t2 failure

[Jason Moxham]

I forgot all about t2 ...

FAIL: t-modlinv
ld.so.1: t-popc: fatal: /usr/local/sparc-solaris-toolchain/lib/libgcc_s.so.1:
wrong ELF class: ELFCLASS32
/bin/bash: line 1: 17505 Killed ${dir}$tst
FAIL: t-popc
ld.so.1: t-parity:
fatal: /usr/local/sparc-solaris-toolchain/lib/libgcc_s.so.1: wrong ELF class:
ELFCLASS32
/bin/bash: line 1: 17525 Killed ${dir}$tst
FAIL: t-parity
ld.so.1: t-sub: fatal: /usr/local/sparc-solaris-toolchain/lib/libgcc_s.so.1:
wrong ELF class: ELFCLASS32
/bin/bash: line 1: 17545 Killed ${dir}$tst
FAIL: t-sub
=============================================================
9 of 9 tests failed
Please report to http://groups.google.co.uk/group/mpir-devel/
=============================================================
make[4]: *** [check-TESTS] Error 1
make[4]: Leaving directory `/tmp/jason/t2/tests'
make[3]: *** [check-am] Error 2

I expect this is a wrong path .
How come so many unix boxen have their compilers set up wrong , or the paths
not set?

[William Hart]

I already did tests for t2 and all passed. I forgot the path now but I
can send it later when home. Yes it is definitely a missing path. It's
something like / usr/local/....toolchain/lib

And it is libstdc++.so.6 it needs.

Bill.

Sent from my iPhone

[Jason Moxham]

Ok
I do get this which I suppose is a NFS problem

-I/home/jasonmoxham/mpir/mpir/trunk/mpn -I.. `test -f 'lshift.asm' ||
echo '/home/jasonmoxham/mpir/mpir/trunk/mpn/'`lshift.asm
make[2]: stat:rshift.asm: Interrupted system call
make[2]: *** No rule to make target `rshift.asm', needed by `rshift.lo'.
Stop.
make[2]: *** Waiting for unfinished jobs....

[William Hart]

Parallel make doesn't work on solaris. Could this be the problem?

Bill.

Sent from my iPhone

[Jason Moxham]

You mean make -j 2 or more may not work?

I didn't have a problem with mark/mark2 , but maybe it's rare.

[Bill Hart]

According to my notes, parallel make and above did not work for me on
mark, menas or varro, it is exceptionally slow on eno, and doesn't
work with Sun CC on fulvia.

I recall parallel make doesn't work on t2.

Bill.

2009/6/5 Jason Moxham <ja...@njkfrudils.plus.com>:

[Jason Moxham]

I used make -j 2 on mark mark2 varro and make -j 4 on fulvia and the rest of
the machines I tested I just used however many processors appeared in
proc/cpuinfo , didn't have a problem which any of them with gcc or cc etc

[Bill Hart]

Sometimes it seems to work. However I tend to do make -j without
specifying a number. Perhaps that is wrong.

2009/6/5 Jason Moxham <ja...@njkfrudils.plus.com>:

[Jason Moxham]

On Friday 05 June 2009 22:56:50 Bill Hart wrote:
> Sometimes it seems to work. However I tend to do make -j without
> specifying a number. Perhaps that is wrong.

Testing on box1 nehalem 4core,2 threads ie 8 virtual cores

make distclean ; time (./configure && make && make check)
real 2m3.288s
user 1m34.790s
sys 0m30.626s

make distclean ; time (./configure && make -j 2 && make -j 2 check)

real 1m18.902s
user 1m35.302s
sys 0m32.178s

make distclean ; time (./configure && make -j 4 && make -j 4 check)


real 1m2.553s
user 1m48.535s
sys 0m34.038s

make distclean ; time (./configure && make -j 6 && make -j 6 check)

real 0m58.596s
user 2m0.740s
sys 0m35.658s

make distclean ; time (./configure && make -j 8 && make -j 8 check)

real 0m57.187s
user 2m6.732s
sys 0m37.018s

make distclean ; time (./configure && make -j 10 && make -j 10 check)

real 0m55.837s
user 2m10.700s
sys 0m37.790s

make distclean ; time (./configure && make -j 12 && make -j 12 check)

real 0m55.485s
user 2m13.480s
sys 0m38.810s

make distclean ; time (./configure && make -j && make -j check)

real 0m55.176s
user 2m17.061s
sys 0m39.362s

Clearly configure and the tests are holding back the times.
I tried making the tests go in parallel but I think you may need the latest
autotools , as all I got was an error message when I did autoreconf

[Jason Moxham]

I added
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/sparc-solaris-toolchain/lib/sparcv9
and the test suite passes for gcc but not for cc
I'll try a non-parellel build just in case it was that

the error is

cc: Warning: -xarch=v9 is deprecated, use -m64 to create 64-bit programs
ld: fatal: file libmpir.so.3: open failed: No such file or directory
make[2]: *** [libmpir.la] Error 1
make[2]: Leaving directory `/home/jasonmoxham/mpir/mpir/trunk/t2'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/jasonmoxham/mpir/mpir/trunk/t2'
make: *** [all] Error 2

[Jason Moxham]

Well , its not a parallel build error , although if do make -j 16 I get

IN_GMP -I.. -xarch=v9 -xchip=ultra -c -o clear.lo clear.c
make[2]: stat:clrbit.lo: Interrupted system call
make[2]: stat:clrbit.c: Interrupted system call
make[2]: *** No rule to make target `clrbit.c', needed by `clrbit.lo'. Stop.

make[2]: *** Waiting for unfinished jobs....

so clearly theres a problem with NFS?

I going to try static,shared builds to see if it's one or the other



On Saturday 06 June 2009 18:53:02 Jason Moxham wrote:
> I added
> LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/sparc-solaris-toolchain/lib/spa

[Jason Moxham]

The static library is fine , it's just the shared library that doesn't build
from configure it looks like the gnu linker is being used whereas we should (I
think) use the sun one.

[Jason Moxham]

setting
LD=/usr/ccs/bin/sparcv9/ld
appears to work , nearly.... it now breaks c++ :(

[Jason Moxham]

Seem to have most stuff working now but we now get this c++ error , as my
knowledge of c++ is <0 , I'll leave for someone else :)


/../libmpirxx.la ../../libmpir.la
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Ambiguous
partial specialization for __gmp_expr<__mpf_struct[1],
__gmp_binary_expr<long, __gmp_expr<__mpf_struct[1], __mpf_struct[1]>,
__gmp_binary_divides>>, __gmp_expr and __gmp_expr.
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Ambiguous
partial specialization for __gmp_expr<__mpf_struct[1],
__gmp_binary_expr<long, __gmp_expr<__mpf_struct[1], __mpf_struct[1]>,
__gmp_binary_divides>>, __gmp_expr and __gmp_expr.
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Ambiguous
partial specialization for __gmp_expr<__mpf_struct[1],
__gmp_binary_expr<__gmp_expr<__mpf_struct[1], __gmp_binary_expr<long,
__gmp_expr<__mpf_struct[1], __mpf_struct[1]>, __gmp_binary_divides>>,
__gmp_expr<__mpf_struct[1], __gmp_binary_expr<long,
__gmp_expr<__mpf_struct[1], __mpf_struct[1]>, __gmp_binary_divides>>,
__gmp_binary_plus>>, __gmp_expr and __gmp_expr.
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Too many
arguments in cast to __gmp_expr<__mpf_struct[1],
__gmp_binary_expr<__gmp_expr<__mpf_struct[1], __gmp_binary_expr<long,
__gmp_expr<__mpf_struct[1], __mpf_struct[1]>, __gmp_binary_divides>>,
__gmp_expr<__mpf_struct[1], __gmp_binary_expr<long,
__gmp_expr<__mpf_struct[1], __mpf_struct[1]>, __gmp_binary_divides>>,
__gmp_binary_plus>>.
4 Error(s) detected.
make[4]: *** [t-prec.o] Error 4
make[4]: *** Waiting for unfinished jobs....
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Ambiguous
partial specialization for __gmp_expr<__mpz_struct[1],
__gmp_binary_expr<__gmp_expr<__mpz_struct[1], __mpz_struct[1]>,
__gmp_expr<__mpz_struct[1], __mpz_struct[1]>, __gmp_binary_plus>>, __gmp_expr
and __gmp_expr.
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Too many
arguments in cast to __gmp_expr<__mpz_struct[1],
__gmp_binary_expr<__gmp_expr<__mpz_struct[1], __mpz_struct[1]>,
__gmp_expr<__mpz_struct[1], __mpz_struct[1]>, __gmp_binary_plus>>.
2 Error(s) detected.
make[4]: *** [t-binary.o] Error 2
CC -xarch=v9 -xchip=ultra -o .libs/t-constr
t-constr.o -L/tmp/jason/t2/.libs ../../tests/.libs/libtests.a ../../.libs/libmpirxx.so /tmp/jason/t2/.libs/libmpir.so ../../.libs/libmpir.so -R/usr/local/lib
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3076: Error: Ambiguous
partial specialization for __gmp_expr<__mpz_struct[1],
__gmp_unary_expr<__gmp_expr<__mpz_struct[1], __mpz_struct[1]>,
__gmp_unary_plus>>, __gmp_expr and __gmp_expr.
"/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3076: Error: The
type "__gmp_expr<__mpz_struct[1],
__gmp_unary_expr<__gmp_expr<__mpz_struct[1], __mpz_struct[1]>,
__gmp_unary_plus>>" is incomplete.
2 Error(s) detected.
make[4]: *** [t-unary.o] Error 2

make[3]: *** [check-am] Error 2

this was built with

LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/sparc-solaris-toolchain/lib/sparcv9
LD=/usr/ccs/bin/sparcv9/ld
CC=cc
CXX=CC
./configure --enable-cxx

[Marc]

On Jun 6, 3:18 pm, Jason Moxham <ja...@njkfrudils.plus.com> wrote:
> Seem to have most stuff working now but we now get this c++ error , as my
> knowledge of c++ is <0 , I'll leave for someone else :)
>
> /../libmpirxx.la ../../libmpir.la
> "/home/jasonmoxham/mpir/mpir/trunk/mpirxx.h", line 3080: Error: Ambiguous
> partial specialization for __gmp_expr<__mpf_struct[1],
> __gmp_binary_expr<long, __gmp_expr<__mpf_struct[1], __mpf_struct[1]>,
> __gmp_binary_divides>>, __gmp_expr and __gmp_expr.

I guess you are using studio 11 or an unpatched studio 12.
http://gmplib.org/list-archives/gmp-bugs/2009-June/001491.html

> LD=/usr/ccs/bin/sparcv9/ld

I think nowadays things tend to work better if you never invoke ld
directly but only through the compiler. I haven't checked exactly
where LD is used in mpir though.

[Jason Moxham]

Without any options nothing would compile , with
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/sparc-solaris-toolchain/lib/sparcv9
I could get gcc to compile c and c++
Sun's cc,CC would compile a c static library but not a shared one
Setting LD=/usr/ccs/bin/sparcv9/ld enable compilation of the shared library
but of then we get the above c++ error

Looks like we have a unpatched studio12

Jason



>

[Bill Hart]

Hi Jason,

t2 is actually not on SkyNet. It is a machine donated to University of
Washington by Sun. I'm CC'ing the relevant people who maintain the
machine.

Bill.

2009/6/7 Jason Moxham <ja...@njkfrudils.plus.com>: