Re: [Gcl-devel] Error compiling functions w/ (declare (integer x)) present

[John Jorgensen]

CLX compiles! There seem to be other problems with CLX though. I'm going
to see if I can chase these problems down.

J*
-------------------------------------------
On 8 Jun 2002, Camm Maguire wrote:

> OK, I've checked in the fix for this.  avma seems to be a pari only
> thing.
> 
> I the process I've located and fixed, I hope, the segfault I saw, as
> well as reimplemented the declare integer to be more in line with what
> Dr. Schleter was originally doing, and alluded to in doc/bignum.
> 
> 1) One cannot use mpz_init, for example, without wrapping the allocated
>    space in a lisp object as done in new_bignum(), lest the garbage
>    collector hose the space.  this is so even though mpz_init uses
>    alloc_contblock to allocate -- apparently the gc walks through
>    objects to get to the contblocks.  For future reference.  again,
>    this is only to the best of my understanding.
> 
> 2) The bottleneck in bignum calculations appeared to be gc in
>    Dr. Schelter's analysis, so he arranged to have the integer
>    declaration bypass the lisp memory management and alloca sufficient
>    space for the bignum temporarily on the function C stack.  I've
>    followed suit, and noticed that it does reduce gc calls.  But in
>    the example given in doc/bignum, gc is not eliminated, as internal
>    functions in gmp_big.c and pari_big.c called after the integer
>    declaration do in turn allocate lisp objects.  Maybe in the future
>    we could enhance performance by using alloca here as well.
> 
> 3) Right now, you must have alloca if you use gmp.  From my
>    understanding, everyone does, but if this becomes a problem, we can
>    code the non-alloca case as well.
> 
> 4) The pi-inv example in doc/bignum works, as does the max0 function
>    in R. Toy's f2cl-lib.lsp.
> 
> 5) Please let me know if clx compiles
> 
> 6) As the avma stuff is truly a bug, I'm going to release 2.4.3
>    quickly.  If other items surface, 2.4.4 will similarly follow
>    shortly if I have time -- I think it important to leave the
>    official release without known bugs as quickly as possible. 
> 
> Take care,
> 
> 
> 
> 
> John Jorgensen <address@hidden> writes:
> 
> > Well, CLX gets further in it's compile, but now with the example in the
> > bignum docs, I get the error;
> > =========================
> > >(compile 'pi-inv)
> > 
> > Compiling gazonk0.lsp.
> > End of Pass 1.  
> > End of Pass 2.  
> > gazonk0.c: In function `L1':
> > gazonk0.c:4139: `save_avma' undeclared (first use in this function)
> > gazonk0.c:4139: (Each undeclared identifier is reported only once
> > gazonk0.c:4139: for each function it appears in.)
> > gazonk0.c:4140: `restore_avma' undeclared (first use in this function)
> > 
> > Correctable error: (SYSTEM "(cd . ;gcc -pipe -fwritable-strings
> > -DVOL=volatile -I/home/jjorgens/gcl/o -fsigned-char  -O -c -I. gazonk0.c
> > -w)") returned a non-zero value 0.
> > Signalled by UNLESS.
> > If continued: Continues anyway.
> > Broken at CERROR.  Type :H for Help.
> > >>
> > =========================
> > Whatever the heck a avma is... CLX stops with the same type of error
> > (save_avma undeclared).
> > 
> > J*
> > 
> > -------------------
> > On 7 Jun 2002, Camm Maguire wrote:
> > 
> > > Greetings!  OK, I've committed a fix for this.  Please check it out.
> > > And R.Toy, if you're listening, you should be able to uncomment
> > > (declare (integer... too.
> > > 
> > > The solution I've implemented doesn't seem very optimal, but rather
> > > uses the existing integer/bignum handling macros that were in place
> > > for pari support and the original custom mp support.  The lisp
> > > compiler spits out these C macros, and I'm far less comfortable
> > > modifying the lisp compiler code.  
> > > 
> > > I thought declarations were supposed to help the compiler.  But all it
> > > seems that they accomplish is the copying of integer data back and
> > > forth needlessly.
> > > 
> > > Consider:
> > > 
> > > (defun max0 (x y &rest z)
> > >   (declare (integer x y))
> > >   (apply #'max x y z))
> > > 
> > > Here is the C code without the declaration :
> > > 
> > >   {object V1;
> > >   object V2;
> > >   object V3;
> > >   V1=(base[0]);
> > >   V2=(base[1]);
> > >   vs_base=vs_base+2;
> > >   vs_top[0]=Cnil;
> > >   {object *p=vs_top;
> > >    for(;p>vs_base;p--)p[-1]=MMcons(p[-1],p[0]);}
> > >   V3=(base[2]);
> > >   vs_top=sup;
> > >   base[3]= (V1);
> > >   base[4]= (V2);
> > >   {object V4;
> > >   V4= (V3);
> > >    vs_top=base+5;
> > >    while(V4!=Cnil)
> > >    {vs_push((V4)->c.c_car);V4=(V4)->c.c_cdr;}
> > >   vs_base=base+3;}
> > >   Lmax();
> > >   return;
> > >   }
> > > 
> > > 
> > > 
> > > And here is the code with the declaration:
> > > 
> > > 
> > > 
> > >   {IDECL(GEN V1,V1space,V1alloc);
> > >   IDECL(GEN V2,V2space,V2alloc);
> > >   object V3;
> > >   SETQ_IO(V1,V1alloc,(base[0]));
> > >   SETQ_IO(V2,V2alloc,(base[1]));
> > >   vs_base=vs_base+2;
> > >   vs_top[0]=Cnil;
> > >   {object *p=vs_top;
> > >    for(;p>vs_base;p--)p[-1]=MMcons(p[-1],p[0]);}
> > >   V3=(base[2]);
> > >   vs_top=sup;
> > >   base[3]= make_integer(V1);
> > >   base[4]= make_integer(V2);
> > >   {object V4;
> > >   V4= (V3);
> > >    vs_top=base+5;
> > >    while(V4!=Cnil)
> > >    {vs_push((V4)->c.c_car);V4=(V4)->c.c_cdr;}
> > >   vs_base=base+3;}
> > >   Lmax();
> > >   return;
> > >        }
> > > 
> > > 
> > > What do we gain here, considering that we have to go back to objects
> > > anyway when calling Lmax?
> > > 
> > > Take care,
> > > 
> > > 
> > > John Jorgensen <address@hidden> writes:
> > > 
> > > > When I try to compile functions that declare integers, I get errors. 
> > > > They
> > > > interpret just fine, but they don't compile. here is an example;
> > > > ====================================
> > > > (defun pi-inv (bits &aux (m 0))
> > > >   (declare (integer bits m))
> > > >   (let* ((n (+ bits (integer-length bits) 11))
> > > >          (tt (truncate (ash 1 n) 882))
> > > >          (d (* 4 882 882))
> > > >          (s 0))
> > > >     (declare (integer s d tt n))
> > > >     (do ((i 2 (+ i 2))
> > > >          (j 1123 (+ j 21460)))
> > > >         ((zerop tt) (cons s (- (+ n 2))))
> > > >       (declare (integer i j))
> > > >         (setq s (+ s (* j tt))
> > > >               m (- (* (- i 1) (- (* 2 i) 1) (- (* 2 i) 3)))
> > > >               tt (truncate (* m tt) (* d (the integer (expt i 3))))))))
> > > > ==========================
> > > > 
> > > > Right from the bugnum examples in the GCL documentation. running
> > > > (compile 'pi-inv) produces the following errors;
> > > > ==========================
> > > > >(compile 'pi-inv)
> > > > 
> > > > Compiling gazonk1.lsp.
> > > > End of Pass 1.  
> > > > End of Pass 2.  
> > > > gazonk1.c: In function `L1':
> > > > gazonk1.c:3002: `GEN' undeclared (first use in this function)
> > > > gazonk1.c:3002: (Each undeclared identifier is reported only once
> > > > gazonk1.c:3002: for each function it appears in.)
> > > > gazonk1.c:3002: parse error before `V1'
> > > > gazonk1.c:3004: `V1' undeclared (first use in this function)
> > > > gazonk1.c:3004: `V1alloc' undeclared (first use in this function)
> > > > gazonk1.c:3007: parse error before `register'
> > > > gazonk1.c:3008: `V2' undeclared (first use in this function)
> > > > gazonk1.c:3008: `V2alloc' undeclared (first use in this function)
> > > > gazonk1.c:3009: parse error before `register'
> > > > gazonk1.c:3010: parse error before `register'
> > > > gazonk1.c:3011: parse error before `V5'
> > > > gazonk1.c:3012: parse error before `register'
> > > > gazonk1.c:3023: `save_avma' undeclared (first use in this function)
> > > > gazonk1.c:3024: `V3' undeclared (first use in this function)
> > > > gazonk1.c:3024: `V3alloc' undeclared (first use in this function)
> > > > gazonk1.c:3024: `restore_avma' undeclared (first use in this function)
> > > > gazonk1.c:3036: `V4' undeclared (first use in this function)
> > > > gazonk1.c:3036: `V4alloc' undeclared (first use in this function)
> > > > gazonk1.c:3044: `V5' undeclared (first use in this function)
> > > > gazonk1.c:3044: `V5alloc' undeclared (first use in this function)
> > > > gazonk1.c:3045: `V6' undeclared (first use in this function)
> > > > gazonk1.c:3045: `V6alloc' undeclared (first use in this function)
> > > > gazonk1.c:3046: parse error before `register'
> > > > gazonk1.c:3047: parse error before `register'
> > > > gazonk1.c:3048: `V7' undeclared (first use in this function)
> > > > gazonk1.c:3048: `V7alloc' undeclared (first use in this function)
> > > > gazonk1.c:3049: `V8' undeclared (first use in this function)
> > > > gazonk1.c:3049: `V8alloc' undeclared (first use in this function)
> > > > 
> > > > Correctable error: (SYSTEM "(cd . ;gcc -pipe -fwritable-strings
> > > > -DVOL=volatile -I/opt/dl/gcl/o -fsigned-char  -O -c -I. gazonk1.c
> > > > -w)") returned a non-zero value 0.
> > > > Signalled by UNLESS.
> > > > If continued: Continues anyway.
> > > > Broken at CERROR.  Type :H for Help.
> > > > >>
> > > > ============================
> > > > 
> > > > If I remove the declares, it compiles just fine. It looks like this is
> > > > what is keeping CLX from compiling. GEN looks like it's supposed to be a
> > > > long, but I'm not sure.
> > > > 
> > > > J*
> > > > 
> > > > 
> > > > _______________________________________________
> > > > Gcl-devel mailing list
> > > > address@hidden
> > > > http://mail.gnu.org/mailman/listinfo/gcl-devel
> > > > 
> > > > 
> > > 
> > > 
> > 
> > 
> > 
> 
>