This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: [PATCH] GCSE after reload - resubmit (SPECint +1.5% fp -0.5%).

From: Kazu Hirata <kazu at cs dot umass dot edu>
To: roger at eyesopen dot com
Cc: dje at watson dot ibm dot com, MUSTAFA at il dot ibm dot com, ZAKS at il dot ibm dot com, gcc-patches at gcc dot gnu dot org
Date: Fri, 19 Mar 2004 02:35:37 -0500 (EST)
Subject: Re: [PATCH] GCSE after reload - resubmit (SPECint +1.5% fp -0.5%).
References: <200403181905.i2IJ5jT07046@makai.watson.ibm.com><Pine.LNX.4.44.0403181130280.21203-100000@www.eyesopen.com>

Hi Roger,

> Do you have any figures on how many partially redundant loads this
> change affects?  Presumably very few given the rare incidence of
> any problems.

I just compiled mainline as of Dec 31, 2003 or so with and without
Mostafa's patch on i686-pc-linux-gnu with -O3 -fomit-frame-pointer
-funroll-all-loops, the same set of options that caused a failure on
H8.  The only difference was found in emit_prefetch_instructions() in
loop.c.  I haven't analyzed if it's miscompiled or not, but it may
well be given the fact that Mostafa's fix did change something.
Anyway, the bug rarely triggers.

Kazu Hirata

References:
- Re: [PATCH] GCSE after reload - resubmit (SPECint +1.5% fp -0.5%).
  - From: David Edelsohn
- Re: [PATCH] GCSE after reload - resubmit (SPECint +1.5% fp -0.5%).
  - From: Roger Sayle

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]