Schnelle Matrizenmultiplikation

This report presents two new ASSEMBLER-subroutines MULR8 and MULC16 for fast multiplication of especially large and not necessarily quadratic matrices with real and complex entries, respectively, developed for the IBM 370/168. Matrix-multiplication is performed by P.C. FISCHER's improvement of the well-known V. STRASSEN-algorithm combined with the method of S. WINOGRAD. By a convenient addition of zero rows and columns, the original matrix dimensions are modified such that anoptimal application of STRASSEN's block method is guaranteed, whereas these zeros are only treated as fictitious, non-arithmetic constants. With respect to execution time, numerical results demonstrate that these new subroutines are superior to other well-known methods for matrix-multiplication, such as the classical algorithm, the procedure of S. WINOGRAD and V. STRASSEN's method