src/audio/SDL_audiocvt.c

/*
  Simple DirectMedia Layer
  Copyright (C) 1997-2024 Sam Lantinga <slouken@libsdl.org>

  This software is provided 'as-is', without any express or implied
  warranty.  In no event will the authors be held liable for any damages
  arising from the use of this software.

  Permission is granted to anyone to use this software for any purpose,
  including commercial applications, and to alter it and redistribute it
  freely, subject to the following restrictions:

  1. The origin of this software must not be misrepresented; you must not
     claim that you wrote the original software. If you use this software
     in a product, an acknowledgment in the product documentation would be
     appreciated but is not required.
  2. Altered source versions must be plainly marked as such, and must not be
     misrepresented as being the original software.
  3. This notice may not be removed or altered from any source distribution.
*/
#include "SDL_internal.h"

#include "SDL_sysaudio.h"

#include "SDL_audioqueue.h"
#include "SDL_audioresample.h"

#ifndef SDL_INT_MAX
#define SDL_INT_MAX ((int)(~0u>>1))
#endif

/*
 * CHANNEL LAYOUTS AS SDL EXPECTS THEM:
 *
 * (Even if the platform expects something else later, that
 * SDL will swizzle between the app and the platform).
 *
 * Abbreviations:
 * - FRONT=single mono speaker
 * - FL=front left speaker
 * - FR=front right speaker
 * - FC=front center speaker
 * - BL=back left speaker
 * - BR=back right speaker
 * - SR=surround right speaker
 * - SL=surround left speaker
 * - BC=back center speaker
 * - LFE=low-frequency speaker
 *
 * These are listed in the order they are laid out in
 * memory, so "FL+FR" means "the front left speaker is
 * layed out in memory first, then the front right, then
 * it repeats for the next audio frame".
 *
 * 1 channel (mono) layout: FRONT
 * 2 channels (stereo) layout: FL+FR
 * 3 channels (2.1) layout: FL+FR+LFE
 * 4 channels (quad) layout: FL+FR+BL+BR
 * 5 channels (4.1) layout: FL+FR+LFE+BL+BR
 * 6 channels (5.1) layout: FL+FR+FC+LFE+BL+BR
 * 7 channels (6.1) layout: FL+FR+FC+LFE+BC+SL+SR
 * 8 channels (7.1) layout: FL+FR+FC+LFE+BL+BR+SL+SR
 */

#ifdef SDL_SSE3_INTRINSICS
// Convert from stereo to mono. Average left and right.
static void SDL_TARGETING("sse3") SDL_ConvertStereoToMono_SSE3(float *dst, const float *src, int num_frames)
{
    LOG_DEBUG_AUDIO_CONVERT("stereo", "mono (using SSE3)");

    const __m128 divby2 = _mm_set1_ps(0.5f);
    int i = num_frames;

    /* Do SSE blocks as long as we have 16 bytes available.
       Just use unaligned load/stores, if the memory at runtime is
       aligned it'll be just as fast on modern processors */
    while (i >= 4) {  // 4 * float32
        _mm_storeu_ps(dst, _mm_mul_ps(_mm_hadd_ps(_mm_loadu_ps(src), _mm_loadu_ps(src + 4)), divby2));
        i -= 4;
        src += 8;
        dst += 4;
    }

    // Finish off any leftovers with scalar operations.
    while (i) {
        *dst = (src[0] + src[1]) * 0.5f;
        dst++;
        i--;
        src += 2;
    }
}
#endif

#ifdef SDL_SSE_INTRINSICS
// Convert from mono to stereo. Duplicate to stereo left and right.
static void SDL_TARGETING("sse") SDL_ConvertMonoToStereo_SSE(float *dst, const float *src, int num_frames)
{
    LOG_DEBUG_AUDIO_CONVERT("mono", "stereo (using SSE)");

    // convert backwards, since output is growing in-place.
    src += (num_frames-4) * 1;
    dst += (num_frames-4) * 2;

    /* Do SSE blocks as long as we have 16 bytes available.
       Just use unaligned load/stores, if the memory at runtime is
       aligned it'll be just as fast on modern processors */
    // convert backwards, since output is growing in-place.
    int i = num_frames;
    while (i >= 4) {                                           // 4 * float32
        const __m128 input = _mm_loadu_ps(src);                // A B C D
        _mm_storeu_ps(dst, _mm_unpacklo_ps(input, input));     // A A B B
        _mm_storeu_ps(dst + 4, _mm_unpackhi_ps(input, input)); // C C D D
        i -= 4;
        src -= 4;
        dst -= 8;
    }

    // Finish off any leftovers with scalar operations.
    src += 3;
    dst += 6;   // adjust for smaller buffers.
    while (i) {  // convert backwards, since output is growing in-place.
        const float srcFC = src[0];
        dst[1] /* FR */ = srcFC;
        dst[0] /* FL */ = srcFC;
        i--;
        src--;
        dst -= 2;
    }
}
#endif

// Include the autogenerated channel converters...
#include "SDL_audio_channel_converters.h"

static SDL_bool SDL_IsSupportedAudioFormat(const SDL_AudioFormat fmt)
{
    switch (fmt) {
    case SDL_AUDIO_U8:
    case SDL_AUDIO_S8:
    case SDL_AUDIO_S16LE:
    case SDL_AUDIO_S16BE:
    case SDL_AUDIO_S32LE:
    case SDL_AUDIO_S32BE:
    case SDL_AUDIO_F32LE:
    case SDL_AUDIO_F32BE:
        return SDL_TRUE;  // supported.

    default:
        break;
    }

    return SDL_FALSE;  // unsupported.
}

static SDL_bool SDL_IsSupportedChannelCount(const int channels)
{
    return ((channels >= 1) && (channels <= 8));
}

// This does type and channel conversions _but not resampling_ (resampling happens in SDL_AudioStream).
// This does not check parameter validity, (beyond asserts), it expects you did that already!
// All of this has to function as if src==dst==scratch (conversion in-place), but as a convenience
// if you're just going to copy the final output elsewhere, you can specify a different output pointer.
//
// The scratch buffer must be able to store `num_frames * CalculateMaxSampleFrameSize(src_format, src_channels, dst_format, dst_channels)` bytes.
// If the scratch buffer is NULL, this restriction applies to the output buffer instead.
void ConvertAudio(int num_frames, const void *src, SDL_AudioFormat src_format, int src_channels,
                  void *dst, SDL_AudioFormat dst_format, int dst_channels, void* scratch)
{
    SDL_assert(src != NULL);
    SDL_assert(dst != NULL);
    SDL_assert(SDL_IsSupportedAudioFormat(src_format));
    SDL_assert(SDL_IsSupportedAudioFormat(dst_format));
    SDL_assert(SDL_IsSupportedChannelCount(src_channels));
    SDL_assert(SDL_IsSupportedChannelCount(dst_channels));

    if (!num_frames) {
        return;  // no data to convert, quit.
    }

#if DEBUG_AUDIO_CONVERT
    SDL_Log("SDL_AUDIO_CONVERT: Convert format %04x->%04x, channels %u->%u", src_format, dst_format, src_channels, dst_channels);
#endif

    const int src_bitsize = (int) SDL_AUDIO_BITSIZE(src_format);
    const int dst_bitsize = (int) SDL_AUDIO_BITSIZE(dst_format);

    const int dst_sample_frame_size = (dst_bitsize / 8) * dst_channels;

    /* Type conversion goes like this now:
        - byteswap to CPU native format first if necessary.
        - convert to native Float32 if necessary.
        - change channel count if necessary.
        - convert to final data format.
        - byteswap back to foreign format if necessary.

       The expectation is we can process data faster in float32
       (possibly with SIMD), and making several passes over the same
       buffer is likely to be CPU cache-friendly, avoiding the
       biggest performance hit in modern times. Previously we had
       (script-generated) custom converters for every data type and
       it was a bloat on SDL compile times and final library size. */

    // see if we can skip float conversion entirely.
    if (src_channels == dst_channels) {
        if (src_format == dst_format) {
            // nothing to do, we're already in the right format, just copy it over if necessary.
            if (src != dst) {
                SDL_memcpy(dst, src, num_frames * dst_sample_frame_size);
            }
            return;
        }

        // just a byteswap needed?
        if ((src_format ^ dst_format) == SDL_AUDIO_MASK_BIG_ENDIAN) {
            ConvertAudioSwapEndian(dst, src, num_frames * src_channels, src_bitsize);
            return;  // all done.
        }
    }

    if (!scratch) {
        scratch = dst;
    }

    const SDL_bool srcconvert = src_format != SDL_AUDIO_F32;
    const SDL_bool channelconvert = src_channels != dst_channels;
    const SDL_bool dstconvert = dst_format != SDL_AUDIO_F32;

    // get us to float format.
    if (srcconvert) {
        void* buf = (channelconvert || dstconvert) ? scratch : dst;
        ConvertAudioToFloat((float *) buf, src, num_frames * src_channels, src_format);
        src = buf;
    }

    // Channel conversion

    if (channelconvert) {
        SDL_AudioChannelConverter channel_converter;
        SDL_AudioChannelConverter override = NULL;

        // SDL_IsSupportedChannelCount should have caught these asserts, or we added a new format and forgot to update the table.
        SDL_assert(src_channels <= SDL_arraysize(channel_converters));
        SDL_assert(dst_channels <= SDL_arraysize(channel_converters[0]));

        channel_converter = channel_converters[src_channels - 1][dst_channels - 1];
        SDL_assert(channel_converter != NULL);

        // swap in some SIMD versions for a few of these.
        if (channel_converter == SDL_ConvertStereoToMono) {
            #ifdef SDL_SSE3_INTRINSICS
            if (!override && SDL_HasSSE3()) { override = SDL_ConvertStereoToMono_SSE3; }
            #endif
        } else if (channel_converter == SDL_ConvertMonoToStereo) {
            #ifdef SDL_SSE_INTRINSICS
            if (!override && SDL_HasSSE()) { override = SDL_ConvertMonoToStereo_SSE; }
            #endif
        }

        if (override) {
            channel_converter = override;
        }

        void* buf = dstconvert ? scratch : dst;
        channel_converter((float *) buf, (const float *) src, num_frames);
        src = buf;
    }

    // Resampling is not done in here. SDL_AudioStream handles that.

    // Move to final data type.
    if (dstconvert) {
        ConvertAudioFromFloat(dst, (const float *) src, num_frames * dst_channels, dst_format);
        src = dst;
    }

    SDL_assert(src == dst);  // if we got here, we _had_ to have done _something_. Otherwise, we should have memcpy'd!
}

// Calculate the largest frame size needed to convert between the two formats.
static int CalculateMaxFrameSize(SDL_AudioFormat src_format, int src_channels, SDL_AudioFormat dst_format, int dst_channels)
{
    const int src_format_size = SDL_AUDIO_BYTESIZE(src_format);
    const int dst_format_size = SDL_AUDIO_BYTESIZE(dst_format);
    const int max_app_format_size = SDL_max(src_format_size, dst_format_size);
    const int max_format_size = SDL_max(max_app_format_size, sizeof (float));  // ConvertAudio and ResampleAudio use floats.
    const int max_channels = SDL_max(src_channels, dst_channels);
    return max_format_size * max_channels;
}

static Sint64 GetAudioStreamResampleRate(SDL_AudioStream* stream, int src_freq, Sint64 resample_offset)
{
    src_freq = (int)((float)src_freq * stream->freq_ratio);

    Sint64 resample_rate = SDL_GetResampleRate(src_freq, stream->dst_spec.freq);

    // If src_freq == dst_freq, and we aren't between frames, don't resample
    if ((resample_rate == 0x100000000) && (resample_offset == 0)) {
        resample_rate = 0;
    }

    return resample_rate;
}

static int UpdateAudioStreamInputSpec(SDL_AudioStream *stream, const SDL_AudioSpec *spec)
{
    if (AUDIO_SPECS_EQUAL(stream->input_spec, *spec)) {
        return 0;
    }

    if (SDL_ResetAudioQueueHistory(stream->queue, SDL_GetResamplerHistoryFrames()) != 0) {
        return -1;
    }

    SDL_copyp(&stream->input_spec, spec);

    return 0;
}

SDL_AudioStream *SDL_CreateAudioStream(const SDL_AudioSpec *src_spec, const SDL_AudioSpec *dst_spec)
{
    SDL_ChooseAudioConverters();
    SDL_SetupAudioResampler();

    SDL_AudioStream *retval = (SDL_AudioStream *)SDL_calloc(1, sizeof(SDL_AudioStream));
    if (!retval) {
        return NULL;
    }

    retval->freq_ratio = 1.0f;
    retval->queue = SDL_CreateAudioQueue(8192);

    if (!retval->queue) {
        SDL_free(retval);
        return NULL;
    }

    retval->lock = SDL_CreateMutex();
    if (!retval->lock) {
        SDL_free(retval->queue);
        SDL_free(retval);
        return NULL;
    }

    OnAudioStreamCreated(retval);

    if (SDL_SetAudioStreamFormat(retval, src_spec, dst_spec) == -1) {
        SDL_DestroyAudioStream(retval);
        return NULL;
    }

    return retval;
}

SDL_PropertiesID SDL_GetAudioStreamProperties(SDL_AudioStream *stream)
{
    if (!stream) {
        SDL_InvalidParamError("stream");
        return 0;
    }
    if (stream->props == 0) {
        stream->props = SDL_CreateProperties();
    }
    return stream->props;
}

int SDL_SetAudioStreamGetCallback(SDL_AudioStream *stream, SDL_AudioStreamCallback callback, void *userdata)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }
    SDL_LockMutex(stream->lock);
    stream->get_callback = callback;
    stream->get_callback_userdata = userdata;
    SDL_UnlockMutex(stream->lock);
    return 0;
}

int SDL_SetAudioStreamPutCallback(SDL_AudioStream *stream, SDL_AudioStreamCallback callback, void *userdata)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }
    SDL_LockMutex(stream->lock);
    stream->put_callback = callback;
    stream->put_callback_userdata = userdata;
    SDL_UnlockMutex(stream->lock);
    return 0;
}

int SDL_LockAudioStream(SDL_AudioStream *stream)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }
    SDL_LockMutex(stream->lock);
    return 0;
}

int SDL_UnlockAudioStream(SDL_AudioStream *stream)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }
    SDL_UnlockMutex(stream->lock);
    return 0;
}

int SDL_GetAudioStreamFormat(SDL_AudioStream *stream, SDL_AudioSpec *src_spec, SDL_AudioSpec *dst_spec)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    SDL_LockMutex(stream->lock);
    if (src_spec) {
        SDL_copyp(src_spec, &stream->src_spec);
    }
    if (dst_spec) {
        SDL_copyp(dst_spec, &stream->dst_spec);
    }
    SDL_UnlockMutex(stream->lock);

    if (src_spec && src_spec->format == 0) {
        return SDL_SetError("Stream has no source format");
    } else if (dst_spec && dst_spec->format == 0) {
        return SDL_SetError("Stream has no destination format");
    }

    return 0;
}

int SDL_SetAudioStreamFormat(SDL_AudioStream *stream, const SDL_AudioSpec *src_spec, const SDL_AudioSpec *dst_spec)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    // Picked mostly arbitrarily.
    static const int min_freq = 4000;
    static const int max_freq = 384000;

    if (src_spec) {
        if (!SDL_IsSupportedAudioFormat(src_spec->format)) {
            return SDL_InvalidParamError("src_spec->format");
        } else if (!SDL_IsSupportedChannelCount(src_spec->channels)) {
            return SDL_InvalidParamError("src_spec->channels");
        } else if (src_spec->freq <= 0) {
            return SDL_InvalidParamError("src_spec->freq");
        } else if (src_spec->freq < min_freq) {
            return SDL_SetError("Source rate is too low");
        } else if (src_spec->freq > max_freq) {
            return SDL_SetError("Source rate is too high");
        }
    }

    if (dst_spec) {
        if (!SDL_IsSupportedAudioFormat(dst_spec->format)) {
            return SDL_InvalidParamError("dst_spec->format");
        } else if (!SDL_IsSupportedChannelCount(dst_spec->channels)) {
            return SDL_InvalidParamError("dst_spec->channels");
        } else if (dst_spec->freq <= 0) {
            return SDL_InvalidParamError("dst_spec->freq");
        } else if (dst_spec->freq < min_freq) {
            return SDL_SetError("Destination rate is too low");
        } else if (dst_spec->freq > max_freq) {
            return SDL_SetError("Destination rate is too high");
        }
    }

    SDL_LockMutex(stream->lock);

    // quietly refuse to change the format of the end currently bound to a device.
    if (stream->bound_device) {
        if (stream->bound_device->physical_device->iscapture) {
            src_spec = NULL;
        } else {
            dst_spec = NULL;
        }
    }

    if (src_spec) {
        SDL_copyp(&stream->src_spec, src_spec);
    }

    if (dst_spec) {
        SDL_copyp(&stream->dst_spec, dst_spec);
    }

    SDL_UnlockMutex(stream->lock);

    return 0;
}

float SDL_GetAudioStreamFrequencyRatio(SDL_AudioStream *stream)
{
    if (!stream) {
        SDL_InvalidParamError("stream");
        return 0.0f;
    }

    SDL_LockMutex(stream->lock);
    const float freq_ratio = stream->freq_ratio;
    SDL_UnlockMutex(stream->lock);

    return freq_ratio;
}

int SDL_SetAudioStreamFrequencyRatio(SDL_AudioStream *stream, float freq_ratio)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    // Picked mostly arbitrarily.
    const float min_freq_ratio = 0.01f;
    const float max_freq_ratio = 100.0f;

    if (freq_ratio < min_freq_ratio) {
        return SDL_SetError("Frequency ratio is too low");
    } else if (freq_ratio > max_freq_ratio) {
        return SDL_SetError("Frequency ratio is too high");
    }

    SDL_LockMutex(stream->lock);
    stream->freq_ratio = freq_ratio;
    SDL_UnlockMutex(stream->lock);

    return 0;
}

static int CheckAudioStreamIsFullySetup(SDL_AudioStream *stream)
{
    if (stream->src_spec.format == 0) {
        return SDL_SetError("Stream has no source format");
    } else if (stream->dst_spec.format == 0) {
        return SDL_SetError("Stream has no destination format");
    }

    return 0;
}

static int PutAudioStreamBuffer(SDL_AudioStream *stream, const void *buf, int len, SDL_ReleaseAudioBufferCallback callback, void* userdata)
{
#if DEBUG_AUDIOSTREAM
    SDL_Log("AUDIOSTREAM: wants to put %d bytes", len);
#endif

    SDL_LockMutex(stream->lock);

    if (CheckAudioStreamIsFullySetup(stream) != 0) {
        SDL_UnlockMutex(stream->lock);
        return -1;
    }

    if ((len % SDL_AUDIO_FRAMESIZE(stream->src_spec)) != 0) {
        SDL_UnlockMutex(stream->lock);
        return SDL_SetError("Can't add partial sample frames");
    }

    SDL_AudioTrack* track = NULL;

    if (callback) {
        track = SDL_CreateAudioTrack(stream->queue, &stream->src_spec, (Uint8 *)buf, len, len, callback, userdata);

        if (!track) {
            SDL_UnlockMutex(stream->lock);
            return -1;
        }
    }

    const int prev_available = stream->put_callback ? SDL_GetAudioStreamAvailable(stream) : 0;

    int retval = 0;

    if (track) {
        SDL_AddTrackToAudioQueue(stream->queue, track);
    } else {
        retval = SDL_WriteToAudioQueue(stream->queue, &stream->src_spec, (const Uint8 *)buf, len);
    }

    if (retval == 0) {
        if (stream->put_callback) {
            const int newavail = SDL_GetAudioStreamAvailable(stream) - prev_available;
            stream->put_callback(stream->put_callback_userdata, stream, newavail, newavail);
        }
    }

    SDL_UnlockMutex(stream->lock);

    return retval;
}

static void SDLCALL FreeAllocatedAudioBuffer(void *userdata, const void *buf, int len)
{
    SDL_free((void*) buf);
}

int SDL_PutAudioStreamData(SDL_AudioStream *stream, const void *buf, int len)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    } else if (!buf) {
        return SDL_InvalidParamError("buf");
    } else if (len < 0) {
        return SDL_InvalidParamError("len");
    } else if (len == 0) {
        return 0; // nothing to do.
    }

    // When copying in large amounts of data, try and do as much work as possible
    // outside of the stream lock, otherwise the output device is likely to be starved.
    const int large_input_thresh = 64 * 1024;

    if (len >= large_input_thresh) {
        void* data = SDL_malloc(len);

        if (!data) {
            return -1;
        }

        SDL_memcpy(data, buf, len);
        buf = data;

        int ret = PutAudioStreamBuffer(stream, buf, len, FreeAllocatedAudioBuffer, NULL);

        if (ret < 0) {
            SDL_free(data);
        }

        return ret;
    }

    return PutAudioStreamBuffer(stream, buf, len, NULL, NULL);
}

int SDL_FlushAudioStream(SDL_AudioStream *stream)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    SDL_LockMutex(stream->lock);
    SDL_FlushAudioQueue(stream->queue);
    SDL_UnlockMutex(stream->lock);

    return 0;
}

/* this does not save the previous contents of stream->work_buffer. It's a work buffer!!
   The returned buffer is aligned/padded for use with SIMD instructions. */
static Uint8 *EnsureAudioStreamWorkBufferSize(SDL_AudioStream *stream, size_t newlen)
{
    if (stream->work_buffer_allocation >= newlen) {
        return stream->work_buffer;
    }

    Uint8 *ptr = (Uint8 *) SDL_aligned_alloc(SDL_GetSIMDAlignment(), newlen);
    if (!ptr) {
        return NULL;  // previous work buffer is still valid!
    }

    SDL_aligned_free(stream->work_buffer);
    stream->work_buffer = ptr;
    stream->work_buffer_allocation = newlen;
    return ptr;
}

static Sint64 NextAudioStreamIter(SDL_AudioStream* stream, void** inout_iter,
    Sint64* inout_resample_offset, SDL_AudioSpec* out_spec, SDL_bool* out_flushed)
{
    SDL_AudioSpec spec;
    SDL_bool flushed;
    size_t queued_bytes = SDL_NextAudioQueueIter(stream->queue, inout_iter, &spec, &flushed);

    if (out_spec) {
        SDL_copyp(out_spec, &spec);
    }

    // There is infinite audio available, whether or not we are resampling
    if (queued_bytes == SDL_SIZE_MAX) {
        *inout_resample_offset = 0;

        if (out_flushed) {
            *out_flushed = SDL_FALSE;
        }

        return SDL_MAX_SINT32;
    }

    Sint64 resample_offset = *inout_resample_offset;
    Sint64 resample_rate = GetAudioStreamResampleRate(stream, spec.freq, resample_offset);
    Sint64 output_frames = (Sint64)(queued_bytes / SDL_AUDIO_FRAMESIZE(spec));

    if (resample_rate) {
        // Resampling requires padding frames to the left and right of the current position.
        // Past the end of the track, the right padding is filled with silence.
        // But we only want to do that if the track is actually finished (flushed).
        if (!flushed) {
            output_frames -= SDL_GetResamplerPaddingFrames(resample_rate);
        }

        output_frames = SDL_GetResamplerOutputFrames(output_frames, resample_rate, &resample_offset);
    }

    if (flushed) {
        resample_offset = 0;
    }

    *inout_resample_offset = resample_offset;

    if (out_flushed) {
        *out_flushed = flushed;
    }

    return output_frames;
}

static Sint64 GetAudioStreamAvailableFrames(SDL_AudioStream* stream, Sint64* out_resample_offset)
{
    void* iter = SDL_BeginAudioQueueIter(stream->queue);

    Sint64 resample_offset = stream->resample_offset;
    Sint64 output_frames = 0;

    while (iter) {
        output_frames += NextAudioStreamIter(stream, &iter, &resample_offset, NULL, NULL);

        // Already got loads of frames. Just clamp it to something reasonable
        if (output_frames >= SDL_MAX_SINT32) {
            output_frames = SDL_MAX_SINT32;
            break;
        }
    }

    if (out_resample_offset) {
        *out_resample_offset = resample_offset;
    }

    return output_frames;
}

static Sint64 GetAudioStreamHead(SDL_AudioStream* stream, SDL_AudioSpec* out_spec, SDL_bool* out_flushed)
{
    void* iter = SDL_BeginAudioQueueIter(stream->queue);

    if (!iter) {
        SDL_zerop(out_spec);
        *out_flushed = SDL_FALSE;
        return 0;
    }

    Sint64 resample_offset = stream->resample_offset;
    return NextAudioStreamIter(stream, &iter, &resample_offset, out_spec, out_flushed);
}

// You must hold stream->lock and validate your parameters before calling this!
// Enough input data MUST be available!
static int GetAudioStreamDataInternal(SDL_AudioStream *stream, void *buf, int output_frames)
{
    const SDL_AudioSpec* src_spec = &stream->input_spec;
    const SDL_AudioSpec* dst_spec = &stream->dst_spec;

    const SDL_AudioFormat src_format = src_spec->format;
    const int src_channels = src_spec->channels;

    const SDL_AudioFormat dst_format = dst_spec->format;
    const int dst_channels = dst_spec->channels;

    const int max_frame_size = CalculateMaxFrameSize(src_format, src_channels, dst_format, dst_channels);
    const Sint64 resample_rate = GetAudioStreamResampleRate(stream, src_spec->freq, stream->resample_offset);

#if DEBUG_AUDIOSTREAM
    SDL_Log("AUDIOSTREAM: asking for %d frames.", output_frames);
#endif

    SDL_assert(output_frames > 0);

    // Not resampling? It's an easy conversion (and maybe not even that!)
    if (resample_rate == 0) {
        Uint8* work_buffer = NULL;

        // Ensure we have enough scratch space for any conversions
        if ((src_format != dst_format) || (src_channels != dst_channels)) {
            work_buffer = EnsureAudioStreamWorkBufferSize(stream, output_frames * max_frame_size);

            if (!work_buffer) {
                return -1;
            }
        }

        if (SDL_ReadFromAudioQueue(stream->queue, buf, dst_format, dst_channels, 0, output_frames, 0, work_buffer) != buf) {
            return SDL_SetError("Not enough data in queue");
        }

        return 0;
    }

    // Time to do some resampling!
    // Calculate the number of input frames necessary for this request.
    // Because resampling happens "between" frames, The same number of output_frames
    // can require a different number of input_frames, depending on the resample_offset.
    // Infact, input_frames can sometimes even be zero when upsampling.
    const int input_frames = (int) SDL_GetResamplerInputFrames(output_frames, resample_rate, stream->resample_offset);

    const int padding_frames = SDL_GetResamplerPaddingFrames(resample_rate);

    const SDL_AudioFormat resample_format = SDL_AUDIO_F32;

    // If increasing channels, do it after resampling, since we'd just
    // do more work to resample duplicate channels. If we're decreasing, do
    // it first so we resample the interpolated data instead of interpolating
    // the resampled data.
    const int resample_channels = SDL_min(src_channels, dst_channels);

    // The size of the frame used when resampling
    const int resample_frame_size = SDL_AUDIO_BYTESIZE(resample_format) * resample_channels;

    // The main portion of the work_buffer can be used to store 3 things:
    // src_sample_frame_size * (left_padding+input_buffer+right_padding)
    //   resample_frame_size * (left_padding+input_buffer+right_padding)
    // dst_sample_frame_size * output_frames
    //
    // ResampleAudio also requires an additional buffer if it can't write straight to the output:
    //   resample_frame_size * output_frames
    //
    // Note, ConvertAudio requires (num_frames * max_sample_frame_size) of scratch space
    const int work_buffer_frames = input_frames + (padding_frames * 2);
    int work_buffer_capacity = work_buffer_frames * max_frame_size;
    int resample_buffer_offset = -1;

    // Check if we can resample directly into the output buffer.
    // Note, this is just to avoid extra copies.
    // Some other formats may fit directly into the output buffer, but i'd rather process data in a SIMD-aligned buffer.
    if ((dst_format != resample_format) || (dst_channels != resample_channels)) {
        // Allocate space for converting the resampled output to the destination format
        int resample_convert_bytes = output_frames * max_frame_size;
        work_buffer_capacity = SDL_max(work_buffer_capacity, resample_convert_bytes);

        // SIMD-align the buffer
        int simd_alignment = (int) SDL_GetSIMDAlignment();
        work_buffer_capacity += simd_alignment - 1;
        work_buffer_capacity -= work_buffer_capacity % simd_alignment;

        // Allocate space for the resampled output
        int resample_bytes = output_frames * resample_frame_size;
        resample_buffer_offset = work_buffer_capacity;
        work_buffer_capacity += resample_bytes;
    }

    Uint8* work_buffer = EnsureAudioStreamWorkBufferSize(stream, work_buffer_capacity);

    if (!work_buffer) {
        return -1;
    }

    const Uint8* input_buffer = SDL_ReadFromAudioQueue(stream->queue,
        NULL, resample_format, resample_channels,
        padding_frames, input_frames, padding_frames, work_buffer);

    if (!input_buffer) {
        return SDL_SetError("Not enough data in queue (resample)");
    }

    input_buffer += padding_frames * resample_frame_size;

    // Decide where the resampled output goes
    void* resample_buffer = (resample_buffer_offset != -1) ? (work_buffer + resample_buffer_offset) : buf;

    SDL_ResampleAudio(resample_channels,
                  (const float *) input_buffer, input_frames,
                  (float*) resample_buffer, output_frames,
                  resample_rate, &stream->resample_offset);

    // Convert to the final format, if necessary
    ConvertAudio(output_frames, resample_buffer, resample_format, resample_channels, buf, dst_format, dst_channels, work_buffer);

    return 0;
}

// get converted/resampled data from the stream
int SDL_GetAudioStreamData(SDL_AudioStream *stream, void *voidbuf, int len)
{
    Uint8 *buf = (Uint8 *) voidbuf;

#if DEBUG_AUDIOSTREAM
    SDL_Log("AUDIOSTREAM: want to get %d converted bytes", len);
#endif

    if (!stream) {
        return SDL_InvalidParamError("stream");
    } else if (!buf) {
        return SDL_InvalidParamError("buf");
    } else if (len < 0) {
        return SDL_InvalidParamError("len");
    } else if (len == 0) {
        return 0; // nothing to do.
    }

    SDL_LockMutex(stream->lock);

    if (CheckAudioStreamIsFullySetup(stream) != 0) {
        SDL_UnlockMutex(stream->lock);
        return -1;
    }

    const int dst_frame_size = SDL_AUDIO_FRAMESIZE(stream->dst_spec);

    len -= len % dst_frame_size;  // chop off any fractional sample frame.

    // give the callback a chance to fill in more stream data if it wants.
    if (stream->get_callback) {
        Sint64 total_request = len / dst_frame_size;  // start with sample frames desired
        Sint64 additional_request = total_request;

        Sint64 resample_offset = 0;
        Sint64 available_frames = GetAudioStreamAvailableFrames(stream, &resample_offset);

        additional_request -= SDL_min(additional_request, available_frames);

        Sint64 resample_rate = GetAudioStreamResampleRate(stream, stream->src_spec.freq, resample_offset);

        if (resample_rate) {
            total_request = SDL_GetResamplerInputFrames(total_request, resample_rate, resample_offset);
            additional_request = SDL_GetResamplerInputFrames(additional_request, resample_rate, resample_offset);
        }

        total_request *= SDL_AUDIO_FRAMESIZE(stream->src_spec);  // convert sample frames to bytes.
        additional_request *= SDL_AUDIO_FRAMESIZE(stream->src_spec);  // convert sample frames to bytes.
        stream->get_callback(stream->get_callback_userdata, stream, (int) SDL_min(additional_request, SDL_INT_MAX), (int) SDL_min(total_request, SDL_INT_MAX));
    }

    // Process the data in chunks to avoid allocating too much memory (and potential integer overflows)
    const int chunk_size = 4096;

    int total = 0;

    while (total < len) {
        // Audio is processed a track at a time.
        SDL_AudioSpec input_spec;
        SDL_bool flushed;
        const Sint64 available_frames = GetAudioStreamHead(stream, &input_spec, &flushed);

        if (available_frames == 0) {
            if (flushed) {
                SDL_PopAudioQueueHead(stream->queue);
                SDL_zero(stream->input_spec);
                stream->resample_offset = 0;
                continue;
            }
            // There are no frames available, but the track hasn't been flushed, so more might be added later.
            break;
        }

        if (UpdateAudioStreamInputSpec(stream, &input_spec) != 0) {
            total = total ? total : -1;
            break;
        }

        // Clamp the output length to the maximum currently available.
        // GetAudioStreamDataInternal requires enough input data is available.
        int output_frames = (len - total) / dst_frame_size;
        output_frames = SDL_min(output_frames, chunk_size);
        output_frames = (int) SDL_min(output_frames, available_frames);

        if (GetAudioStreamDataInternal(stream, &buf[total], output_frames) != 0) {
            total = total ? total : -1;
            break;
        }

        total += output_frames * dst_frame_size;
    }

    SDL_UnlockMutex(stream->lock);

#if DEBUG_AUDIOSTREAM
    SDL_Log("AUDIOSTREAM: Final result was %d", total);
#endif

    return total;
}

// number of converted/resampled bytes available for output
int SDL_GetAudioStreamAvailable(SDL_AudioStream *stream)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    SDL_LockMutex(stream->lock);

    if (CheckAudioStreamIsFullySetup(stream) != 0) {
        SDL_UnlockMutex(stream->lock);
        return 0;
    }

    Sint64 count = GetAudioStreamAvailableFrames(stream, NULL);

    // convert from sample frames to bytes in destination format.
    count *= SDL_AUDIO_FRAMESIZE(stream->dst_spec);

    SDL_UnlockMutex(stream->lock);

    // if this overflows an int, just clamp it to a maximum.
    return (int) SDL_min(count, SDL_INT_MAX);
}

// number of sample frames that are currently queued as input.
int SDL_GetAudioStreamQueued(SDL_AudioStream *stream)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    SDL_LockMutex(stream->lock);

    size_t total = SDL_GetAudioQueueQueued(stream->queue);

    SDL_UnlockMutex(stream->lock);

    // if this overflows an int, just clamp it to a maximum.
    return (int) SDL_min(total, SDL_INT_MAX);
}

int SDL_ClearAudioStream(SDL_AudioStream *stream)
{
    if (!stream) {
        return SDL_InvalidParamError("stream");
    }

    SDL_LockMutex(stream->lock);

    SDL_ClearAudioQueue(stream->queue);
    SDL_zero(stream->input_spec);
    stream->resample_offset = 0;

    SDL_UnlockMutex(stream->lock);
    return 0;
}

void SDL_DestroyAudioStream(SDL_AudioStream *stream)
{
    if (!stream) {
        return;
    }

    SDL_DestroyProperties(stream->props);

    OnAudioStreamDestroy(stream);

    const SDL_bool simplified = stream->simplified;
    if (simplified) {
        if (stream->bound_device) {
            SDL_assert(stream->bound_device->simplified);
            SDL_CloseAudioDevice(stream->bound_device->instance_id);  // this will unbind the stream.
        }
    } else {
        SDL_UnbindAudioStream(stream);
    }

    SDL_aligned_free(stream->work_buffer);
    SDL_DestroyAudioQueue(stream->queue);
    SDL_DestroyMutex(stream->lock);

    SDL_free(stream);
}

static void SDLCALL DontFreeThisAudioBuffer(void *userdata, const void *buf, int len)
{
    // We don't own the buffer, but know it will outlive the stream
}

int SDL_ConvertAudioSamples(const SDL_AudioSpec *src_spec, const Uint8 *src_data, int src_len,
                            const SDL_AudioSpec *dst_spec, Uint8 **dst_data, int *dst_len)
{
    if (dst_data) {
        *dst_data = NULL;
    }

    if (dst_len) {
        *dst_len = 0;
    }

    if (!src_data) {
        return SDL_InvalidParamError("src_data");
    } else if (src_len < 0) {
        return SDL_InvalidParamError("src_len");
    } else if (!dst_data) {
        return SDL_InvalidParamError("dst_data");
    } else if (!dst_len) {
        return SDL_InvalidParamError("dst_len");
    }

    int retval = -1;
    Uint8 *dst = NULL;
    int dstlen = 0;

    SDL_AudioStream *stream = SDL_CreateAudioStream(src_spec, dst_spec);
    if (stream) {
        if ((PutAudioStreamBuffer(stream, src_data, src_len, DontFreeThisAudioBuffer, NULL) == 0) && (SDL_FlushAudioStream(stream) == 0)) {
            dstlen = SDL_GetAudioStreamAvailable(stream);
            if (dstlen >= 0) {
                dst = (Uint8 *)SDL_malloc(dstlen);
                if (dst) {
                    retval = (SDL_GetAudioStreamData(stream, dst, dstlen) == dstlen) ? 0 : -1;
                }
            }
        }
    }

    if (retval == -1) {
        SDL_free(dst);
    } else {
        *dst_data = dst;
        *dst_len = dstlen;
    }

    SDL_DestroyAudioStream(stream);
    return retval;
}