On Thu, 2005-01-06 at 14:12 +0100, Patrick Lauer wrote: > On Thu, 2005-01-06 at 14:51 +0000, Daniel Drake wrote: > > Good idea. Here's a few: > > > [snip] > > Use make -jX to get multiple compilations happening at once. I think most > > places suggest an X value of one more than your CPU count. > Don't overdo that. Using -jX for X>3*CPU you usually get lower > performance (cpu cache trashing, memory exhaustion). > Also, some ebuilds still revert to -j1
Ayup ... anything over -j2 is probably more trouble than it's worth, even on a multi-processor or cluster. You want your CPU utilization *at* 100 percent; you can't get it *over* 100 percent. > > gcc 2.95 is quite a bit smaller than gcc3, uses less memory, and is a fair > > bit > > quicker at compiling C code (although the resulting binaries are perhaps > > not > > quite as optimized). On slow/low-memory environments like my laptop this > > can > > be quite handy. Note that gcc2 sucks for C++ :) > I wonder how big the differences in code generation are? > (correctness, speed, ...) Well ... gosh ... if you want to go that route, why not use the 2.2 kernel? :) Seriously, though, gcc 3 generates much better code than gcc 2 for number-crunching code and "modern" architectures. There is probably a compile-time penalty, but I'm guessing that can be lowered by choosing lower optimization levels for compiling general software. Note ... performance engineering is what I do for a living. Gentoo is supposed to be a *hobby*, people. :) I charge real money for the good stuff. <weg> -- [email protected] mailing list
