Re: [PATCH] Separate jump bypassing pass

[Jan Hubicka <jh at suse dot cz>]

> 
> 
> The following patch splits our current GCSE, so that the optimizations
> that significantly modify a function's control flow graph are performed
> after GCC's loop optimizations.
> 
> The motivation for this patch and SPECcpu2000 timings were presented
> in http://gcc.gnu.org/ml/gcc/2002-09/msg01129.html back in September.
> The patch below is "New Bypass" as described in that post.  I had
> originally intended to post it for review shortly afterwards, but
> Mark Mitchell requested that we concentrate fixing bugs in 3.2.1
> and the 3.3 branch before I got around to it.
> http://gcc.gnu.org/ml/gcc/2002-09/msg01157.html
> 
> This patch provides significant performance advantages on
> SPECfp2000, but primarily it allows more aggressive forms of
> GCSE without adversely affecting performance by confusing GCC's
> loop optimizations with irreducible CFGs or misplaced loop notes.
> 
> 
> The following patch has been tested on i686-pc-linux-gnu by a full
> bootstrap, all languages except Ada and treelang, and a complete
> "make -k check" with no new regressions.
> 
> Ok for mainline?

Roger,
thanks for the jump bypassing!  It helps to the performance more than
I've expected :)
I think you should write also something to gcc-3.3/changes.html and
news, so we manifestate some activity to outside world.

Honza