[kaffe] Kaffe for Big-endian ARM

[Jari Korva]

Thanks! Unfortunately I wasn't lucky: tried compiling
methodcalls.c without -O and then finally ripped off everything I could
from the command line but the result was always the same (see below). I
suppose I need to retry again as soon as crosstool supports gcc beyond
1.4.1 and/or new patches that might help...

armv5b-softfloat-linux-gcc -DHAVE_CONFIG_H -I. -I. -I../../../config
-I../../../include/kaffe -I../../../kaffe/kaffevm/systems/unix-pthreads
-I../../../kaffe/kaffevm/jni -I.. -I./.. -DINTERPRETER -I../../../include
-I/wrk/ipaq/u/jko/kaffe-cvs-head/./kaffe/kaffevm
-I/wrk/ipaq/u/jko/kaffe-cvs-head/./kaffe/kaffevm/intrp
-I/wrk/ipaq/u/jko/kaffe-cvs-head/./kaffe/kaffevm/systems/unix-pthreads
-I../../../binreloc -I../../../config -I../../../include -g -pthread
-DENABLE_BINRELOC -MF .deps/methodcalls.Tpo -c methodcalls.c -o
.libs/methodcalls.o
methodcalls.c: In function `engine_callMethod':
methodcalls.c:44: warning: initialization makes pointer from integer
without a cast
methodcalls.c:136: error: insn does not satisfy its constraints:
(insn 666 462 463 32 ../../../config/arm/sysdepCallMethod.h:73 (set
(reg:DF 1 r1)
        (reg/v:DF 16 f0 [ f0 ])) 145 {*movdf_soft_insn} (nil)
    (nil))
methodcalls.c:136: internal compiler error: in
reload_cse_simplify_operands, at postreload.c:378
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

- Jari

On Thu, 26 Aug 2004, Luca Saiu wrote:
>    You hit a GCC bug. I don't know whether Kaffe developers know a
> Kaffe-specific method to work around this one, but in cases like yours
> you can nearly always work around the bug disabling optimizations or
> using a lower optimization level.
>    Try manually compiling just the file where compilation blown without
> optimizazion, and then calling make again.
>
>    You can also try usign a different GCC version, but beware of the
> incompatibilities of different ABIs; using 3.4.1 (and 3.4.2 which is
> scheduled to be released on 1st September) is most probably safe. The
> *minor* GCC versions should be the same for two versions to be C++
> ABI-compatible.
>
>    Regards,
>
> --
> Luca Saiu, maintainer of GNU epsilon
> http://www.gnu.org/software/epsilon
>