This is the mail archive of the gcc-regression@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: 14 GCC HEAD regressions, 1 new, with your patch on 2005-09-12T08:41:23Z.

From: Andrew Pinski <pinskia at physics dot uc dot edu>
To: Andrew Pinski <pinskia at physics dot uc dot edu>
Cc: jakub at redhat dot com, Dale Johannesen <dalej at apple dot com>, edelsohn at gnu dot org, ebotcazou at adacore dot com, rth at redhat dot com, amodra at bigpond dot net dot au, bkoz at redhat dot com, gcc-regression at gcc dot gnu dot org, coudert at clipper dot ens dot fr
Date: Tue, 13 Sep 2005 02:01:17 -0400
Subject: Re: 14 GCC HEAD regressions, 1 new, with your patch on 2005-09-12T08:41:23Z.
References: <20050913022707.BD4C51A215CCC@gcc-regress.apple.com> <449374c1e40c9acddf750bf0ed2b1055@physics.uc.edu> <d4289557a5b0c139bc2271d65f6bccd8@physics.uc.edu>

On Sep 13, 2005, at 1:23 AM, Andrew Pinski wrote:

The new failures are:
native gcc.sum gcc.c-torture/execute/960512-1.c

This is a big mess.
This latent bug was introduced by:
       * config/rs6000/rs6000.md: Add missing int register
        target case to movdf_low.


I needed to lay down for a minute to get an idea which seems to
fix this testcase (and should fix others which could have came up).

But basically instead of emitting the following sequence for the load:
addis r2,r31,ha16(LC0-"L00000000001$pb")
---
lwz r3,lo16(LC0-"L00000000001$pb")(r2)
addis r4,r31,ha16(LC0-"L00000000001$pb"+4)
lwz r4,lo16(LC0-"L00000000001$pb"+4)(r4)

Where --- was the load for r31 had already happened.
We should emit:
addis r2,r31,ha16(LC0-"L00000000001$pb")
---
lwz r3,lo16(LC0-"L00000000001$pb")(r2)
la r4,lo16(LC0-"L00000000001$pb")(r2)
lwz r4,4(r4)
Which is exactly the same number of registers used and same length as
before.  It also simplifies the mess in darwin.md a little further.

I have not bootstrapped this yet. I can do that tomorrow if people
this is the correct approach.

Thanks,
Andrew Pinski

Attachment: t.diff.txt
Description: Text document

References:
- 14 GCC HEAD regressions, 1 new, with your patch on 2005-09-12T08:41:23Z.
  - From: GCC regression checker
- Re: 14 GCC HEAD regressions, 1 new, with your patch on 2005-09-12T08:41:23Z.
  - From: Andrew Pinski
- Re: 14 GCC HEAD regressions, 1 new, with your patch on 2005-09-12T08:41:23Z.
  - From: Andrew Pinski

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