BUG in 2.95.3 - wrong asm code provokes an input variable to be nullified

[Maxime Austruy]

In certain conditions (still not clear) gcc overwrites a parameter
given to a function. The attached file is the preprocessed (ready
to be compiled) file which generates the error.

If we disassembled the function...
   (gdb) disass FaultyFunction
   Dump of assembler code for function FaultyFunction:
   0x13e30 <FaultyFunction>:   push   %ebp
   0x13e31 <FaultyFunction+1>: mov    %esp,%ebp
   0x13e33 <FaultyFunction+3>: sub    $0x17c,%esp
   0x13e39 <FaultyFunction+9>: push   %edi
   0x13e3a <FaultyFunction+10>:        push   %esi
   0x13e3b <FaultyFunction+11>:        push   %ebx
   0x13e3c <FaultyFunction+12>:        mov    $0x22764,%edx
   0x13e41 <FaultyFunction+17>:        call   0x13e42
<FaultyFunction+18>
   0x13e46 <FaultyFunction+22>:        mov    0xc(%ebp),%edi
   0x13e49 <FaultyFunction+25>:        mov    0x10(%ebp),%ebx
   0x13e4c <FaultyFunction+28>:        mov    0x14(%ebp),%eax // ?
   0x13e4f <FaultyFunction+31>:        xor    %edx,%edx       // XXX
   0x13e51 <FaultyFunction+33>:        mov    %eax,0x14(%ebp) // ?
   0x13e54 <FaultyFunction+36>:        mov    %edx,0x18(%ebp) // XXX


A part of this code is generic to all the functions. We see that the
processor clears edx, then put it into ebp[0x18] which corresponds to
the address of the fifth input variable, M. And this looks like a bug.
Apart from this, it generates some weird unusefull instructions,
like just putting a value from memory to eax, and then restoring this
same information to memory.

/******************************************************************/

Here's the .S file generated when we compile using -S -g3 -O2
   [...]
.text
 .align 4
.stabs "FaultyFunction:F(0,19)",36,0,3433,FaultyFunction
.stabs "s1:p(34,32)",160,0,3428,8
.stabs "s2:p(34,32)",160,0,3429,12
.stabs "m1:p(53,7)",160,0,3430,16
.stabs "m2:p(53,7)",160,0,3431,20
.stabs "M:p(53,7)",160,0,3432,24
.globl FaultyFunction
 .type  FaultyFunction,@function
FaultyFunction:
.LBB189:
 pushl %ebp
 movl %esp,%ebp
 subl $380,%esp
 pushl %edi
 pushl %esi
 pushl %ebx
.data
 .align 4
.LP100:
 .long 0
.text
 movl $.LP100,%edx
 call mcount
.stabn 68,0,3433,.LM1059-FaultyFunction
.LM1059:
 movl 12(%ebp),%edi
 movl 16(%ebp),%ebx
 movl 20(%ebp),%eax   // ??
 xorl %edx,%edx       // XXX
 movl %eax,20(%ebp)   // ??
 movl %edx,24(%ebp)   // XXX
   [...]

------------------------------------
[maustruy@valhalla vmx]$ gcc -v
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/2.95.3/specs

gcc version 2.95.3 20010315 (release)
------------------------------------
(I use the RPM of gcc 2.95.3 from PLD)

If this is a known issue and a workaround exists, please let me know!
Thanks,

Maxime
ps: you have to compile the attached file with -O2 for the bug to
appear.
with gcc 2.7.2.3, everything is fine....