[herton@moen herton]$ gcc -v
Reading specs from /usr/lib/gcc-lib/i386-conectiva-linux/3.3.3/specs
Configured with: ../configure --prefix=/usr --bindir=/usr/bin --libdir=/usr/lib
--mandir=/usr/share/man --infodir=/usr/share/info --datadir=/usr/share
--enable-shared --disable-checking --enable-long-long --enable-__cxa_atexit
--enable-threads=posix --disable-libunwind-exceptions --enable-libgcj
--enable-languages=c,c++,f77,objc,java --with-system-zlib
--host=i386-conectiva-linux --target=i386-conectiva-linux
--with-gxx_include_dir=/usr/include/c++/3.3.3
Thread model: posix
gcc version 3.3.3 (release)

Ops, I forgot to include the steps to compile and reproduce the error:

1) Edit Makefile and Makefile.target with provided CFLAGS (for example using sed
as already showed).
2) Run configure: ./configure --prefix=/usr
3) make -j1

Please attach the preprocessed source obtained by adding -save-temps.

Created attachment 6625 [details]
preprocessed file

Here goes, console output, compressed .i file attached:

gcc -v -save-temps -O2 -frename-registers -fomit-frame-pointer -march=pentium4
-I. -I/tmp/qemu/BUILD/qemu-0.5.5/target-i386
-I/tmp/qemu/BUILD/qemu-0.5.5 -D_GNU_SOURCE -c -o helper.o
/tmp/qemu/BUILD/qemu-0.5.5/target-i386/helper.c
Reading specs from /usr/lib/gcc-lib/i386-conectiva-linux/3.3.3/specs
Configured with: ../configure --prefix=/usr --bindir=/usr/bin --libdir=/usr/lib
--mandir=/usr/share/man --infodir=/usr/share
/info --datadir=/usr/share --enable-shared --disable-checking
--enable-long-long --enable-__cxa_atexit --enable-threads=posi
x --disable-libunwind-exceptions --enable-libgcj
--enable-languages=c,c++,f77,objc,java --with-system-zlib --host=i386-conec
tiva-linux --target=i386-conectiva-linux
--with-gxx_include_dir=/usr/include/c++/3.3.3
Thread model: posix
gcc version 3.3.3 (release)
 /usr/lib/gcc-lib/i386-conectiva-linux/3.3.3/cc1 -E -quiet -v -I.
-I/tmp/qemu/BUILD/qemu-0.5.5/target-i386 -I/tmp/qemu/BUILD
/qemu-0.5.5 -D__GNUC__=3 -D__GNUC_MINOR__=3 -D__GNUC_PATCHLEVEL__=3
-D_GNU_SOURCE /tmp/qemu/BUILD/qemu-0.5.5/target-i386/hel
per.c -march=pentium4 -frename-registers -fomit-frame-pointer -O2 helper.i
ignoring nonexistent directory "/usr/i386-conectiva-linux/include"
#include "..." search starts here:
#include <...> search starts here:
 .
 /tmp/qemu/BUILD/qemu-0.5.5/target-i386
 /tmp/qemu/BUILD/qemu-0.5.5
 /usr/local/include
 /usr/lib/gcc-lib/i386-conectiva-linux/3.3.3/include
 /usr/include
End of search list.
 /usr/lib/gcc-lib/i386-conectiva-linux/3.3.3/cc1 -fpreprocessed helper.i -quiet
-dumpbase helper.c -march=pentium4 -auxbase-
strip helper.o -O2 -version -frename-registers -fomit-frame-pointer -o helper.s

GNU C version 3.3.3 (release) (i386-conectiva-linux)
	compiled by GNU C version 3.3.3 (release).
GGC heuristics: --param ggc-min-expand=31 --param ggc-min-heapsize=4096
In file included from /tmp/qemu/BUILD/qemu-0.5.5/target-i386/helper.c:2492:
/tmp/qemu/BUILD/qemu-0.5.5/softmmu_template.h: In function `slow_ldq_mmu':
/tmp/qemu/BUILD/qemu-0.5.5/softmmu_template.h:179: error: unable to find a
register to spill in class `GENERAL_REGS'
/tmp/qemu/BUILD/qemu-0.5.5/softmmu_template.h:179: error: this is the insn:
(insn:HI 167 164 168 6 0x4071f3f4 (parallel [
	    (set (reg:DI 112)
		(lshiftrt:DI (reg/v:DI 63)
		    (subreg:QI (reg:SI 111) 0)))
	    (clobber (scratch:SI))
	    (clobber (reg:CC 17 flags))
	]) 320 {lshrdi3_1} (insn_list 136 (insn_list 164 (nil)))
    (expr_list:REG_DEAD (reg/v:DI 63)
	(expr_list:REG_UNUSED (scratch:SI)
	    (expr_list:REG_UNUSED (reg:CC 17 flags)
		(nil)))))
/tmp/qemu/BUILD/qemu-0.5.5/softmmu_template.h:179: confused by earlier errors,
bailing out
make[1]: ** [helper.o] Erro 1
make[1]: Leaving directory `/tmp/qemu/BUILD/qemu-0.5.5/i386'
make: ** [all] Erro 1

small testcase:
-------------------------------------------------------------------------------
register T0 asm ("ebx");
register T1 asm ("esi");
register T2 asm ("edi");

void foo (int a, int b, long long c)
{
	*(long long *) (a + b) = c;
}
-------------------------------------------------------------------------------

Confirmed, a regression from 2.6.3 and 2.7.2.

x86 is known to be register stared and global registers just makes it worse because they make the 
registers fixed.

Caused by this patch:

Fri May 15 21:40:06 1998  John Wehle  (john@feith.com)

	* i386.md (movdi-1, movdi): Rewrite based on SI move patterns.

*** Bug 21469 has been marked as a duplicate of this bug. ***

*** Bug 22080 has been marked as a duplicate of this bug. ***

Taking three or more registers as in the test cases from comment #5 and 
from Bug 21469, or using regparam, is just asking for trouble on a target 
with only six registers available, total.  While this is a regression, we 
also have cases that didn't compile with old compilers and now do.  It is 
just a matter of luck if you're going to fix registers -- even the order 
in which the statements are expanded to RTL can matter for the register 
pressure, and in this case y'all are so unlucky to need a larger number of 
registers than what you've left available to the compiler. 
 
I believe this bug should be closed as SUSPEND unless test cases exist 
that use neither regparam nor global register variables. 
 

*** Bug 24292 has been marked as a duplicate of this bug. ***

(In reply to comment #5)
> small testcase:

This particular testcase doesn't fail for me anymore on mainline
as of 2006-02-24. However, QEMU 0.8.0 still doesn't build with
this compiler (you'll have to configure with --disable-gcc-check).

FWIW, I get:
--------------------------- 8< ---------------------------
/home/ranmath/src/gcc/build/gcc/xgcc -B/home/ranmath/src/gcc/build/gcc/ -Wall -O
2 -g -fno-strict-aliasing -fomit-frame-pointer -mpreferred-stack-boundary=2 -fal
ign-functions=0 -fno-gcse -fno-reorder-blocks -fno-optimize-sibling-calls -I. -I
/home/ranmath/tmp/qemu-0.8.0/target-i386 -I/home/ranmath/tmp/qemu-0.8.0 -I/home/
ranmath/tmp/qemu-0.8.0/linux-user -I/home/ranmath/tmp/qemu-0.8.0/linux-user/i386
 -D_GNU_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -I/home/ranmath/tmp/qe
mu-0.8.0/fpu -DHAS_AUDIO -I/home/ranmath/tmp/qemu-0.8.0/slirp -c -o op.o /home/r
anmath/tmp/qemu-0.8.0/target-i386/op.c
/home/ranmath/tmp/qemu-0.8.0/target-i386/ops_sse.h: In function 'op_pshufw_mmx':
/home/ranmath/tmp/qemu-0.8.0/target-i386/ops_sse.h:574: error: unable to find a
register to spill in class 'GENERAL_REGS'
/home/ranmath/tmp/qemu-0.8.0/target-i386/ops_sse.h:574: error: this is the insn:
(insn:HI 17 16 18 2 /home/ranmath/tmp/qemu-0.8.0/target-i386/ops_sse.h:569 (set
(strict_low_part (subreg:HI (reg/v:DI 63 [ r ]) 0))
        (mem/s/j:HI (plus:SI (mult:SI (reg:SI 64)
                    (const_int 2 [0x2]))
                (reg/v/f:SI 59 [ s ])) [0 <variable>._w S2 A16])) 40 {*movstrict
hi_1} (insn_list:REG_DEP_TRUE 49 (insn_list:REG_DEP_TRUE 12 (insn_list:REG_DEP_T
RUE 16 (nil))))
    (expr_list:REG_DEAD (reg:SI 64)
        (nil)))
/home/ranmath/tmp/qemu-0.8.0/target-i386/ops_sse.h:574: confused by earlier erro
rs, bailing out
make[1]: *** [op.o] Error 1
make[1]: Leaving directory `/home/ranmath/tmp/qemu-0.8.0/i386-user'
make: *** [all] Error 1
--------------------------- 8< ---------------------------

I am at SVN revision 111414.

*** Bug 27431 has been marked as a duplicate of this bug. ***

*** Bug 27967 has been marked as a duplicate of this bug. ***

*** Bug 31753 has been marked as a duplicate of this bug. ***

*** Bug 35135 has been marked as a duplicate of this bug. ***

A recent patch in mainline should fix this problem [1]. The testcase works OK for 4.3.4, 4.4.2 and 4.5.0. Please re-test with current mainline SVN and open a new PR if the test still fails.

[1] http://gcc.gnu.org/ml/gcc-patches/2009-09/msg00003.html