This is the mail archive of the
mailing list for the GCC project.
Re: gcc compile-time performance
- From: Marc Espie <espie at quatramaran dot ens dot fr>
- To: gcc at gcc dot gnu dot org
- Date: Sat, 18 May 2002 17:49:33 +0200
- Subject: Re: gcc compile-time performance
- Organization: Ecole Normale Superieure (quatramaran)
In article <20020518143705.73153F2A59@nile.gnat.com> you write:
>But to take the position that a 20% increase in compilation time is a
>disastrous regression seems excessive rhetoric to me. I still remain
>more concerned that the runtime peformance of GCC seems to be falling
>further behind proprietary compilers on some architectures.
20% using -O1 with the modern compiler, and -O2 with the old.
I call this disastrous because I would imagine that old gcc doing -O2
is doing *more* work at -O2 than modern gcc at -O1.
Show me proof of the contrary, and I'll stop calling that disastrous.
Remember that the first result was that -O2 compilers were 70% slower
on gcc 3.1 than gcc 2.95.3.
I don't have any hard data left between gcc 2.8 and gcc 2.95, but I recall
that there was a significant slowdown there as well...
It means that, between gcc 2.8 and gcc 3.1, compilation time with somewhat
comparable options, has doubled. Even trimming down the options for gcc 3.1
still shows gcc 3.1 as measurably slower than gcc 2.8.
I'm thinking you are a bit too quick to dismiss those hard facts. 20% is
definitely not a statistical anomaly. I will try benching more code. I would
expect new gcc to be anywhere between 10% and 30% slower than the old one
(pitting -O1 vs. -O2), and probably 50% to 100% slower at -O2 levels...