function with side effects ignored

[Matthew Woehlke]

Using gcc-4.3.2-7.x86_64 (Fedora 10).

I have these functions:

void my_ntoh_float(float *f)
{
   int p = (int*)f;
   *p = ntohl(*p);
}

float my_get_float(const void * buffer)
{
   float value;
   memcpy(&val, buffer, sizeof(float);
   my_ntoh_float(&value);
   return value;
}

...which is compiled (-O2) like so:
<my_get_float+0>:   push   %ebp
<my_get_float+1>:   mov    %esp,%ebp
<my_get_float+3>:   sub    $0x10,%esp
<my_get_float+6>:   mov    0x8(%ebp),%eax
<my_get_float+9>:   mov    (%eax),%eax
<my_get_float+11>:  mov    %eax,-0x4(%ebp)
<my_ntoh_float+0>:  mov    -0x4(%ebp),%eax
<my_get_float+17>:  flds   -0x4(%ebp)
<my_ntoh_float+6>:  ror    $0x8,%ax
<my_ntoh_float+10>: ror    $0x10,%eax
<my_ntoh_float+13>: ror    $0x8,%ax
<my_ntoh_float+17>: mov    %eax,-0x4(%ebp)
<my_get_float+34>:  leave
<my_get_float+35>:  ret

Now, obviously this is not going to work correctly; the result of the 
call to my_ntoh_float is effectively discarded. Is the bug in the code, 
or in gcc? In either case, is there a way (besides dropping 
optimizations) to fix this that doesn't involve non-portable code?

(Incidentally, there is at least one gcc "bug" here, in that the 
instruction at my_ntoh_float+0 is superfluous.)

-- 
Matthew
Please do not quote my e-mail address unobfuscated in message bodies.
-- 
"It's not easy for [Microsoft] to accept [competing fairly]"
   -- Dave Heiner (a Microsoft VP, paraphrased)