rmb@omepd.UUCP (05/15/87)
I have posted the source of the "Doduc" benchmark to net.sources (in 8 parts!). This is one of the benchmarks which John Mashey reported on in his most recent MIPS performance brief, but is not (I believe) commonly available in the benchmarking community. The author, Nhuan Doduc, describes it as follows: The Fortran program simulates the time evolution of a thermo- hydraulical modelization ("hydrocode") for a nuclear reactor's component . The benchmark is created from the computational kernel of the original program ; it has the following profile : - 5300 cards in 40 subroutines , strictly ANS (66/77) , - program size : 330 kbytes or 130 kwords(octal) (CDC) , - (typically) 286 CP seconds of an IBM 3033-U ; - the benchmark will essentially run in 64-bit precision mode eg. in single precision on 60- or 64-bit machines or in dou- ble precision on 'smaller word' (even 16-bit) machines . - the program is best described as an heavy computational (CP profile) with negligible I/O code ; almost all the processing is done with floating point in double precision; but the code doesn't include much array processing that can be vectorized (about 2 %), instead it has an iterative structure with an abundance of short branches and small loops that could be best served by features like cache memory .... - it is important to note that, while minor modifications could be made which would improve its performance quite significantly on a particular machine, it should be run strictly AS IS. The performance metric is a normalized number, R, which is the relative speed of the target machine - hence the bigger the value of R the better. R is computed as 48671/(total cpu time in seconds). An IBM 370/168-3 has an R rating of 100, an IBM 3033-U is 170. Results from the benchmark should be sent to the author, NOT to me. His address is : Nhuan Doduc FRAMENTEC La Boursidiere F 92350 Le Plessis Robinson uucp : ...seismo!mcvax!ftcsun3!ndoduc 'ARPA' : 'mcvax!ftcsun3!ndoduc'@SEISMO.CSS.GOV Bit/Earn : NDODUC at FRSAC11