Re: inline-asm/9570: [3.3/3.4 regression] Assember error with -finline-functions with g++-3.3

[Adam Lackorzynski <adam at os dot inf dot tu-dresden dot de>]

Hi,

On Sat Mar 15, 2003 at 17:24:14 -0000, jason at gcc dot gnu dot org wrote:
> Synopsis: [3.3/3.4 regression] Assember error with -finline-functions with g++-3.3
> 
> State-Changed-From-To: analyzed->closed
> State-Changed-By: jason
> State-Changed-When: Sat Mar 15 17:24:13 2003
> State-Changed-Why:
>     This is not a bug.  Your code assumes that the inline assembly
>     will be emitted after the definition of foo(), but if foo() is
>     inline (as a result of -finline-functions), it is deferred until
>     EOF, so the inline assembly is emitted after the definition of i.
>     i lives in .bss, so you end up trying to emit executable code into
>     .bss, which doesn't work very well.  You probably want to wrap
>     that code in .pushsection ".text" and .popsection
> 
> http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=9570

Just a small annotation: I think the problem is the location of the "b"
definition. If I pull the definition of "b" out of the
function, the test case compiles. Looking at the asm-code, the place of
the definition (or whatever it exactly is) of b differs and if it's
below the inline asm code, it doesn't work.

--- x.c Sat Mar 15 19:56:23 2003
+++ x2.c        Sat Mar 15 19:56:36 2003
@@ -1,7 +1,7 @@
 int i = 0;
 
-void foo() {
   static bool b;
+void foo() {
   if (! i)
     asm("movl %%esp, %0" : "=r" (i));
 }


Adding .pushsection and .popsection around the call does seem to fix it
as well.



Thanks,
Adam
-- 
Adam                 adam at os dot inf dot tu-dresden dot de
  Lackorzynski         http://os.inf.tu-dresden.de/~adam/