martou@fynu.ucl.ac.be (Jean-Luc Martou) (05/22/91)
Hello! We have a big problem concerning the file <stdarg.h> with the AT&T C++ 2.1 translator we have ported on SPARC. In fact, all the macro va_start,... are wrong for the sparc. Who has build a right version of this header file? Please send me your answer on my electronic-mail : martou@fynu.ucl.ac.be, and I shall summarize all the answers. Thanks for your help. Jean-Luc MARTOU Computer Staff Nuclear Physic Department Universite Catholique de Louvain B-1348 Louvain-La-Neuve BELGIUM E-MAIL : martou@fynu.ucl.ac.be
joost@cadlab.de (Michael Joosten) (05/27/91)
Well, there are two solutions:
1) Either you cope around the problem, as demonstrated in NIHCL (Class.c)
by renaming the last parameter before the ellipsis to '__builtin_va_alist'
2) Or you do the Right Thing: Solve the problem. But this is
machine-dependent (better: RISC-dependent) and you have to patch the cfront
sources. I have done it one DECStation a year ago (cfront 1.2) and for a
Pyramid and a SPARC last month for cfront 2.0 .
Short Description:
The MIPS and Sun-SPARC compiler recognize their 'magic word'
(__builtin_va_alist for SPARC, va_alist for MIPS) and emit code to copy the
input window/registers into contigous memory, i.e. the stack, so that
va_arg could extract whatever it wants.
(Sorry, no diffs, I don't have the originals handy...)
First what AT&T forgot: Set the right preprocessor flags in CC.
The (undocumented, of course) variable 'Y' keeps the flags for the
preprocessor, while 'X' is responsible for the 'cc' flags.
One should do this for other machines, too !
Here are my adds in bin/CC:
Y="-Dsun -Dsparc -Dunix -D__BUILTIN_VA_ARG_INCR" # forgotten by AT&T... for SUN4
Don't forget to delete the statement 'Y=' in the block around line 65:
TMPDIR=${TMPDIR-"/usr/tmp"}
TEMPDIR=$TMPDIR/CC.$$
C= EE=0 FON= XON= O=
PON= R= SUF= X= Y= <------------ !!!
DASHR= PLUSI= STRIP=
Z= E=0 OF= P=
Now the changes for stargs:
-------------------------stdarg.h----------------------------------------------
..........
.........
#ifndef va_start
#ifdef pyr
typedef int va_buf[3];
typedef va_buf *va_list;
extern "C" {
extern void _vastart(va_list, char*);
extern char *_vaarg(va_list, int);
}
#define va_start(ap, parmN) {\
va_buf _va;\
_vastart(ap = (va_list)_va, (char *)&parmN + sizeof parmN)
#define va_end(ap) }
#define va_arg(ap, mode) *((mode *)_vaarg(ap, sizeof (mode)))
#else
typedef char *va_list;
#define va_end(ap)
#ifdef sparc
#define va_start(ap, parmN) \
(ap = ((char *) &(parmN) + (((sizeof (parmN) + \
sizeof (int) - 1) / sizeof (int)) * sizeof (int))))
#ifdef __BUILTIN_VA_ARG_INCR
# define va_arg(list,mode) ((mode*)__builtin_va_arg_incr((mode *)list))[0]
#else
# define va_arg(list,mode) ((mode *)(list += sizeof(mode)))[-1]
#endif
extern "C" void *__builtin_va_arg_incr(...);
#else
#ifdef u370
#define va_start(ap, parmN) ap =\
(char *) ((int)&parmN + 2*sizeof(parmN) - 1 & -sizeof(parmN))
#define va_arg(ap, mode) ((mode *)(ap = \
(char *) ((int)ap + 2*sizeof(mode) - 1 & -sizeof(mode))))[-1]
#else
#ifdef hp9000s800
#define va_start(ap, parmN) (ap = (char *)&parmN + sizeof(parmN),(void)va_arg(ap,int))
#define va_arg(ap, mode) ((mode *)(ap = (char *)((int)(ap-sizeof(mode)) & ~(sizeof(mode)-1))))[0]
#else
#define va_start(ap, parmN) ap = (char *)( &parmN+1 )
#define va_arg(ap, mode) ((mode *)(ap += sizeof(mode)))[-1]
#endif
#endif
#endif
#endif
...............
...............
-------------------------------------------------------------------------------
And now a patch for print2.c:
In method fct::dcl_print(), add/replace the following at the end:
.........
.........
else {
// print argument names followed by argument type declarations:
if (body && Cast==0) {
for (nn=at; nn;) {
nn->print();
if (nn=nn->n_list) puttok(CM); else break;
}
#ifdef pyr
if (nargs_known == ELLIPSIS) putstring(",__0va0,__0va1,_
_0va2,__0va3,__0va4,__0va5,__0va6,__0va7,__0va8,__0va9,__0va10,__0va11");
#endif
#ifdef mips
if (nargs_known == ELLIPSIS) {
putst(",va_alist) int va_alist;");
} else
#endif
#ifdef sparc
if (nargs_known == ELLIPSIS) {
putst(",__builtin_va_alist) int __builtin_va_ali
st;");
} else
#endif
putch(')');
}
else
putch(')');
}
}
-------------------------------------------------------------------------------
Good Luck Michael
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| Michael Joosten | Tel. : (+49) (+) 5251-284 120 |
| CADLAB | Fax : (+49) (+) 5251-284 140 |
| Bahnhofstr. 32 | E-Mail: joost@cadlab.uucp |
| D-4790 Paderborn | joost@cadlab.cadlab.de |
| FRG | ...!uunet!unido!cadlab!joost |
| | Mass mail to: joost@pbinfo.uni-paderborn.de |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| Michael Joosten | Tel. : (+49) (+) 5251-284 120 |
| CADLAB | Fax : (+49) (+) 5251-284 140 |
| Bahnhofstr. 32 | E-Mail: joost@cadlab.uucp |
| D-4790 Paderborn | joost@cadlab.cadlab.de |
| FRG | ...!uunet!unido!cadlab!joost |
| | Mass mail to: joost@pbinfo.uni-paderborn.de |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~rfg@NCD.COM (Ron Guilmette) (05/31/91)
In article <1991May22.144659.13993@info-sparc1.info.ucl.ac.be> martou@fynu.ucl.ac.be (Jean-Luc Martou) writes: >Hello! > >We have a big problem concerning the file <stdarg.h> with the AT&T C++ 2.1 >translator we have ported on SPARC. In fact, all the macro va_start,... >are wrong for the sparc. Who has build a right version of this header >file? On this same topic, it seems worthy of note that users of any "binary only" version of cfront should be very carful about using the right underlying C compiler. I learned recently that (contrary to my prior belief) you cannot simply substitute *any* C compiler (e.g. GNU C) for the particular C compiler that your (binary) copy of cfront was targeted for. If you try to do that, you may get into serious trouble because the "other" C compiler may have a totally different way of handling so-called "varargs" functions. Just a word to the wise. Tread carefully. -- // Ron ("Loose Cannon") Guilmette // Internet: rfg@ncd.com uucp: ...uunet!lupine!rfg // New motto: If it ain't broke, try using a bigger hammer.