Support start trimming post-decoding. Use it with FFmpegDemuxer.

media/base/audio_discard_helper_unittest.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 "base/memory/scoped_ptr.h"
 #include "media/base/audio_buffer.h"
 #include "media/base/audio_bus.h"
 #include "media/base/audio_discard_helper.h"
 #include "media/base/buffers.h"
 #include "media/base/decoder_buffer.h"
@@ skipping 28 matching lines @@
 static float ExtractDecodedData(const scoped_refptr<AudioBuffer>& buffer,
                                 int index) {
   // This is really inefficient, but we can't access the raw AudioBuffer if any
   // start trimming has been applied.
   scoped_ptr<AudioBus> temp_bus = AudioBus::Create(buffer->channel_count(), 1);
   buffer->ReadFrames(1, index, 0, temp_bus.get());
   return temp_bus->channel(0)[0];
 }
 
 TEST(AudioDiscardHelperTest, TimeDeltaToFrames) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
 
   EXPECT_EQ(0u, discard_helper.TimeDeltaToFrames(base::TimeDelta()));
   EXPECT_EQ(
       kSampleRate / 100,
       discard_helper.TimeDeltaToFrames(base::TimeDelta::FromMilliseconds(10)));
 
   // Ensure partial frames are rounded down correctly.  The equation below
   // calculates a frame count with a fractional part < 0.5.
   const int small_remainder =
       base::Time::kMicrosecondsPerSecond * (kSampleRate - 0.9) / kSampleRate;
   EXPECT_EQ(kSampleRate - 1,
             discard_helper.TimeDeltaToFrames(
                 base::TimeDelta::FromMicroseconds(small_remainder)));
 
   // Ditto, but rounded up using a fractional part > 0.5.
   const int large_remainder =
       base::Time::kMicrosecondsPerSecond * (kSampleRate - 0.4) / kSampleRate;
   EXPECT_EQ(kSampleRate,
             discard_helper.TimeDeltaToFrames(
                 base::TimeDelta::FromMicroseconds(large_remainder)));
 }
 
 TEST(AudioDiscardHelperTest, BasicProcessBuffers) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
 
   // Use an estimated duration which doesn't match the number of decoded frames
   // to ensure the helper is correctly setting durations based on output frames.
   const base::TimeDelta kEstimatedDuration =
       base::TimeDelta::FromMilliseconds(9);
   const base::TimeDelta kActualDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kActualDuration);
@@ skipping 11 matching lines @@
 
   // Verify a Reset() takes us back to an uninitialized state.
   discard_helper.Reset(0);
   ASSERT_FALSE(discard_helper.initialized());
 
   // Verify a NULL output buffer returns false.
   ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, NULL));
 }
 
 TEST(AudioDiscardHelperTest, NegativeTimestampClampsToZero) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = -base::TimeDelta::FromSeconds(1);
   const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
 
   scoped_refptr<DecoderBuffer> encoded_buffer =
       CreateEncodedBuffer(kTimestamp, kDuration);
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   // Verify the basic case where nothing is discarded.
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
   EXPECT_EQ(base::TimeDelta(), decoded_buffer->timestamp());
   EXPECT_EQ(kDuration, decoded_buffer->duration());
   EXPECT_EQ(kTestFrames, decoded_buffer->frame_count());
 }
 
 TEST(AudioDiscardHelperTest, ProcessBuffersWithInitialDiscard) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
   const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
 
   // Tell the helper we want to discard half of the initial frames.
   const int kDiscardFrames = kTestFrames / 2;
   discard_helper.Reset(kDiscardFrames);
 
   scoped_refptr<DecoderBuffer> encoded_buffer =
       CreateEncodedBuffer(kTimestamp, kDuration);
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   // Verify half the frames end up discarded.
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
   EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
   EXPECT_EQ(kDiscardFrames, decoded_buffer->frame_count());
   ASSERT_FLOAT_EQ(kDiscardFrames * kDataStep,
                   ExtractDecodedData(decoded_buffer, 0));
 }
 
 TEST(AudioDiscardHelperTest, ProcessBuffersWithLargeInitialDiscard) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
   const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
 
   // Tell the helper we want to discard 1.5 buffers worth of frames.
   discard_helper.Reset(kTestFrames * 1.5);
 
   scoped_refptr<DecoderBuffer> encoded_buffer =
@@ skipping 12 matching lines @@
   // The timestamp should match that of the initial buffer.
   const int kDiscardFrames = kTestFrames / 2;
   EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
   EXPECT_EQ(kDiscardFrames, decoded_buffer->frame_count());
   ASSERT_FLOAT_EQ(kDiscardFrames * kDataStep,
                   ExtractDecodedData(decoded_buffer, 0));
 }
 
 TEST(AudioDiscardHelperTest, AllowNonMonotonicTimestamps) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
   const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
 
   scoped_refptr<DecoderBuffer> encoded_buffer =
       CreateEncodedBuffer(kTimestamp, kDuration);
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
   EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration, decoded_buffer->duration());
   EXPECT_EQ(kTestFrames, decoded_buffer->frame_count());
 
   // Process the same input buffer again to ensure input timestamps which go
   // backwards in time are not errors.
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   EXPECT_EQ(kTimestamp + kDuration, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration, decoded_buffer->duration());
   EXPECT_EQ(kTestFrames, decoded_buffer->frame_count());
 }
 
-TEST(AudioDiscardHelperTest, DiscardPadding) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+TEST(AudioDiscardHelperTest, DiscardEndPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
   const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
 
   scoped_refptr<DecoderBuffer> encoded_buffer =
       CreateEncodedBuffer(kTimestamp, kDuration);
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   // Set a discard padding equivalent to half the buffer.
-  encoded_buffer->set_discard_padding(kDuration / 2);
+  encoded_buffer->set_discard_padding(
+      std::make_pair(base::TimeDelta(), kDuration / 2));
 
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
   EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
   EXPECT_EQ(kTestFrames / 2, decoded_buffer->frame_count());
-  ASSERT_FLOAT_EQ(0, ExtractDecodedData(decoded_buffer, 0));
 }
 
-TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPadding) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+TEST(AudioDiscardHelperTest, BadDiscardEndPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Set a discard padding equivalent to double the buffer size.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(base::TimeDelta(), kDuration * 2));
+
+  // Verify the end discard padding is rejected.
+  ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  ASSERT_TRUE(discard_helper.initialized());
+}
+
+TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardEndPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
+  ASSERT_FALSE(discard_helper.initialized());
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
   const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
   const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
 
   scoped_refptr<DecoderBuffer> encoded_buffer =
       CreateEncodedBuffer(kTimestamp, kDuration);
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   // Set a discard padding equivalent to a quarter of the buffer.
-  encoded_buffer->set_discard_padding(kDuration / 4);
+  encoded_buffer->set_discard_padding(
+      std::make_pair(base::TimeDelta(), kDuration / 4));
 
   // Set an initial discard of a quarter of the buffer.
   const int kDiscardFrames = kTestFrames / 4;
   discard_helper.Reset(kDiscardFrames);
 
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
   EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
   EXPECT_EQ(kTestFrames / 2, decoded_buffer->frame_count());
   ASSERT_FLOAT_EQ(kDiscardFrames * kDataStep,
                   ExtractDecodedData(decoded_buffer, 0));
 }
 
+TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
+  ASSERT_FALSE(discard_helper.initialized());
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Set a discard padding equivalent to a quarter of the buffer.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(kDuration / 4, kDuration / 4));

[acolwell GONE FROM CHROMIUM, 2014/05/01 01:08:34]

How about using different fractions for the start, end, and initial discard? I
feel like that would make it easier to see that one of these values isn't simply
being copied. How about something like 1/16, 1/8, and 1/4? I believe that should
result in 9/16 left right?

[DaleCurtis, 2014/05/01 19:21:27]

Done.

+
+  // Set an initial discard of a quarter of the buffer.
+  const int kDiscardFrames = kTestFrames / 4;
+  discard_helper.Reset(kDiscardFrames);
+
+  ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  ASSERT_TRUE(discard_helper.initialized());
+  EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
+  EXPECT_EQ(kDuration / 4, decoded_buffer->duration());
+  EXPECT_EQ(kTestFrames / 4, decoded_buffer->frame_count());
+}
+
+TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPaddingAndCodecDelay) {
+  // Use a codec delay of 5ms.
+  const int kCodecDelay = kSampleRate / 100 / 2;

[acolwell GONE FROM CHROMIUM, 2014/05/01 01:08:34]

How about making this 80ms, but keeping the 10ms decode buffers so it can
simulate a possible situation in Opus?

[DaleCurtis, 2014/05/01 19:21:27]

As discussed this doesn't apply for opus and for complexity reasons isn't
allowed in the code.  See the various DCHECKs around ensuring the discard
padding is within the current buffer and that if a buffer is entirely discarded
it contains no discard padding.

+  AudioDiscardHelper discard_helper(kSampleRate, kCodecDelay);
+  ASSERT_FALSE(discard_helper.initialized());
+  discard_helper.Reset(kCodecDelay);
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Set a discard padding equivalent to half of the buffer.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(kDuration / 2, base::TimeDelta()));
+
+  // All of the first buffer should be discarded.
+  ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  ASSERT_TRUE(discard_helper.initialized());
+
+  // Processing another buffer (with the same discard padding) should discard
+  // the back half of the buffer since kCodecDelay is half a buffer.
+  encoded_buffer->set_timestamp(kTimestamp + kDuration);
+  decoded_buffer = CreateDecodedBuffer(kTestFrames);
+  ASSERT_FLOAT_EQ(0.0f, ExtractDecodedData(decoded_buffer, 0));
+  ASSERT_NEAR(kCodecDelay * kDataStep,
+              ExtractDecodedData(decoded_buffer, kCodecDelay),
+              kDataStep * 1000);
+  ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
+  EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
+  EXPECT_EQ(kTestFrames / 2, decoded_buffer->frame_count());
+
+  // Verify it was actually the latter half of the buffer that was removed.
+  ASSERT_FLOAT_EQ(0.0f, ExtractDecodedData(decoded_buffer, 0));
+}
+
+TEST(AudioDiscardHelperTest, DelayedDiscardInitialDiscardAndDiscardPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
+  ASSERT_FALSE(discard_helper.initialized());
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+
+  // Set a discard padding equivalent to a quarter of the buffer.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(kDuration / 4, kDuration / 4));

[acolwell GONE FROM CHROMIUM, 2014/05/01 01:08:34]

Ditto w/ respect to using different fractions here as well.

[DaleCurtis, 2014/05/01 19:21:27]

Done.

+
+  // Set an initial discard of a quarter of the buffer.
+  discard_helper.Reset(kTestFrames / 4);
+
+  // Verify nothing is output for the first buffer, yet initialized is true.
+  ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, NULL));
+  ASSERT_TRUE(discard_helper.initialized());
+
+  // Create an encoded buffer with no discard padding.
+  encoded_buffer = CreateEncodedBuffer(kTimestamp + kDuration, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Verify that when the decoded buffer is consumed, the discards from the
+  // previous encoded buffer are applied.
+  ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
+  EXPECT_EQ(kDuration / 4, decoded_buffer->duration());
+  EXPECT_EQ(kTestFrames / 4, decoded_buffer->frame_count());
+}
+
 }  // namespace media