This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Modulo-scheduling improvements. Patch 2 of 2.
- From: Mircea Namolaru <NAMOLARU at il dot ibm dot com>
- To: Andrey Belevantsev <abel at ispras dot ru>
- Cc: Andrey Belevantsev <andrey dot belevantsev at gmail dot com>, Ayal Zaks <ZAKS at il dot ibm dot com>, David Edelsohn <dje at watson dot ibm dot com>, Dorit Nuzman <dorit at il dot ibm dot com>, gcc-patches at gcc dot gnu dot org, Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>, Vladimir Yanovsky <yanov at il dot ibm dot com>, vmakarov at toronto dot redhat dot com, Vladimir Yanovsky <volodyan at gmail dot com>
- Date: Sun, 24 Jun 2007 00:02:54 +0200
- Subject: Re: [PATCH] Modulo-scheduling improvements. Patch 2 of 2.
Andrey Belevantsev <andrey.belevantsev@gmail.com> wrote on 22/06/2007
22:12:39:
>
>
> I'm not an spu expert, so I don't know is it important to schedule the
> decrement or not. If yes, then the current scheme should be fine,
> otherwise with a normal split maybe the second scheduler will tidy the
> decrement after SMS.
>
> Andrey
If the representation on which SMS is working is not accurate, it
may lead to a sub-optimal scheduling (if the doloop instruction
is split after SMS). Not always the regular scheduling will succeed
to fix this. It is hard to say in which measure this affects the
performance, but it seems preferable to do SMS on a representation
as close as possible to the code to be executed.
Splitting is done before scheduling and SMS. To handle the "doloop"
instruction differently and to split it after SMS doesn't seem
a good option.
Mircea