[Bug c/15491] New: ICE (Internal compiler error) while compiling glibc to vax-linux target

[Bug c/15491] New: ICE (Internal compiler error) while compiling glibc to vax-linux target

[gabucino at mplayerhq dot hu]

See http://cab.bio.u-szeged.hu/~gabucino/gccbug

-- 
           Summary: ICE (Internal compiler error) while compiling glibc to
                    vax-linux target
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: critical
          Priority: P1
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: gabucino at mplayerhq dot hu
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-host_pc-linux-gnu
  GCC host triplet: vax-dec-linux-gnu
GCC target triplet: vax-dec-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug c/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[gabucino at mplayerhq dot hu]

------- Additional Comments From gabucino at mplayerhq dot hu  2004-05-17 10:37 -------
Created an attachment (id=6315)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6315&action=view)
Compiler outbut


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug c/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[gabucino at mplayerhq dot hu]

------- Additional Comments From gabucino at mplayerhq dot hu  2004-05-17 10:38 -------
Created an attachment (id=6316)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6316&action=view)
.i file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jbglaw at lug-owl dot de


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal
          Component|c                           |target
           Keywords|                            |ice-on-valid-code
            Version|unknown                     |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-05-27 12:12 -------
Created an attachment (id=6399)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6399&action=view)
gdb backtrace of cc1's SIGSEGV

Attached you find a gdb bt from cc1's SIGSEGV. As you see, it breaks
in ./gcc/config/vax/vax.c' RTX cost calculation. However, I'm not all
that familiar with gcc's internals, so I fear I can't help by more than
typing commands you tell me, or by giving access to the machine, ...

For now, I've placed a "if (!op) return c+5;" between
"register rtx op = XEXP (x, i++);" and "code = GET_CODE (op);" at around
line 710, but I'm sure that's not the correct fix:)

MfG, JBG


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-05-27 13:17 -------
(From update of attachment 6399)
>Program received signal SIGSEGV, Segmentation fault.
>vax_rtx_costs_1 (x=0x40273ea0, code=SYMBOL_REF, outer_code=INSN_LIST)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/config/vax/vax.c:709
>709           code = GET_CODE (op);
>(gdb) bt
>#0  vax_rtx_costs_1 (x=0x40273ea0, code=SYMBOL_REF, outer_code=INSN_LIST)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/config/vax/vax.c:709
>#1  0x082ee7cf in vax_rtx_costs (x=0x2, code=2, outer_code=2, total=0x4f)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/config/vax/vax.c:764
>#2  0x082a0998 in rtx_cost (x=0x40273ea0, outer_code=INCLUDE)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/rtlanal.c:3815
>#3  0x08160241 in notreg_cost (x=0x40273ea0, outer=INCLUDE)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cse.c:809
>#4  0x0816841b in cse_insn (insn=0x402adb84, libcall_insn=0x0)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cse.c:5246
>#5  0x0816a5c6 in cse_basic_block (from=0x40270e34, to=0x402add68, next_branch=0x84304b0, around_loop=1)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cse.c:7124
>#6  0x0816a240 in cse_main (f=0x0, nregs=-1073749120, after_loop=0, file=0x0)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cse.c:6985
>#7  0x082eb0d2 in rest_of_handle_cse (decl=0x2, insns=0x40270e34)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/passes.c:1121
>#8  0x082ebc2e in rest_of_compilation (decl=0x4029d414)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/passes.c:1562
>#9  0x080c60b2 in tree_rest_of_compilation (fndecl=0x4029d414, nested_p=false)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/tree-optimize.c:590
>#10 0x08063a68 in c_expand_body_1 (fndecl=0x4029d414, nested_p=0)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/c-decl.c:6380
>#11 0x08063c03 in c_expand_body (fndecl=0x4f) at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/c-decl.c:6403
>#12 0x082f7242 in cgraph_expand_function (node=0x402a5e80)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cgraphunit.c:794
>#13 0x082f629a in cgraph_assemble_pending_functions ()
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cgraphunit.c:295
>#14 0x082f63d6 in cgraph_finalize_function (decl=0x4029d414, nested=false)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/cgraphunit.c:376
>#15 0x080633ef in c_finalize (fndecl=0x4029d414)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/c-decl.c:6222
>#16 0x08063638 in finish_function () at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/c-decl.c:6345
>#17 0x0804a3c7 in yyparse () at c-parse.y:372
>#18 0x0805191b in c_parse_file () at c-parse.y:3015
>#19 0x0809765b in c_common_parse_file (set_yydebug=2)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/c-opts.c:1239
>#20 0x082bcf7a in compile_file () at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/toplev.c:1653
>#21 0x082be485 in do_compile () at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/toplev.c:2699
>#22 0x082be4f5 in toplev_main (argc=2, argv=0xbfffeef4)
>    at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/toplev.c:2731
>#23 0x080add1b in main (argc=2, argv=0x2) at /home/jbglaw/vax-linux/toolchain/__temp/source/gcc-HEAD/gcc/main.c:35
>(gdb) print op
>$1 = 0x0

So this segv happens in cost estimation. Could someone please review the code a
bit? In pristine sources, this is
right after while (*fmt++ == 'e') (in vax_rtx_costs_1()): register rtx op =
XEXP (x, i++); results in a NULL ptr,
which the next statement (code = GET_CODE (op);;) then tries to dereference ...

I think the NetBSD guys could be interested in this, too...

MfG, JBG


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-27 14:45 -------
    case PLUS:
      /* Check for small negative integer operand: subl2 can be used with
         a short positive constant instead.  */
      if (GET_CODE (XEXP (x, 1)) == CONST_INT)
        if ((unsigned)(INTVAL (XEXP (x, 1)) + 63) < 127)
          fmt = "e";
    case MINUS:
      c = (mode == DFmode) ? 13 : 8;    /* 6/8 on VAX 9000, 16/15 on VAX 2 */
    case IOR:
    case XOR:
      c = 3;
      break;
This is werid. No breaks.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-05-27 14:54 -------
So this sounds like I'd place two extra breaks in the PLUS and MINUS case?
However, that won't "fix" my initial problem, right? Uh, or will it? "fmt"
is used in the following while () loop ...

I'd add breaks and do another build. On my machine, that only takes some
15 min ...

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-05-27 15:20 -------
Just adding two breaks doesn't fix the segv, so I'll keep those two breaks,
as well as the extra check.

So this bug isn't yet fixed :(

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-05-28 14:46 -------
Created an attachment (id=6416)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6416&action=view)
Patch to add missing "break;"s in gcc/config/vax/vax.c for RTX calculation

Hi!

The attached patch adds two breaks, which I now truely believe are missing.
This was initially spotted by Andrew Pinsky <pinskia from gcc, gnu, org>.

It does still not fix the SIGSEGV, nor does this patch contain the workaround
for that problem. It just fixes cost calculation.

MfG, JBG


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[danglin at gcc dot gnu dot org]

------- Additional Comments From danglin at gcc dot gnu dot org  2004-06-25 03:04 -------
I think the code in rtx_cost should look like this (totally untested):

    case PLUS:
    case MINUS:
      c = (mode == DFmode) ? 13 : 8;    /* 6/8 on VAX 9000, 16/15 on VAX 2 */
      /* Check for small negative integer operand: subl2 can be used with
         a short positive constant instead.  */
      if (GET_CODE (XEXP (x, 1)) == CONST_INT)
        if ((unsigned)(INTVAL (XEXP (x, 1)) + 63) < 127)
          fmt = "e";
      break;

Adding a break after the code currently in the "PLUS" case is wrong as
c isn't defined.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-06-26 15:12 -------
Subject: Re:  ICE (Internal compiler error) while compiling glibc to vax-linux target

On Fri, 2004-06-25 03:04:30 -0000, danglin at gcc dot gnu dot org <gcc-bugzilla@gcc.gnu.org>
wrote in message <20040625030430.2794.qmail@sourceware.org>:
> ------- Additional Comments From danglin at gcc dot gnu dot org  2004-06-25 03:04 -------
> I think the code in rtx_cost should look like this (totally untested):
> 
>     case PLUS:
>     case MINUS:
>       c = (mode == DFmode) ? 13 : 8;    /* 6/8 on VAX 9000, 16/15 on VAX 2 */
>       /* Check for small negative integer operand: subl2 can be used with
>          a short positive constant instead.  */
>       if (GET_CODE (XEXP (x, 1)) == CONST_INT)
>         if ((unsigned)(INTVAL (XEXP (x, 1)) + 63) < 127)
>           fmt = "e";
>       break;
> 
> Adding a break after the code currently in the "PLUS" case is wrong as
> c isn't defined.

> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

At least, it looks even a bit better and does make some sense:) For now,
I don't think that cycle costs are the largest problem for VAX (the VAX
backend was even mentioned to be removed completely...).

I'll take your code fragment and put it into my patchqueue.

MfG, JBG



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-29 03:01 -------
This is funny as vax-linux is not officially part of gcc yet and it sounds like VAX might be removed 
from gcc as noone is willing to maintain it at all.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[danglin at gcc dot gnu dot org]

------- Additional Comments From danglin at gcc dot gnu dot org  2004-06-30 02:36 -------
If you look at the PR's submitted since the beginning of the year for VAX,
you will see that interest in this target isn't totally dead.  There has
been a fair bit of work done on binutils for the netbsd and linux targets.
Clearly, this was done with the assumption that the VAX target wouldn't
be removed from GCC.  However, I'm going to stay out of the debate regarding
the removal of VAX.

My personal issues with trying to maintain VAX are:

1) It's much lower priority than hppa and my time is limited.
2) The VAX that I have has a rather limited amount of memory and runs
   ultrix (until recently, this machine was used for remote data access
   from two scientific monitors and there wasn't the time or money available
   to upgrade the software used for this activity).  While the machine
   could be switched over to a newer OS, it would still be hobbled by
   its limited memory.  Hopefully, one of the people interested in netbsd
   or linux will take it over, although I know they fear the learning
   curve.

With the mainline, C would build after Jan's patch to fix a case falls
though problem until late January.  Then, another change broke this
again.  At the moment, there are new issues such as:

if gcc -DHAVE_CONFIG_H -I. -I../../gcc/libcpp -I.  -I../../gcc/libcpp/../include
 -I../../gcc/libcpp -I../../gcc/libcpp/include -DLOCALEDIR="\"/usr/local/gcc/gcc
-3.5.0/share/locale\""   -g -O2 -MT files.o -MD -MP -MF ".deps/files.Tpo" -c -o
files.o ../../gcc/libcpp/files.c; \
then mv -f ".deps/files.Tpo" ".deps/files.Po"; else rm -f ".deps/files.Tpo"; exi
t 1; fi
../../gcc/libcpp/files.c: In function `read_file_guts':
../../gcc/libcpp/files.c:503: error: `ssize_t' undeclared (first use in this fun
ction)
../../gcc/libcpp/files.c:503: error: (Each undeclared identifier is reported onl
y once
../../gcc/libcpp/files.c:503: error: for each function it appears in.)
../../gcc/libcpp/files.c:503: error: parse error before "size"
../../gcc/libcpp/files.c:524: error: parse error before numeric constant
../../gcc/libcpp/files.c:524: error: parse error before numeric constant
../../gcc/libcpp/files.c:524: error: parse error before numeric constant
...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[danglin at gcc dot gnu dot org]

------- Additional Comments From danglin at gcc dot gnu dot org  2004-06-30 02:55 -------
> At least, it looks even a bit better and does make some sense:) For now,
> I don't think that cycle costs are the largest problem for VAX (the VAX
> backend was even mentioned to be removed completely...).

I wasn't too concerned about the cost calculation.  If the code fragment
doesn't fix the original problem, can you try in gdb

p debug_rtx (x)

so that we can see in more detail the rtx that caused the problem?  It
might be necessary to break on vax_rtx_costs_1 with x==0x40273ea0 and
step though the routine.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-06-30 06:17 -------
Subject: Re:  ICE (Internal compiler error) while compiling glibc to vax-linux target

On Tue, 2004-06-29 03:01:07 -0000, pinskia at gcc dot gnu dot org <gcc-bugzilla@gcc.gnu.org>
wrote in message <20040629030107.7992.qmail@sourceware.org>:
> 
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-29 03:01 -------
> This is funny as vax-linux is not officially part of gcc yet and it sounds like VAX might be removed 
> from gcc as noone is willing to maintain it at all.

That's quite right, VAX was suggested to be removed. From the amount of
work involved, I'm like stepping forward to take maintainership.
However, I'm not skilled enough with GCC's internals to do that right
now. I'm learning ... :)

MfG, JBG



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-06-30 06:22 -------
Subject: Re:  ICE (Internal compiler error) while compiling glibc to vax-linux target

On Wed, 2004-06-30 02:36:05 -0000, danglin at gcc dot gnu dot org <gcc-bugzilla@gcc.gnu.org>
wrote in message <20040630023605.3485.qmail@sourceware.org>:
> you will see that interest in this target isn't totally dead.  There has
> been a fair bit of work done on binutils for the netbsd and linux targets.
> Clearly, this was done with the assumption that the VAX target wouldn't
> be removed from GCC.  However, I'm going to stay out of the debate regarding

Quite right!

> 1) It's much lower priority than hppa and my time is limited.
> 2) The VAX that I have has a rather limited amount of memory and runs
>    ultrix (until recently, this machine was used for remote data access
>    from two scientific monitors and there wasn't the time or money available
>    to upgrade the software used for this activity).  While the machine
>    could be switched over to a newer OS, it would still be hobbled by
>    its limited memory.  Hopefully, one of the people interested in netbsd
>    or linux will take it over, although I know they fear the learning
>    curve.

Do you already know Linux or *BSD? ...and how much memory is installed
in your box?

Booting a VAX with Linux isn't all that hard. Depending on the model,
there may or may not be a SCSI driver available, but network interfaces
are supported. So at least, you can MOP boot and run into NFS-root.

> With the mainline, C would build after Jan's patch to fix a case falls
> though problem until late January.  Then, another change broke this
> again.  At the moment, there are new issues such as:

Hu? I've done builds after January for sure:)

> if gcc -DHAVE_CONFIG_H -I. -I../../gcc/libcpp -I.  -I../../gcc/libcpp/../include
>  -I../../gcc/libcpp -I../../gcc/libcpp/include -DLOCALEDIR="\"/usr/local/gcc/gcc
> -3.5.0/share/locale\""   -g -O2 -MT files.o -MD -MP -MF ".deps/files.Tpo" -c -o
> files.o ../../gcc/libcpp/files.c; \
> then mv -f ".deps/files.Tpo" ".deps/files.Po"; else rm -f ".deps/files.Tpo"; exi
> t 1; fi
> ../../gcc/libcpp/files.c: In function `read_file_guts':
> ../../gcc/libcpp/files.c:503: error: `ssize_t' undeclared (first use in this fun
> ction)
> ../../gcc/libcpp/files.c:503: error: (Each undeclared identifier is reported onl
> y once
> ../../gcc/libcpp/files.c:503: error: for each function it appears in.)
> ../../gcc/libcpp/files.c:503: error: parse error before "size"
> ../../gcc/libcpp/files.c:524: error: parse error before numeric constant
> ../../gcc/libcpp/files.c:524: error: parse error before numeric constant
> ../../gcc/libcpp/files.c:524: error: parse error before numeric constant
> ...

I'll start another build right now to check your claim. How did you
configure gcc and binutils? On what machine are you building on?

MfG, JBG



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-06-30 06:25 -------
Subject: Re:  ICE (Internal compiler error) while compiling glibc to vax-linux target

On Wed, 2004-06-30 02:55:52 -0000, danglin at gcc dot gnu dot org <gcc-bugzilla@gcc.gnu.org>
wrote in message <20040630025552.9449.qmail@sourceware.org>:
> 
> ------- Additional Comments From danglin at gcc dot gnu dot org  2004-06-30 02:55 -------
> > At least, it looks even a bit better and does make some sense:) For now,
> > I don't think that cycle costs are the largest problem for VAX (the VAX
> > backend was even mentioned to be removed completely...).
> 
> I wasn't too concerned about the cost calculation.  If the code fragment
> doesn't fix the original problem, can you try in gdb
> 
> p debug_rtx (x)
> 
> so that we can see in more detail the rtx that caused the problem?  It
> might be necessary to break on vax_rtx_costs_1 with x==0x40273ea0 and
> step though the routine.

A good first introduction into GCC debugging:) Thanks. Any chance to
organize some hacker's meeting in Europe to let you teach first steps in
GCC hacking? Maybe you'd come over to the Oldenburger Linuxtag's Kernel
Hacker's meeting (mostly Linux), that would take care for proper
audience.

MfG, JBG



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[jbglaw at lug-owl dot de]

------- Additional Comments From jbglaw at lug-owl dot de  2004-06-30 06:53 -------
Subject: Re:  ICE (Internal compiler error) while compiling glibc to vax-linux target

On Wed, 2004-06-30 02:36:05 -0000, danglin at gcc dot gnu dot org <gcc-bugzilla@gcc.gnu.org>
wrote in message <20040630023605.3485.qmail@sourceware.org>:
> if gcc -DHAVE_CONFIG_H -I. -I../../gcc/libcpp -I.  -I../../gcc/libcpp/../include
>  -I../../gcc/libcpp -I../../gcc/libcpp/include -DLOCALEDIR="\"/usr/local/gcc/gcc
> -3.5.0/share/locale\""   -g -O2 -MT files.o -MD -MP -MF ".deps/files.Tpo" -c -o
> files.o ../../gcc/libcpp/files.c; \
> then mv -f ".deps/files.Tpo" ".deps/files.Po"; else rm -f ".deps/files.Tpo"; exi
> t 1; fi
> ../../gcc/libcpp/files.c: In function `read_file_guts':
> ../../gcc/libcpp/files.c:503: error: `ssize_t' undeclared (first use in this fun
> ction)
> ../../gcc/libcpp/files.c:503: error: (Each undeclared identifier is reported onl
> y once
> ../../gcc/libcpp/files.c:503: error: for each function it appears in.)
> ../../gcc/libcpp/files.c:503: error: parse error before "size"
> ../../gcc/libcpp/files.c:524: error: parse error before numeric constant
> ../../gcc/libcpp/files.c:524: error: parse error before numeric constant
> ../../gcc/libcpp/files.c:524: error: parse error before numeric constant
> ...

My build is still compiling, however, if you care enough, please go to
http://cvs.sourceforge.net/viewcvs.py/linux-vax/toolchain/experimental_toolchain/
and download the current version of experimental_toolchain.sh .

Then:
$ mkdir ~/some_temp_dir
$ cd ~/some_temp_dir
$ mkdir temp_for_build install
$ ..../experimental_toolahcin.sh --prepare --install --jbglaw ./temp_for_build/ ./install/

At least on my Linux system, that'll produce a vax-linux toolchain
(binutils + C-only gcc) and break somewhere in glibc some time later
(this is where I'm currently fixing:)

Oh, please read the comments inside the script; they'll provide some
important information (eg. you need to have some CVS pserver sources in
your ~/.cvspass).

MfG, JBG



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[danglin at gcc dot gnu dot org]

------- Additional Comments From danglin at gcc dot gnu dot org  2005-01-09 15:45 -------
Created an attachment (id=7912)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7912&action=view)
Patch

For main.  Should also apply to 3.4.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-01-09 15:58:53
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[danglin at gcc dot gnu dot org]

------- Additional Comments From danglin at gcc dot gnu dot org  2005-01-09 16:08 -------
Could you test this change?  I've done a full bootstrap with an x86 cross
to vax-ultrix but this doesn't provide a full test.

The ICE was introduced in the 3.4 branch when the macro RTX_COSTS was
removed.  This macro contained a switch statement containing a list of
codes that vax_rtx_costs could handle.  This list appeared redundant
but it was in fact protecting vax_rtx_costs from being exposed to rts's
such EXPR_LIST that it wasn't designed to handle.  I've now reworked
vax_rtx_costs to eliminate an unnecessary calling layer and to return
false for rtx codes that it can't handle.

In doing this rework, it became clear that vax_rtx_costs was broken
in a number of ways:

1) Various rtx codes used in the machine definition were missing,
2) The POST_INC and PRE_DEC codes were not accessible in 3.3 and
   earlier,
3) The operands for NEG and NOT were not being included in the cost
   estimate.

I've tried to fix the ICE and the above problems.  The missing
codes are given a default cost of 3 which certainly isn't right.
However, as can be seen from the comments in the function, the
cost estimates are only approximate and variations occur from
cpu to cpu.

I will try to test the patch latter but my vax is currently
busy doing a 3.3 build.



-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-03-25 17:31 -------
Subject: Bug 15491

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-4_0-branch
Changes by:	danglin@gcc.gnu.org	2005-03-25 17:31:41

Modified files:
	gcc            : ChangeLog 
	gcc/config/vax : vax.c 

Log message:
	PR target/15491
	* vax.c (vax_rtx_costs_1): Merge with vax_rtx_costs.
	(vax_rtx_costs): Return false when passed unsupported rtx's.  Handle
	FLOAT_EXTEND, FLOAT_TRUNCATE and TRUNCATE.  Fix costs for POST_INC,
	PRE_DEC, NEG and NOT.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=2.7592.2.84&r2=2.7592.2.85
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/vax/vax.c.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=1.57&r2=1.57.10.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-03-25 17:55 -------
Subject: Bug 15491

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	danglin@gcc.gnu.org	2005-03-25 17:54:59

Modified files:
	gcc            : ChangeLog 
	gcc/config/vax : vax.c 

Log message:
	PR target/15491
	* vax.c (vax_rtx_costs_1): Merge with vax_rtx_costs.
	(vax_rtx_costs): Return false when passed unsupported rtx's.  Handle
	FLOAT_EXTEND, FLOAT_TRUNCATE and TRUNCATE.  Fix costs for POST_INC,
	PRE_DEC, NEG and NOT.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7991&r2=2.7992
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/vax/vax.c.diff?cvsroot=gcc&r1=1.57&r2=1.58



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-03-25 18:31 -------
Subject: Bug 15491

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_4-branch
Changes by:	danglin@gcc.gnu.org	2005-03-25 18:30:58

Modified files:
	gcc            : ChangeLog 
	gcc/config/vax : vax.c 

Log message:
	PR target/15491
	* vax.c (vax_rtx_costs_1): Merge with vax_rtx_costs.
	(vax_rtx_costs): Return false when passed unsupported rtx's.  Handle
	FLOAT_EXTEND, FLOAT_TRUNCATE and TRUNCATE.  Fix costs for POST_INC,
	PRE_DEC, NEG and NOT.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=2.2326.2.826&r2=2.2326.2.827
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/vax/vax.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.52.6.1&r2=1.52.6.2



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[danglin at gcc dot gnu dot org]

------- Additional Comments From danglin at gcc dot gnu dot org  2005-03-25 19:28 -------
Fixed in CVS.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491

[Bug target/15491] ICE (Internal compiler error) while compiling glibc to vax-linux target

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15491