Net2/usr/src/usr.bin/gcc/cc1/grot/README-VMS
Return-Path: <info-gcc-request@prep.ai.mit.edu>
Date: 11 Sep 90 14:07:21 GMT
From: news@operations.dccs.upenn.edu (USENET News System)
Organization: University of Pennsylvania
Subject: Building gcc 1.37.1 with Vax C
References: <9009102236.AA15872@wubios.wustl.edu>
Sender: info-gcc-request@prep.ai.mit.edu
To: info-gcc@prep.ai.mit.edu
during it; hopefully they'll come in handy.
To build the new gcc-cpp and gcc-cc1 that come with Gnu C
version 1.37.1 with Vax C and NOT a previous version of gcc,
do the following:
I - Building gcc-cpp
A - Modify cccp.c
1 - Replace
#include <sys/types.h>
#include <sys/stat.h>
with
#ifndef VMS
#include <sys/types.h>
#include <sys/stat.h>
#else
#include <types.h>
#include <stat.h>
#endif
[It is claimed that
$ define sys sys$library
will solve the same problem without source changes.]
2 - In the VMS-specific section (has perror.h etc in it),
add the following:
#include <descrip.h> /* for dsc$descriptor_s */
#include <string.h> /* for strchr & strrchr */
#define index strchr
#define rindex strrchr
3 - You have to replace all occurences of:
extern char *index(), *rindex();
as well as any where they're separated (e.g. extern char *index()
by itself or extern char *rindex() by itself), with the following:
#ifndef VMS
extern char *index(); /* or whatever was here */
#endif
There's a total of three: one in main(), one in do_include(), and
one in hack_vms_include_specification().
B - You have to have alloca.mar for cccp.c; it was distributed
with vmsgcc1.34; it's also in the bison distribution. (both 1.03
and 1.06)
C - After you've compiled alloca.mar (MACRO ALLOCA.MAR), follow
the instructions at the top of MAKE-CC1.COM (namely, change the
CC := line to say cc/noopt instead of gcc, take the
cc1_options="-mpcc-alignment" part out of the CFLAGS line, and change
the LIBS line to say:
$ LIBS := alloca,sys$share:vaxcrtl/libr
D - Since it doesn't come with the distribution, you have
to generate cexp_tab.c yourself, either with bison on either
a Vax or a Unix box, or yacc in Unix. (@make below will die if you
don't have bison on your Vax or don't have this file)
E - Type:
@make
and watch it hum. It should go through okay; it'll die on the LINK
line, but don't worry about it.
F - Now link the whole thing with:
link /nomap/exe=gcc-cpp cccp,cexp,version,alloca,-
sys$share:vaxcrtl/libr
G - You should be screaming for joy right now; if you're not, then
give up on life. <snicker>
II - Building gcc-cc1
A - Change MAKE-CC1.COM to contain the line:
$ libs :== sys$share:vaxcrtl/libr,alloca
Yep, you have to have alloca.obj here too.
B - Modify toplev.c:
1 - Replace
#include <sys/types.h>
#include <sys/stat.h>
with
#ifndef VMS
#include <sys/types.h>
#include <sys/stat.h>
#else
#include <types.h>
#include <stat.h>
#endif
(You'll see this again.)
2 - Do:
Repl: rtx_equal_function_value_matters
With: rtx_equ_func_value_matters
Num: once
This is cuz you can't have a name over 31 chars with the
Vax C compiler; and yep you guessed it this is over that
limit. You're gonna be doing this a few more times.
C - Modify gcc.c:
1 - This is optional, since you never compile it (I'm
working on that):
Replace
#include <stdio.h>
#include <sys/types.h>
#include <signal.h>
#include <sys/file.h>
with
#include <stdio.h>
#include <signal.h>
#ifdef VMS
#include <types.h>
#include <file.h>
#else
#include <sys/types.h>
#include <sys/file.h>
#endif
D - Modify c-parse_tab.c (if this isn't there, you'll have
to create it with bison or yacc):
Repl: expand_start_loop_continue_elsewhere
With: expand_startloop_cont_elsewhere
Num: twice
E - Modify tree.h:
Repl: expand_start_loop_continue_elsewhere
With: expand_startloop_cont_elsewhere
Num: once
F - Modify stmt.c:
Repl: expand_start_loop_continue_elsewhere
With: expand_startloop_cont_elsewhere
Num: twice
Repl: current_function_returns_pcc_struct
With: cur_func_ret_pcc_struct
Num: nine
Repl: current_function_returns_pointer
With: cur_func_ret_ptr
Num: 2
Repl: current_function_pretend_args_size
With: cur_func_pretend_args_size
Num: 3
G - Modify rtanal.c:
Repl: rtx_equal_function_value_matters
With: rtx_equ_func_values_matters
Num: twice
H - Modify output.h:
Repl: current_function_returns_pcc_struct
With: cur_func_ret_pcc_struct
Num: once
Repl: current_function_returns_pointer
With: cur_func_ret_ptr
Num: once
Repl: current_function_pretend_args_size
With: cur_func_pretend_args_size
Num: once
I - Modify tree.def:
Change the line that reads:
DEFTREECODE (FILE_TYPE, "file_type", "t", 0)
to:
#ifdef VMS
DEFTREECODE (FILE_TYPE_GNU, "file_type", "t", 0)
#else
DEFTREECODE (FILE_TYPE, "file_type", "t", 0)
#endif
This is cuz FILE_TYPE is defined in stdio.h as being
struct _iobuf *, which totally screws everything up.
J - Modify expr.c:
Change the line that reads:
if (code == FILE_TYPE)
to:
#ifdef VMS
if (code == FILE_TYPE_GNU)
#else
if (code == FILE_TYPE)
#endif
K - Modify expmed.c:
Change the line that reads:
trymode && GET_MODE_SIZE (trymode) <= maxsize;
to:
(trymode != 0) && (GET_MODE_SIZE (trymode) <= maxsize);
Enclosing the second half of the and in parenthesis may be
overkill. By the time I thought of it it was too late to
change it (read: I didn't feel like going back), but you
may want to give it a try.
L - Modify symout.c:
Change the line that has #include <stddef.h> on it to:
#ifndef VMS
#include <stddef.h>
#else
#define NULL (void *) 0
#endif
M - Modify insn-emit.c:
Line #1011 (in mine) is a 376-character long line that's
one HUGE return (it calls a function a few zillion times
embedded upon each other). This is a tad too big for Vax C
to chew; try cutting it down to a few lines. I made it look
like this:
rtx
gen_casesi (operand0, operand1, operand2, operand3)
rtx operand0;
rtx operand1;
rtx operand2;
rtx operand3;
{
return gen_rtx (SET, VOIDmode, pc_rtx,
gen_rtx (IF_THEN_ELSE, VOIDmode,
gen_rtx (LE, VOIDmode,
gen_rtx (MINUS, SImode, operand0, operand1),
operand2),
gen_rtx (PLUS, SImode,
gen_rtx (SIGN_EXTEND, SImode,
gen_rtx (MEM, HImode,
gen_rtx (PLUS, SImode, pc_rtx,
gen_rtx (MINUS, SImode, operand0, operand1)))),
gen_rtx (LABEL_REF, SImode, operand3)),
pc_rtx));
}
Kinda smacks of lisp, no?
N - Modify final.c:
Change the line that reads:
rtx final_sequence;
to:
#ifndef VMS
rtx final_sequence;
#endif
This is cuz rtx final_sequence already appears in
output.h, and Vax C screams when it sees this.
O - You have to have alloca(), bcopy(), bcmp(), and
bzero(); all are in the VMS Gcc 1.34 distribution; only
bcopy() is in the bison distribution. Since they're pretty
short, I'm gonna include 'em here. These were sent under the
Gnu public license.
alloca.mar:
.PSECT $CODE,LONG,PIC,REL,SHR,EXE,RD,NOWRT
.ENTRY ALLOCA,^M<>
SUBL2 4(AP),SP
MOVL 16(FP),R1
MOVQ 8(FP),AP
BICL2 #3,SP
ADDL2 #28,SP
MOVL SP,R0
JMP (R1)
.END
bcopy.mar:
.PSECT $CODE,LONG,PIC,REL,SHR,EXE,RD,NOWRT
; bcopy(from, to, size)
.ENTRY BCOPY,^M<R2,R3,R4,R5,R6>
MOVL 4(AP),R1
MOVL 8(AP),R3
MOVL 12(AP),R6
CMPL R1,R3
BGTR 2$ ; NORMAL FORWARD CASE
BLSS 3$ ; OVERLAPPING, MUST DO BACKWARDS
RET ; EQUAL, NOTHING TO DO
1$: SUBL2 R0,R6
MOVC3 R0,(R1),(R3)
2$: MOVZWL #65535,R0
CMPL R6,R0
BGTR 1$
MOVC3 R6,(R1),(R3)
RET
3$: ADDL2 R6,R1
ADDL2 R6,R3
MOVZWL #65535,R0
BRW 5$
4$: SUBL2 R0,R6
SUBL2 R0,R1
SUBL2 R0,R3
MOVC3 R0,(R1),(R3)
MOVZWL #65535,R0
SUBL2 R0,R1
SUBL2 R0,R3
5$: CMPL R6,R0
BGTR 4$
SUBL2 R6,R1
SUBL2 R6,R3
MOVC3 R6,(R1),(R3)
RET
.END
bcmp.mar:
.PSECT $CODE,LONG,PIC,REL,SHR,EXE,RD,NOWRT
; bcmp(s1, s2, n)
.ENTRY BCMP,^M<R2,R3,R4,R5>
MOVL 4(AP),R1
MOVL 8(AP),R3
MOVL 12(AP),R4
1$: MOVZWL #65535,R0
CMPL R4,R0
BLEQ 2$
SUBL2 R0,R4
CMPC3 R0,(R1),(R3)
BEQL 1$
ADDL2 R4,R0
RET
2$: CMPC3 R4,(R1),(R3)
RET
.END
bzero.mar:
.PSECT $CODE,LONG,PIC,REL,SHR,EXE,RD,NOWRT
; bzero(ptr, size)
.ENTRY BZERO,^M<R2,R3,R4,R5>
MOVL 4(AP),R3
BRB 2$
1$: SUBL2 R0,8(AP)
MOVC5 #0,(R3),#0,R0,(R3)
2$: MOVZWL #65535,R0
CMPL 8(AP),R0
BGTR 1$
MOVC5 #0,(R3),#0,8(AP),(R3)
RET
.END
P - On the last lines (where it's got all of the link shit),
change it so it reads:
...blahblahblah...
...blahblah,insn-extract,insn-output,obstack,-
integrate,caller-save,-
bcopy,bcmp,bzero
$!
So now it'll link the bcopy, bcmp, and bzero routines in.
Q - You should be screaming for joy right now; if you're
not, then give up on life again. Hey, I don't get paid for my humor.
R - Finally, you have to use old versions of:
GCC.EXE GCC-AS.EXE
GCC.COM GCC.CLD
GCCLIB.OLB
to make the package work properly. I've only been able to make the
new preprocessors make properly; since it wasn't even in the make
files that came with it, I don't think gcc.exe 1.37.1 was intended
to be built for the Vax .. we'll see.
In case you're wondering, I could never get vmsgcc134 to work properly;
that's why I did this with Vax C.
Good luck! This only worked under 5.3-1 (and the latest version of
Vax C...3.0? 3.1?), and isn't guaranteed in any way shape or form. If
you have stumbling blocks and think I may have come upon it during all
of this, feel free to mail me at kehoe@scotty.dccs.upenn.edu.
--
Brendan Kehoe | Soon: brendan@cs.widener.edu
For now: kehoe@scotty.dccs.upenn.edu | Or: bkehoe@widener.bitnet
Last resort: brendan.kehoe@cyber.widener.edu
Brendan Kehoe | Soon: brendan@cs.widener.edu
For now: kehoe@scotty.dccs.upenn.edu | Or: bkehoe@widener.bitnet
Last resort: brendan.kehoe@cyber.widener.edu