H264 HW encode using VideoToolbox

media/cast/sender/h264_vt_encoder.cc

 // Copyright 2014 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/cast/sender/h264_vt_encoder.h"
 
 #include <stddef.h>
 
 #include <string>
 #include <vector>
@@ skipping 25 matching lines @@
   const VideoEncoder::FrameEncodedCallback frame_encoded_callback;
 
   InProgressFrameEncode(RtpTimeTicks rtp,
                         base::TimeTicks r_time,
                         VideoEncoder::FrameEncodedCallback callback)
       : rtp_timestamp(rtp),
         reference_time(r_time),
         frame_encoded_callback(callback) {}
 };
 
-base::ScopedCFTypeRef<CFDictionaryRef>
-DictionaryWithKeysAndValues(CFTypeRef* keys, CFTypeRef* values, size_t size) {
-  return base::ScopedCFTypeRef<CFDictionaryRef>(CFDictionaryCreate(
-      kCFAllocatorDefault, keys, values, size, &kCFTypeDictionaryKeyCallBacks,
-      &kCFTypeDictionaryValueCallBacks));
-}
-
-base::ScopedCFTypeRef<CFDictionaryRef> DictionaryWithKeyValue(CFTypeRef key,
-                                                              CFTypeRef value) {
-  CFTypeRef keys[1] = {key};
-  CFTypeRef values[1] = {value};
-  return DictionaryWithKeysAndValues(keys, values, 1);
-}
-
-base::ScopedCFTypeRef<CFArrayRef> ArrayWithIntegers(const int* v, size_t size) {
-  std::vector<CFNumberRef> numbers;
-  numbers.reserve(size);
-  for (const int* end = v + size; v < end; ++v)
-    numbers.push_back(CFNumberCreate(nullptr, kCFNumberSInt32Type, v));
-  base::ScopedCFTypeRef<CFArrayRef> array(CFArrayCreate(
-      kCFAllocatorDefault, reinterpret_cast<const void**>(&numbers[0]),
-      numbers.size(), &kCFTypeArrayCallBacks));
-  for (auto& number : numbers) {
-    CFRelease(number);
-  }
-  return array;
-}
-
-template <typename NalSizeType>
-void CopyNalsToAnnexB(char* avcc_buffer,
-                      const size_t avcc_size,
-                      std::string* annexb_buffer) {
-  static_assert(sizeof(NalSizeType) == 1 || sizeof(NalSizeType) == 2 ||
-                    sizeof(NalSizeType) == 4,
-                "NAL size type has unsupported size");
-  static const char startcode_3[3] = {0, 0, 1};
-  DCHECK(avcc_buffer);
-  DCHECK(annexb_buffer);
-  size_t bytes_left = avcc_size;
-  while (bytes_left > 0) {
-    DCHECK_GT(bytes_left, sizeof(NalSizeType));
-    NalSizeType nal_size;
-    base::ReadBigEndian(avcc_buffer, &nal_size);
-    bytes_left -= sizeof(NalSizeType);
-    avcc_buffer += sizeof(NalSizeType);
-
-    DCHECK_GE(bytes_left, nal_size);
-    annexb_buffer->append(startcode_3, sizeof(startcode_3));
-    annexb_buffer->append(avcc_buffer, nal_size);
-    bytes_left -= nal_size;
-    avcc_buffer += nal_size;
-  }
-}
-
-// Copy a H.264 frame stored in a CM sample buffer to an Annex B buffer. Copies
-// parameter sets for keyframes before the frame data as well.
-void CopySampleBufferToAnnexBBuffer(CoreMediaGlue::CMSampleBufferRef sbuf,
-                                    std::string* annexb_buffer,
-                                    bool keyframe) {
-  // Perform two pass, one to figure out the total output size, and another to
-  // copy the data after having performed a single output allocation. Note that
-  // we'll allocate a bit more because we'll count 4 bytes instead of 3 for
-  // video NALs.
-
-  OSStatus status;
-
-  // Get the sample buffer's block buffer and format description.
-  auto bb = CoreMediaGlue::CMSampleBufferGetDataBuffer(sbuf);
-  DCHECK(bb);
-  auto fdesc = CoreMediaGlue::CMSampleBufferGetFormatDescription(sbuf);
-  DCHECK(fdesc);
-
-  size_t bb_size = CoreMediaGlue::CMBlockBufferGetDataLength(bb);
-  size_t total_bytes = bb_size;
-
-  size_t pset_count;
-  int nal_size_field_bytes;
-  status = CoreMediaGlue::CMVideoFormatDescriptionGetH264ParameterSetAtIndex(
-      fdesc, 0, nullptr, nullptr, &pset_count, &nal_size_field_bytes);
-  if (status ==
-      CoreMediaGlue::kCMFormatDescriptionBridgeError_InvalidParameter) {
-    DLOG(WARNING) << " assuming 2 parameter sets and 4 bytes NAL length header";
-    pset_count = 2;
-    nal_size_field_bytes = 4;
-  } else if (status != noErr) {
-    DLOG(ERROR)
-        << " CMVideoFormatDescriptionGetH264ParameterSetAtIndex failed: "
-        << status;
-    return;
-  }
-
-  if (keyframe) {
-    const uint8_t* pset;
-    size_t pset_size;
-    for (size_t pset_i = 0; pset_i < pset_count; ++pset_i) {
-      status =
-          CoreMediaGlue::CMVideoFormatDescriptionGetH264ParameterSetAtIndex(
-              fdesc, pset_i, &pset, &pset_size, nullptr, nullptr);
-      if (status != noErr) {
-        DLOG(ERROR)
-            << " CMVideoFormatDescriptionGetH264ParameterSetAtIndex failed: "
-            << status;
-        return;
-      }
-      total_bytes += pset_size + nal_size_field_bytes;
-    }
-  }
-
-  annexb_buffer->reserve(total_bytes);
-
-  // Copy all parameter sets before keyframes.
-  if (keyframe) {
-    const uint8_t* pset;
-    size_t pset_size;
-    for (size_t pset_i = 0; pset_i < pset_count; ++pset_i) {
-      status =
-          CoreMediaGlue::CMVideoFormatDescriptionGetH264ParameterSetAtIndex(
-              fdesc, pset_i, &pset, &pset_size, nullptr, nullptr);
-      if (status != noErr) {
-        DLOG(ERROR)
-            << " CMVideoFormatDescriptionGetH264ParameterSetAtIndex failed: "
-            << status;
-        return;
-      }
-      static const char startcode_4[4] = {0, 0, 0, 1};
-      annexb_buffer->append(startcode_4, sizeof(startcode_4));
-      annexb_buffer->append(reinterpret_cast<const char*>(pset), pset_size);
-    }
-  }
-
-  // Block buffers can be composed of non-contiguous chunks. For the sake of
-  // keeping this code simple, flatten non-contiguous block buffers.
-  base::ScopedCFTypeRef<CoreMediaGlue::CMBlockBufferRef> contiguous_bb(
-      bb, base::scoped_policy::RETAIN);
-  if (!CoreMediaGlue::CMBlockBufferIsRangeContiguous(bb, 0, 0)) {
-    contiguous_bb.reset();
-    status = CoreMediaGlue::CMBlockBufferCreateContiguous(
-        kCFAllocatorDefault, bb, kCFAllocatorDefault, nullptr, 0, 0, 0,
-        contiguous_bb.InitializeInto());
-    if (status != noErr) {
-      DLOG(ERROR) << " CMBlockBufferCreateContiguous failed: " << status;
-      return;
-    }
-  }
-
-  // Copy all the NAL units. In the process convert them from AVCC format
-  // (length header) to AnnexB format (start code).
-  char* bb_data;
-  status = CoreMediaGlue::CMBlockBufferGetDataPointer(contiguous_bb, 0, nullptr,
-                                                      nullptr, &bb_data);
-  if (status != noErr) {
-    DLOG(ERROR) << " CMBlockBufferGetDataPointer failed: " << status;
-    return;
-  }
-
-  if (nal_size_field_bytes == 1) {
-    CopyNalsToAnnexB<uint8_t>(bb_data, bb_size, annexb_buffer);
-  } else if (nal_size_field_bytes == 2) {
-    CopyNalsToAnnexB<uint16_t>(bb_data, bb_size, annexb_buffer);
-  } else if (nal_size_field_bytes == 4) {
-    CopyNalsToAnnexB<uint32_t>(bb_data, bb_size, annexb_buffer);
-  } else {
-    NOTREACHED();
-  }
-}
-
 }  // namespace
 
 class H264VideoToolboxEncoder::VideoFrameFactoryImpl
     : public base::RefCountedThreadSafe<VideoFrameFactoryImpl>,
       public VideoFrameFactory {
  public:
   // Type that proxies the VideoFrameFactory interface to this class.
   class Proxy;
 
   VideoFrameFactoryImpl(const base::WeakPtr<H264VideoToolboxEncoder>& encoder,
@@ skipping 162 matching lines @@
       CastEnvironment::MAIN, FROM_HERE,
       base::Bind(status_change_cb_, STATUS_CODEC_REINIT_PENDING));
 
   // Destroy the current session, if any.
   DestroyCompressionSession();
 
   // On OS X, allow the hardware encoder. Don't require it, it does not support
   // all configurations (some of which are used for testing).
   base::ScopedCFTypeRef<CFDictionaryRef> encoder_spec;
 #if !defined(OS_IOS)
-  encoder_spec = DictionaryWithKeyValue(
+  encoder_spec = video_toolbox::DictionaryWithKeyValue(
       videotoolbox_glue_
           ->kVTVideoEncoderSpecification_EnableHardwareAcceleratedVideoEncoder(),
       kCFBooleanTrue);
 #endif
 
   // Force 420v so that clients can easily use these buffers as GPU textures.
   const int format[] = {
       CoreVideoGlue::kCVPixelFormatType_420YpCbCr8BiPlanarVideoRange};
 
   // Keep these attachment settings in-sync with those in ConfigureSession().
   CFTypeRef attachments_keys[] = {kCVImageBufferColorPrimariesKey,
                                   kCVImageBufferTransferFunctionKey,
                                   kCVImageBufferYCbCrMatrixKey};
   CFTypeRef attachments_values[] = {kCVImageBufferColorPrimaries_ITU_R_709_2,
                                     kCVImageBufferTransferFunction_ITU_R_709_2,
                                     kCVImageBufferYCbCrMatrix_ITU_R_709_2};
   CFTypeRef buffer_attributes_keys[] = {kCVPixelBufferPixelFormatTypeKey,
                                         kCVBufferPropagatedAttachmentsKey};
   CFTypeRef buffer_attributes_values[] = {
-      ArrayWithIntegers(format, arraysize(format)).release(),
-      DictionaryWithKeysAndValues(attachments_keys, attachments_values,
-                                  arraysize(attachments_keys)).release()};
+      video_toolbox::ArrayWithIntegers(format, arraysize(format)).release(),
+      video_toolbox::DictionaryWithKeysAndValues(
+          attachments_keys, attachments_values, arraysize(attachments_keys))
+          .release()};
   const base::ScopedCFTypeRef<CFDictionaryRef> buffer_attributes =
-      DictionaryWithKeysAndValues(buffer_attributes_keys,
-                                  buffer_attributes_values,
-                                  arraysize(buffer_attributes_keys));
+      video_toolbox::DictionaryWithKeysAndValues(
+          buffer_attributes_keys, buffer_attributes_values,
+          arraysize(buffer_attributes_keys));
   for (auto& v : buffer_attributes_values)
     CFRelease(v);
 
   // Create the compression session.
 
   // Note that the encoder object is given to the compression session as the
   // callback context using a raw pointer. The C API does not allow us to use a
   // smart pointer, nor is this encoder ref counted. However, this is still
   // safe, because we 1) we own the compression session and 2) we tear it down
   // safely. When destructing the encoder, the compression session is flushed
@@ skipping 26 matching lines @@
       base::scoped_policy::RETAIN);
   video_frame_factory_->Update(pool, frame_size_);
 
   // Notify that reinitialization is done.
   cast_environment_->PostTask(
       CastEnvironment::MAIN, FROM_HERE,
       base::Bind(status_change_cb_, STATUS_INITIALIZED));
 }
 
 void H264VideoToolboxEncoder::ConfigureCompressionSession() {
-  SetSessionProperty(
+  video_toolbox::SessionPropertySetter session_property_setter(
+      compression_session_, videotoolbox_glue_);
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_ProfileLevel(),
       videotoolbox_glue_->kVTProfileLevel_H264_Main_AutoLevel());
-  SetSessionProperty(videotoolbox_glue_->kVTCompressionPropertyKey_RealTime(),
-                     true);
-  SetSessionProperty(
+  session_property_setter.Set(
+      videotoolbox_glue_->kVTCompressionPropertyKey_RealTime(), true);
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_AllowFrameReordering(),
       false);
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_MaxKeyFrameInterval(), 240);
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_
           ->kVTCompressionPropertyKey_MaxKeyFrameIntervalDuration(),
       240);
   // TODO(jfroy): implement better bitrate control
   //              https://crbug.com/425352
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_AverageBitRate(),
       (video_config_.min_bitrate + video_config_.max_bitrate) / 2);
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_ExpectedFrameRate(),
       video_config_.max_frame_rate);
   // Keep these attachment settings in-sync with those in Initialize().
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_ColorPrimaries(),
       kCVImageBufferColorPrimaries_ITU_R_709_2);
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_TransferFunction(),
       kCVImageBufferTransferFunction_ITU_R_709_2);
-  SetSessionProperty(
+  session_property_setter.Set(
       videotoolbox_glue_->kVTCompressionPropertyKey_YCbCrMatrix(),
       kCVImageBufferYCbCrMatrix_ITU_R_709_2);
   if (video_config_.max_number_of_video_buffers_used > 0) {
-    SetSessionProperty(
+    session_property_setter.Set(
         videotoolbox_glue_->kVTCompressionPropertyKey_MaxFrameDelayCount(),
         video_config_.max_number_of_video_buffers_used);
   }
 }
 
 void H264VideoToolboxEncoder::DestroyCompressionSession() {
   DCHECK(thread_checker_.CalledOnValidThread());
 
   // If the compression session exists, invalidate it. This blocks until all
   // pending output callbacks have returned and any internal threads have
@@ skipping 48 matching lines @@
   // frame into memory visible by the hardware encoder. The VideoFrame's
   // lifetime is extended for the lifetime of the returned CVPixelBuffer.
   auto pixel_buffer = media::WrapVideoFrameInCVPixelBuffer(*video_frame);
   if (!pixel_buffer) {
     DLOG(ERROR) << "WrapVideoFrameInCVPixelBuffer failed.";
     return false;
   }
 
   // Convert the frame timestamp to CMTime.
   auto timestamp_cm = CoreMediaGlue::CMTimeMake(
-      (reference_time - base::TimeTicks()).InMicroseconds(), USEC_PER_SEC);
+      video_frame->timestamp().InMicroseconds(), USEC_PER_SEC);
 
   // Wrap information we'll need after the frame is encoded in a heap object.
   // We'll get the pointer back from the VideoToolbox completion callback.
   scoped_ptr<InProgressFrameEncode> request(new InProgressFrameEncode(
       RtpTimeTicks::FromTimeDelta(video_frame->timestamp(), kVideoFrequency),
       reference_time, frame_encoded_callback));
 
   // Build a suitable frame properties dictionary for keyframes.
   base::ScopedCFTypeRef<CFDictionaryRef> frame_props;
   if (encode_next_frame_as_keyframe_) {
-    frame_props = DictionaryWithKeyValue(
+    frame_props = video_toolbox::DictionaryWithKeyValue(
         videotoolbox_glue_->kVTEncodeFrameOptionKey_ForceKeyFrame(),
         kCFBooleanTrue);
     encode_next_frame_as_keyframe_ = false;
   }
 
   // Submit the frame to the compression session. The function returns as soon
   // as the frame has been enqueued.
   OSStatus status = videotoolbox_glue_->VTCompressionSessionEncodeFrame(
       compression_session_, pixel_buffer, timestamp_cm,
       CoreMediaGlue::CMTime{0, 0, 0, 0}, frame_props,
@@ skipping 75 matching lines @@
   // Reset the compression session only if the frame size is not zero (which
   // will obviously fail). It is possible for the frame size to be zero if no
   // frame was submitted for encoding or requested from the video frame factory
   // before suspension.
   if (!frame_size_.IsEmpty()) {
     VLOG(1) << "OnResume: Resetting compression session.";
     ResetCompressionSession();
   }
 }
 
-bool H264VideoToolboxEncoder::SetSessionProperty(CFStringRef key,
-                                                 int32_t value) {
-  base::ScopedCFTypeRef<CFNumberRef> cfvalue(
-      CFNumberCreate(nullptr, kCFNumberSInt32Type, &value));
-  return videotoolbox_glue_->VTSessionSetProperty(compression_session_, key,
-                                                  cfvalue) == noErr;
-}
-
-bool H264VideoToolboxEncoder::SetSessionProperty(CFStringRef key, bool value) {
-  CFBooleanRef cfvalue = (value) ? kCFBooleanTrue : kCFBooleanFalse;
-  return videotoolbox_glue_->VTSessionSetProperty(compression_session_, key,
-                                                  cfvalue) == noErr;
-}
-
-bool H264VideoToolboxEncoder::SetSessionProperty(CFStringRef key,
-                                                 CFStringRef value) {
-  return videotoolbox_glue_->VTSessionSetProperty(compression_session_, key,
-                                                  value) == noErr;
-}
-
 void H264VideoToolboxEncoder::CompressionCallback(void* encoder_opaque,
                                                   void* request_opaque,
                                                   OSStatus status,
                                                   VTEncodeInfoFlags info,
                                                   CMSampleBufferRef sbuf) {
   auto encoder = reinterpret_cast<H264VideoToolboxEncoder*>(encoder_opaque);
   const scoped_ptr<InProgressFrameEncode> request(
       reinterpret_cast<InProgressFrameEncode*>(request_opaque));
   bool keyframe = false;
   bool has_frame_data = false;
@@ skipping 38 matching lines @@
     // doesn't support the concept of forward-referencing frame dependencies or
     // multiple frame dependencies; so pretend that all frames are only
     // decodable after their immediately preceding frame is decoded. This will
     // ensure a Cast receiver only attempts to decode the frames sequentially
     // and in order. Furthermore, the encoder is configured to never use forward
     // references (see |kVTCompressionPropertyKey_AllowFrameReordering|). There
     // is no way to prevent multiple reference frames.
     encoded_frame->referenced_frame_id = frame_id - 1;
   }
 
-  if (has_frame_data)
-    CopySampleBufferToAnnexBBuffer(sbuf, &encoded_frame->data, keyframe);
+  if (has_frame_data) {
+    video_toolbox::CopySampleBufferToAnnexBBuffer(sbuf, keyframe,
+                                                  &encoded_frame->data);
+  }
 
   // TODO(miu): Compute and populate the |deadline_utilization| and
   // |lossy_utilization| performance metrics in |encoded_frame|.
 
   encoded_frame->encode_completion_time =
       encoder->cast_environment_->Clock()->NowTicks();
   encoder->cast_environment_->PostTask(
       CastEnvironment::MAIN, FROM_HERE,
       base::Bind(request->frame_encoded_callback,
                  base::Passed(&encoded_frame)));
 }
 
 }  // namespace cast
 }  // namespace media