Improved method of operation in parallel of several computing units, in particular in the processing of images, and corresponding architecture.

摘要

When several computing units intended to evaluate in an iterative manner and the values of convergence of a plurality of variables respectively associated with the various points of a mesh of the predetermined each variable in order to a pre-established iterative relationship between this - same and n variables are associated with n points of the adjacent mesh (p2 - p5), & br / - it has a memory means for each point of the grid, intended to be stored successively all the values of the variable associated with this point of the grid, & br / - it distributes the calculation units so as to have any point of connection to a computation unit and two computing units separate two adjacent points chosen of the grid, & br / - is operated, all the calculation units in parallel and independently of one another, & br / and for each current point of the grid, & br / - it is calculated by the computing unit assigned to the current point each successive value of the variable associated with the current point on the basis of the value contained in the memory means allocated to the current point and of those available and from the memory area assigned to the adjacent points of the current point, regardless of the rows of iteration which correspond to these values are available, and & br / - it is stored, by the calculation unit, each new value thus calculated in the memory means associated with the current point.