need more constant constraints

need more constant constraints

[Aldy Hernandez]

hi guys.

we have I,J,K,L,M,N,O,P as possible constant constraints to be used in
CONST_OK_FOR_LETTER_P and friends.

i'm working on altivec support for powerpc and all these letters are
already used.

can i add new cases to:

    recog.c
    regclass.c
    reload.c
    
see patch below.

what is the recommended approach, or is this correct?

thanks
aldy

Index: recog.c
===================================================================
RCS file: /cvs/uberbaum/gcc/recog.c,v
retrieving revision 1.124
diff -c -r1.124 recog.c
*** recog.c     2001/09/21 12:55:15     1.124
--- recog.c     2001/09/27 19:01:38
***************
*** 2503,2508 ****
--- 2503,2512 ----
                  win = 1;
                break;
  
+             case 'A':
+             case 'B':
+             case 'C':
+             case 'D':
              case 'I':
              case 'J':
              case 'K':


-- 
Aldy Hernandez					E-mail: aldyh@redhat.com
Professional Gypsy on a Motorcycle
Red Hat, Inc.

Re: need more constant constraints

[Geoff Keating]

Aldy Hernandez <aldyh@redhat.com> writes:

> hi guys.
> 
> we have I,J,K,L,M,N,O,P as possible constant constraints to be used in
> CONST_OK_FOR_LETTER_P and friends.
> 
> i'm working on altivec support for powerpc and all these letters are
> already used.
> 
> can i add new cases to:
> 
>     recog.c
>     regclass.c
>     reload.c
>     
> see patch below.
> 
> what is the recommended approach, or is this correct?

This is probably not correct, unless you've checked that no port uses
'A' ... 'D' for other things.

Have you tried using EXTRA_CONSTRAINT?

It's also possible that some of the new constraints would be better
expressed as predicates.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: need more constant constraints

[David Edelsohn]

	First, before you do any work on Altivec, please coordinate with
Apple's GCC group and with me.  It sounds like you are duplicating work
already being developed by others.

	We probably can get rid of the 'S' and 'T' EXTRA_CONSTRAINTS in
the PowerPC port.  I don't know why Kenner originally defined 'S'.  I
stayed with the precedent for 'T', but I don't think either serve a useful
purpose other than consistency check.

David

Re: need more constant constraints

[Aldy Hernandez]

On Thu, 2001-09-27 at 22:53, David Edelsohn wrote:
> 	First, before you do any work on Altivec, please coordinate with
> Apple's GCC group and with me.  It sounds like you are duplicating work
> already being developed by others.

yes yes yes.  i would love that, but... it turns out that apple is
basing their code on motorola's original patch, and they haven't
submitted the paperwork for the copyright.

motorola has a present and future copyright assignment but only for 68k
and 88k.  so unless motorola signs off, we can't use the apple/motorola
code base.

i would however, love to coordinate work.

> 
> 	We probably can get rid of the 'S' and 'T' EXTRA_CONSTRAINTS in
> the PowerPC port.  I don't know why Kenner originally defined 'S'.  I
> stayed with the precedent for 'T', but I don't think either serve a useful
> purpose other than consistency check.
> 
> David
-- 
Aldy Hernandez					E-mail: aldyh@redhat.com
Professional Gypsy on a Motorcycle
Red Hat, Inc.

Re: need more constant constraints

[Stan Shebs]

Aldy Hernandez wrote:
> 
> On Thu, 2001-09-27 at 22:53, David Edelsohn wrote:
> >       First, before you do any work on Altivec, please coordinate with
> > Apple's GCC group and with me.  It sounds like you are duplicating work
> > already being developed by others.
> 
> yes yes yes.  i would love that, but... it turns out that apple is
> basing their code on motorola's original patch, and they haven't
> submitted the paperwork for the copyright.
> 
> motorola has a present and future copyright assignment but only for 68k
> and 88k.  so unless motorola signs off, we can't use the apple/motorola
> code base.

This contradicts http://gcc.gnu.org/ml/gcc/2001-07/msg01007.html ,
which Kumar Gala sent out after getting a copy of Motorola's
actual assignment and looking at it.

Given that Motorola's assignment has been repeatedly cast into doubt
by non-Motorola people (myself included), I'd say that at this point
no statement from Motorola will be sufficient, and we're going to
need an explicit ruling by the FSF.  The situation is ridiculous, is
interfering with progress, and I'd like us to put an end to it once
and for all.

Stan

Re: need more constant constraints

[David Edelsohn]

>>>>> Aldy Hernandez writes:

Aldy> yes yes yes.  i would love that, but... it turns out that apple is
Aldy> basing their code on motorola's original patch, and they haven't
Aldy> submitted the paperwork for the copyright.

Aldy> motorola has a present and future copyright assignment but only for 68k
Aldy> and 88k.  so unless motorola signs off, we can't use the apple/motorola
Aldy> code base.

	Your information about the starting point for the work is
completely incorrect.  Apple is not using Motorola's original
implementation exactly because of the copyright issue.

David

Re: need more constant constraints

[Aldy Hernandez]

> 	Your information about the starting point for the work is
> completely incorrect.  Apple is not using Motorola's original
> implementation exactly because of the copyright issue.

uhhh, i thight that's what i unedrstood from stan.  perhaps i'm wrong.

stan?

> 
> David
-- 
Aldy Hernandez					E-mail: aldyh@redhat.com
Professional Gypsy on a Motorcycle
Red Hat, Inc.

Re: need more constant constraints

[Aldy Hernandez]

> This contradicts http://gcc.gnu.org/ml/gcc/2001-07/msg01007.html ,
> which Kumar Gala sent out after getting a copy of Motorola's
> actual assignment and looking at it.

and i quote kumar:

    1. As far as the copyright is concerned.  Motorola did sign a future
    assignment back in August of 1993 for gcc.  I have a copy of it sent
    to me
    by Brian Youmans at FSF.  The 68k/88k references are in Schedule A
    not in
    the actual assignment (it looks like the standard assignment from
    that
    point in time).
    
that would be great.  i was just a bit hesitant because of the 68k/88k
reference.  but someone needs to clarify all this pronto.


    
> Given that Motorola's assignment has been repeatedly cast into doubt
> by non-Motorola people (myself included), I'd say that at this point
> no statement from Motorola will be sufficient, and we're going to
> need an explicit ruling by the FSF.  The situation is ridiculous, is
> interfering with progress, and I'd like us to put an end to it once
> and for all.
> 
> Stan
-- 
Aldy Hernandez					E-mail: aldyh@redhat.com
Professional Gypsy on a Motorcycle
Red Hat, Inc.

Re: need more constant constraints

[Stan Shebs]

David Edelsohn wrote:
> 
> >>>>> Aldy Hernandez writes:
> 
> Aldy> yes yes yes.  i would love that, but... it turns out that apple is
> Aldy> basing their code on motorola's original patch, and they haven't
> Aldy> submitted the paperwork for the copyright.
> 
> Aldy> motorola has a present and future copyright assignment but only for 68k
> Aldy> and 88k.  so unless motorola signs off, we can't use the apple/motorola
> Aldy> code base.
> 
>         Your information about the starting point for the work is
> completely incorrect.  Apple is not using Motorola's original
> implementation exactly because of the copyright issue.

Uh, right now we are.  At one point I was planning to redo everything
because I thought that they didn't have a copyright assignment, but
in July Kumar Gala assured us that Motorola did have one.  Here's a
copy of the message I just sent off to assign@gnu.org, in the hopes
of getting a definitive ruling that everybody will accept:

Subject: Motorola's copyright assignment for GCC
Date: Fri, 28 Sep 2001 07:09:46 -0700
From: Stan Shebs <shebs@apple.com>
To: assign@gnu.org

Hi, a question has arisen on the GCC list about the extent of the
copyright assignment that Motorola has on file.

A couple of years ago, Motorola wrote a set of patches to GCC to
provide support for the AltiVec vector unit in the PowerPC, and
made them generally available.  Recently a number of us have been
working on integrating them into GCC as distributed by the FSF.
However, some persons have expressed doubt that Motorola's
current copyright assignment covers these patches - apparently
the assignment, which is from 1993 and thus predates the PowerPC,
only mentions changes for the 68k and 88k processors, which were
the only two that Motorola offered at the time.

Recently you sent a copy of the old assignment to Kumar Gala of
Motorola, who has said publicly that the processors are only
mentioned in schedule A, but that the body of the assignment
doesn't limit Motorola's assignment to just what is in the
schedule.  So the situation is very confused right now, and GCC
maintainers can't act because they don't know whether the copyright
to the AltiVec patches has or has not already been assigned to the
FSF.

We would greatly appreciate if you would have somebody authoritative
review the assignment, and post the decision about whether the
AltiVec patches are covered by it to gcc@gcc.gnu.org.

Stan Shebs
shebs@apple.com

Re: need more constant constraints

[law]

  In message <3BB47F81.DD4769@apple.com>you write:
  > Given that Motorola's assignment has been repeatedly cast into doubt
  > by non-Motorola people (myself included), I'd say that at this point
  > no statement from Motorola will be sufficient, and we're going to
  > need an explicit ruling by the FSF.  The situation is ridiculous, is
  > interfering with progress, and I'd like us to put an end to it once
  > and for all.
Agreed.
jeff