以mp4decoder为例，大概流程如下：
1 AwesomePlayer::initVideoDecoder()
mVideoSource = OMXCodec::Create(...)
2 OMXCodec::Create()
实际会执行下面代码
#if BUILD_WITH_FULL_STAGEFRIGHT
sp softwareCodec =
InstantiateSoftwareCodec(componentName, source);
#endif
FACTORY_CREATE(MP3Decoder)
FACTORY_CREATE(AMRNBDecoder)
FACTORY_CREATE(AMRWBDecoder)
FACTORY_CREATE(AACDecoder)
FACTORY_CREATE(AVCDecoder)
FACTORY_CREATE(M4vH263Decoder)
FACTORY_CREATE(VorbisDecoder)
FACTORY_CREATE(AMRNBEncoder)
上述decoder里找不到时，会执行如下部分，下面代码才是调到opencore的omx部分
status_t err = omx->allocateNode(componentName, observer, &node);
if (err == OK) {
LOGV("Successfully allocated OMX node '%s'", componentName);
sp codec = new OMXCodec(
omx, node, getComponentQuirks(componentName),
createEncoder, mime, componentName,
source);
observer->setCodec(codec);
err = codec->configureCodec(meta);
if (err == OK) {
return codec;
}
LOGV("Failed to configure codec '%s'", componentName);
}
3 关于read函数
实际read时，先读取数据再调用解码，这一点是与opencore处理机制不同的，如果系统采用软解码，要考虑read加decode作为一个原子执行是否会影响效率，如果decode部分有硬解码则到无妨。
status_t M4vH263Decoder::read(MediaBuffer **out, const ReadOptions *options) {
...
status_t err = mSource->read(&inputBuffer, options);
...
if (PVDecodeVideoFrame(
mHandle, &bitstream, ×tamp, &bufferSize,
&useExtTimestamp,
(uint8_t *)mFrames[mNumSamplesOutput & 0x01]->data())
!= PV_TRUE) {
LOGE("failed to decode video frame.");
inputBuffer->release();
inputBuffer = NULL;
return UNKNOWN_ERROR;
}
...
}
