Compiler parallelization of an elliptic grid generator for 1990 Gordon Bell prize

The paper presents a case study involving the applz­ cation of an automatically parallelizing compiler to a numerically intensive Fortran program, and mchtdes a detai led analysIs of the result. The paper IS based all work performed as part of an entry that won the 1990 Gordon Bell Pnze for Compzler Parallelization. The compiler involved IS the CM Fortran 1.0 com­ piler, which targets the 2048 nodes/FPUs avazlable 111 a full size CM-2 parallel computer. The applicatIOn code implements grid generatIOn for numencal simu­ lation. The paper presents a parallel speedup of 4770 over a serial run on a Sun workstatIOn (which only has 1 FPUj, and includes a cycle counting analysIs of how such a speedup is possible with 2048 FPUs. A full 2048 node/FPU CM-2 is also compared to a 1 node/FPU CM-2, and the measured speedup of 1900 is discussed.