MSE: Add StreamParser buffer remuxing utility and tests

media/base/stream_parser.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/base/stream_parser.h"
 
+#include "media/base/buffers.h"
+#include "media/base/stream_parser_buffer.h"
+
 namespace media {
 
 StreamParser::StreamParser() {}
 
 StreamParser::~StreamParser() {}
 
+static bool MergeBufferQueuesInternal(
+    const std::vector<const StreamParser::BufferQueue*>& buffer_queues,
+    StreamParser::BufferQueue* merged_buffers) {
+  // Instead of std::merge usage, this method implements a custom merge because:
+  // 1) |buffer_queues| may contain N queues,
+  // 2) we must detect and return false if any of the queues in |buffer_queues|
+  // is unsorted, and
+  // 3) we must detect and return false if any of the buffers in |buffer_queues|
+  // has a decode timestamp prior to the last, if any, buffer in
+  // |merged_buffers|.
+  // TODO(wolenetz/acolwell): Refactor stream parsers to eliminate need for
+  // this large grain merge. See http://crbug.com/338484.
+
+  // Done if no inputs to merge.
+  if (buffer_queues.empty())
+    return true;
+
+  // Build a vector of iterators, one for each input, to traverse inputs.
+  // The union of these iterators points to the set of candidate buffers
+  // for being appended to |merged_buffers|.
+  std::vector<StreamParser::BufferQueue::const_iterator> itrs;

[xhwang, 2014/02/06 00:36:32]

Use vector::vector(size_type n) to avoid expanding the vector in the following
for loop.

[wolenetz, 2014/02/06 23:56:03]

Done.

+  for (size_t i = 0; i < buffer_queues.size(); ++i)
+    itrs.push_back(buffer_queues[i]->begin());
+
+  // |last_decode_timestamp| tracks the lower bound, if any, that all candidate
+  // buffers must not be less than. If |merged_buffers| already has buffers,
+  // initialize |last_decode_timestamp| to the decode timestamp of the last
+  // buffer in it.
+  base::TimeDelta last_decode_timestamp = kNoTimestamp();
+  if (!merged_buffers->empty())
+    last_decode_timestamp = merged_buffers->back()->GetDecodeTimestamp();
+
+  // Repeatedly select and append the next buffer from the candidate buffers
+  // until either:
+  // 1) returning false, to indicate detection of decreasing DTS in some queue,
+  //    when a candidate buffer has decode timestamp below
+  //    |last_decode_timestamp|, which means either an input buffer wasn't
+  //    sorted correctly or had a buffer with decode timestamp below the last
+  //    buffer, if any, in |merged_buffers|, or
+  // 2) returning true when all buffers have been merged successfully;
+  //    equivalently, when all of the iterators in |itrs| have reached the end
+  //    of their respective queue from |buffer_queues|.
+  while (true) {
+    // Tracks which queue's iterator is pointing to the candidate buffer to
+    // append next, or -1 if no candidate buffers found. This indexes |itrs|.
+    int index_of_queue_with_next_decode_timestamp = -1;
+    base::TimeDelta next_decode_timestamp = kNoTimestamp();
+
+    // Scan each of the iterators for |buffer_queues| to find the candidate
+    // buffer, if any, that has the lowest decode timestamp.
+    for (size_t i = 0; i < buffer_queues.size(); ++i) {
+      if (itrs[i] != buffer_queues[i]->end()) {

[xhwang, 2014/02/06 00:36:32]

if (itrs[i] == buffer_queues[i]->end())
  continue;

...

[wolenetz, 2014/02/06 23:56:03]

Done.

+        // Extract the candidate buffer's decode timestamp.
+        base::TimeDelta ts = (*itrs[i])->GetDecodeTimestamp();
+
+        if (last_decode_timestamp != kNoTimestamp() &&
+            ts < last_decode_timestamp)
+          return false;
+
+        if (ts < next_decode_timestamp ||
+            next_decode_timestamp == kNoTimestamp()) {
+          // Remember the decode timestamp and queue iterator index for this
+          // potentially winning candidate buffer.
+          next_decode_timestamp = ts;
+          index_of_queue_with_next_decode_timestamp = i;
+        }
+      }
+    }

[xhwang, 2014/02/06 00:36:32]

This block is inefficient. Ideally, we should use a heap to store the head of
all queues, so that we can pop the head with lowest decode timestamp in log(N)
time. Since we don't have a lot of queues and this code is going away, I think
this implementation is okay. But can you add a comment just in case people
wonder?

[wolenetz, 2014/02/06 23:56:03]

Done.

+
+    // All done if no further candidate buffers exist.
+    if (index_of_queue_with_next_decode_timestamp == -1)
+      return true;
+
+    // Otherwise, append the winning candidate buffer to |merged_buffers|,
+    // remember its decode timestamp as |last_decode_timestamp| now that it is
+    // the last buffer in |merged_buffers|, advance the corresponding
+    // input BufferQueue iterator, and continue.
+    scoped_refptr<StreamParserBuffer> buffer =
+        *itrs[index_of_queue_with_next_decode_timestamp];
+    last_decode_timestamp = buffer->GetDecodeTimestamp();
+    merged_buffers->push_back(buffer);
+    ++itrs[index_of_queue_with_next_decode_timestamp];
+  }
+}
+
+bool MergeBufferQueues(const StreamParser::BufferQueue& audio_buffers,
+                       const StreamParser::BufferQueue& video_buffers,
+                       const StreamParser::TextBufferQueueMap& text_buffers,
+                       StreamParser::BufferQueue* merged_buffers) {
+  DCHECK(merged_buffers);
+
+  // Prepare vector containing pointers to any provided non-empty buffer queues.
+  std::vector<const StreamParser::BufferQueue*> buffer_queues;
+  if (!audio_buffers.empty())
+    buffer_queues.push_back(&audio_buffers);
+  if (!video_buffers.empty())
+    buffer_queues.push_back(&video_buffers);
+  for (StreamParser::TextBufferQueueMap::const_iterator map_itr =
+           text_buffers.begin();
+       map_itr != text_buffers.end();
+       map_itr++) {
+    if (!map_itr->second.empty())
+      buffer_queues.push_back(&(map_itr->second));
+  }
+
+  // Do the merge.
+  return MergeBufferQueuesInternal(buffer_queues, merged_buffers);
+}
+
 }  // namespace media