[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

nfib figure for the *prototype* Glasgow Haskell compiler



I don't have time to run a full set of benchmark tests at the moment
(there are some early results in "The Implementation of Haskell Type
Classes", Hammond and Blott, Proc. 1989 Glasgow Workshop on Functional
Programming, Springer Verlag).  However, I have run nfib (26) on our
pet 4Mb Sun-3 using the prototype Glasgow Haskell compiler.

Our results are (Unix "time" averaged over 5 runs, unloaded Sun 3/50):

User   System   Elapsed*  Utilisation   Memory    IO      Page Faults
13.44u 0.24s    0:13      98.75%        0+448k    0+0io   0pf+0w

So our performance is 29229 nfib/s.  This is using an old (0.9?)
Chalmers LML code generator; the Spineless-Tagless code generator is a
rather faster.  

I believe the figures above don't suffer from the methodological flaws
which were cited in another message.  In particular there is no page
faulting and low heap consumption.  Stopwatch timings confirm the
user/elapsed time.  The "Maximum real memory usage" was 57K.  Unshared
data use (reported above) amounted to 448K.

Kevin

NB	Take *all* sequential nfib figures with a dose of salt.

*	This was 0:14 on 2 occasions.  An average here is probably 
	meaningless given the low precision of the raw value.