1. from each element are evaluated one-by-one, This, unfortunately. results in vector lengths of the size of the element matrices, typically about 8-16 long. On the VPS 32. vector lengths of a t least 60 are needed for vectorization payoff and vector lengths of thousands or more are considered optiml. Obviously. alternatives to traditional finite element programing strategies must be considered for the vectorization to be beneficial. Two factors are important in the development of a vectorizing scheme: (a) the need to operate on long vectors, and (b) the extent of vectorization that needs to be achieved in the entire program. Fig. 4 shows the ratio of computational speed to maximum computational speed attainable for various levels of vectorization for computers with vector/scalar computational speeds of 5. 10 and 20. It can be seen for the VPS 32 (vectorlscalar speed ratio of 20). a program that is 90% vectorized will only run a t about 30% of need for a "global' vectorization scheme. the maximum attainable speed. This underlines the
2. Procedure
3. Fig. 6 shows the pressure contours on the symaetry plane of the nozzle a t 2-0. The presence of the expansion waves a t the inlet sections and th? formation of the shock wave and its subsequent reflection from the symmetry plane a t x-17 i s apparent frm this figure. Figs. 7 and 8 compare the finite element solutions w i t h solutions obtained frm a reference plane finite difference procedure 8. Fig. 7 shows the axial variation of pressure a t the intersection of the planes of symmetry (y-2-01. The sharp increase in pressure a t x-17 occurs due to the intersection of shock waves emanating from the walls of four