Hi,
> On Thursday 16 January 2003 23:26, Quentin Mason wrote:
> > Hi,
> > I was wondering whether it was common knowledge or whether anyone else has
> > noticed how much faster it is to compile things on MandrakePPC over an
> > Intel machine ?
> > For example:
> > 500 Mhz 1Mb L2 Cache, 384Mb RAM Wallstreet Powerlogix G3 upgrade, g++ 2.95
> > ~3mins Dual 933 MHz Pentium 3, 1Gb RAM ,
> > g++ 3.2.1 2.25 Hrs
>
> You should try this again using the SAME compiler version on both machines.
> gcc 2.95 is known to run circles around 3.x on the same machine, as 3.x had a
> full API change and is now waaay slower than gcc 2.95, though it produces
> faster and smaller binaries.
That has not been my experience -- I have found that g++3.2.1 is somewhat
faster (and smaller memory+stackspace footprint whilst compiling) than
2.95.3. I found an older compiler on the office machines, and it did
indeed take longer than 3.2.1:
Dual 933Mhz P3 g++2.95.3 3.5 hrs
Dual 933Mhz P3 g++3.2.1 2.25 hrs
500Mhz Wallstreet g++2.95.3 ~3mins
I am compiling over NFS -- surely that cannot be the problem for such a
large file and long compile time ?
One of the advantages of the 3.x series is the "-Q" flag on the compiler
-- it tells you exactly what template instantiation or subroutine that it
is working on [watch how compiler-expensive all those #includes are --
what pre-compiled headers could do...], and spits out a time spent
analysis on all the steps (including optimisation). However I do not have
a g++3 for PPC [anyone have a {ppc,src}.rpm to coexist with system
compiler ?] to compare where the time is spent.
Q.
