Re: A scheduler/aliase bug?

[Jeffrey A Law <law at cygnus dot com>]

  In message <m101bGH-00038dC@ocean.lucon.org>you write:
  > With egcs 1.1.1 and 1999-01-12 on 586, I got
  > 
  > # gcc -B/home/work/build/bin/egcs-1.1.1/gcc/ -O2 -fPIC 980506-2a.c -mcpu=i5
  > 86
  > # a.out
  > zsh: 28220 abort      ./a.out
  > 
  > The bug may be in scheduler or aliase analysis. Add -fno-schedule-insns2
  > hides the bug. The problem is around
  > 
  >         call self@PLT
  >         movl $0,-4(%ebp)
  >         movl (%eax),%eax
  > 
  > schedule_block () changes it to
  > 
  >         call self@PLT
  >         movl (%eax),%eax
  >         movl $0,-4(%ebp)
  > 
  > But it doesn't notice -4(%ebp) == (%eax). BTW, this testcase is almost
  > the same as gcc.c-torture/execute/980506-2.c.
  > 
  > The interesting thing is it only happens with -mcpu=i586. i386, i486 and
  > i686 are ok.
You might try Mark Mitchell's patch for MEM_IN_STRUCT_P (which I haven't had
a chance to look at yet).

jeff