Problem with stack-adjustment code

[Jan Hubicka]

> On Sat, Apr 29, 2000 at 12:57:05AM +0200, Jan Hubicka wrote:
> > BTW the calls.c seems to be big trouble currently. I would like to know
> > your opinion about my patches.  On the one hand I am trying to cleanup
> > the design on the other hand I am causing breakage here and there.  Do
> > you think my attempts are worts that?
> 
> I believe that cleaning up calls.c is still a worthy cause.  We just
> need to be more careful in the writing and reviewing.  And make the
> changes as small as possible.
Thanks.
I propose as the next step the code permutation I have in the queue.
It splits most of code out of the sib loop, that makes it easier
to split whole expand_call into multiple functions.  I hope this will
avoid problems with placing variaous function calls to other areas.
I will prepare patch shortly.
> 
> > BTW2 my plans about the whole stack alignment issue whas to redesign
> > it basically to do the alignment and neccesary calculation right before
> > the first push instruction is emitted (called from store_one_arg
> 
> I'll remain agnostic to this, not having thought it all the way through.
> If it reduces the number of places we have to care about this mess then
> it's probably a good thing.
I believe so.
> 
> It's probably the case that merging expand_call and emit_library_call
> is the cleanup with the biggest payoff in that code currently.  If we're
> to do more changing in this code, lets tackle that one first, so that
> we can make the other desired cleanups in one place instead of two.

OK. I think I am ready to start - abort we've just fixed was latest hight
priority issue on my TODO list (expce for the Jeff's PA-64 failure I hope to
get kind of testcase soon).  First problem that stopped me from
the idea of consturcting tree node for the calls is the converison of
rtx modes to the tree types. Is it even possible to do such converison?
(ie is there for every possible mode an tree type?)

I still have backdoors with filling the args array.  My code permutation
patch makes this possible.

I am going to sleep now. It is 3pm, but tommorow I will re-update the
permutation patch if I will find the copy of that file on my hdd,
I think I have it only in Prague, otherwise I will try some other work
for weekend.

Thanks,
Honza