[m-dev.] x86 shared library performance

[Tyson Richard DOWD]

> Hi,
> 
> I did some benchmarking of Mercury's shared libraries on Linux.
> Some of you may be interested in the results.

Indeed. These results are somewhat unexpected, considering shared
libraries throw away 50% of the available registers! (sounds more
impressive than 1 register, doesn't it).

> However, when compiling with `-O6 --intermodule-optimization',
> the slowdowns disappeared into the timing noise.

What about -O6 by itself? Is intermodule-optimization on such simple
code reducing the pressure on registers? (they do few intermodule calls
anyway, so perhaps not). Do you have register use profiling results
between O6 and O2 handy?

More importantly, do we know how it will scale?

-- 
       Tyson Dowd           #          Another great idea from the 
                            #            people who brought you
      trd at .cs.mu.oz.au      #               Beer Milkshakes!
http://www.cs.mu.oz.au/~trd #	         Confidence --- Red Dwarf