Re: gcc.c-torture/execute/frame-address.c

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Aug 29, 2005, at 11:11 AM, Maciej W. Rozycki wrote:

> On Mon, 29 Aug 2005, Alan Modra wrote:
>
> > This test fails on powerpc64-linux, because the function invoking
> > __builtin_frame_address doesn't have a stack frame.  Fixed by taking
> > the address of a local var.
> >
> > 	* gcc.c-torture/execute/frame-address.c: Ensure function
> > 	invoking __builtin_frame_address has a stack frame.
> >
> > OK to apply?
>
>  Well, the function already has a local variable.  Besides it's never
> inlined, so it should always have a frame (i.e. an address local 
> variables
> are/would be offset from), even if empty.  Are you sure
> __builtin_frame_address() does the right thing for your target?  
> Tracking
> down problems with the builtin was the objective of the test case, so
> please make sure you don't just hide the symptoms...


It does not have a frame because we sibling call optimize check_fa_work
which enables us not to have a frame pointer (or a stack pointer change)
in check_fa_mid.  This is not hiding the symptom, this is fixing the
testcase for assuming things wrong.

-- Pinski