Re: [Gcl-devel] FIXNUM error

[Camm Maguire]

Greetings!  OK, native mips64 object relocation support appears fairly
solid at the moment.  I'm compiling the full set of applications on
gcc42 on the gcc compile farm, which is very similar to your sicortex
setup.

There are two niggles on your machine only which require manual
intervention at the moment:

1) the gcc link stage fails saying the libgcl.a library found with
-lgcl is incompatible.  Simply execute the failed command with the
library name in place of -lgcl, and all proceeds normally.

2) Your linker does not produce a working stub address for _IO_getc
for some reason.  I've disabled the two small gcl optimizations making
use of this:

./saved_ansi_gcl
GCL (GNU Common Lisp)  2.6.8 ANSI    Oct 26 2010 00:05:04
Source License: LGPL(gcl,gmp), GPL(unexec,bfd,xgcl)
Binary License:  GPL due to GPL'ed components: (XGCL READLINE UNEXEC)
Modifications of this banner must retain notice of a compatible license
Dedicated to the memory of W. Schelter

Use (help) to get some basic information on how to use GCL.
Temporary directory for compiler files set to /tmp/

>(remprop 'read-byte 'compiler::co1)
(remprop 'read-byte 'compiler::co1)

T

>(remprop 'read-char 'compiler::co1)
(remprop 'read-char 'compiler::co1)

T

>(si::save-system "ff")

mv ff saved_ansi_gcl

With this maxima build and passes all tests.  The other applications
are currently running.

The gcl source is in gcln, with an ansi build in place.  Export
PATH=....gcln/bin:$PATH to use 'gcl'.

I've also built a working copy of nesl in nesl_2 on this machine.
There were two small patches required, at the head of
neslsrc/build.lisp and in vcode/vstack.c.  The image is dumped --
bin/runnesl works.

Please let me know if any problems persist.  And perhaps if you have a
moment you can do a mini-tutorial on the wonderful things you can do
with nesl, time permitting of course.

Take care,

Frederick Isaac <address@hidden> writes:

> Hi agian,
>  
> looking at the build in gcl it appears that the build did not have 
> --enable-ansi is this true or am I reading
> this wrongly?
>
> On Sat, Oct 9, 2010 at 3:40 AM, Frederick Isaac <address@hidden> wrote:
>
>     Hi Camm,
>      
>     Apologies for not getting back to you sooner. I appreciate all the hard 
> work you have put into this and
>     very much appreciate it. I unfortunately have been swamped at work and 
> was unable to get back to you
>     sooner.
>      
>     I think that the solution you propose with regards using compiler::link 
> seem more than satisfactory for
>     now. After reading your posts under mips64 with regards the GOT and PLT I 
> have been better able to
>     appreciate how the clisp system operates with regards saving a running 
> image.
>      
>     With regards to updating the gcc/binutils I would like to say that these 
> sicortex systems (which has
>     since gone bust) I am not sure. I would ordinarily have few reservations 
> but given the these systems are
>     quite specialized and their use of the pathscale compilers (something I 
> would like to keep) I would
>     probably prefer to post a message to the sicortex user group about in 
> order the get a good solid opinion
>     about. I understand that you would like to support clisp on a mips64 
> architecture but I am sure that you
>     can appreciate my concerns with respect to changing a fundamental feature 
> such a the loader etc.
>      
>     With respect to nesl - you said that you got it working, I have had no 
> such luck. I have tried on linux
>     etc. and the vinterp.serial fails.
>      
>     What part of nesl did you test and did you test it on my machine or yours?
>      
>     Cheers
>      
>      
>    
>     On Fri, Oct 1, 2010 at 2:40 PM, Camm Maguire <address@hidden> wrote:
>    
>         Greetings!
>        
>         Frederick Isaac <address@hidden> writes:
>        
>         > Hi,
>         >  
>         > how are you doing?
>         >  
>         > Do you have much of an idea how much longer I will need to leave 
> this machine open to you?
>         >  
>         > I have seen your posts under mips64 and assume that they are 
> related.
>         >  
>         > Any idea how close your are?
>         >  
>
>         I took the liberty of doing a full optimized build without native
>         object relocation.  This is under the standard directory (not gcln) if
>         you care to use it temporarily.  It can build all applications, but at
>         times one needs to save images with compiler::link instead of
>         si::save-system.  This is described in the documentation.  I can
>         provide details if you need.  gcl builds all the big lisp apps on
>         debian using this method on ia64 and hppa, so it is well supported.
>        
>         I now know exactly what to do for the native object relocation.  What
>         I am unsure of is the timing -- apparently there is a forthcoming gcc
>         improvement which will make things much easier, and I don't know if
>         its worth the wait.  Are you able to upgrade your gcc/binutils/libc to
>         the latest?  Or do so in a chroot?  Or do you need this particular
>         version of the os?  Unless you are an expert, I would recommend
>         sticking with your current set of tools and we take the more
>         cumbersome route.  Please let me know.
>        
>         (In short, it has become clear that the only way to support mips64
>         with this version of the tools is to append stubs to each loaded .o
>         file to guide any calls to external symbols in shared libraries.  This
>         is straightforward, but unfortunate as it complicates the code in a
>         highly non-portable manner.  Minor considerations if you need this
>         functionality, though.)
>        
>         Take care,
>
>         > Cheers
>         >
>         >  
>         > On Sun, Sep 19, 2010 at 4:29 AM, Frederick Isaac <address@hidden> 
> wrote:
>         >
>         >     Of course when you say mips compiler expert what exactly is 
> your question
>         >      
>         >     are you asking about the MIPS command set or the n64 definition 
> which are two different things?
>         >      
>         >     Maybe you are looking for a frame pointer. This is not 
> available under the n32 and n64 mips
>         definition.
>         >      
>         >     Cheers
>         >      
>         >
>         >      
>         >     On Sun, Sep 19, 2010 at 4:24 AM, Frederick Isaac 
> <address@hidden> wrote:
>         >
>         >         Hi,
>         >          
>         >         not really (more of an ARMH person myself)! what is your 
> roadblock - maybe I can look into
>         it for you
>         >         or at least post a message on the sicortex users group and 
> there are a few guys that are
>         mips experts
>         >         on that.
>         >          
>         >         To answer your previous questions - I chose NESL besause 
> the sicortex has quite a uniqie
>         >          interconnect and I wanted that would play into that. I 
> also have a background in N-body
>         problems and
>         >         computational geometry and thus thought NESL would be a 
> good place to start with putting
>         the kautz
>         >         digraph through its paces.
>         >          
>         >         I never anticipated so many problems with getting gcl up 
> and running.
>         >          
>         >         BTW when you say that you tested nesl was this on a IA86 
> linux box and did you try both
>         neslseqsrc
>         >         and neslsrc.
>         >          
>         >         I have been able to run nesseqlsrc on the sicortex but 
> neslsrc is still throwing errors.
>         >          
>         >         Cheers
>         >          
>         >
>         >          
>         >         On Thu, Sep 16, 2010 at 10:35 AM, Camm Maguire 
> <address@hidden> wrote:
>         >
>         >             Greetings!  Just checking -- you don't happen to be a 
> MIPS assembler
>         >             expert, do you?  I have hit a fork in the road and am 
> not sure which
>         >             branch to take.  I can provide details if they happen 
> to be meaningful
>         >             to you.
>         >
>         >             Take care,
>         >             --
>         >             Camm Maguire                                       
> address@hidden
>         >             
> ==========================================================================
>         >             "The earth is but one country, and mankind its 
> citizens."  --  Baha'u'llah
>         >
>        
>         --
>         Camm Maguire                                       address@hidden
>         
> ==========================================================================
>         "The earth is but one country, and mankind its citizens."  --  
> Baha'u'llah
>

-- 
Camm Maguire                                        address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah