Frame and Fields

In SDK terminology, a frame (or frame surface, interchangeably) contains either a progressive frame or a complementary field pair. If the frame is a complementary field pair, the odd lines of the surface buffer store the top fields and the even lines of the surface buffer store the bottom fields.

 

During encoding, decoding or video processing, cases arise that require reserving input or output frames for future use. In the case of decoding, for example, a frame that is ready for output must remain as a reference frame until the current sequence pattern ends. The usual approach is to cache the frames internally. This method requires a copy operation, which can significantly reduce performance.

SDK functions define a frame-locking mechanism to avoid the need for copy operations. This mechanism is as follows:

1.    The application allocates a pool of frame surfaces large enough to include SDK function I/O frame surfaces and internal cache needs. Each frame surface maintains a Locked counter, part of the mfxFrameData structure. Initially, the Locked counter is set to zero.

2.    The application calls an SDK function with frame surfaces from the pool, whose Locked counter is zero. If the SDK function needs to reserve any frame surface, the SDK function increases the Locked counter of the frame surface. A non-zero Locked counter indicates that the calling application must treat the frame surface as “in use.” That is, the application can read, but cannot alter, move, delete or free the frame surface.

3.    In subsequent SDK executions, if the frame surface is no longer in use, the SDK decreases the Locked counter. When the Locked counter reaches zero, the application is free to do as it wishes with the frame surface.

In general, the application must not increase or decrease the Locked counter, since the SDK manages this field. If, for some reason, the application needs to modify the Locked counter, the operation must be atomic to avoid race condition. Modifying the Locked counter is not recommended.

 

For more complete information about compiler optimizations, see our Optimization Notice.