Re: Tutorial Time

Re: Tutorial Time

[Mark Butcher]

Hi Richard

mcore-pe.h:#define CC1_SPEC "-funsigned-bitfields %{!DIN_GCC:-fno-rtti}
%{!DIN_GCC:-fno-exceptions}" is indeed in gcc 3.1 but (sorry forgot to
specify) I had the problem with gcc 3.0.4, where its isn't.

gcc 3.1 was released a few weeks back but I have been working on the
problem from before then !! Have just loaded gcc 3.1 and will continue the
experiments with that...

Cheers

Mark

Tutorial Time

[Mark Butcher]

Hi All

This is both a request for help and a discussion of internal workings.

The background is that I am (still) having problems building a cross
compiler for the Mcore target and have decided to learn as much as possible
about the procedure and environment so that I can solve them myself (still
beginner status with regards to just about everything).

I have configured to build c and c++ and the build is failing while
configuring internally to build the c++ library "libstdc++".



Here is the script which is failing while trying to identify the exception
model:



/************************************ beginning of script part causing
error ************************************************/
  echo $ac_n "checking for exception model to use""... $ac_c" 1>&6
echo "configure:3700: checking for exception model to use" >&5
  
  ac_ext=C
# CXXFLAGS is not in ac_cpp because -g, -O, etc. are not valid cpp options.
ac_cpp='$CXXCPP $CPPFLAGS'
ac_compile='${CXX-g++} -c $CXXFLAGS $CPPFLAGS conftest.$ac_ext 1>&5'
ac_link='${CXX-g++} -o conftest${ac_exeext} $CXXFLAGS $CPPFLAGS $LDFLAGS
conftest.$ac_ext $LIBS 1>&5'
cross_compiling=$ac_cv_prog_cxx_cross

  # Check whether --enable-sjlj-exceptions or --disable-sjlj-exceptions was
given.
if test "${enable_sjlj_exceptions+set}" = set; then
  enableval="$enable_sjlj_exceptions"
  :
else
                       cat > conftest.$ac_ext << EOF
#line 3715 "configure"
struct S { ~S(); };
void bar();
void foo()
{
  S s;
  bar();
}
EOF
   old_CXXFLAGS="$CXXFLAGS"  
   CXXFLAGS=-S
   if { (eval echo configure:3726: \"$ac_compile\") 1>&5; (eval
$ac_compile) 2>&5; }; then
     if grep _Unwind_SjLj_Resume conftest.s >/dev/null 2>&1 ; then
       enable_sjlj_exceptions=yes
     elif grep _Unwind_Resume conftest.s >/dev/null 2>&1 ; then
       enable_sjlj_exceptions=no
     fi
   fi
   CXXFLAGS="$old_CXXFLAGS"
   rm -f conftest*
fi

   if test x$enable_sjlj_exceptions = xyes; then
     cat >> confdefs.h <<\EOF
#define _GLIBCPP_SJLJ_EXCEPTIONS 1
EOF

     ac_exception_model_name=sjlj
   elif test x$enable_sjlj_exceptions = xno; then
     ac_exception_model_name="call frame"
   else
     { echo "configure: error: unable to detect exception model" 1>&2; exit
1; }
   fi

/************************************ end of script part causing error
************************************************/


After analysing this I have identified that a test file called conftest.c
is temporarily created with the following contents

#line 3715 "configure"
struct S { ~S(); };
void bar();
void foo()
{
  S s;
  bar();
}

This is then compiled to test what the compiler generates - the exact
command being:

/usr/local/src/gnu/BUILD/gcc/gcc/xgcc -B/usr/local/src/gnu/BUILD/gcc/gcc/
-nostdinc++  -L/usr/local/src/gnu/BUILD/gcc/mcore-elf/libstdc++-v3/src
-L/usr/local/src/gnu/BUILD/gcc/mcore-elf/libstdc++-v3/src/.libs
-B/usr/local/mcore/mcore-elf/bin/ -B/usr/local/mcore/mcore-elf/lib/
-isystem /usr/local/mcore/mcore-elf/include -c -S  conftest.C

Now, if I actually compile this file with the said command I see that the
generated assembler file is correct for the desired target (Mcore) - the
content is:

        .file   "conftest.c"
        .text
        .align  1
        .export _Z3foov
        .type   _Z3foov,@function
_Z3foov:
        subi    sp,16
        stw     r15,(sp,8)
        stw     r8,(sp,12)
        mov     r8,sp
        jbsr    _Z3barv
        mov     r2,r8
        jbsr    _ZN1SD1Ev
        mov     sp,r8
        ldw     r15,(sp,8)
        ldw     r8,(sp,12)
        addi    sp,16
        jmp     r15
.Lfe1:
        .size   _Z3foov,.Lfe1-_Z3foov
        .ident  "GCC: (GNU) 3.0.4"


If I am correctly interpreting the script, grep is used to search for
information about two possible exception models, either the string
_Unwind_SjLj_Resume or 
_Unwind_Resume
should be somewhere in the S-file.

Since none of these are in the file, the configuration can not identify the
model and aborts.


Now to the questions:

1. Are these strings (which are searched for) added by some compilers and
not by the mcore cross compiler ?
2. Could this error be avoided by configuring with
"--disable-sjlj-exceptions"
3. What does sjij mean ?
4. Can anyone explain what this 'exception' model is, or where to get the
necessary details about it ?
5. I have scanned various documents and have not yet found a mention of the
Motorola Mcore being supported by GCC (I didn't find anything in the latest
version of the gcc compiler manual). Have I missed something or is this
really the case ?
6. Does anyone know details about the target configuration (for example use
of --disable-sjlj-exceptions, if necessary) ? I checked out lists of test
reports and found no mention of the Mcore there.
7. Am I mad ? (if not, certainly on the way...)

Thanks for any remarks

Cheers

Mark in Switzerland

Re: Tutorial Time

[Richard Henderson]

On Wed, Jun 12, 2002 at 06:47:02PM -0400, Mark Butcher wrote:
> 2. Could this error be avoided by configuring with
> "--disable-sjlj-exceptions"

No.  The problem is 

mcore-pe.h:#define CC1_SPEC "-funsigned-bitfields %{!DIN_GCC:-fno-rtti} %{!DIN_GCC:-fno-exceptions}"

Exceptions have been disabled for this target entirely.
I have no idea why this was done.


r~

Re: Tutorial Time

[mike stump]

> Date: Thu, 13 Jun 2002 09:04:16 -0700
> From: Richard Henderson <rth@redhat.com>
> To: Mark Butcher <M_J_BUTCHER@compuserve.com>

> On Wed, Jun 12, 2002 at 06:47:02PM -0400, Mark Butcher wrote:
> > 2. Could this error be avoided by configuring with
> > "--disable-sjlj-exceptions"

> No.  The problem is 

> mcore-pe.h:#define CC1_SPEC "-funsigned-bitfields %{!DIN_GCC:-fno-rtti} %{!DIN_GCC:-fno-exceptions}"

> Exceptions have been disabled for this target entirely.  I have no
> idea why this was done.

I can guess.  Someone was too lazy, or wasn't paid to make it work and
QA it.  :-)  Certainly I've seen mcore work with sjlj exceptions before
and pass the entire C++ testsuite with around 3 failures in-house, so I
know it can be made to work.

Re: Tutorial Time

[Gokhan Kisacikoglu]

> 3. What does sjij mean ?

Check out the following links;

http://www.rtems.com/rtems/maillistArchives/rtems-users/1999/april/thrd1.html#00016

http://www.rtems.com/rtems/maillistArchives/rtems-users/1999/april/msg00016.html

"D. V. Henkel-Wallace" <gumby@zembu.com> says;

sjlj-exceptions are exceptions implemented via setjump and longjump.  I 
don't advise using them, especially since ELF exceptions should work.

The fact that sjlj-exceptions only work part of the time suggests that
you 
may have a different problem than the exceptions themselves.

Contact him if you need more info, there is also a number of people
there who covered the topic there...

> 5. ... not yet found a mention of the Motorola Mcore being supported ...

Check out the following link: 

http://gcc.gnu.org/gcc-3.0/features.html

Search for Motorola, you will find the following;

	New Targets and Target Specific Improvements:
	o Port of GCC to Motorola's MCore 210 and 340 contributed. 

So, I guess it depends which mcore you are using...

> 7. Am I mad ?

It depends on the observer, everything is relative.

Gokhan