Self-diagnostic tools of the APEmille parallel machine

The authors describe the self-diagnostic tools of the APEmille SIMD machine, whose logical architecture is a three-dimensional torus of processors. The tools are aimed at implementing system-level diagnosis using a comparison model. The diagnostic model accounts for some critical features of the APEmille architecture, and has been validated by means of VHDL simulation. Essentially, diagnostic tools force all processors to synchronously execute the same test program using the same data set, and compare the outputs of adjacent processors in real time. The diagnostic tools also implement a preliminary test session aimed at covering faults that might disrupt the comparison model. The diagnosis algorithm and the test programs used in comparison tests are also described.