This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: 3.3-branch QA assessment
- From: Matt Austern <austern at apple dot com>
- To: Joe Buck <jbuck at synopsys dot com>
- Cc: Roger Sayle <roger at www dot eyesopen dot com>, gcc at gcc dot gnu dot org, Neil Booth <neil at daikokuya dot co dot uk>
- Date: Mon, 27 Jan 2003 11:38:57 -0800
- Subject: Re: 3.3-branch QA assessment
On Monday, January 27, 2003, at 09:37 AM, Joe Buck wrote:
On Sat, Jan 25, 2003 at 09:11:54AM -0700, Roger Sayle wrote:
The latest graphs show that bootstrap times are decreasing:
http://www.suse.de/~aj/SPEC/CINT/d-permanent/Bootstrap-time_big.png
Ah, but what caused that enormous jump around Dec. 18 or so?
Can anyone pinpoint the day it occurred?
I'm not convinced that bootstrap times are a useful measurement
of compile speed, since they mung together too many things:
compile speed, which source is compiled (which can in turn depend
on default config parameters), etc. I'd be more interested in seeing
a graph of time to compile some specific constant piece of code.
As before, I recommend a file from qt. I can post preprocessed
source for the file I was using for measurements, if anyone is
interested. It's long enough to be interesting, and I think it's
representative of at least one common style of C++ programming.
My preliminary findings are the same as Andi's: the biggest regression
was in the time spent in the gc. Perhaps this is just a straightforward
bug somewhere in the collector. Wouldn't that be nice?
--Matt