High Performance Fortran
|
Space (elements of arrays A,B) in problem is mapped into space (i.e. arrays preserved) in Virtual Problem |
---> Compiler can recognize data parallelism |
---> Good parallel performance on all machines |
Fortran77 Do 1 I = 1, 10000
|
Space in problem is mapped into time (successive iterations of DO loop) in Virtual Problem |
---> Compiler has to "undo" this to recognize parallelism and generally fails |
---> Good languages preserve problem architecture |