
        {
            // Sound has finished playing on this channel,
            // but sound data has not been released to cache

            ReleaseSoundOnChannel(i);
        }
    }
}

static void I_SDL_ShutdownSound(void)
{
    if (!sound_initialized)
    {
        return;
    }

    Mix_CloseAudio();
    SDL_QuitSubSystem(SDL_INIT_AUDIO);

    sound_initialized = false;
}

// Calculate slice size, based on snd_maxslicetime_ms.
// The result must be a power of two.

static int GetSliceSize(void)
{
    int limit;
    int n;

    limit = (snd_samplerate * snd_maxslicetime_ms) / 1000;

    // Try all powers of two, not exceeding the limit.

    for (n=0;; ++n)
    {
        // 2^n <= limit < 2^n+1 ?

        if ((1 << (n + 1)) > limit)
        {
            return (1 << n);
        }
    }

    // Should never happen?

    return 1024;
}

static boolean I_SDL_InitSound(boolean _use_sfx_prefix)
{
    int i;

    // SDL 2.0.6 has a bug that makes it unusable.
    if (SDL_COMPILEDVERSION == SDL_VERSIONNUM(2, 0, 6))
    {
        I_Error(
            "I_SDL_InitSound: "
            "You are trying to launch with SDL 2.0.6 which has a known bug "
            "that makes the game crash. Please either downgrade to "
            "SDL 2.0.5 or upgrade to 2.0.7. See the following bug for some "
            "additional context:\n"
            "<https://github.com/chocolate-doom/chocolate-doom/issues/945>");
    }

    use_sfx_prefix = _use_sfx_prefix;

    // No sounds yet
    for (i=0; i<NUM_CHANNELS; ++i)
    {
        channels_playing[i] = NULL;
    }

    if (SDL_Init(SDL_INIT_AUDIO) < 0)
    {
        fprintf(stderr, "Unable to set up sound.\n");
        return false;
    }

    if (Mix_OpenAudio(snd_samplerate, AUDIO_S16SYS, 2, GetSliceSize()) < 0)
    {
        fprintf(stderr, "Error initialising SDL_mixer: %s\n", Mix_GetError());
        return false;
    }

    ExpandSoundData = ExpandSoundData_SDL;

    Mix_QuerySpec(&mixer_freq, &mixer_format, &mixer_channels);

#ifdef HAVE_LIBSAMPLERATE
    if (use_libsamplerate != 0)
    {
        if (SRC_ConversionMode() < 0)
        {
            I_Error("I_SDL_InitSound: Invalid value for use_libsamplerate: %i",
                    use_libsamplerate);
        }

        ExpandSoundData = ExpandSoundData_SRC;
    }
#else
    if (use_libsamplerate != 0)
    {
        fprintf(stderr, "I_SDL_InitSound: use_libsamplerate=%i, but "
                        "libsamplerate support not compiled in.\n",
                        use_libsamplerate);
    }
#endif

    Mix_AllocateChannels(NUM_CHANNELS);

    SDL_PauseAudio(0);

    sound_initialized = true;

    return true;
}

static snddevice_t sound_sdl_devices[] = 
{
    SNDDEVICE_SB,
    SNDDEVICE_PAS,
    SNDDEVICE_GUS,
    SNDDEVICE_WAVEBLASTER,
    SNDDEVICE_SOUNDCANVAS,
    SNDDEVICE_AWE32,
};

sound_module_t sound_sdl_module = 
{
    sound_sdl_devices,
    arrlen(sound_sdl_devices),
    I_SDL_InitSound,
    I_SDL_ShutdownSound,
    I_SDL_GetSfxLumpNum,
    I_SDL_UpdateSound,
    I_SDL_UpdateSoundParams,
    I_SDL_StartSound,
    I_SDL_StopSound,
    I_SDL_SoundIsPlaying,
    I_SDL_PrecacheSounds,
};