Mp4Muxer: disable sample batching and copying by default.

When sample batching is disabled, copying of the ByteBuffer data
is not necessary as samples are written as they arrive.
Copying of the BufferInfo is necessary because the info is needed
for writing the moov atom.

The input ByteBuffer can be in little endian order, or have its
position set. AnnexBUtils now ensures big endian order before
inspecting bytes, and supports reading from a non-zero position.

This change reduces the amount of memory allocations by Mp4Muxer
in its default configuration

PiperOrigin-RevId: 723401822

Author: ychaparov

RELEASENOTES.md

@@ -24,6 +24,7 @@
 *   DRM:
 *   Effect:
 *   Muxers:
+    *   Disable `Mp4Muxer` sample batching and copying by default.
 *   IMA extension:
 *   Session:
 *   UI:

libraries/muxer/src/main/java/androidx/media3/muxer/AnnexBToAvccConverter.java

@@ -15,8 +15,6 @@
  */
 package androidx.media3.muxer;
 
-import static androidx.media3.common.util.Assertions.checkArgument;
-
 import androidx.media3.common.util.UnstableApi;
 import com.google.common.collect.ImmutableList;
 import java.nio.ByteBuffer;
@@ -35,9 +33,6 @@ public interface AnnexBToAvccConverter {
           return inputBuffer;
         }
 
-        checkArgument(
-            inputBuffer.position() == 0, "The input buffer should have position set to 0.");
-
         ImmutableList<ByteBuffer> nalUnitList = AnnexBUtils.findNalUnits(inputBuffer);
 
         int totalBytesNeeded = 0;

libraries/muxer/src/main/java/androidx/media3/muxer/AnnexBUtils.java

@@ -20,6 +20,7 @@
 import androidx.media3.common.MimeTypes;
 import com.google.common.collect.ImmutableList;
 import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
 
 /** NAL unit utilities for start codes and emulation prevention. */
 /* package */ final class AnnexBUtils {
@@ -40,11 +41,13 @@ public static ImmutableList findNalUnits(ByteBuffer input) {
     if (input.remaining() == 0) {
       return ImmutableList.of();
     }
+    input = input.asReadOnlyBuffer();
+    input.order(ByteOrder.BIG_ENDIAN);
 
     // The algorithm always searches for 0x000001 start code but it will work for 0x00000001 start
     // code as well because the first 0 will be considered as a leading 0 and will be skipped.
 
-    int nalStartCodeIndex = skipLeadingZerosAndFindNalStartCodeIndex(input, /* currentIndex= */ 0);
+    int nalStartCodeIndex = skipLeadingZerosAndFindNalStartCodeIndex(input, input.position());
 
     int nalStartIndex = nalStartCodeIndex + THREE_BYTE_NAL_START_CODE_SIZE;
     boolean readingNalUnit = true;
@@ -69,7 +72,6 @@ public static ImmutableList findNalUnits(ByteBuffer input) {
       }
     }
 
-    input.rewind();
     return nalUnits.build();
   }
 

libraries/muxer/src/main/java/androidx/media3/muxer/Mp4Muxer.java

@@ -219,8 +219,6 @@ public Builder(FileOutputStream outputStream) {
       this.outputStream = outputStream;
       lastSampleDurationBehavior =
           LAST_SAMPLE_DURATION_BEHAVIOR_SET_FROM_END_OF_STREAM_BUFFER_OR_DUPLICATE_PREVIOUS;
-      sampleCopyEnabled = true;
-      sampleBatchingEnabled = true;
       attemptStreamableOutputEnabled = true;
       outputFileFormat = FILE_FORMAT_DEFAULT;
     }
@@ -259,7 +257,11 @@ public Mp4Muxer.Builder setAnnexBToAvccConverter(AnnexBToAvccConverter annexBToA
      * to reuse them immediately. Otherwise, the muxer takes ownership of the {@link ByteBuffer} and
      * the {@link BufferInfo} and the caller must not modify them.
      *
-     * The default value is {@code true}.
+     * When {@linkplain #setSampleBatchingEnabled(boolean) sample batching} is disabled, samples
+     * are written as they {@linkplain #writeSampleData(int, ByteBuffer, BufferInfo) arrive} and
+     * sample copying is disabled.
+     *
+     * The default value is {@code false}.
      */
     @CanIgnoreReturnValue
     public Mp4Muxer.Builder setSampleCopyingEnabled(boolean enabled) {
@@ -274,7 +276,13 @@ public Mp4Muxer.Builder setSampleCopyingEnabled(boolean enabled) {
      * samples are written as they {@linkplain #writeSampleData(int, ByteBuffer, BufferInfo)
      * arrive}.
      *
-     * The default value is {@code true}.
+     * When sample batching is enabled, and {@linkplain #setSampleCopyingEnabled(boolean) sample
+     * copying} is disabled the {@link ByteBuffer} contents provided to {@link #writeSampleData(int,
+     * ByteBuffer, BufferInfo)} should not be modified. Otherwise, if sample batching is disabled or
+     * sample copying is enabled, the {@linkplain ByteBuffer sample data} contents can be modified
+     * after calling {@link #writeSampleData(int, ByteBuffer, BufferInfo)}.
+     *
+     * The default value is {@code false}.
      */
     @CanIgnoreReturnValue
     public Mp4Muxer.Builder setSampleBatchingEnabled(boolean enabled) {
@@ -391,7 +399,7 @@ private Mp4Muxer(
     outputChannel = outputStream.getChannel();
     this.lastSampleDurationBehavior = lastFrameDurationBehavior;
     this.annexBToAvccConverter = annexBToAvccConverter;
-    this.sampleCopyEnabled = sampleCopyEnabled;
+    this.sampleCopyEnabled = sampleBatchingEnabled && sampleCopyEnabled;
     this.sampleBatchingEnabled = sampleBatchingEnabled;
     this.attemptStreamableOutputEnabled = attemptStreamableOutputEnabled;
     this.outputFileFormat = outputFileFormat;
@@ -470,14 +478,6 @@ && isAuxiliaryTrack(format)) {
   /**
    * Writes encoded sample data.
    *
-   * When sample batching is {@linkplain Mp4Muxer.Builder#setSampleBatchingEnabled(boolean)
-   * enabled}, provide sample data ({@link ByteBuffer}, {@link BufferInfo}) that won't be modified
-   * after calling the {@link #writeSampleData(int, ByteBuffer, BufferInfo)} method, unless sample
-   * copying is also {@linkplain Mp4Muxer.Builder#setSampleCopyingEnabled(boolean) enabled}. This
-   * ensures data integrity within the batch. If sample copying is {@linkplain
-   * Mp4Muxer.Builder#setSampleCopyingEnabled(boolean) enabled}, it's safe to modify the data after
-   * the method returns, as the muxer internally creates a sample copy.
-   *
    * @param trackId The track id for which this sample is being written.
    * @param byteBuffer The encoded sample. The muxer takes ownership of the buffer if {@link
    *     Builder#setSampleCopyingEnabled(boolean) sample copying} is disabled. Otherwise, the

libraries/muxer/src/main/java/androidx/media3/muxer/Track.java

@@ -92,22 +92,21 @@ public void writeSampleData(ByteBuffer byteBuffer, BufferInfo bufferInfo) {
     }
 
     ByteBuffer byteBufferToAdd = byteBuffer;
-    BufferInfo bufferInfoToAdd = bufferInfo;
-
     if (sampleCopyEnabled) {
       // Copy sample data and release the original buffer.
       byteBufferToAdd = ByteBuffer.allocateDirect(byteBuffer.remaining());
       byteBufferToAdd.put(byteBuffer);
       byteBufferToAdd.rewind();
-
-      bufferInfoToAdd = new BufferInfo();
-      bufferInfoToAdd.set(
-          /* newOffset= */ byteBufferToAdd.position(),
-          /* newSize= */ byteBufferToAdd.remaining(),
-          bufferInfo.presentationTimeUs,
-          bufferInfo.flags);
     }
 
+    // Always copy the buffer info as it is retained until the track is finalized.
+    BufferInfo bufferInfoToAdd = new BufferInfo();
+    bufferInfoToAdd.set(
+        /* newOffset= */ byteBufferToAdd.position(),
+        /* newSize= */ byteBufferToAdd.remaining(),
+        bufferInfo.presentationTimeUs,
+        bufferInfo.flags);
+
     pendingSamplesBufferInfo.addLast(bufferInfoToAdd);
     pendingSamplesByteBuffer.addLast(byteBufferToAdd);
   }

libraries/muxer/src/test/java/androidx/media3/muxer/AnnexBUtilsTest.java

@@ -22,6 +22,7 @@
 import androidx.test.ext.junit.runners.AndroidJUnit4;
 import com.google.common.collect.ImmutableList;
 import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
 import org.junit.Test;
 import org.junit.runner.RunWith;
 
@@ -53,6 +54,17 @@ public void findNalUnits_singleNalUnitWithFourByteStartCode_returnsSingleElement
     assertThat(components).containsExactly(ByteBuffer.wrap(getBytesFromHexString("ABCDEF")));
   }
 
+  @Test
+  public void
+      findNalUnits_singleNalUnitWithFourByteStartCodeAndLittleEndianOrder_returnsSingleElement() {
+    ByteBuffer buffer =
+        ByteBuffer.wrap(getBytesFromHexString("00000001ABCDEF")).order(ByteOrder.LITTLE_ENDIAN);
+
+    ImmutableList<ByteBuffer> components = AnnexBUtils.findNalUnits(buffer);
+
+    assertThat(components).containsExactly(ByteBuffer.wrap(getBytesFromHexString("ABCDEF")));
+  }
+
   @Test
   public void findNalUnits_singleNalUnitWithThreeByteStartCode_returnsSingleElement() {
     ByteBuffer buffer = ByteBuffer.wrap(getBytesFromHexString("000001ABCDEF"));
@@ -100,6 +112,34 @@ public void findNalUnits_withTrainingZeroesFollowedByGarbageData_throws() {
     assertThrows(IllegalStateException.class, () -> AnnexBUtils.findNalUnits(buffer));
   }
 
+  @Test
+  public void findNalUnits_withPositionAndLimitSet_returnsNalUnit() {
+    ByteBuffer buffer =
+        ByteBuffer.wrap(getBytesFromHexString("12345600000001ABCDEF002233445566778899"));
+    buffer.position(3);
+    buffer.limit(10);
+
+    ImmutableList<ByteBuffer> components = AnnexBUtils.findNalUnits(buffer);
+
+    assertThat(components).containsExactly(ByteBuffer.wrap(getBytesFromHexString("ABCDEF")));
+  }
+
+  @Test
+  public void findNalUnits_withMultipleNalUnitsAndPositionSet_returnsAllNalUnits() {
+    ByteBuffer buffer =
+        ByteBuffer.wrap(getBytesFromHexString("123456000001ABCDEF000001DDCC000001BBAA"));
+    buffer.position(3);
+
+    ImmutableList<ByteBuffer> components = AnnexBUtils.findNalUnits(buffer);
+
+    assertThat(components)
+        .containsExactly(
+            ByteBuffer.wrap(getBytesFromHexString("ABCDEF")),
+            ByteBuffer.wrap(getBytesFromHexString("DDCC")),
+            ByteBuffer.wrap(getBytesFromHexString("BBAA")))
+        .inOrder();
+  }
+
   @Test
   public void findNalUnits_withTrailingZeroes_stripsTrailingZeroes() {
     // The first NAL unit has some training zeroes.

libraries/muxer/src/test/java/androidx/media3/muxer/Mp4MuxerEndToEndTest.java

@@ -138,8 +138,10 @@ public void createMp4File_muxerNotClosed_createsPartiallyWrittenValidFile() thro
 
     // Muxer not closed.
 
-    // Audio sample written = 192 out of 195.
-    // Video sample written = 125 out of 127.
+    // The output depends on Mp4Muxer.MOOV_BOX_UPDATE_INTERVAL_US and whether or not
+    // sample batching is enabled.
+    // Audio sample written = 187 out of 195.
+    // Video sample written = 93 out of 127.
     // Output is still a valid MP4 file.
     FakeExtractorOutput fakeExtractorOutput =
         TestUtil.extractAllSamplesFromFilePath(

libraries/muxer/src/test/java/androidx/media3/muxer/Mp4MuxerMetadataTest.java

@@ -20,7 +20,6 @@
 import static androidx.media3.container.MdtaMetadataEntry.TYPE_INDICATOR_STRING;
 import static androidx.media3.muxer.MuxerTestUtil.FAKE_VIDEO_FORMAT;
 import static androidx.media3.muxer.MuxerTestUtil.XMP_SAMPLE_DATA;
-import static com.google.common.truth.Truth.assertThat;
 
 import android.content.Context;
 import android.media.MediaCodec.BufferInfo;
@@ -292,7 +291,6 @@ public void writeMp4File_addManyLargeStringMetadata_doesNotThrow() throws Except
 
     try {
       muxer.writeSampleData(trackId, sampleAndSampleInfo.first, sampleAndSampleInfo.second);
-      assertThat(sampleAndSampleInfo.first.remaining()).isEqualTo(0);
     } finally {
       muxer.close();
     }

libraries/test_data/src/test/assets/muxerdumps/hdr10-720p.mp4.dump

@@ -3,8 +3,8 @@ seekMap:
   duration = 4236600
   getPosition(0) = [[timeUs=0, position=400052]]
   getPosition(1) = [[timeUs=0, position=400052], [timeUs=1002955, position=430177]]
-  getPosition(2118300) = [[timeUs=2003566, position=2503758], [timeUs=3003444, position=4426916]]
-  getPosition(4236600) = [[timeUs=4003277, position=4459282]]
+  getPosition(2118300) = [[timeUs=2003566, position=2562813], [timeUs=3003444, position=4426916]]
+  getPosition(4236600) = [[timeUs=4003277, position=6370233]]
 numberOfTracks = 2
 track 0:
   total output bytes = 7944083

libraries/test_data/src/test/assets/muxerdumps/mp4_with_auxiliary_tracks_in_axte.box.dump

@@ -1,6 +1,6 @@
 ftyp (28 bytes):
   Data = length 20, hash EF896440
-moov (881 bytes):
+moov (913 bytes):
   mvhd (108 bytes):
     Data = length 100, hash 2613A5C
   meta (191 bytes):
@@ -10,40 +10,40 @@ moov (881 bytes):
       Data = length 70, hash C5147150
     ilst (72 bytes):
       Data = length 64, hash 3B286CD9
-  trak (574 bytes):
+  trak (606 bytes):
     tkhd (92 bytes):
       Data = length 84, hash 3D79758F
-    mdia (474 bytes):
+    mdia (506 bytes):
       mdhd (32 bytes):
         Data = length 24, hash 41542D81
       hdlr (44 bytes):
         Data = length 36, hash A0852FF2
-      minf (390 bytes):
+      minf (422 bytes):
         vmhd (20 bytes):
           Data = length 12, hash EE830681
         dinf (36 bytes):
           Data = length 28, hash D535436B
-        stbl (326 bytes):
+        stbl (358 bytes):
           stsd (166 bytes):
             Data = length 158, hash 11532063
           stts (24 bytes):
             Data = length 16, hash E534C287
           stsz (40 bytes):
             Data = length 32, hash B3F09E
           stsc (28 bytes):
-            Data = length 20, hash 8FA6E089
-          co64 (24 bytes):
-            Data = length 16, hash E4EE6662
+            Data = length 20, hash 8F6E8285
+          co64 (56 bytes):
+            Data = length 48, hash A4175473
           stss (36 bytes):
             Data = length 28, hash 53024615
-free (399127 bytes):
-  Data = length 399119, hash 82DEBDDF
+free (399095 bytes):
+  Data = length 399087, hash A215F9DF
 mdat (296 bytes):
   Data = length 280, hash 8DCFD2E3
 axte (400628 bytes):
   ftyp (28 bytes):
     Data = length 20, hash EF896440
-  moov (1446 bytes):
+  moov (1510 bytes):
     mvhd (108 bytes):
       Data = length 100, hash 2613A5D
     meta (182 bytes):
@@ -53,59 +53,59 @@ axte (400628 bytes):
         Data = length 72, hash 4E5C3894
       ilst (61 bytes):
         Data = length 53, hash 75D49FBD
-    trak (574 bytes):
+    trak (606 bytes):
       tkhd (92 bytes):
         Data = length 84, hash 3D79758F
-      mdia (474 bytes):
+      mdia (506 bytes):
         mdhd (32 bytes):
           Data = length 24, hash 41542D81
         hdlr (44 bytes):
           Data = length 36, hash A0852FF2
-        minf (390 bytes):
+        minf (422 bytes):
           vmhd (20 bytes):
             Data = length 12, hash EE830681
           dinf (36 bytes):
             Data = length 28, hash D535436B
-          stbl (326 bytes):
+          stbl (358 bytes):
             stsd (166 bytes):
               Data = length 158, hash 11532063
             stts (24 bytes):
               Data = length 16, hash E534C287
             stsz (40 bytes):
               Data = length 32, hash B3F09E
             stsc (28 bytes):
-              Data = length 20, hash 8FA6E089
-            co64 (24 bytes):
-              Data = length 16, hash E4EE6662
+              Data = length 20, hash 8F6E8285
+            co64 (56 bytes):
+              Data = length 48, hash A4175473
             stss (36 bytes):
               Data = length 28, hash 53024615
-    trak (574 bytes):
+    trak (606 bytes):
       tkhd (92 bytes):
         Data = length 84, hash 2ECB0510
-      mdia (474 bytes):
+      mdia (506 bytes):
         mdhd (32 bytes):
           Data = length 24, hash 41542D81
         hdlr (44 bytes):
           Data = length 36, hash A0852FF2
-        minf (390 bytes):
+        minf (422 bytes):
           vmhd (20 bytes):
             Data = length 12, hash EE830681
           dinf (36 bytes):
             Data = length 28, hash D535436B
-          stbl (326 bytes):
+          stbl (358 bytes):
             stsd (166 bytes):
               Data = length 158, hash 11532063
             stts (24 bytes):
               Data = length 16, hash E534C287
             stsz (40 bytes):
               Data = length 32, hash B3F09E
             stsc (28 bytes):
-              Data = length 20, hash 8FA6E089
-            co64 (24 bytes):
-              Data = length 16, hash E4EE6699
+              Data = length 20, hash 8F6E8285
+            co64 (56 bytes):
+              Data = length 48, hash 44DD22A5
             stss (36 bytes):
               Data = length 28, hash 53024615
-  free (398562 bytes):
-    Data = length 398554, hash C1D2F8C1
+  free (398498 bytes):
+    Data = length 398490, hash B07980C1
   mdat (576 bytes):
     Data = length 560, hash 9E0D5FC5