This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] bnw-simple-branch now usable
- From: Diego Novillo <dnovillo at redhat dot com>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: Jason Merrill <jason at redhat dot com>, Daniel Berlin <dberlin at dberlin dot org>, Richard Henderson <rth at redhat dot com>, Ben Elliston <bje at redhat dot com>, "Frank Ch. Eigler" <fche at redhat dot com>, Gabriel Dos Reis <gdr at codesourcery dot com>, PopSébastian <pop at gauvain dot u-strasbg dot fr>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Date: 12 Sep 2002 11:37:53 -0400
- Subject: Re: [tree-ssa] bnw-simple-branch now usable
- References: <200209121512.LAA28772@makai.watson.ibm.com>
On Thu, 2002-09-12 at 11:12, David Edelsohn wrote:
> Would you please quantify the compile-time performance of this new
> functionality relative to the current compiler? How efficient are the
> tree simplifier rewrite and new IR? What impact do they have on the speed
> of the compiler?
>
The idea was not so much address performance problems but make the tree
IR a bit more reasonable (especially the C one). From what I
understand, the new IR should not have any performance impact, but
that's easy to track. We keep track of bootstrap and SPEC build times
in the SPEC95 page at
http://people.redhat.com/dnovillo/SPEC/tree-ssa-branch/
We also run SPEC2k periodically. I'm planning to put the SPEC2k
tracking page on a public server soon.
The compile-time performance problems on the branch are largely due to
the DFA and SSA passes. Also, the simplifier leaves many small trees
that translate to quite a bit of RTL. Once we have a working set of
simple scalar cleanup passes (pre, ccp, dce, copy-prop, etc) we should
start looking at profiling with more detail.
Diego.