[cmucl-help] 19c alpha [Was Re: CMUCL 18c building on tru64 5.1]

[Fausto Saporito]

Hello Ray,

is it possible create a Linux version of lisp without linkage-table,
modular-arith, etc... just to build a cross-compiler with more or less
the same alpha features ?

If so, how can I do that?


2014-09-10 0:55 GMT+02:00 Fausto Saporito <fausto.saporito at gmail.com>:
> I did such change... but I have always the same error...
> I checked in the source code and that function should be included only
> with modular-arith!
>
> ;; Not sure this is 100% right, but the defoptimizer for ash only
> ;; calls ash-left-mod32 when the COUNT is known to be a (UNSIGNED-BYTE
> ;; 5), which is what %ASHL wants.
> #+modular-arith
> (defun vm::ash-left-mod32 (integer count)
>   (declare (type integer integer)
>            (type (integer 0 31) count))
>   (bignum::%ashl (ldb (byte 32 0) integer) count))
>
> if I understood well the #-syntax.
>
> Maybe, the cross compiler override such definition... I don't know...
>
>
> 2014-09-10 0:22 GMT+02:00 Raymond Toy <toy.raymond at gmail.com>:
>>
>>
>> On Tue, Sep 9, 2014 at 3:14 PM, Fausto Saporito <fausto.saporito at gmail.com>
>> wrote:
>>>
>>> ok... this is interesting:
>>>
>>> ldb> print $1
>>> $1=       0x30182847: other pointer
>>>             header: 0x000003ba: fdefn
>>> $2=         name: 0x30182817: ASH-LEFT-MOD32
>>> $nil=       function: 0x2800000b: NIL
>>>             raw_addr: 0x00010140: 16464
>>>
>>> I remember ASH-LEFT-MOD32... is something related to modular-arithmetic ?
>>> I disabled such feature, but I'm not sure if this could be related.
>>
>>
>> Yes, that's modular arithmetic.  Something is leaking.
>> Maybe in your cross script, add something like
>>
>> (setf *features* (remove :modular-arith *features*))
>>
>> after c::new-backend. It might be good to do that with :double-double and
>> :linkage-table. Not sure though, but probably harmless if you do since alpha
>> doesn't support that.
>>
>>>
>>>
>>>
>>> 2014-09-10 0:04 GMT+02:00 Raymond Toy <toy.raymond at gmail.com>:
>>> >
>>> >
>>> > On Tue, Sep 9, 2014 at 3:00 PM, Fausto Saporito
>>> > <fausto.saporito at gmail.com>
>>> > wrote:
>>> >>
>>> >> Now lisp is more verbose:
>>> >>
>>> >> axpvm01.gitanes.taz> ./lisp -core kernel.core
>>> >> In initial-function, and running.
>>> >> GLOBALDB-INIT
>>> >> FDEFN-INIT
>>> >> Error: Undefined symbol.
>>> >>     SC: 16, Offset: 23  $1=       0x30182847: other pointer
>>> >> LDB monitor
>>> >> ldb>
>>> >>
>>> >> in lisp.map there's no address 30182847
>>> >
>>> >
>>> > Because you're somewhere in Lisp code.
>>> >
>>> > Type "help" to get some help on how to use ldb.  At this point, you can
>>> > probably learn a lot by doing
>>> >
>>> > print $1 (or print 0x30182847)
>>> >
>>> > This should tell you at least the symbol name. And since it's undefined,
>>> > we'll have to figure out why it's undefined.  Could be an issue in the
>>> > 19c
>>> > source code, or perhaps something bad happening with cross-compiling
>>> > from
>>> > 19c linux with :linkage-table enabled.
>>> >
>>> > But this progress is really encouraging!
>>> >
>>> >>
>>> >>
>>> >> I'lll try to understand which is the missing function.
>>> >>
>>> >>
>>> >> 2014-09-09 23:47 GMT+02:00 Raymond Toy <toy.raymond at gmail.com>:
>>> >> >
>>> >> >
>>> >> > On Tue, Sep 9, 2014 at 2:07 PM, Fausto Saporito
>>> >> > <fausto.saporito at gmail.com>
>>> >> > wrote:
>>> >> >>
>>> >> >> Hello all,
>>> >> >>
>>> >> >> so we have targeted the right process :-) I got always a sigsegv...
>>> >> >> but it's different....
>>> >> >> debug_print is working... is there a way to have those messages
>>> >> >> printed on the screen ? At the moment I can see them only in dbx :)
>>> >> >
>>> >> >
>>> >> > They should be showing up on the screen. debug_print just does
>>> >> > printf/fflush, so they should be there.
>>> >> >
>>> >> >>
>>> >> >>
>>> >> >> Yes... I have to go back to rebuild kernel.core after generated
>>> >> >> lisp.nm, but it seems not all the functions are present in lisp.nm
>>> >> >> The most obvious are missing! Like sin, cos, tan, killpg, etc ...
>>> >> >> and
>>> >> >> they are defined in undefineds.h
>>> >> >
>>> >> >
>>> >> > Some of these should be defined since they're used in kernel.core.
>>> >> > FWIW,
>>> >> > killpg is not in lisp.nm on my linux build (normal compile).
>>> >> >
>>> >> >
>>> >> > --
>>> >> > Ray
>>> >
>>> >
>>> >
>>> >
>>> > --
>>> > Ray
>>
>>
>>
>>
>> --
>> Ray