[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Profile of current guile HEAD
From: |
Ludovic Courtès |
Subject: |
Re: Profile of current guile HEAD |
Date: |
Mon, 06 Aug 2007 10:24:27 +0200 |
User-agent: |
Gnus/5.11 (Gnus v5.11) Emacs/22.1 (gnu/linux) |
Hey!
Andy Wingo <address@hidden> writes:
> Some more data points. I just profiled a run of guile -c 1 from HEAD.
> This is what callgrind says about which functions are taking the most
> time:
>
> cumulative self total
> percent percent calls file:function
> 23.15 19.30 151231
> ports.c:scm_getc[/opt/guile-head/lib/libguile.so.18.0.0]
This one ought to be inline, it's really just a few instructions.
> I still think that the TLS patch can help with the pthread_getspecific
> stuff; I should profile with that patch. Later maybe.
Wrt. TLS: GCC's default TLS access model (called `global-dynamic')
involves one function call for each TLS access, making it comparable to
using `pthread_getspecific ()' performance-wise. I'm investigating
whether we can rightfully use one of the faster access models.
Hopefully more to come soon!
> FWIW, here's a profile of (use-modules (oop goops)):
[...]
> Reading is still probably the highest cost here, but seeing ceval so
> high normally means that most of the time is spent in Scheme code.
We must be spending time in `(oop goops compile)' or similar?
Cheers,
Ludovic.