[Bug middle-end/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[Bug middle-end/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.10.0

[Bug middle-end/61949] New: [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

            Bug ID: 61949
           Summary: [4.10 regression] SEGV compiling
                    gcc.dg/pch/import-[12].c
           Product: gcc
           Version: 4.10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ro at gcc dot gnu.org
                CC: rguenth at gcc dot gnu.org
              Host: i386-pc-solaris2.10
            Target: i386-pc-solaris2.10
             Build: i386-pc-solaris2.10

Between 20140711 (r212451) and 20140718 (r15669), the gcc.dg/pch/import-[12].c
tests started to FAIL on Solaris 10/x86 (only, Solaris 11 is unaffected):

FAIL: gcc.dg/pch/import-1.c  -O0 -g -I. -Dwith_PCH (internal compiler error)
FAIL: gcc.dg/pch/import-1.c  -O0 -g -I. -Dwith_PCH (test for excess errors)
FAIL: gcc.dg/pch/import-1.c -O0 -g assembly comparison

and all others, both 32 and 64-bit.

cc1 SEGVs here:

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 1 (LWP 1)]
0x08c34226 in md5_read_ctx (resbuf=0x8047278, ctx=0x8047140)
    at /vol/gcc/src/hg/trunk/local/libiberty/md5.c:91
91        memcpy (resbuf, buffer, 16);
1: x/i $pc
=> 0x8c34226 <md5_finish_ctx+150>:      movaps -0x28(%ebp),%xmm0

-0x28(%ebp) isn't 16-byte aligned as required.

This only happens with the stage3 cc1, stages 1 and 2 are unaffected.

A reghunt revealed that this is caused by this patch:

2014-07-11  Richard Biener  <rguenther@suse.de>

       PR middle-end/61473
       * builtins.c (fold_builtin_memory_op): Inline memory moves
       that can be implemented with a single load followed by a
       single store.
       (c_strlen): Only warn when only_value is not 2.

  Rainer

[Bug middle-end/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2014-07-30
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
it's odd that stage2 is not affected...  can you provide preprocessed source
of libiberty/md5.c as compiled for the host?

[Bug middle-end/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[ro at CeBiTec dot Uni-Bielefeld.DE]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

--- Comment #2 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> ---
> --- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
> it's odd that stage2 is not affected...  can you provide preprocessed source

I lied: if you generate the .gch file with the stage2 cc1, both stage2
and 3 cc1 SEGV the same way.

> of libiberty/md5.c as compiled for the host?

Attached.

    Rainer

[Bug middle-end/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

--- Comment #3 from Rainer Orth <ro at gcc dot gnu.org> ---
Created attachment 33213
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33213&action=edit
preprocessed libiberty/md5.c

[Bug target/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Reduced testcase, compile with -m32 -msse -O2 -fno-omit-frame-pointer
(works on a x86_64-linux host, no 32bit host available right now):

struct md5_ctx
{
  unsigned int A;
  unsigned int B;
  unsigned int C;
  unsigned int D;

  unsigned int total[2];
  unsigned int buflen;
  char buffer[128] __attribute__ ((__aligned__ (__alignof__ (unsigned int))));
};

void * __attribute__((noinline,noclone))
md5_read_ctx (const struct md5_ctx *ctx, void *resbuf)
{
  unsigned int buffer[4];

  buffer[0] = (ctx->A);
  buffer[1] = (ctx->B);
  buffer[2] = (ctx->C);
  buffer[3] = (ctx->D);

  __builtin_memcpy (resbuf, buffer, 16);

  return resbuf;
}


int main()
{
  char resbuf[16];
  struct md5_ctx c;
  md5_read_ctx (&c, resbuf);
  return 0;
}

Does this testcase fail on Solaris 10?

[Bug target/61949] [4.10 regression] SEGV compiling gcc.dg/pch/import-[12].c

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

--- Comment #6 from H.J. Lu <hjl.tools at gmail dot com> ---
How is GCC configured for i386-pc-solaris2.10?
i386 won't use SSE instructions by default.

[Bug target/61949] [5 regression] SEGV compiling gcc.dg/pch/import-[12].c

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Judging from gcc-testresults, this was showing up in late July and early
August, but already can't be seen e.g. in r213757.  Has it been fixed?  Can
this be closed?

[Bug target/61949] [5 regression] SEGV compiling gcc.dg/pch/import-[12].c

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61949

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|---                         |FIXED

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Assuming fixed based on the gcc-testsuite results.  Please reopen if that is
not the case.