optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()

optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()

[gcc]

>Number:         7460
>Category:       optimization
>Synopsis:       gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          ice-on-legal-code
>Submitter-Id:   net
>Arrival-Date:   Wed Jul 31 15:26:01 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     tbp
>Release:        gcc version 3.1.1 20020718 (prerelease)
>Organization:
>Environment:
w2k/cygwin on an athlon.
Reading specs from /usr/lib/gcc-lib/i686-pc-cygwin/3.1.1/specs
Configured with: /netrel/src/gcc-3.1.1-4/configure --enable-languages=c,c++,f77,java --enable-libgcj --enable-threads=po
six --with-system-zlib --enable-nls --without-included-gettext --enable-interpreter --disable-sjlj-exceptions --disable-
version-specific-runtime-libs --enable-shared --build=i686-pc-linux --host=i686-pc-cygwin --target=i686-pc-cygwin --enab
le-haifa --prefix=/usr --exec-prefix=/usr --sysconfdir=/etc --libdir=/usr/lib --includedir=/nonexistent/include --libexe
cdir=/usr/sbin
Thread model: posix
gcc version 3.1.1 20020718 (prerelease)
>Description:
--> gcc -O2 -march=athlon bug.c -o bug.exe
bug.c: In function `main':
bug.c:5: internal error: Segmentation fault
Please submit a full bug report,
>How-To-Repeat:
conjunction of __builtin_ia32_femms() and -Ox (x>1) when compiling for -march=athlon triggers an internal error.

test case:
int main(int argc, char *argv[]) {
	__builtin_ia32_femms();
	exit(0);
}
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted:

Re: optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()

[bangerth]

Synopsis: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()

State-Changed-From-To: open->closed
State-Changed-By: bangerth
State-Changed-When: Fri Dec  6 13:38:32 2002
State-Changed-Why:
    I can confirm this problem with gcc3.2. However, since the
    respective builtin did not exist in 3.0, this is not a 
    regression and will certainly not be fixed for 3.2.2. It
    works on 3.3CVS, so I close this report. It will be fixed
    for the next major release, gcc3.3

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7460

Re: optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()

[Richard Henderson]

The following reply was made to PR optimization/7460; it has been noted by GNATS.

From: Richard Henderson <rth@redhat.com>
To: Graham Stott <graham.stott@btinternet.com>
Cc: gcc@tbp.dyndns.org, gcc-gnats@gcc.gnu.org, gcc-patches@gcc.gnu.org
Subject: Re: optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()
Date: Thu, 1 Aug 2002 16:44:28 -0700

 On Thu, Aug 01, 2002 at 12:53:36AM +0100, Graham Stott wrote:
 > This doesn't fix the generic problem with the define_attr "memory"
 > definition which is assuming an insn always has two operands.
 
 This is a feature, really.
 
 >         * config/i386/i386.md ("femms"): Add "memory" attr "none".
 
 Ok for mainline now.  I don't believe the 3.2 branch is
 open at present, but once it is, it's ok there too.
 
 
 r~

Re: optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault on __builtin_ia32_femms()

[Graham Stott]

The following reply was made to PR optimization/7460; it has been noted by GNATS.

From: Graham Stott <graham.stott@btinternet.com>
To: gcc@tbp.dyndns.org
Cc: gcc-gnats@gcc.gnu.org,  gcc-patches@gcc.gnu.org
Subject: Re: optimization/7460: gcc/g++ 3.1.1 20020714/20020718 (cygwin) segfault
 on __builtin_ia32_femms()
Date: Thu, 01 Aug 2002 00:53:36 +0100

 All,
 
 Confirmed also present CVS mainline and not cygwin specific.
 
 The problem here is the define_attr "memory" in i386.md assumes that an insn always has two operands
 which is not the case for femms so when get_attr_memory tries to determine the "memory" attribute
 for the femms insn it tries to access operand 2 which doesn't exist.
 
 The patch fixes this problem for femms by explicitly specifying the "memory" attrib on the insn itself
 so that get_attr_memory nevers gets called for this insn.
 
 This doesn't fix the generic problem with the define_attr "memory" definition which is assuming an insn
 always has two operands.
 
 bootstrapped i686-pc-linux-gnu all languages no regressions.
 
 OK for mainline and 3.2?
 
 Graham
 
 ChangeLog
 
 2002-07-31  Graham Stott  <graham.stott@btinternet.com>
 
         * config/i386/i386.md ("femms"): Add "memory" attr "none".
 
 -------------------------------------------------------------
 Index: i386.md
 ===================================================================
 RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.md,v
 retrieving revision 1.378
 diff -c -p -r1.378 i386.md
 *** i386.md     16 Jul 2002 10:24:11 -0000      1.378
 --- i386.md     31 Jul 2002 23:35:13 -0000
 *** 19819,19825 ****
       (clobber (reg:DI 36))]
      "TARGET_3DNOW"
      "femms"
 !   [(set_attr "type" "mmx")])
 
    (define_insn "pf2id"
      [(set (match_operand:V2SI 0 "register_operand" "=y")
 --- 19819,19826 ----
       (clobber (reg:DI 36))]
      "TARGET_3DNOW"
      "femms"
 !   [(set_attr "type" "mmx")
 !    (set_attr "memory" "none")])
 
    (define_insn "pf2id"
      [(set (match_operand:V2SI 0 "register_operand" "=y")
 -----------------------------------------------------------