This class implements the JPEG 2000 entropy coder, which codes stripes in code-blocks. This entropy coding engine can function in a single-threaded mode where one code-block is encoded at a time, or in a multi-threaded mode where multiple code-blocks are entropy coded in parallel. The interface presented by this class is the same in both modes.

The number of threads used by this entropy coder is specified by the "jj2000.j2k.entropy.encoder.StdEntropyCoder.nthreads" Java system property. If set to "0" the single threaded implementation is used. If set to 'n' ('n' larger than 0) then 'n' extra threads are started by this class which are used to encode the code-blocks in parallel (i.e. ideally 'n' code-blocks will be encoded in parallel at a time). On multiprocessor machines under a "native threads" Java Virtual Machine implementation each one of these threads can run on a separate processor speeding up the encoding time. By default the single-threaded implementation is used. The multi-threaded implementation currently assumes that the vast majority of consecutive calls to 'getNextCodeBlock()' will be done on the same component. If this is not the case, the speed-up that can be expected on multiprocessor machines might be significantly decreased.

The code-blocks are rectangular, with dimensions which must be powers of 2. Each dimension has to be no smaller than 4 and no larger than 256. The product of the two dimensions (i.e. area of the code-block) may not exceed 4096.

Context 0 of the MQ-coder is used as the uniform one (uniform, non-adaptive probability distribution). Context 1 is used for RLC coding. Contexts 2-10 are used for zero-coding (ZC), contexts 11-15 are used for sign-coding (SC) and contexts 16-18 are used for magnitude-refinement (MR).

This implementation buffers the symbols and calls the MQ coder only once per stripe and per coding pass, to reduce the method call overhead.

This implementation also provides some timing features. They can be enabled by setting the 'DO_TIMING' constant of this class to true and recompiling. The timing uses the 'System.currentTimeMillis()' Java API call, which returns wall clock time, not the actual CPU time used. The timing results will be printed on the message output. Since the times reported are wall clock times and not CPU usage times they can not be added to find the total used time (i.e. some time might be counted in several places). When timing is disabled ('DO_TIMING' is false) there is no penalty if the compiler performs some basic optimizations. Even if not the penalty should be negligeable.

The source module must implement the CBlkQuantDataSrcEnc interface and code-block's data is received in a CBlkWTData instance. This modules sends code-block's information in a CBlkRateDistStats instance.

Inheritance: CSJ2K.j2k.entropy.encoder.EntropyCoder
コード例 #1
0
ファイル: StdEntropyCoder.cs プロジェクト: RavenB/gridsearch
			/// <summary>The cumulative wall time for this compressor, for each
			/// component. 
			/// </summary>
			//private long[] time;
			
			/// <summary> Creates a new compressor object with the given index.
			/// 
			/// </summary>
			/// <param name="idx">The index of this compressor.
			/// 
			/// </param>
			internal Compressor(StdEntropyCoder enclosingInstance, int idx)
			{
				InitBlock(enclosingInstance);
				this.idx = idx;
#if DO_TIMING
				time = new long[Enclosing_Instance.src.NumComps];
#endif
			}
コード例 #2
0
ファイル: StdEntropyCoder.cs プロジェクト: RavenB/gridsearch
			private void  InitBlock(StdEntropyCoder enclosingInstance)
			{
				this.enclosingInstance = enclosingInstance;
			}