[Bug target/30370] New: [regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[Bug target/30370] New: [regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[rask at sygehus dot dk]

On an i686-pc-linux-gnu host, I configure GCC 4.3 (revision 120287) like this:
../../src/gcc/configure --enable-sim --with-newlib --disable-nls
--enable-e500_double --prefix=${HOME} --enable-checking=yes,rtl
--disable-threads --disable-shared --without-long-double-128
--target=powerpc-unknown-eabispe

(I also tried --with-long-double-128 as well as not specifying this option at
all with the same result.)

The build fails compiling libgcc/__powitf2.o:

/home/rask/build/gcc-ppc/./gcc/xgcc -B/home/rask/build/gcc-ppc/./gcc/ -nostdinc
-B/home/rask/build/gcc-ppc/powerpc-unknown-eabispe/newlib/ -isystem
/home/rask/build/gcc-ppc/powerpc-unknown-eabispe/newlib/targ-include -isystem
/home/rask/src/gcc/newlib/libc/include
-B/home/rask/powerpc-unknown-eabispe/bin/
-B/home/rask/powerpc-unknown-eabispe/lib/ -isystem
/home/rask/powerpc-unknown-eabispe/include -isystem
/home/rask/powerpc-unknown-eabispe/sys-include -L/home/rask/build/gcc-ppc/./ld
-O2  -O2 -g -O2  -DIN_GCC -DCROSS_DIRECTORY_STRUCTURE   -W -Wall
-Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition
 -isystem ./include   -specs=ldblspecs -g  -DIN_LIBGCC2
-D__GCC_FLOAT_NOT_NEEDED -Dinhibit_libc -I. -I. -I../../../src/gcc/gcc
-I../../../src/gcc/gcc/. -I../../../src/gcc/gcc/../include
-I../../../src/gcc/gcc/../libcpp/include 
-I../../../src/gcc/gcc/../libdecnumber -I../libdecnumber -DL_powitf2 -c
../../../src/gcc/gcc/libgcc2.c -o libgcc/./_powitf2.o
../../../src/gcc/gcc/libgcc2.c: In function '__powitf2':
../../../src/gcc/gcc/libgcc2.c:1735: internal compiler error: in gen_reg_rtx,
at emit-rtl.c:770

This is a regression from GCC 4.1, where GCC built for this target:
<URL:http://gcc.gnu.org/ml/gcc-testresults/2005-04/msg01907.html>


-- 
           Summary: [regression] Build failure in libgcc2 powitf2 with ICE
                    in gen_reg_rtx
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rask at sygehus dot dk
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: powerpc-unknown-eabispe


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[joseph at codesourcery dot com]

------- Comment #1 from joseph at codesourcery dot com  2007-01-04 20:34 -------
Subject: Re:   New: [regression] Build failure in libgcc2
 powitf2 with ICE in gen_reg_rtx

On Thu, 4 Jan 2007, rask at sygehus dot dk wrote:

>  -isystem ./include   -specs=ldblspecs -g  -DIN_LIBGCC2

There's your problem: -specs=ldblspecs should not be used for this target, 
where did it come from?

-specs=ldblspecs is a hack formerly used to disable 128-bit long double 
for soft float but enable it for hard float when building libgcc.  Ever 
since it was added, building libgcc for E500 GNU/Linux has been broken 
because -mlong-double-128 doesn't work there without my patch, so 
resulting in undefined references in libgcc_s, so meaning that it was 
necessary to disable -mlong-double-128 for E500 as well.  However, 
-spec=ldblspecs should never have been used on EABI, only on GNU/Linux; 
EABI doesn't build the darwin-ldouble.c support file needed.  With 
toplevel libgcc, -specs=ldblspecs is no more; instead 
libgcc/config/rs6000/t-ldbl128 adds -mlong-double-128 more directly for 
hard float (with the same breakage for E500), but again only for certain 
targets not including EABI.

To have long double actually work for E500 you'll need my patch 
<http://gcc.gnu.org/ml/gcc-patches/2006-12/msg01388.html>.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2007-01-05 05:03 -------
This whole SPE port is a mess and needs really a different target rather than
being part of the rs6000 backend or a big rewrite (well both are big rewrites).


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  BugsThisDependsOn|                            |30259
            Summary|[regression] Build failure  |Build failure in libgcc2
                   |in libgcc2 powitf2 with ICE |powitf2 with ICE in
                   |in gen_reg_rtx              |gen_reg_rtx


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[rask at sygehus dot dk]

------- Comment #3 from rask at sygehus dot dk  2007-01-05 17:29 -------
Created an attachment (id=12861)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12861&action=view)
patch in testing to fix typo

The option -specs=ldblspecs is added by gcc/config/rs6000/t-ppccomm when
compiling libgcc2 for GNU systems, but there is a typo in the condition. The
build is now in progress past the failure point after applying the attached
patch.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[joseph at codesourcery dot com]

------- Comment #4 from joseph at codesourcery dot com  2007-01-05 17:52 -------
Subject: Re:  Build failure in libgcc2 powitf2 with ICE in
 gen_reg_rtx

On Fri, 5 Jan 2007, rask at sygehus dot dk wrote:

> The option -specs=ldblspecs is added by gcc/config/rs6000/t-ppccomm when
> compiling libgcc2 for GNU systems, but there is a typo in the condition. The
> build is now in progress past the failure point after applying the attached
> patch.

Thanks, I'd just presumed that the condition in t-ppccomm, though 
peculiarly written, did work.

If we add

ifeq (,$(findstring gnuspe,$(target)))

to the condition as well, then that should complete letting 4.1 and 4.2 
build properly out of the box for SPE.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[rask at sygehus dot dk]

------- Comment #5 from rask at sygehus dot dk  2007-01-06 14:28 -------
The condition in t-ppccomm is written just like the example in the GNU Make
documentation ("Conditionals that Test Flags"), except for the missing
parentheses.

Indeed toplevel libgcc fixes this, revision 120505 doesn't have this problem.

joseph@codesourcery.com wrote:
> If we add

> ifeq (,$(findstring gnuspe,$(target)))

> to the condition as well, then that should complete letting 4.1 and 4.2
> build properly out of the box for SPE.

Like
ifneq (,$(findstring gnu,$(target)))
ifeq (,$(findstring gnuspe,$(target)))
...
endif
endif
?

How about s/gnuspe/spe/ in case someone comes up with a powerpc-gnufubarspe
target?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

Re: [Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[Joseph S. Myers]

On Sat, 6 Jan 2007, rask at sygehus dot dk wrote:

> Like
> ifneq (,$(findstring gnu,$(target)))
> ifeq (,$(findstring gnuspe,$(target)))
> ...
> endif
> endif
> ?

Yes, just like that.

> How about s/gnuspe/spe/ in case someone comes up with a powerpc-gnufubarspe
> target?

Seems reasonable.

-- 
Joseph S. Myers
joseph@codesourcery.com

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[joseph at codesourcery dot com]

------- Comment #6 from joseph at codesourcery dot com  2007-01-06 15:08 -------
Subject: Re:  Build failure in libgcc2 powitf2 with ICE in
 gen_reg_rtx

On Sat, 6 Jan 2007, rask at sygehus dot dk wrote:

> Like
> ifneq (,$(findstring gnu,$(target)))
> ifeq (,$(findstring gnuspe,$(target)))
> ...
> endif
> endif
> ?

Yes, just like that.

> How about s/gnuspe/spe/ in case someone comes up with a powerpc-gnufubarspe
> target?

Seems reasonable.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[rask at sygehus dot dk]

------- Comment #7 from rask at sygehus dot dk  2007-01-07 14:07 -------
I tried to build several versions to find out which ones work:

GCC 4.0.3 works (dies later compiling newlib/libc/math/e_j0.c).
GCC 4.1.0 unknown (fails to compile rs6000.c).
GCC 4.1.1 works (dies later with bug 27075).
GCC 4.1 branch fails (dies on multc3 shortly after compiling powitf2).
GCC 4.2 branch fails (dies on multc3 shortly after compiling powitf2).
GCC 4.3 works.

The 4.1/4.2 failure is an assertion at emit_move_multiword, expr.c:3188, but
the root cause is the same -specs=ldblspecs option.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[rask at sygehus dot dk]

------- Comment #8 from rask at sygehus dot dk  2007-01-07 14:16 -------
Created an attachment (id=12869)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12869&action=view)
updated patch in testing

Here's a new patch as discussed. The GCC 4.2 branch now manages to build
libgcc2. I haven't tried 4.1 yet with the patch.


-- 

rask at sygehus dot dk changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #12861|0                           |1
        is obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[drow at gcc dot gnu dot org]

------- Comment #9 from drow at gcc dot gnu dot org  2007-01-07 23:42 -------
For mainline, I would suggest using configure to check whether
-mlong-double-128 can be used, for targets where it would be used by the
makefile fragments.  That will handle multilibs properly, et cetera.  Unless
after Joseph's patch we should be using the flag for e500, of course.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[mmitchel at gcc dot gnu dot org]

------- Comment #10 from mmitchel at gcc dot gnu dot org  2007-01-31 01:20 -------
This is a regression in 4.1.2, relative to 4.1.1.


-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
            Summary|Build failure in libgcc2    |[4.1 regression] Build
                   |powitf2 with ICE in         |failure in libgcc2 powitf2
                   |gen_reg_rtx                 |with ICE in gen_reg_rtx


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[dje at gcc dot gnu dot org]

------- Comment #11 from dje at gcc dot gnu dot org  2007-02-02 16:06 -------
I do not have a problem with the patch in principle.  However, prior to the
recent merge of the long-double-128 soft float support, one of the earlier
patches caused GCC to fail to build on GNU/Linux distros without
long-double-128 support in Glibc, such as SLES 9.  I would want to know that
this patch does not cause regressions on GNU/Linux before feeling completely
comfortable about the patch.


-- 

dje at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2007-02-02 16:06:26
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[jsm28 at gcc dot gnu dot org]

-- 

jsm28 at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.1.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[mmitchel at gcc dot gnu dot org]

------- Comment #12 from mmitchel at gcc dot gnu dot org  2007-02-05 06:05 -------
Richard, would you be able to test the patch on SLES9, as David suggests?

Thanks,

-- Mark


-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenther at suse dot de,
                   |                            |mmitchel at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[dje at gcc dot gnu dot org]

------- Comment #13 from dje at gcc dot gnu dot org  2007-02-05 17:35 -------
If the patch is okay on SLES 9, as Richard says,

http://gcc.gnu.org/ml/gcc/2007-02/msg00068.html

then it's fine with me.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[mmitchel at gcc dot gnu dot org]

------- Comment #14 from mmitchel at gcc dot gnu dot org  2007-02-05 19:17 -------
Richard, thank you for the testing.  Since David is now comfortable, I think we
should include this patch on the 4.1 branch, so that we fix this problem for
4.1.2.  If nobody beats me to it, I will apply the patch this afternoon.

Thanks to all for comments and feedback.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[mmitchel at gcc dot gnu dot org]

------- Comment #15 from mmitchel at gcc dot gnu dot org  2007-02-06 20:50 -------
Subject: Bug 30370

Author: mmitchel
Date: Tue Feb  6 20:50:37 2007
New Revision: 121668

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=121668
Log:
        PR 30370 
        * config/rs6000/t-ppccomm: Correct Makefile typo.

Modified:
    branches/gcc-4_1-branch/gcc/ChangeLog
    branches/gcc-4_1-branch/gcc/config/rs6000/t-ppccomm


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[mmitchel at gcc dot gnu dot org]

------- Comment #16 from mmitchel at gcc dot gnu dot org  2007-02-06 20:58 -------
Fixed in 4.1.2.


-- 

mmitchel at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370

[Bug target/30370] [4.1 regression] Build failure in libgcc2 powitf2 with ICE in gen_reg_rtx

[mmitchel at gcc dot gnu dot org]

------- Comment #17 from mmitchel at gcc dot gnu dot org  2007-02-06 21:06 -------
Subject: Bug 30370

Author: mmitchel
Date: Tue Feb  6 21:06:36 2007
New Revision: 121669

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=121669
Log:
        PR 30370 
        * config/rs6000/t-ppccomm: Correct Makefile typo.

Modified:
    branches/gcc-4_2-branch/gcc/ChangeLog
    branches/gcc-4_2-branch/gcc/config/rs6000/t-ppccomm


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30370