An improved method of parallel operation of several computing units, in particular in image processing, and corresponding architecture.

摘要

P When several computational units intended together to evaluate iteratively and cellularly the convergence values of a plurality of variables respectively associated with the different points of a predetermined mesh each variable to satisfy a preestablished iterative relationship between itself and n neighboring variables associated with n neighboring mesh points (P2-P5), BR/ - one memory means is allocated per mesh point, for successively storing all the values of the variable associated with this mesh point, BR/ - the calculation units are distributed in such a way as to allocate any mesh point to a calculation unit and two distinct calculation units to two selected neighboring points of the mesh, BR/ - all the units of the grid are operated computation in parallel and independently of each other, BR/ and for each current point of the mesh, BR/ - it is computed by the calculation unit assigned to this point c for each successive value of the variable associated with this current point from the value contained in the memory means assigned to the current point and from those available and coming from the memory means assigned to the points adjacent to the current point, whatever the ranks. iterations to which correspond these available values, and BR/ - is made to store, by said calculation unit, each new value thus calculated in the memory means associated with the current point. / P