path: root/media/libcubeb/src/cubeb_winmm.c

/*
 * Copyright © 2011 Mozilla Foundation
 *
 * This program is made available under an ISC-style license.  See the
 * accompanying file LICENSE for details.
 */
#undef WINVER
#define WINVER 0x0501
#undef WIN32_LEAN_AND_MEAN

#include "cubeb-internal.h"
#include "cubeb/cubeb.h"
#include <malloc.h>
#include <math.h>
#include <process.h>
#include <stdio.h>
#include <stdlib.h>
#include <windows.h>

/* clang-format off */
/* These need to be included after windows.h */
#include <mmreg.h>
#include <mmsystem.h>
/* clang-format on */

/* This is missing from the MinGW headers. Use a safe fallback. */
#if !defined(MEMORY_ALLOCATION_ALIGNMENT)
#define MEMORY_ALLOCATION_ALIGNMENT 16
#endif

/**This is also missing from the MinGW headers. It  also appears to be
 * undocumented by Microsoft.*/
#ifndef WAVE_FORMAT_48M08
#define WAVE_FORMAT_48M08 0x00001000 /* 48     kHz, Mono, 8-bit */
#endif
#ifndef WAVE_FORMAT_48M16
#define WAVE_FORMAT_48M16 0x00002000 /* 48     kHz, Mono, 16-bit */
#endif
#ifndef WAVE_FORMAT_48S08
#define WAVE_FORMAT_48S08 0x00004000 /* 48     kHz, Stereo, 8-bit */
#endif
#ifndef WAVE_FORMAT_48S16
#define WAVE_FORMAT_48S16 0x00008000 /* 48     kHz, Stereo, 16-bit */
#endif
#ifndef WAVE_FORMAT_96M08
#define WAVE_FORMAT_96M08 0x00010000 /* 96     kHz, Mono, 8-bit */
#endif
#ifndef WAVE_FORMAT_96M16
#define WAVE_FORMAT_96M16 0x00020000 /* 96     kHz, Mono, 16-bit */
#endif
#ifndef WAVE_FORMAT_96S08
#define WAVE_FORMAT_96S08 0x00040000 /* 96     kHz, Stereo, 8-bit */
#endif
#ifndef WAVE_FORMAT_96S16
#define WAVE_FORMAT_96S16 0x00080000 /* 96     kHz, Stereo, 16-bit */
#endif

/**Taken from winbase.h, also not in MinGW.*/
#ifndef STACK_SIZE_PARAM_IS_A_RESERVATION
#define STACK_SIZE_PARAM_IS_A_RESERVATION 0x00010000 // Threads only
#endif

#ifndef DRVM_MAPPER
#define DRVM_MAPPER (0x2000)
#endif
#ifndef DRVM_MAPPER_PREFERRED_GET
#define DRVM_MAPPER_PREFERRED_GET (DRVM_MAPPER + 21)
#endif
#ifndef DRVM_MAPPER_CONSOLEVOICECOM_GET
#define DRVM_MAPPER_CONSOLEVOICECOM_GET (DRVM_MAPPER + 23)
#endif

#define CUBEB_STREAM_MAX 32
#define NBUFS 4

struct cubeb_stream_item {
  SLIST_ENTRY head;
  cubeb_stream * stream;
};

static struct cubeb_ops const winmm_ops;

struct cubeb {
  struct cubeb_ops const * ops;
  HANDLE event;
  HANDLE thread;
  int shutdown;
  PSLIST_HEADER work;
  CRITICAL_SECTION lock;
  unsigned int active_streams;
  unsigned int minimum_latency_ms;
};

struct cubeb_stream {
  /* Note: Must match cubeb_stream layout in cubeb.c. */
  cubeb * context;
  void * user_ptr;
  /**/
  cubeb_stream_params params;
  cubeb_data_callback data_callback;
  cubeb_state_callback state_callback;
  WAVEHDR buffers[NBUFS];
  size_t buffer_size;
  int next_buffer;
  int free_buffers;
  int shutdown;
  int draining;
  int error;
  HANDLE event;
  HWAVEOUT waveout;
  CRITICAL_SECTION lock;
  uint64_t written;
  /* number of frames written during preroll */
  uint64_t position_base;
  float soft_volume;
  /* For position wrap-around handling: */
  size_t frame_size;
  DWORD prev_pos_lo_dword;
  DWORD pos_hi_dword;
};

static size_t
bytes_per_frame(cubeb_stream_params params)
{
  size_t bytes;

  switch (params.format) {
  case CUBEB_SAMPLE_S16LE:
    bytes = sizeof(signed short);
    break;
  case CUBEB_SAMPLE_FLOAT32LE:
    bytes = sizeof(float);
    break;
  default:
    XASSERT(0);
  }

  return bytes * params.channels;
}

static WAVEHDR *
winmm_get_next_buffer(cubeb_stream * stm)
{
  WAVEHDR * hdr = NULL;

  XASSERT(stm->free_buffers > 0 && stm->free_buffers <= NBUFS);
  hdr = &stm->buffers[stm->next_buffer];
  XASSERT(hdr->dwFlags & WHDR_PREPARED ||
          (hdr->dwFlags & WHDR_DONE && !(hdr->dwFlags & WHDR_INQUEUE)));
  stm->next_buffer = (stm->next_buffer + 1) % NBUFS;
  stm->free_buffers -= 1;

  return hdr;
}

static long
preroll_callback(cubeb_stream * stream, void * user, const void * inputbuffer,
                 void * outputbuffer, long nframes)
{
  memset((uint8_t *)outputbuffer, 0, nframes * bytes_per_frame(stream->params));
  return nframes;
}

static void
winmm_refill_stream(cubeb_stream * stm)
{
  WAVEHDR * hdr;
  long got;
  long wanted;
  MMRESULT r;

  ALOG("winmm_refill_stream");

  EnterCriticalSection(&stm->lock);
  if (stm->error) {
    LeaveCriticalSection(&stm->lock);
    return;
  }
  stm->free_buffers += 1;
  XASSERT(stm->free_buffers > 0 && stm->free_buffers <= NBUFS);

  if (stm->draining) {
    LeaveCriticalSection(&stm->lock);
    if (stm->free_buffers == NBUFS) {
      ALOG("winmm_refill_stream draining");
      stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_DRAINED);
    }
    SetEvent(stm->event);
    return;
  }

  if (stm->shutdown) {
    LeaveCriticalSection(&stm->lock);
    SetEvent(stm->event);
    return;
  }

  hdr = winmm_get_next_buffer(stm);

  wanted = (DWORD)stm->buffer_size / bytes_per_frame(stm->params);

  /* It is assumed that the caller is holding this lock.  It must be dropped
     during the callback to avoid deadlocks. */
  LeaveCriticalSection(&stm->lock);
  got = stm->data_callback(stm, stm->user_ptr, NULL, hdr->lpData, wanted);
  EnterCriticalSection(&stm->lock);
  if (got < 0) {
    stm->error = 1;
    LeaveCriticalSection(&stm->lock);
    SetEvent(stm->event);
    stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
    return;
  } else if (got < wanted) {
    stm->draining = 1;
  }
  stm->written += got;

  XASSERT(hdr->dwFlags & WHDR_PREPARED);

  hdr->dwBufferLength = got * bytes_per_frame(stm->params);
  XASSERT(hdr->dwBufferLength <= stm->buffer_size);

  if (stm->soft_volume != -1.0) {
    if (stm->params.format == CUBEB_SAMPLE_FLOAT32NE) {
      float * b = (float *)hdr->lpData;
      uint32_t i;
      for (i = 0; i < got * stm->params.channels; i++) {
        b[i] *= stm->soft_volume;
      }
    } else {
      short * b = (short *)hdr->lpData;
      uint32_t i;
      for (i = 0; i < got * stm->params.channels; i++) {
        b[i] = (short)(b[i] * stm->soft_volume);
      }
    }
  }

  r = waveOutWrite(stm->waveout, hdr, sizeof(*hdr));
  if (r != MMSYSERR_NOERROR) {
    LeaveCriticalSection(&stm->lock);
    stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
    return;
  }

  ALOG("winmm_refill_stream %ld frames", got);

  LeaveCriticalSection(&stm->lock);
}

static unsigned __stdcall winmm_buffer_thread(void * user_ptr)
{
  cubeb * ctx = (cubeb *)user_ptr;
  XASSERT(ctx);

  for (;;) {
    DWORD r;
    PSLIST_ENTRY item;

    r = WaitForSingleObject(ctx->event, INFINITE);
    XASSERT(r == WAIT_OBJECT_0);

    /* Process work items in batches so that a single stream can't
       starve the others by continuously adding new work to the top of
       the work item stack. */
    item = InterlockedFlushSList(ctx->work);
    while (item != NULL) {
      PSLIST_ENTRY tmp = item;
      winmm_refill_stream(((struct cubeb_stream_item *)tmp)->stream);
      item = item->Next;
      _aligned_free(tmp);
    }

    if (ctx->shutdown) {
      break;
    }
  }

  return 0;
}

static void CALLBACK
winmm_buffer_callback(HWAVEOUT waveout, UINT msg, DWORD_PTR user_ptr,
                      DWORD_PTR p1, DWORD_PTR p2)
{
  cubeb_stream * stm = (cubeb_stream *)user_ptr;
  struct cubeb_stream_item * item;

  if (msg != WOM_DONE) {
    return;
  }

  item = _aligned_malloc(sizeof(struct cubeb_stream_item),
                         MEMORY_ALLOCATION_ALIGNMENT);
  XASSERT(item);
  item->stream = stm;
  InterlockedPushEntrySList(stm->context->work, &item->head);

  SetEvent(stm->context->event);
}

static unsigned int
calculate_minimum_latency(void)
{
  OSVERSIONINFOEX osvi;
  DWORDLONG mask;

  /* Running under Terminal Services results in underruns with low latency. */
  if (GetSystemMetrics(SM_REMOTESESSION) == TRUE) {
    return 500;
  }

  /* Vista's WinMM implementation underruns when less than 200ms of audio is
   * buffered. */
  memset(&osvi, 0, sizeof(OSVERSIONINFOEX));
  osvi.dwOSVersionInfoSize = sizeof(OSVERSIONINFOEX);
  osvi.dwMajorVersion = 6;
  osvi.dwMinorVersion = 0;

  mask = 0;
  VER_SET_CONDITION(mask, VER_MAJORVERSION, VER_EQUAL);
  VER_SET_CONDITION(mask, VER_MINORVERSION, VER_EQUAL);

  if (VerifyVersionInfo(&osvi, VER_MAJORVERSION | VER_MINORVERSION, mask) !=
      0) {
    return 200;
  }

  return 100;
}

static void
winmm_destroy(cubeb * ctx);

/*static*/ int
winmm_init(cubeb ** context, char const * context_name)
{
  cubeb * ctx;

  XASSERT(context);
  *context = NULL;

  /* Don't initialize a context if there are no devices available. */
  if (waveOutGetNumDevs() == 0) {
    return CUBEB_ERROR;
  }

  ctx = calloc(1, sizeof(*ctx));
  XASSERT(ctx);

  ctx->ops = &winmm_ops;

  ctx->work = _aligned_malloc(sizeof(*ctx->work), MEMORY_ALLOCATION_ALIGNMENT);
  XASSERT(ctx->work);
  InitializeSListHead(ctx->work);

  ctx->event = CreateEvent(NULL, FALSE, FALSE, NULL);
  if (!ctx->event) {
    winmm_destroy(ctx);
    return CUBEB_ERROR;
  }

  ctx->thread =
      (HANDLE)_beginthreadex(NULL, 256 * 1024, winmm_buffer_thread, ctx,
                             STACK_SIZE_PARAM_IS_A_RESERVATION, NULL);
  if (!ctx->thread) {
    winmm_destroy(ctx);
    return CUBEB_ERROR;
  }

  SetThreadPriority(ctx->thread, THREAD_PRIORITY_TIME_CRITICAL);

  InitializeCriticalSection(&ctx->lock);
  ctx->active_streams = 0;

  ctx->minimum_latency_ms = calculate_minimum_latency();

  *context = ctx;

  return CUBEB_OK;
}

static char const *
winmm_get_backend_id(cubeb * ctx)
{
  return "winmm";
}

static void
winmm_destroy(cubeb * ctx)
{
  DWORD r;

  XASSERT(ctx->active_streams == 0);
  XASSERT(!InterlockedPopEntrySList(ctx->work));

  DeleteCriticalSection(&ctx->lock);

  if (ctx->thread) {
    ctx->shutdown = 1;
    SetEvent(ctx->event);
    r = WaitForSingleObject(ctx->thread, INFINITE);
    XASSERT(r == WAIT_OBJECT_0);
    CloseHandle(ctx->thread);
  }

  if (ctx->event) {
    CloseHandle(ctx->event);
  }

  _aligned_free(ctx->work);

  free(ctx);
}

static void
winmm_stream_destroy(cubeb_stream * stm);

static int
winmm_stream_init(cubeb * context, cubeb_stream ** stream,
                  char const * stream_name, cubeb_devid input_device,
                  cubeb_stream_params * input_stream_params,
                  cubeb_devid output_device,
                  cubeb_stream_params * output_stream_params,
                  unsigned int latency_frames,
                  cubeb_data_callback data_callback,
                  cubeb_state_callback state_callback, void * user_ptr)
{
  MMRESULT r;
  WAVEFORMATEXTENSIBLE wfx;
  cubeb_stream * stm;
  int i;
  size_t bufsz;

  XASSERT(context);
  XASSERT(stream);
  XASSERT(output_stream_params);

  if (input_stream_params) {
    /* Capture support not yet implemented. */
    return CUBEB_ERROR_NOT_SUPPORTED;
  }

  if (input_device || output_device) {
    /* Device selection not yet implemented. */
    return CUBEB_ERROR_DEVICE_UNAVAILABLE;
  }

  if (output_stream_params->prefs & CUBEB_STREAM_PREF_LOOPBACK) {
    /* Loopback is not supported */
    return CUBEB_ERROR_NOT_SUPPORTED;
  }

  *stream = NULL;

  memset(&wfx, 0, sizeof(wfx));
  if (output_stream_params->channels > 2) {
    wfx.Format.wFormatTag = WAVE_FORMAT_EXTENSIBLE;
    wfx.Format.cbSize = sizeof(wfx) - sizeof(wfx.Format);
  } else {
    wfx.Format.wFormatTag = WAVE_FORMAT_PCM;
    if (output_stream_params->format == CUBEB_SAMPLE_FLOAT32LE) {
      wfx.Format.wFormatTag = WAVE_FORMAT_IEEE_FLOAT;
    }
    wfx.Format.cbSize = 0;
  }
  wfx.Format.nChannels = output_stream_params->channels;
  wfx.Format.nSamplesPerSec = output_stream_params->rate;

  /* XXX fix channel mappings */
  wfx.dwChannelMask = SPEAKER_FRONT_LEFT | SPEAKER_FRONT_RIGHT;

  switch (output_stream_params->format) {
  case CUBEB_SAMPLE_S16LE:
    wfx.Format.wBitsPerSample = 16;
    wfx.SubFormat = KSDATAFORMAT_SUBTYPE_PCM;
    break;
  case CUBEB_SAMPLE_FLOAT32LE:
    wfx.Format.wBitsPerSample = 32;
    wfx.SubFormat = KSDATAFORMAT_SUBTYPE_IEEE_FLOAT;
    break;
  default:
    return CUBEB_ERROR_INVALID_FORMAT;
  }

  wfx.Format.nBlockAlign =
      (wfx.Format.wBitsPerSample * wfx.Format.nChannels) / 8;
  wfx.Format.nAvgBytesPerSec =
      wfx.Format.nSamplesPerSec * wfx.Format.nBlockAlign;
  wfx.Samples.wValidBitsPerSample = wfx.Format.wBitsPerSample;

  EnterCriticalSection(&context->lock);
  /* CUBEB_STREAM_MAX is a horrible hack to avoid a situation where, when
     many streams are active at once, a subset of them will not consume (via
     playback) or release (via waveOutReset) their buffers. */
  if (context->active_streams >= CUBEB_STREAM_MAX) {
    LeaveCriticalSection(&context->lock);
    return CUBEB_ERROR;
  }
  context->active_streams += 1;
  LeaveCriticalSection(&context->lock);

  stm = calloc(1, sizeof(*stm));
  XASSERT(stm);

  stm->context = context;

  stm->params = *output_stream_params;

  // Data callback is set to the user-provided data callback after
  // the initialization and potential preroll callback calls are done, because
  // cubeb users don't expect the data callback to be called during
  // initialization.
  stm->data_callback = preroll_callback;
  stm->state_callback = state_callback;
  stm->user_ptr = user_ptr;
  stm->written = 0;

  uint32_t latency_ms = latency_frames * 1000 / output_stream_params->rate;

  if (latency_ms < context->minimum_latency_ms) {
    latency_ms = context->minimum_latency_ms;
  }

  bufsz = (size_t)(stm->params.rate / 1000.0 * latency_ms *
                   bytes_per_frame(stm->params) / NBUFS);
  if (bufsz % bytes_per_frame(stm->params) != 0) {
    bufsz +=
        bytes_per_frame(stm->params) - (bufsz % bytes_per_frame(stm->params));
  }
  XASSERT(bufsz % bytes_per_frame(stm->params) == 0);

  stm->buffer_size = bufsz;

  InitializeCriticalSection(&stm->lock);

  stm->event = CreateEvent(NULL, FALSE, FALSE, NULL);
  if (!stm->event) {
    winmm_stream_destroy(stm);
    return CUBEB_ERROR;
  }

  stm->soft_volume = -1.0;

  /* winmm_buffer_callback will be called during waveOutOpen, so all
     other initialization must be complete before calling it. */
  r = waveOutOpen(&stm->waveout, WAVE_MAPPER, &wfx.Format,
                  (DWORD_PTR)winmm_buffer_callback, (DWORD_PTR)stm,
                  CALLBACK_FUNCTION);
  if (r != MMSYSERR_NOERROR) {
    winmm_stream_destroy(stm);
    return CUBEB_ERROR;
  }

  r = waveOutPause(stm->waveout);
  if (r != MMSYSERR_NOERROR) {
    winmm_stream_destroy(stm);
    return CUBEB_ERROR;
  }

  for (i = 0; i < NBUFS; ++i) {
    WAVEHDR * hdr = &stm->buffers[i];

    hdr->lpData = calloc(1, bufsz);
    XASSERT(hdr->lpData);
    hdr->dwBufferLength = bufsz;
    hdr->dwFlags = 0;

    r = waveOutPrepareHeader(stm->waveout, hdr, sizeof(*hdr));
    if (r != MMSYSERR_NOERROR) {
      winmm_stream_destroy(stm);
      return CUBEB_ERROR;
    }

    winmm_refill_stream(stm);
  }

  stm->frame_size = bytes_per_frame(stm->params);
  stm->prev_pos_lo_dword = 0;
  stm->pos_hi_dword = 0;
  // Set the user data callback now that preroll has finished.
  stm->data_callback = data_callback;
  stm->position_base = 0;

  // Offset the position by the number of frames written during preroll.
  stm->position_base = stm->written;
  stm->written = 0;

  *stream = stm;

  LOG("winmm_stream_init OK");

  return CUBEB_OK;
}

static void
winmm_stream_destroy(cubeb_stream * stm)
{
  int i;

  if (stm->waveout) {
    MMTIME time;
    MMRESULT r;
    int device_valid;
    int enqueued;

    EnterCriticalSection(&stm->lock);
    stm->shutdown = 1;

    waveOutReset(stm->waveout);

    /* Don't need this value, we just want the result to detect invalid
       handle/no device errors than waveOutReset doesn't seem to report. */
    time.wType = TIME_SAMPLES;
    r = waveOutGetPosition(stm->waveout, &time, sizeof(time));
    device_valid = !(r == MMSYSERR_INVALHANDLE || r == MMSYSERR_NODRIVER);

    enqueued = NBUFS - stm->free_buffers;
    LeaveCriticalSection(&stm->lock);

    /* Wait for all blocks to complete. */
    while (device_valid && enqueued > 0 && !stm->error) {
      DWORD rv = WaitForSingleObject(stm->event, INFINITE);
      XASSERT(rv == WAIT_OBJECT_0);

      EnterCriticalSection(&stm->lock);
      enqueued = NBUFS - stm->free_buffers;
      LeaveCriticalSection(&stm->lock);
    }

    EnterCriticalSection(&stm->lock);

    for (i = 0; i < NBUFS; ++i) {
      if (stm->buffers[i].dwFlags & WHDR_PREPARED) {
        waveOutUnprepareHeader(stm->waveout, &stm->buffers[i],
                               sizeof(stm->buffers[i]));
      }
    }

    waveOutClose(stm->waveout);

    LeaveCriticalSection(&stm->lock);
  }

  if (stm->event) {
    CloseHandle(stm->event);
  }

  DeleteCriticalSection(&stm->lock);

  for (i = 0; i < NBUFS; ++i) {
    free(stm->buffers[i].lpData);
  }

  EnterCriticalSection(&stm->context->lock);
  XASSERT(stm->context->active_streams >= 1);
  stm->context->active_streams -= 1;
  LeaveCriticalSection(&stm->context->lock);

  free(stm);
}

static int
winmm_get_max_channel_count(cubeb * ctx, uint32_t * max_channels)
{
  XASSERT(ctx && max_channels);

  /* We don't support more than two channels in this backend. */
  *max_channels = 2;

  return CUBEB_OK;
}

static int
winmm_get_min_latency(cubeb * ctx, cubeb_stream_params params,
                      uint32_t * latency)
{
  // 100ms minimum, if we are not in a bizarre configuration.
  *latency = ctx->minimum_latency_ms * params.rate / 1000;

  return CUBEB_OK;
}

static int
winmm_get_preferred_sample_rate(cubeb * ctx, uint32_t * rate)
{
  WAVEOUTCAPS woc;
  MMRESULT r;

  r = waveOutGetDevCaps(WAVE_MAPPER, &woc, sizeof(WAVEOUTCAPS));
  if (r != MMSYSERR_NOERROR) {
    return CUBEB_ERROR;
  }

  /* Check if we support 48kHz, but not 44.1kHz. */
  if (!(woc.dwFormats & WAVE_FORMAT_4S16) &&
      woc.dwFormats & WAVE_FORMAT_48S16) {
    *rate = 48000;
    return CUBEB_OK;
  }
  /* Prefer 44.1kHz between 44.1kHz and 48kHz. */
  *rate = 44100;

  return CUBEB_OK;
}

static int
winmm_stream_start(cubeb_stream * stm)
{
  MMRESULT r;

  EnterCriticalSection(&stm->lock);
  r = waveOutRestart(stm->waveout);
  LeaveCriticalSection(&stm->lock);

  if (r != MMSYSERR_NOERROR) {
    return CUBEB_ERROR;
  }

  stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_STARTED);

  return CUBEB_OK;
}

static int
winmm_stream_stop(cubeb_stream * stm)
{
  MMRESULT r;

  EnterCriticalSection(&stm->lock);
  r = waveOutPause(stm->waveout);
  LeaveCriticalSection(&stm->lock);

  if (r != MMSYSERR_NOERROR) {
    return CUBEB_ERROR;
  }

  stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_STOPPED);

  return CUBEB_OK;
}

/*
Microsoft wave audio docs say "samples are the preferred time format in which
to represent the current position", but relying on this causes problems on
Windows XP, the only OS cubeb_winmm is used on.

While the wdmaud.sys driver internally tracks a 64-bit position and ensures no
backward movement, the WinMM API limits the position returned from
waveOutGetPosition() to a 32-bit DWORD (this applies equally to XP x64). The
higher 32 bits are chopped off, and to an API consumer the position can appear
to move backward.

In theory, even a 32-bit TIME_SAMPLES position should provide plenty of
playback time for typical use cases before this pseudo wrap-around, e.g:
    (2^32 - 1)/48000 = ~24:51:18 for 48.0 kHz stereo;
    (2^32 - 1)/44100 = ~27:03:12 for 44.1 kHz stereo.
In reality, wdmaud.sys doesn't provide a TIME_SAMPLES position at all, only a
32-bit TIME_BYTES position, from which wdmaud.drv derives TIME_SAMPLES:
    SamplePos = (BytePos * 8) / BitsPerFrame,
    where BitsPerFrame = Channels * BitsPerSample,
Per dom\media\AudioSampleFormat.h, desktop builds always use 32-bit FLOAT32
samples, so the maximum for TIME_SAMPLES should be:
    (2^29 - 1)/48000 = ~03:06:25;
    (2^29 - 1)/44100 = ~03:22:54.
This might still be OK for typical browser usage, but there's also a bug in the
formula above: BytePos * 8 (BytePos << 3) is done on a 32-bit BytePos, without
first casting it to 64 bits, so the highest 3 bits, if set, would get shifted
out, and the maximum possible TIME_SAMPLES drops unacceptably low:
    (2^26 - 1)/48000 = ~00:23:18;
    (2^26 - 1)/44100 = ~00:25:22.

To work around these limitations, we just get the position in TIME_BYTES,
recover the 64-bit value, and do our own conversion to samples.
*/

/* Convert chopped 32-bit waveOutGetPosition() into 64-bit true position. */
static uint64_t
update_64bit_position(cubeb_stream * stm, DWORD pos_lo_dword)
{
  /* Caller should be holding stm->lock. */
  if (pos_lo_dword < stm->prev_pos_lo_dword) {
    stm->pos_hi_dword++;
    LOG("waveOutGetPosition() has wrapped around: %#lx -> %#lx",
        stm->prev_pos_lo_dword, pos_lo_dword);
    LOG("Wrap-around count = %#lx", stm->pos_hi_dword);
    LOG("Current 64-bit position = %#llx",
        (((uint64_t)stm->pos_hi_dword) << 32) | ((uint64_t)pos_lo_dword));
  }
  stm->prev_pos_lo_dword = pos_lo_dword;

  return (((uint64_t)stm->pos_hi_dword) << 32) | ((uint64_t)pos_lo_dword);
}

static int
winmm_stream_get_position(cubeb_stream * stm, uint64_t * position)
{
  MMRESULT r;
  MMTIME time;

  EnterCriticalSection(&stm->lock);
  /* See the long comment above for why not just use TIME_SAMPLES here. */
  time.wType = TIME_BYTES;
  r = waveOutGetPosition(stm->waveout, &time, sizeof(time));

  if (r != MMSYSERR_NOERROR || time.wType != TIME_BYTES) {
    LeaveCriticalSection(&stm->lock);
    return CUBEB_ERROR;
  }

  uint64_t position_not_adjusted =
      update_64bit_position(stm, time.u.cb) / stm->frame_size;

  // Subtract the number of frames that were written while prerolling, during
  // initialization.
  if (position_not_adjusted < stm->position_base) {
    *position = 0;
  } else {
    *position = position_not_adjusted - stm->position_base;
  }

  LeaveCriticalSection(&stm->lock);

  return CUBEB_OK;
}

static int
winmm_stream_get_latency(cubeb_stream * stm, uint32_t * latency)
{
  MMRESULT r;
  MMTIME time;
  uint64_t written, position;

  int rv = winmm_stream_get_position(stm, &position);
  if (rv != CUBEB_OK) {
    return rv;
  }

  EnterCriticalSection(&stm->lock);
  written = stm->written;
  LeaveCriticalSection(&stm->lock);

  XASSERT((written - (position / stm->frame_size)) <= UINT32_MAX);
  *latency = (uint32_t)(written - (position / stm->frame_size));

  return CUBEB_OK;
}

static int
winmm_stream_set_volume(cubeb_stream * stm, float volume)
{
  EnterCriticalSection(&stm->lock);
  stm->soft_volume = volume;
  LeaveCriticalSection(&stm->lock);
  return CUBEB_OK;
}

#define MM_11025HZ_MASK                                                        \
  (WAVE_FORMAT_1M08 | WAVE_FORMAT_1M16 | WAVE_FORMAT_1S08 | WAVE_FORMAT_1S16)
#define MM_22050HZ_MASK                                                        \
  (WAVE_FORMAT_2M08 | WAVE_FORMAT_2M16 | WAVE_FORMAT_2S08 | WAVE_FORMAT_2S16)
#define MM_44100HZ_MASK                                                        \
  (WAVE_FORMAT_4M08 | WAVE_FORMAT_4M16 | WAVE_FORMAT_4S08 | WAVE_FORMAT_4S16)
#define MM_48000HZ_MASK                                                        \
  (WAVE_FORMAT_48M08 | WAVE_FORMAT_48M16 | WAVE_FORMAT_48S08 |                 \
   WAVE_FORMAT_48S16)
#define MM_96000HZ_MASK                                                        \
  (WAVE_FORMAT_96M08 | WAVE_FORMAT_96M16 | WAVE_FORMAT_96S08 |                 \
   WAVE_FORMAT_96S16)
static void
winmm_calculate_device_rate(cubeb_device_info * info, DWORD formats)
{
  if (formats & MM_11025HZ_MASK) {
    info->min_rate = 11025;
    info->default_rate = 11025;
    info->max_rate = 11025;
  }
  if (formats & MM_22050HZ_MASK) {
    if (info->min_rate == 0)
      info->min_rate = 22050;
    info->max_rate = 22050;
    info->default_rate = 22050;
  }
  if (formats & MM_44100HZ_MASK) {
    if (info->min_rate == 0)
      info->min_rate = 44100;
    info->max_rate = 44100;
    info->default_rate = 44100;
  }
  if (formats & MM_48000HZ_MASK) {
    if (info->min_rate == 0)
      info->min_rate = 48000;
    info->max_rate = 48000;
    info->default_rate = 48000;
  }
  if (formats & MM_96000HZ_MASK) {
    if (info->min_rate == 0) {
      info->min_rate = 96000;
      info->default_rate = 96000;
    }
    info->max_rate = 96000;
  }
}

#define MM_S16_MASK                                                            \
  (WAVE_FORMAT_1M16 | WAVE_FORMAT_1S16 | WAVE_FORMAT_2M16 | WAVE_FORMAT_2S16 | \
   WAVE_FORMAT_4M16 | WAVE_FORMAT_4S16 | WAVE_FORMAT_48M16 |                   \
   WAVE_FORMAT_48S16 | WAVE_FORMAT_96M16 | WAVE_FORMAT_96S16)
static int
winmm_query_supported_formats(UINT devid, DWORD formats,
                              cubeb_device_fmt * supfmt,
                              cubeb_device_fmt * deffmt)
{
  WAVEFORMATEXTENSIBLE wfx;

  if (formats & MM_S16_MASK)
    *deffmt = *supfmt = CUBEB_DEVICE_FMT_S16LE;
  else
    *deffmt = *supfmt = 0;

  ZeroMemory(&wfx, sizeof(WAVEFORMATEXTENSIBLE));
  wfx.Format.wFormatTag = WAVE_FORMAT_EXTENSIBLE;
  wfx.Format.nChannels = 2;
  wfx.Format.nSamplesPerSec = 44100;
  wfx.Format.wBitsPerSample = 32;
  wfx.Format.nBlockAlign =
      (wfx.Format.wBitsPerSample * wfx.Format.nChannels) / 8;
  wfx.Format.nAvgBytesPerSec =
      wfx.Format.nSamplesPerSec * wfx.Format.nBlockAlign;
  wfx.Format.cbSize = 22;
  wfx.Samples.wValidBitsPerSample = wfx.Format.wBitsPerSample;
  wfx.dwChannelMask = SPEAKER_FRONT_LEFT | SPEAKER_FRONT_RIGHT;
  wfx.SubFormat = KSDATAFORMAT_SUBTYPE_IEEE_FLOAT;
  if (waveOutOpen(NULL, devid, &wfx.Format, 0, 0, WAVE_FORMAT_QUERY) ==
      MMSYSERR_NOERROR)
    *supfmt = (cubeb_device_fmt)(*supfmt | CUBEB_DEVICE_FMT_F32LE);

  return (*deffmt != 0) ? CUBEB_OK : CUBEB_ERROR;
}

static char *
guid_to_cstr(LPGUID guid)
{
  char * ret = malloc(40);
  if (!ret) {
    return NULL;
  }
  _snprintf_s(ret, 40, _TRUNCATE,
              "{%08X-%04X-%04X-%02X%02X-%02X%02X%02X%02X%02X%02X}", guid->Data1,
              guid->Data2, guid->Data3, guid->Data4[0], guid->Data4[1],
              guid->Data4[2], guid->Data4[3], guid->Data4[4], guid->Data4[5],
              guid->Data4[6], guid->Data4[7]);
  return ret;
}

static cubeb_device_pref
winmm_query_preferred_out_device(UINT devid)
{
  DWORD mmpref = WAVE_MAPPER, compref = WAVE_MAPPER, status;
  cubeb_device_pref ret = CUBEB_DEVICE_PREF_NONE;

  if (waveOutMessage((HWAVEOUT)WAVE_MAPPER, DRVM_MAPPER_PREFERRED_GET,
                     (DWORD_PTR)&mmpref,
                     (DWORD_PTR)&status) == MMSYSERR_NOERROR &&
      devid == mmpref)
    ret |= CUBEB_DEVICE_PREF_MULTIMEDIA | CUBEB_DEVICE_PREF_NOTIFICATION;

  if (waveOutMessage((HWAVEOUT)WAVE_MAPPER, DRVM_MAPPER_CONSOLEVOICECOM_GET,
                     (DWORD_PTR)&compref,
                     (DWORD_PTR)&status) == MMSYSERR_NOERROR &&
      devid == compref)
    ret |= CUBEB_DEVICE_PREF_VOICE;

  return ret;
}

static char *
device_id_idx(UINT devid)
{
  char * ret = malloc(16);
  if (!ret) {
    return NULL;
  }
  _snprintf_s(ret, 16, _TRUNCATE, "%u", devid);
  return ret;
}

static void
winmm_create_device_from_outcaps2(cubeb_device_info * ret, LPWAVEOUTCAPS2A caps,
                                  UINT devid)
{
  XASSERT(ret);
  ret->devid = (cubeb_devid)devid;
  ret->device_id = device_id_idx(devid);
  ret->friendly_name = _strdup(caps->szPname);
  ret->group_id = guid_to_cstr(&caps->ProductGuid);
  ret->vendor_name = guid_to_cstr(&caps->ManufacturerGuid);

  ret->type = CUBEB_DEVICE_TYPE_OUTPUT;
  ret->state = CUBEB_DEVICE_STATE_ENABLED;
  ret->preferred = winmm_query_preferred_out_device(devid);

  ret->max_channels = caps->wChannels;
  winmm_calculate_device_rate(ret, caps->dwFormats);
  winmm_query_supported_formats(devid, caps->dwFormats, &ret->format,
                                &ret->default_format);

  /* Hardcoded latency estimates... */
  ret->latency_lo = 100 * ret->default_rate / 1000;
  ret->latency_hi = 200 * ret->default_rate / 1000;
}

static void
winmm_create_device_from_outcaps(cubeb_device_info * ret, LPWAVEOUTCAPSA caps,
                                 UINT devid)
{
  XASSERT(ret);
  ret->devid = (cubeb_devid)devid;
  ret->device_id = device_id_idx(devid);
  ret->friendly_name = _strdup(caps->szPname);
  ret->group_id = NULL;
  ret->vendor_name = NULL;

  ret->type = CUBEB_DEVICE_TYPE_OUTPUT;
  ret->state = CUBEB_DEVICE_STATE_ENABLED;
  ret->preferred = winmm_query_preferred_out_device(devid);

  ret->max_channels = caps->wChannels;
  winmm_calculate_device_rate(ret, caps->dwFormats);
  winmm_query_supported_formats(devid, caps->dwFormats, &ret->format,
                                &ret->default_format);

  /* Hardcoded latency estimates... */
  ret->latency_lo = 100 * ret->default_rate / 1000;
  ret->latency_hi = 200 * ret->default_rate / 1000;
}

static cubeb_device_pref
winmm_query_preferred_in_device(UINT devid)
{
  DWORD mmpref = WAVE_MAPPER, compref = WAVE_MAPPER, status;
  cubeb_device_pref ret = CUBEB_DEVICE_PREF_NONE;

  if (waveInMessage((HWAVEIN)WAVE_MAPPER, DRVM_MAPPER_PREFERRED_GET,
                    (DWORD_PTR)&mmpref,
                    (DWORD_PTR)&status) == MMSYSERR_NOERROR &&
      devid == mmpref)
    ret |= CUBEB_DEVICE_PREF_MULTIMEDIA | CUBEB_DEVICE_PREF_NOTIFICATION;

  if (waveInMessage((HWAVEIN)WAVE_MAPPER, DRVM_MAPPER_CONSOLEVOICECOM_GET,
                    (DWORD_PTR)&compref,
                    (DWORD_PTR)&status) == MMSYSERR_NOERROR &&
      devid == compref)
    ret |= CUBEB_DEVICE_PREF_VOICE;

  return ret;
}

static void
winmm_create_device_from_incaps2(cubeb_device_info * ret, LPWAVEINCAPS2A caps,
                                 UINT devid)
{
  XASSERT(ret);
  ret->devid = (cubeb_devid)devid;
  ret->device_id = device_id_idx(devid);
  ret->friendly_name = _strdup(caps->szPname);
  ret->group_id = guid_to_cstr(&caps->ProductGuid);
  ret->vendor_name = guid_to_cstr(&caps->ManufacturerGuid);

  ret->type = CUBEB_DEVICE_TYPE_INPUT;
  ret->state = CUBEB_DEVICE_STATE_ENABLED;
  ret->preferred = winmm_query_preferred_in_device(devid);

  ret->max_channels = caps->wChannels;
  winmm_calculate_device_rate(ret, caps->dwFormats);
  winmm_query_supported_formats(devid, caps->dwFormats, &ret->format,
                                &ret->default_format);

  /* Hardcoded latency estimates... */
  ret->latency_lo = 100 * ret->default_rate / 1000;
  ret->latency_hi = 200 * ret->default_rate / 1000;
}

static void
winmm_create_device_from_incaps(cubeb_device_info * ret, LPWAVEINCAPSA caps,
                                UINT devid)
{
  XASSERT(ret);
  ret->devid = (cubeb_devid)devid;
  ret->device_id = device_id_idx(devid);
  ret->friendly_name = _strdup(caps->szPname);
  ret->group_id = NULL;
  ret->vendor_name = NULL;

  ret->type = CUBEB_DEVICE_TYPE_INPUT;
  ret->state = CUBEB_DEVICE_STATE_ENABLED;
  ret->preferred = winmm_query_preferred_in_device(devid);

  ret->max_channels = caps->wChannels;
  winmm_calculate_device_rate(ret, caps->dwFormats);
  winmm_query_supported_formats(devid, caps->dwFormats, &ret->format,
                                &ret->default_format);

  /* Hardcoded latency estimates... */
  ret->latency_lo = 100 * ret->default_rate / 1000;
  ret->latency_hi = 200 * ret->default_rate / 1000;
}

static int
winmm_enumerate_devices(cubeb * context, cubeb_device_type type,
                        cubeb_device_collection * collection)
{
  UINT i, incount, outcount, total;
  cubeb_device_info * devices;
  cubeb_device_info * dev;

  outcount = waveOutGetNumDevs();
  incount = waveInGetNumDevs();
  total = outcount + incount;

  devices = calloc(total, sizeof(cubeb_device_info));
  collection->count = 0;

  if (type & CUBEB_DEVICE_TYPE_OUTPUT) {
    WAVEOUTCAPSA woc;
    WAVEOUTCAPS2A woc2;

    ZeroMemory(&woc, sizeof(woc));
    ZeroMemory(&woc2, sizeof(woc2));

    for (i = 0; i < outcount; i++) {
      dev = &devices[collection->count];
      if (waveOutGetDevCapsA(i, (LPWAVEOUTCAPSA)&woc2, sizeof(woc2)) ==
          MMSYSERR_NOERROR) {
        winmm_create_device_from_outcaps2(dev, &woc2, i);
        collection->count += 1;
      } else if (waveOutGetDevCapsA(i, &woc, sizeof(woc)) == MMSYSERR_NOERROR) {
        winmm_create_device_from_outcaps(dev, &woc, i);
        collection->count += 1;
      }
    }
  }

  if (type & CUBEB_DEVICE_TYPE_INPUT) {
    WAVEINCAPSA wic;
    WAVEINCAPS2A wic2;

    ZeroMemory(&wic, sizeof(wic));
    ZeroMemory(&wic2, sizeof(wic2));

    for (i = 0; i < incount; i++) {
      dev = &devices[collection->count];
      if (waveInGetDevCapsA(i, (LPWAVEINCAPSA)&wic2, sizeof(wic2)) ==
          MMSYSERR_NOERROR) {
        winmm_create_device_from_incaps2(dev, &wic2, i);
        collection->count += 1;
      } else if (waveInGetDevCapsA(i, &wic, sizeof(wic)) == MMSYSERR_NOERROR) {
        winmm_create_device_from_incaps(dev, &wic, i);
        collection->count += 1;
      }
    }
  }

  collection->device = devices;

  return CUBEB_OK;
}

static int
winmm_device_collection_destroy(cubeb * ctx,
                                cubeb_device_collection * collection)
{
  uint32_t i;
  XASSERT(collection);

  (void)ctx;

  for (i = 0; i < collection->count; i++) {
    free((void *)collection->device[i].device_id);
    free((void *)collection->device[i].friendly_name);
    free((void *)collection->device[i].group_id);
    free((void *)collection->device[i].vendor_name);
  }

  free(collection->device);
  return CUBEB_OK;
}

static struct cubeb_ops const winmm_ops = {
    /*.init =*/winmm_init,
    /*.get_backend_id =*/winmm_get_backend_id,
    /*.get_max_channel_count=*/winmm_get_max_channel_count,
    /*.get_min_latency=*/winmm_get_min_latency,
    /*.get_preferred_sample_rate =*/winmm_get_preferred_sample_rate,
    /*.enumerate_devices =*/winmm_enumerate_devices,
    /*.device_collection_destroy =*/winmm_device_collection_destroy,
    /*.destroy =*/winmm_destroy,
    /*.stream_init =*/winmm_stream_init,
    /*.stream_destroy =*/winmm_stream_destroy,
    /*.stream_start =*/winmm_stream_start,
    /*.stream_stop =*/winmm_stream_stop,
    /*.stream_get_position =*/winmm_stream_get_position,
    /*.stream_get_latency = */ winmm_stream_get_latency,
    /*.stream_get_input_latency = */ NULL,
    /*.stream_set_volume =*/winmm_stream_set_volume,
    /*.stream_set_name =*/NULL,
    /*.stream_get_current_device =*/NULL,
    /*.stream_device_destroy =*/NULL,
    /*.stream_register_device_changed_callback=*/NULL,
    /*.register_device_collection_changed =*/NULL};