This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: testing Octave


> Date: Wed, 12 Aug 1998 18:50:28 -0500
> From: Mumit Khan <khan@xraylith.wisc.edu>

> I typically do test blitz since the code is close to what I do. My code
> is quite a bit easier on compilers than blitz, and has been working
> happily under egcs snapshots for a the past 8 weeks or so. The big
> problem with both mine and blitz is that I can't use the optimizer;
> eg., if you specify -O[1234] when compiling code using blitz arrays
> for example, egcs will quite likely eat up all the RAM. I've seen upto
> 800M or so, until the system or I kill it. Kai eats up about 100M or
> so on the same code using the nastiest optim flags I could find, and
> *very slowly* does the job.

This reminds by of the guys that said that on machines where a float
operation takes the exact same time as a double operation, in some
codes, floats are twice as fast as doubles, because you can fit twice
as many into the cache.

A friend once said to me that the most compact form of saving away the
template inlines was to just use text, as it is smaller than anything
else, and can be processed quickly.

What are your times (memory) like if you use -O9 and
-fno-inline-functions and let us know how it goes.  I am thinking that
you'll get much better code that -O0.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]