positron@cosmic.berkeley.edu (Shigeki Misawa) (09/26/90)
The physics group that I am working for is searching for a computer to do numerically intensive data analysis. If anyone has had experience porting fortran code from a VAX to a DECstation, we would be interested in hearing about porting problems and performance gains or losses. Also, if anyone has any information concerning the SPEC ratings of the various DEC risc boxes and VAX boxes, we would appreciate it if you could email us that information. Thanks in advance. Shigeki Misawa - UCB Physics Department.
GEOMAGIC@kcgl1.eng.ohio-state.edu (Daniel OConnell) (09/26/90)
> The physics group that I am working for is searching for a >computer to do numerically intensive data analysis. If anyone has >had experience porting fortran code from a VAX to a DECstation, we >would be interested in hearing about porting problems and performance >gains or losses. Also, if anyone has any information concerning the >SPEC ratings of the various DEC risc boxes and VAX boxes, we would >appreciate it if you could email us that information. Thanks in advance. > >Shigeki Misawa - UCB Physics Department. Configuration DEC3100 vs VAX3100: DEC3100 was ~ 4 times faster than the VAX3100 using -O2 FORTRAN optimization. The code was mostly double precision linear inversion (singular value decomposition, matrix multiplies, ray tracing, etc). HOWEVER, I had to compile my ~4000 line main program on our DEC5400 (32 MB memory vs 16MB on the DEC3100) when optimizing because the optimizer needed 27 MB of memory to avoid paging to oblivion. The code recompiled with some VAX FORTRAN extensions without any problems, except a CHARACTER formatting problem in a FORMAT statement. A ~200,000 line VAX specific program recompiled and worked, although some FORTRAN 66 H character formats needed fixing. In summary, the DEC3100, etc Ultrix f77 2.1 compiler appears to compile VAX fortran code reliably, many times without any modification. Certain things still scare me, like program continuation after encountering NANS and NO ERROR MESSAGES to alert the user, arcsin( > 1 ) = 0?, sqrt( x < 0 ) = 0?, etc. You get the idea. Happy hunting, Dan O'Connell geomagic@kcgl1.eng.ohio-state.edu VMS geomagic@geo1s.mps.ohio-state.edu Ultrix geomagic@you.must.be.joking WYLBUR