Re: Using qsort(3) in the kernel

[Johnny Billquist <bqt%softjar.se@localhost>]

Matt Thomas wrote:
> On Nov 17, 2008, at 2:13 AM, Johnny Billquist wrote:
>
> > Andrew Reilly wrote:
> > > On Sun, Nov 16, 2008 at 06:32:15PM -0800, Matt Thomas wrote:
> > > > And not all architectures can do tail call optimizations...
> > > Really?  I thought that a tail call optimization essentially
> > > turned a return into a goto (plus some variable (re)binding).
> >
> > Well, strictly it should replace a call followed by a return, into a 
> > goto. :-)
> >
> > But yeah, I agree. I can't see why any architecture should (could?) 
> > make any difference here.
>
> VAX.  The use of CALLS/RET instruction pretty inhibit the use of tall 
> call optimizations since you can't guarantee the existing call frame has 
> preserved the needed saved registers of the function you will transfer 
> control to.  And let's ignore manufacturing the needed argument list.

For tail call *recursion* the optimization is definitely possible, even 
if you use CALLS/RET on a VAX.

Actually, I only had my mind focused on tail call recursion 
optimization. For general tail call optimization, there are more traps 
for the unwary. And that is probably true for more architectures than VAX.

        Johnny