Image coding or decoding device and method involving multithreading of processing operations over a plurality of processors, and corresponding computer program and synchronisation signal
View Patent ↗A method and apparatus are provided for coding or decoding an image comprising macro-blocks which are distributed in lines and columns. The processing of at least one given macro-block requires the pre-processing of at least one other macro-block on which said dependent macro-block depends. Moreover, the macro-blocks are processed sequentially line by line or column by column. Processing of the macro-blocks is multithreaded over N processors, N≧ 2 . The image is separated into N vertical bands each comprising a plurality of lines and at least one column of macro-blocks if the macro-block is processed sequentially line by line, or into N horizontal bands each comprising a plurality of columns and at least one line of macro-blocks if the macro-block is processed sequentially column by column. One of the N bands is processed by each processor, and the processing operations performed by the N processors is synchronized.
1. Method for coding or decoding an image including macroblocks distributed in lines and columns, comprising:
processing of the macroblocks sequentially line by line or column by column, wherein the processing of at least one given macroblock, referred to as a dependent macroblock, requires prior processing of at least one other macroblock on which said dependent macroblock depends; and
multithreading the processing of the macroblocks over N processors, with N≧2, including the following steps:
separation of the image:
into N vertical bands each including a plurality of lines and at least one column of macroblocks, if the processing of macroblocks is performed sequentially line by line;
into N horizontal bands each including a plurality of columns and at least one line of macroblocks, if the processing of macroblocks is performed sequentially column by column;
processing by each processor of one of the N bands; and
synchronisation of the processing operations carried out by the N processors.
2. Method according to claim 1 , wherein the synchronisation of the processing operations carried out by the N processors comprises, for each processor of each pair of processors that process two adjacent bands:
informing the other processor of said pair about each completed processing of a macroblock on which a dependent macroblock included in the band processed by said other processor depends; and
verifying, before processing a dependent macroblock, that said at least one other macroblock on which the dependent macroblock depends has previously been processed by said processor or said other processor.
3. Method according to claim 1 , wherein said image belongs to the group including:
video sequence images formed by a series of images; and
fixed images.
4. Method according to claim 1 , wherein the processing of the macroblocks is carried out sequentially, line by line, from left to right over the same line and from top to bottom from one line to the next.
5. Method according to claim 4 , wherein the processing of each dependent macroblock requires the prior processing, when it exists, of at least the macroblock located to the left of said dependent macroblock, and wherein, in each pair of processors that process two adjacent vertical bands, the processor that processes the left vertical band informs the other processor of each completed processing of the macroblock of the right end of one of the lines of macroblocks of said left vertical band.
6. Method according to claim 4 , wherein the processing of each dependent macroblock requires the prior processing, when it exists, of at least the macroblock located above and to the right of said dependent macroblock, and wherein, in each pair of processors that process two adjacent vertical bands, the processor that processes the right vertical band informs the other processor of each completed processing of the macroblock of the left end of one of the lines of macroblocks of said right vertical band.
7. Method according to claim 1 , wherein the coding or decoding is consistent with a standard belonging to the group including: H.263, H.263+, H264 and MPEG-4 Video.
8. A non-transitory computer readable medium containing a computer program, comprising program code instructions to implement a method, when said program is run on a computer, for coding or decoding an image including macroblocks distributed in lines and columns, the method comprising:
processing the macroblocks sequentially line by line or column by column, wherein the processing of at least one given macroblock, referred to as a dependent macroblock, requires prior processing of at least one other macroblock on which said dependent macroblock depends; and
multithreading the processing of the macroblocks over N processors, with N≧2, including the following steps:
separation of the image:
into N vertical bands each including a plurality of lines and at least one column of macroblocks, if the processing of macroblocks is performed sequentially line by line;
into N horizontal bands each including a plurality of columns and at least one line of macroblocks, if the processing of macroblocks is performed sequentially column by column;
processing by each processor of one of the N bands; and
synchronisation of the processing operations carried out by the N processors.
9. Device for coding or decoding an image including macroblocks distributed in lines and columns-wherein said device comprises:
N processors, with N≧2, which process the macroblocks, wherein processing of at least one given macroblock, referred to as a dependent macroblock, requires prior processing of at least one other macroblock on which said dependent macroblock depends, wherein the processing of the macroblocks is performed sequentially line by line or column by column; and
means for multithreading the processing of macroblocks over said N processors, which means for multithreading the processing include:
means for separating the image:
into N vertical bands each including a plurality of lines and at least one column of macroblocks, if the processing of macroblocks is performed sequentially line by line; and
into N horizontal bands each including a plurality of columns and at least one line of macroblocks, if the processing of macroblocks is performed sequentially column by column;
in each processor, means for processing one of the N bands; and
means for synchronising the processing operations performed by the N processors.
10. A method comprising:
generating a synchronisation signal to be transmitted from a first to a second processor of a pair of processors that process two adjacent bands of an image, wherein said image includes macroblocks distributed in lines and columns and is separated:
into N vertical bands each including a plurality of lines and at least one column of macroblocks, if the processing of macroblocks is performed sequentially line by line; or
into N horizontal bands each including a plurality of columns and at least one line of macroblocks, if the processing of macroblocks is performed sequentially column by column;
wherein processing of at least one given macroblock, referred to as a dependent macroblock, requires prior processing of at least one other macroblock on which said dependent macroblock depends, said first and second processors belonging to a set of N processors each simultaneously processing one of the bands of the image, in order to code or decode the image;
including into said synchronisation signal information by way of which said first processor informs the second processor of the completed processing by the first processor of a macroblock on which a dependent macroblock included in the band processed by the second processor depends, so that the second processor can verify, before processing said dependent macroblock, that said at least one other macroblock on which the dependent macroblock depends has previously been processed by the first processor; and
transmitting the synchronization signal from the first processor to the second processor.