Re: target/2427

Re: target/2427

[mmitchel]

The following reply was made to PR target/2427; it has been noted by GNATS.

From: mmitchel@gcc.gnu.org
To: gcc-gnats@gcc.gnu.org, jsm28@cam.ac.uk, nobody@gcc.gnu.org
Cc:  
Subject: Re: target/2427
Date: 5 Jun 2001 23:10:17 -0000

 Synopsis: Extra loop failures with -mcpu=athlon
 
 State-Changed-From-To: open->feedback
 State-Changed-By: mmitchel
 State-Changed-When: Tue Jun  5 16:10:17 2001
 State-Changed-Why:
     Please identify all of the the specific failures -- both 
     those that appear with -mcpu=k6 and -mcpu=athlon.  The former 
     are high-priority; the latter are not.  Until we know
     which failures are which, we cannot know when to downgrade
     the report.
 
 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=2427&database=gcc

Re: target/2427

[Joseph S. Myers]

The following reply was made to PR target/2427; it has been noted by GNATS.

From: "Joseph S. Myers" <jsm28@cam.ac.uk>
To: <mmitchel@gcc.gnu.org>
Cc: <gcc-gnats@gcc.gnu.org>
Subject: Re: target/2427
Date: Wed, 6 Jun 2001 00:31:34 +0100 (BST)

 On 5 Jun 2001 mmitchel@gcc.gnu.org wrote:
 
 >     Please identify all of the the specific failures -- both
 >     those that appear with -mcpu=k6 and -mcpu=athlon.  The former
 >     are high-priority; the latter are not.  Until we know
 >     which failures are which, we cannot know when to downgrade
 >     the report.
 
 The following are the failures with 3.0:
 
 loop-2b.c -mcpu=k6 -O3 -fomit-frame-pointer
 loop-2b.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-loops
 loop-2b.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions
 loop-2b.c -mcpu=k6 -O3 -g
 loop-2b.c -mcpu=k6 -Os
 loop-2b.c -mcpu=k6 -O2
 loop-2b.c -mcpu=athlon -O3 -fomit-frame-pointer
 loop-2b.c -mcpu=athlon -O3 -fomit-frame-pointer -funroll-loops
 loop-2b.c -mcpu=athlon -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions
 loop-2b.c -mcpu=athlon -O3 -g
 loop-2b.c -mcpu=athlon -Os
 loop-2b.c -mcpu=athlon -O2
 loop-2e.c -mcpu=athlon -O3 -fomit-frame-pointer
 loop-2e.c -mcpu=athlon -O3 -g
 loop-2e.c -mcpu=athlon -Os
 loop-2e.c -mcpu=athlon -O2
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-loops
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions
 loop-3c.c -mcpu=k6 -O3 -g
 loop-3c.c -mcpu=k6 -Os
 loop-3c.c -mcpu=k6 -O2
 loop-3c.c -mcpu=athlon -O3 -fomit-frame-pointer
 loop-3c.c -mcpu=athlon -O3 -fomit-frame-pointer -funroll-loops
 loop-3c.c -mcpu=athlon -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions
 loop-3c.c -mcpu=athlon -O3 -g
 loop-3c.c -mcpu=athlon -Os
 loop-3c.c -mcpu=athlon -O2
 
 Of these, the following are regressions:
 
 loop-2b.c -mcpu=k6 -Os
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-loops
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions
 loop-3c.c -mcpu=k6 -O3 -g
 loop-3c.c -mcpu=k6 -Os
 loop-3c.c -mcpu=k6 -O2
 
 Two testcases appear to hang:
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-loops
 loop-3c.c -mcpu=k6 -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions
 
 The rest of the failures abort quickly.
 
 PR target/656 also discusses the same problem.
 
 -- 
 Joseph S. Myers
 jsm28@cam.ac.uk