PICAP 3. A coarse-grained linear SIMD-array

Gives a brief description of the overall architecture of PICAP 3. The redesigned processor module has full floating-point arithmetic so that a 32 module machine will have a peak performance of 320 MFLOP. The authors show how the linear organization and the local address modification can be used efficiently for algorithms like FFT. Transposition, Matrix multiplication, Histogramming, Convolution and Binary image processing. PICAP 3 is orders of magnitudes faster than most commercially available systems.<
>