🚀 go-pugleaf

Andrew Koenig  (ark@research.att.com) wrote on 05.10.95
about "Re: Bounds inference [Was: Re: Is the STL too authentic... ?]":

> > An interesting sidelight: in all of the cases where I've been called
> > in because of performance problems, the problem has been memory
> > management.
> I was once able to obtain significant (i.e. obviously perceptible to users)
> performance improvement in a frequently used program by rewriting one
> inner loop so that it caused the compiler to generate three instructions
> instead of five.  This involved allocating one extra byte for an array so as
> to be able to put a sentinel at the end.  I don't know whether you want to
> count that as a memory management problem or not.  I do know that if the
> compiler had done any kind of bounds checking, the performance would have
> been dramatically worse.

Andrew, can you imagine that bounds checking can be switched off ?
Take a good profiler and take a look at your functions. Optimize the few
functions which are critical. For all other functions: use bounds
checking.
And: the use of a good algorithm can speed up your program much more
than all compiler-switches or turning of bounds checking.
We need to provide quality !
---
Walter Sommergruber
email: wsog@site71.ping.at