全新系列文章已更新：
- Android Media Framework - 开篇
- Android Media Framework（一）OpenMAX 框架简介
- Android Media Framework（二）OpenMAX 类型阅读与分析
- Android Media Framework（三）OpenMAX API阅读与分析
- Android Media Framework（四）Non-Tunneled组件的状态转换与buffer分配过程分析
- Android Media Framework（五）Tunnel Mode
- Android Media Framework（六）插件式编程与OMXStore
- Android Media Framework（七）MediaCodecService
- Android Media Framework（八）OMXNodeInstance - Ⅰ
- Android Media Framework（九）OMXNodeInstance - Ⅱ
- Android Media Framework（十）OMXNodeInstance - Ⅲ
前一节我们了解了input buffer写入的流程，知道了起播写前几笔数据时会先获取graphic buffer，这一节我们就一起来了解下dequeueBufferFromNativeWindow是如何工作的。
1、dequeueBufferFromNativeWindow
ACodec::BufferInfo *ACodec::dequeueBufferFromNativeWindow() {
// 判断是否有 native window
ANativeWindowBuffer *buf;
CHECK(mNativeWindow.get() != NULL);
// tunnel mode无需获取graphic buffer
if (mTunneled) {
ALOGW("dequeueBufferFromNativeWindow() should not be called in tunnel"
" video playback mode mode!");
return NULL;
}
if (mFatalError) {
ALOGW("not dequeuing from native window due to fatal error");
return NULL;
}
int fenceFd = -1;
do {
// 从 native window 获取 buffer
status_t err = mNativeWindow->dequeueBuffer(mNativeWindow.get(), &buf, &fenceFd);
if (err != 0) {
ALOGE("dequeueBuffer failed: %s(%d).", asString(err), err);
return NULL;
}
bool stale = false;
for (size_t i = mBuffers[kPortIndexOutput].size(); i > 0;) {
i--;
// 获取一个 BufferInfo，index从大到小
BufferInfo *info = &mBuffers[kPortIndexOutput].editItemAt(i);
// 如果bufferinfo中的mGraphicBuffer不为null，且mGraphicBuffer的handle等于获取的buffer的handle
if (info->mGraphicBuffer != NULL &&
info->mGraphicBuffer->handle == buf->handle) {
// Since consumers can attach buffers to BufferQueues, it is possible
// that a known yet stale buffer can return from a surface that we
// once used. We can simply ignore this as we have already dequeued
// this buffer properly. NOTE: this does not eliminate all cases,
// e.g. it is possible that we have queued the valid buffer to the
// NW, and a stale copy of the same buffer gets dequeued - which will
// be treated as the valid buffer by ACodec.
// 如果buffer的状态不是属于native window，则说明这个buffer是过时的buffer
if (info->mStatus != BufferInfo::OWNED_BY_NATIVE_WINDOW) {
ALOGI("dequeued stale buffer %p. discarding", buf);
stale = true;
break;
}
ALOGV("dequeued buffer #%u with age %u, graphicBuffer %p",
(unsigned)(info - &mBuffers[kPortIndexOutput][0]),
mDequeueCounter - info->mDequeuedAt,
info->mGraphicBuffer->handle);
// 否则直接将bufferinfo返回，同时给bufferinfo设置write fence
info->mStatus = BufferInfo::OWNED_BY_US;
info->setWriteFence(fenceFd, "dequeueBufferFromNativeWindow");
updateRenderInfoForDequeuedBuffer(buf, fenceFd, info);
return info;
}
}
// It is also possible to receive a previously unregistered buffer
// in non-meta mode. These should be treated as stale buffers. The
// same is possible in meta mode, in which case, it will be treated
// as a normal buffer, which is not desirable.
// TODO: fix this.
if (!stale && !storingMetadataInDecodedBuffers()) {
ALOGI("dequeued unrecognized (stale) buffer %p. discarding", buf);
stale = true;
}
// 如果是过时的buffer则重新从native window获取graphic buffer
if (stale) {
// TODO: detach stale buffer, but there is no API yet to do it.
buf = NULL;
}
} while (buf == NULL);
// get oldest undequeued buffer
// 计算没有使用时间最长的buffer
BufferInfo *oldest = NULL;
for (size_t i = mBuffers[kPortIndexOutput].size(); i > 0;) {
i--;
BufferInfo *info =
&mBuffers[kPortIndexOutput].editItemAt(i);
if (info->mStatus == BufferInfo::OWNED_BY_NATIVE_WINDOW &&
(oldest == NULL ||
// avoid potential issues from counter rolling over
mDequeueCounter - info->mDequeuedAt >
mDequeueCounter - oldest->mDequeuedAt)) {
oldest = info;
}
}
// it is impossible dequeue a buffer when there are no buffers with ANW
CHECK(oldest != NULL);
// it is impossible to dequeue an unknown buffer in non-meta mode, as the
// while loop above does not complete
CHECK(storingMetadataInDecodedBuffers());
// 将获取到的graphic buffer绑定到buffer info上，并且标注上是新的graphic buffer
// discard buffer in LRU info and replace with new buffer
oldest->mGraphicBuffer = GraphicBuffer::from(buf);
oldest->mNewGraphicBuffer = true;
oldest->mStatus = BufferInfo::OWNED_BY_US;
// 设置 write fence
oldest->setWriteFence(fenceFd, "dequeueBufferFromNativeWindow for oldest");
mRenderTracker.untrackFrame(oldest->mRenderInfo);
oldest->mRenderInfo = NULL;
ALOGV("replaced oldest buffer #%u with age %u, graphicBuffer %p",
(unsigned)(oldest - &mBuffers[kPortIndexOutput][0]),
mDequeueCounter - oldest->mDequeuedAt,
oldest->mGraphicBuffer->handle);
updateRenderInfoForDequeuedBuffer(buf, fenceFd, oldest);
return oldest;
}
dequeueBufferFromNativeWindow 的代码比较长，但是如果了解buffer机制很容易就知道这段代码是什么意思了，接下来就来谈谈我的理解。
在开始解码流程之初，我们都知道output buffer没有指向实际的graphic buffer，而正常工作的状态下，一个 BufferInfo 会对应有一个 graphic buffer，但是这里的绑定关系可能会变化的，因为从native window 中获取的 buffer 可能是新的 buffer，这时候就要选择绑定到某一个 BufferInfo 上了。了解了这个内容，我们再来看代码：
- 调用mNativeWindow->dequeueBuffer获取graphic buffer；
- 遍历所有的 BufferInfo，查找是否有handle相同的BufferInfo，如果有就说明这个graphic buffer已经绑定过 BufferInfo 了，接下来就要检查 graphic buffer的状态，如果BufferInfo存储的状态不是OWNED_BY_NATIVE_WINDOW，就说明这块buffer已经被使用了，还没有还给native window，这块buffer是不能够再被使用的，因此需要重新dequeueBuffer；
- 如果dequeue出来的graphic buffer没有对应的 BufferInfo，说明这是一块新的 graphic buffer，需要绑定到一个 BufferInfo，绑定到哪一个上面呢？
- ACodec 给出的方法是绑定到一个最久没有被使用的 BufferInfo 上；如何判断最久没有被使用呢？遍历所有的BufferInfo，判断归属于OWNED_BY_NATIVE_WINDOW，并且mDequeuedAt最小的BufferInfo；ACodec 有一个成员
mDequeueCounter，decoder每次输出一帧，这个计数值会加一；BufferInfo有一个成员mDequeuedAt，这个值用于记录当前BufferInfo在第几帧被使用（填上输出）；有了这两个值，就可以知道当前哪个 BufferInfo 最久没有被使用了。 - 最新绑定的 BufferInfo 还有一个成员
mNewGraphicBuffer会被置为 true，说明这是一个新的 graphic buffer，需要注册给 OMX 组件。
- ACodec 给出的方法是绑定到一个最久没有被使用的 BufferInfo 上；如何判断最久没有被使用呢？遍历所有的BufferInfo，判断归属于OWNED_BY_NATIVE_WINDOW，并且mDequeuedAt最小的BufferInfo；ACodec 有一个成员
- dequeueBuffer 获取 graphic buffer的同时会拿到一个 fence，fence翻译为篱笆，其实就是保护的意思；为什么要这个东西呢？dequeueBuffer 有一种特殊情况，graphic buffer已经被填充返回给 native window，但是buffer中的内容还没有被消费，这时候就重新被获取，并让decoder去填充；但是这很明显是有问题的，decoder需要等待 buffer 中的内容被消费完才能填充这个 buffer，但是graphic buffer的填充和使用是在两个不同的进程当中的，要如何知道buffer被消费完成呢？或者知道buffer被填充完成呢？这里就用fence来实现，填充完成时用fence加锁，等到消费完成解锁，才能继续填充，大致就是这个意思。
- FrameRenderTracker我们后面再研究是干什么用的。
到这 dequeueBufferFromNativeWindow 就分析完成了，拿到graphic buffer之后，就要把它送给decoder使用了。
2、fillBuffer
status_t ACodec::fillBuffer(BufferInfo *info) {
status_t err;
// Even in dynamic ANW buffer mode, if the graphic buffer is not changing,
// send sPreset instead of the same graphic buffer, so that OMX server
// side doesn't update the meta. In theory it should make no difference,
// however when the same buffer is parcelled again, a new handle could be
// created on server side, and some decoder doesn't recognize the handle
// even if it's the same buffer.
if (!storingMetadataInDecodedBuffers() || !info->mNewGraphicBuffer) {
err = mOMXNode->fillBuffer(
info->mBufferID, OMXBuffer::sPreset, info->mFenceFd);
} else {
err = mOMXNode->fillBuffer(
info->mBufferID, info->mGraphicBuffer, info->mFenceFd);
}
info->mNewGraphicBuffer = false;
info->mFenceFd = -1;
if (err == OK) {
info->mStatus = BufferInfo::OWNED_BY_COMPONENT;
}
return err;
}
fillBuffer 的代码比较简单，就是调用 OMXNode 的 fillBuffer 方法，但是分为两种情况：
- 当 BufferInfo 没有发生变化时，调用 fillBuffer 时传入参数为 OMXBuffer::sPreset；什么叫没有发生变化，当output buffer 使用普通buffer，buffer就不会发生变化；当graphic buffer没有发生变化，BufferInfo也视作为没有发生变化；
- 当 BufferInfo 发生变化时，调用 fillBuffer 传入参数为info->mGraphicBuffer；当BufferInfo中绑定了新的graphic buffer时，需要把graphic buffer重新绑定给decoder，因此需要作为参数传递给OMX。
3、OMXNodeInstance::fillBuffer
status_t OMXNodeInstance::fillBuffer(
IOMX::buffer_id buffer, const OMXBuffer &omxBuffer, int fenceFd) {
Mutex::Autolock autoLock(mLock);
if (mHandle == NULL) {
return DEAD_OBJECT;
}
// 找到id对应的 Buffer header
OMX_BUFFERHEADERTYPE *header = findBufferHeader(buffer, kPortIndexOutput);
if (header == NULL) {
ALOGE("b/25884056");
return BAD_VALUE;
}
// 判断buffer type是否为kBufferTypeANWBuffer
if (omxBuffer.mBufferType == OMXBuffer::kBufferTypeANWBuffer) {
// 将 graphic buffer 绑定给 bufferMeta
status_t err = updateGraphicBufferInMeta_l(
kPortIndexOutput, omxBuffer.mGraphicBuffer, buffer, header);
if (err != OK) {
CLOG_ERROR(fillBuffer, err, FULL_BUFFER(
(intptr_t)header->pBuffer, header, fenceFd));
return err;
}
} else if (omxBuffer.mBufferType != OMXBuffer::kBufferTypePreset) {
return BAD_VALUE;
}
// 重置buffer状态
header->nFilledLen = 0;
header->nOffset = 0;
header->nFlags = 0;
// 等待fence释放
// meta now owns fenceFd
status_t res = storeFenceInMeta_l(header, fenceFd, kPortIndexOutput);
if (res != OK) {
CLOG_ERROR(fillBuffer::storeFenceInMeta, res, EMPTY_BUFFER(buffer, header, fenceFd));
return res;
}
{
Mutex::Autolock _l(mDebugLock);
mOutputBuffersWithCodec.add(header);
CLOG_BUMPED_BUFFER(fillBuffer, WITH_STATS(EMPTY_BUFFER(buffer, header, fenceFd)));
}
// 传递给 OMX 组件
OMX_ERRORTYPE err = OMX_FillThisBuffer(mHandle, header);
if (err != OMX_ErrorNone) {
CLOG_ERROR(fillBuffer, err, EMPTY_BUFFER(buffer, header, fenceFd));
Mutex::Autolock _l(mDebugLock);
mOutputBuffersWithCodec.remove(header);
}
return StatusFromOMXError(err);
}
- 当ACodec有新的graphic buffer传递下来时，OMXNodeInstance 检查到 buffer type 为 kBufferTypeANWBuffer，会将新的graphic buffer 绑定到 BufferMeta上，同时也会将 graphic buffer绑定到 BufferHeader 中。
关注公众号《青山渺渺》阅读完整内容； 如有问题可在公众号后台私信，也可进入音视频开发技术分享群一起讨论！
1065
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
