[Cbe-oss-dev] [HACK 1:9] spufs: Workaround gcc (4.1.1) problem with __weak 'empty' functions

[Ulrich Weigand]

Segher Boessenkool wrote:

> > Workaround gcc (4.1.1) problem with __weak 'empty' functions
> 
> 4.1.x with x < 2
> 
> > The compiler is optimizing weak functions that are empty.
> 
> It considers for inlining all weak functions that are pure
> functions.

Actually, it has nothing to do with inlining.  (Therefore, -fno-inline
etc. have no effect whatsoever on this problem!)

The compiler simply knows that a pure function has no side
effects, and hence is called only for using its result.
A call to a pure function where the result is unused can 
therefore always be eliminated (no matter whether the
function is empty or not).

All this is in fact a completely valid optimization *if the
function that is called is actually the function that was
recognized as being pure*!    This is not the case if
the function is marked weak; that this check was missing
is the bug.


B.t.w. the next release of ppu-gcc will contain a backport
of the fix for this problem.


Mit freundlichen Gruessen / Best Regards

Ulrich Weigand

-- 
  Dr. Ulrich Weigand | Phone: +49-7031/16-3727
  GNU compiler/toolchain for Linux on System z and Cell BE
  IBM Deutschland Entwicklung GmbH
  Vorsitzender des Aufsichtsrats: Martin Jetter | Geschäftsführung: 
Herbert Kircher
  Sitz der Gesellschaft: Böblingen | Registergericht: Amtsgericht 
Stuttgart, HRB 243294