Support floating-point audio output for Windows7+

media/audio/win/audio_low_latency_output_win.cc

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "media/audio/win/audio_low_latency_output_win.h"
 
 #include <Functiondiscoverykeys_devpkey.h>
 
 #include "base/command_line.h"
 #include "base/logging.h"
 #include "base/macros.h"
 #include "base/metrics/histogram.h"
 #include "base/strings/utf_string_conversions.h"
 #include "base/time/time.h"
 #include "base/trace_event/trace_event.h"
 #include "base/win/scoped_propvariant.h"
 #include "media/audio/audio_device_description.h"
 #include "media/audio/win/audio_manager_win.h"
 #include "media/audio/win/avrt_wrapper_win.h"
 #include "media/audio/win/core_audio_util_win.h"
+#include "media/base/audio_sample_types.h"
 #include "media/base/limits.h"
 #include "media/base/media_switches.h"
 
 using base::win::ScopedComPtr;
 using base::win::ScopedCOMInitializer;
 using base::win::ScopedCoMem;
 
 namespace media {
 
 // static
@@ skipping 34 matching lines @@
       opened_(false),
       volume_(1.0),
       packet_size_frames_(0),
       packet_size_bytes_(0),
       endpoint_buffer_size_frames_(0),
       device_id_(device_id),
       device_role_(device_role),
       share_mode_(GetShareMode()),
       num_written_frames_(0),
       source_(NULL),
-      audio_bus_(AudioBus::Create(params)) {
+      audio_bus_(AudioBus::Create(AudioParameters(
+          params.format(),
+          params.channel_layout(),
+          params.sample_rate(),
+          // Ignore the given bits per sample; we want 32 because
+          // we're outputting floats.
+          32,
+          params.frames_per_buffer()))) {
   DCHECK(manager_);
 
   // The empty string is used to indicate a default device and the
   // |device_role_| member controls whether that's the default or default
   // communications device.
   DCHECK_NE(device_id_, AudioDeviceDescription::kDefaultDeviceId);
   DCHECK_NE(device_id_, AudioDeviceDescription::kCommunicationsDeviceId);
 
   DVLOG(1) << "WASAPIAudioOutputStream::WASAPIAudioOutputStream()";
   DVLOG_IF(1, share_mode_ == AUDCLNT_SHAREMODE_EXCLUSIVE)
        << "Core Audio (WASAPI) EXCLUSIVE MODE is enabled.";
 
   // Load the Avrt DLL if not already loaded. Required to support MMCSS.
   bool avrt_init = avrt::Initialize();
   DCHECK(avrt_init) << "Failed to load the avrt.dll";
 
   // Set up the desired render format specified by the client. We use the
   // WAVE_FORMAT_EXTENSIBLE structure to ensure that multiple channel ordering
   // and high precision data can be supported.
 
   // Begin with the WAVEFORMATEX structure that specifies the basic format.
   WAVEFORMATEX* format = &format_.Format;
   format->wFormatTag = WAVE_FORMAT_EXTENSIBLE;
   format->nChannels = params.channels();
   format->nSamplesPerSec = params.sample_rate();
-  format->wBitsPerSample = params.bits_per_sample();
+  // Always want 32 bits because we're using floats.
+  format->wBitsPerSample = 32;

[tommi (sloooow) - chröme, 2016/11/02 20:23:04]

nit: Add a constant at the top of the file for this? Just so that readers will
be more aware of where this value needs to be set.

[Raymond Toy, 2016/11/02 21:26:14]

Done. Please up choose a good name if you don't like kBitsPerFloatSample.

   format->nBlockAlign = (format->wBitsPerSample / 8) * format->nChannels;
   format->nAvgBytesPerSec = format->nSamplesPerSec * format->nBlockAlign;
   format->cbSize = sizeof(WAVEFORMATEXTENSIBLE) - sizeof(WAVEFORMATEX);
 
   // Add the parts which are unique to WAVE_FORMAT_EXTENSIBLE.
-  format_.Samples.wValidBitsPerSample = params.bits_per_sample();
+  format_.Samples.wValidBitsPerSample = 32;
   format_.dwChannelMask = CoreAudioUtil::GetChannelConfig(device_id, eRender);
-  format_.SubFormat = KSDATAFORMAT_SUBTYPE_PCM;
+  format_.SubFormat = KSDATAFORMAT_SUBTYPE_IEEE_FLOAT;

[tommi (sloooow) - chröme, 2016/11/02 20:23:04]

that's it? :-|

[Raymond Toy, 2016/11/02 21:26:14]

I think so. :-)  I can hear output now that wasn't heard before on my windows
machine.

 
   // Store size (in different units) of audio packets which we expect to
   // get from the audio endpoint device in each render event.
   packet_size_frames_ = params.frames_per_buffer();
   packet_size_bytes_ = params.GetBytesPerBuffer();
   DVLOG(1) << "Number of bytes per audio frame  : " << format->nBlockAlign;
   DVLOG(1) << "Number of audio frames per packet: " << packet_size_frames_;
   DVLOG(1) << "Number of bytes per packet       : " << packet_size_bytes_;
   DVLOG(1) << "Number of milliseconds per packet: "
           << params.GetBufferDuration().InMillisecondsF();
@@ skipping 422 matching lines @@
     int frames_filled =
         source_->OnMoreData(delay, delay_timestamp, 0, audio_bus_.get());
     uint32_t num_filled_bytes = frames_filled * format_.Format.nBlockAlign;
     DCHECK_LE(num_filled_bytes, packet_size_bytes_);
 
     // Note: If this ever changes to output raw float the data must be
     // clipped and sanitized since it may come from an untrusted
     // source such as NaCl.
     const int bytes_per_sample = format_.Format.wBitsPerSample >> 3;
     audio_bus_->Scale(volume_);
-    audio_bus_->ToInterleaved(
-        frames_filled, bytes_per_sample, audio_data);
+    audio_bus_->ToInterleaved<Float32SampleTypeTraits>(
+        frames_filled, reinterpret_cast<float*>(audio_data));
 
     // Release the buffer space acquired in the GetBuffer() call.
     // Render silence if we were not able to fill up the buffer totally.
     DWORD flags = (num_filled_bytes < packet_size_bytes_) ?
         AUDCLNT_BUFFERFLAGS_SILENT : 0;
     audio_render_client_->ReleaseBuffer(packet_size_frames_, flags);
 
     num_written_frames_ += packet_size_frames_;
   }
 
@@ skipping 93 matching lines @@
 
     // Ensure that we don't quit the main thread loop immediately next
     // time Start() is called.
     ResetEvent(stop_render_event_.Get());
   }
 
   source_ = NULL;
 }
 
 }  // namespace media