Re: High Level Language vs Assembly

> > When I took a compiler class in Grad School, my professor was
> > adamant about the efficiency of code produced by new high level
> > language compilers. He stated that a human assembly language writer
> > was no longer able to compete with these compilers (unless, of course
>


<snip lots of discussion about human vs machine compilers>


I would think talking about how good compilers are in this news group
is like singing to the choir, so let me ask this; how much improvement
is there between non-optimized and highly optimized code? I ask
because in two areas where performance is critical, DSPs and
supercomputing, compilers have a really poor reputation. I've noticed
the difference between non-optimized and optimized to be much greater
for these types of machines than general purpose cpus. If I turn on
all the bells, whistles and preprocessors, and go in there and hand
tune I might get a 30% improvement of code on a workstation. If I do
the same thing on a parallel machine I might find a 500% difference.
30% difference is nice, I really don't care if I gain it or not. So
I'll never write assembly code for a workstation. But 500% is
critical. If I can get that kind of improvement by mucking around in
the code then it's worth it. So I use MPI. Another aspect of this
problem is that sometimes the compiler can get the 500% improvement,
but not always, and usually not without someone reverse engineering
the compiler to give it the code it wants. So maybe the question is
will compilers ever mature enough to not lose the 500% on contorted
architectures like DSPs and parallel computers?


Matt

Return to the comp.compilers page.
Search the comp.compilers archives again.