Overview

For maximum performance, video decompressor components should be marked as multi-buffer aware by setting the subCodecIsMultiBufferAware flag to true.

By setting this flag, the decompressor informs the Image Compression Manager that it is able to draw each frame to a new pixel buffer. This contract with the Image Compression Manager reduces the amount of buffer copies necessary to play through Core Video.

What You Need To Do

In your image decompressor's ImageCodecInitialize function, make sure to set the subCodecIsMultiBufferAware flag to true. This flag is part of the ImageSubCodecDecompressCapabilities structure passed to the Initialize function. See Listing 1.

Listing 1  ImageCodecInitialize function.

ComponentResult MyDecompressor_Initialize(MyDecompressorGlobals *storage,

                                          ImageSubCodecDecompressCapabilities *cap)

{

    // Specify the size of the ImageSubCodecDecompressRecord structure

    cap->decompressRecordSize = sizeof(MyDecompressRecord);

    // Say we can support asynchronous decompression - With the help of the base

    // image decompressor, any image decompressor that uses only interrupt-safe

    // calls for decompression operations can support asynchronous decompression

    cap->canAsync = true;

    // set other capabilities as appropriate

    .

    .

    .

    // subCodecIsMultiBufferAware added in QuickTime 7

    // tell the base codec that we are "multi-buffer aware" - This promises that we

    // always draw using ImageSubCodecDecompressRecord.baseAddr/rowBytes

    // passed to our ImageCodecDrawBand function, and that we always overwrite every

    // pixel in the buffer - it is important to set this in order to get optimal

    // performance when playing through Core Video

    //

    // add this line of code

    cap->subCodecIsMultiBufferAware = true;

    .

    .

    .

}

By setting this flag, a decompressor MUST honor a contract with the Image Compression Manager and be able to do the following:

• Always draw using the ImageSubCodecDecompressRecord.baseAddr/rowBytes passed to ImageCodecDrawBand. Do not save the values of these fields in ImageCodecBeginBand, because that function may be called before the actual destination buffer is known.

• Ignore the base address and row bytes in the port and dstPixMap passed to ImageCodecBeginBand as part of the CodecDecompressParams struct. During multi-buffer playback, these will not point to the correct buffer. In particular, the decompressor must not call QuickDraw to draw to the port.

  NOTE: It's OK to use the dstPixMap to find the pixel format of the output buffers during the ImageCodecPreflight function.

• Always write every pixel in the buffer.

Decompressors that do not set this flag will use a compatibility mode which involves an extra buffer copy per frame when playing to a multi-buffer destination through Core Video.

Some old decompressors (particularly those that do not perform motion compensation), do not write every pixel of difference frames, relying on the old pixels staying present. These decompressors should NOT set the subCodecIsMultiBufferAware flag.

Compatibility Notes

While subCodecIsMultiBufferAware is new field added to the ImageSubCodecDecompressCapabilities structure for QuickTime 7 and Mac OS 10.4, it is completely safe to set this field all the way back to QuickTime 5.0.1. This is because this byte reuses a former padding field. On releases before QuickTime 7 this field is simply ignored.

Sample Code

The ExampleIPBCodec sample demonstrates modern (QuickTime 7 and greater) techniques for writing both Image Compressors and Decompressors.