Polyhedron PB05/PPC problems
IainS
developer@sandoe-acoustics.co.uk
Wed Jun 18 09:20:00 GMT 2008
A "heads up" on some problems with ppc (at least on G5).
Apologies that the info is not more detailed, but time is very
limited right now.
I've being trying to run Polyhedron PB05 on two platforms and running
into these errors (PPC).
(The i386 code built from the same tree passes all these tests)
gcc version 4.4.0 20080617 (experimental) (GCC)
Configured with: /Volumes/UFSScratch/GCC/gcc-4-4-trunk/configure --
prefix=/usr --build=powerpc-apple-darwin8 --enable-languages=c,c+
+,objc,obj-c++,fortran --with-slibdir='/usr/lib' --with-gmp=/
Developer/SDKs/MacOSX10.4us.sdk/usr/local/gmp-4-2-2 --with-mpfr=/
Developer/SDKs/MacOSX10.4us.sdk/usr/local/mpfr-2-3-1 --enable-threads
--disable-nls --enable-version-specific-runtime-libs --mandir='/usr/
share/man' --infodir='/usr/share/info' --program-suffix=-4.4.0 --
enable-checking=release --enable-serial-configure --host=powerpc-
apple-darwin8 --target=powerpc-apple-darwin8 --enable-werror --enable-
bootstrap
=====
gas_dynam:
O0
OK
O2
OK
O3
> Value=0.86723800000E-06 Target=0.86774900000E-06
Tolerance=0.10000000000E-10
gas_dyn FAILED 1 fails and 0 passes
-ffast-math -funroll-loops -ftree-loop-linear -ftree-vectorize -O3
OK
========
linpk
O0
OK
O2
> Value= 26.254094400 Target= 23.100000000 Tolerance=
2.0000000000
linpk FAILED 1 fails and 4 passes
O3
> Value= 26.254094400 Target= 23.100000000 Tolerance=
2.0000000000
linpk FAILED 1 fails and 4 passes
-ffast-math -funroll-loops -ftree-loop-linear -ftree-vectorize -O3
> Value= 26.254094400 Target= 23.100000000 Tolerance=
2.0000000000
linpk FAILED 1 fails and 4 passes
========
capacita
O0
OK
O2
OK
O3
OK
-ffast-math -funroll-loops -ftree-loop-linear -ftree-vectorize -O3
OK
-mcpu=970 -mtune=970
produced code that didn't complete (after >8 hours) :-
( [ compare with 241 secs compiled -O0]
sampler says that the code is spending
100% __solve_cap_MOD_gcstab
98% __solve_cap_MOD_mv_prod
55% __solve_cap_MOD_prod1
44% __solve_cap_MOD_preco
<1% __solv_cap_MOD_mv_prod
<1% _gfortrani_reshape_packed
2% __solve_cap_MOD_gcstab
----
Iain
More information about the Fortran
mailing list