no joy with OSX10.6.4/HDF ?

[koppenhoefer]

Hi guys! (gals?).

short: Has anyone succeed in using icc 11.1 under OSX10.6.4 when building HDF5-1.8.5?

long: I am trying to BUILD and CHECK the latest HDF5-1.8.5 library with the latest INTEL-11.1 compiler Mac OSX-10.6.4 and despite a successful build, I am having difficulties at the 'make check' phase. We think there may be a bug in Intel icc 11.1 under 10.6.4. ?? Is there?

We do know that HD5-1.8.5 compile-checking has succeeded under MACOS10.6.3(Intel) as reported by @pahra (http://www.hdfgroup.org/HDF5/release/platforms5.html) but we have no reports of anyone succeeding with MACOS10.6.4(Intel) + Intel-11.1

Although compilation succeeds (with a few warnings),
when I do the 'make check' I get errors about "Error: hyperslab values don't match"

Building and checking with gcc works perfectly but not icc.
Any help would be appreciated :-)

shawn
p.s.
I don't know if it matters, but reading here:
http://software.intel.com/en-us/articles/performance-tools-for-software-developers-building-hdf5-with-intel-compilers/
I thought I understood that HDF5 uses an Autoconf "configure" script to determine the build environment and so I could get away with this:

export CC=icc
export CXX=icpc
export AR=xiar
export CFLAGS=''
export CXXFLAGS=''
export LDFLAGS=''
export LIBS=''

and then ./configure --prefix=/usr/local --with-szlib=/usr/local

[mecej4]

You may have set the bar too high: responders should have an interest in compiling a specific source version of HDF on a specific platform with a specific compiler, despite a prebuilt binary release of HDF being available.

I suggest that you obtain the binary release, and build/run the test problems against the distributed libraries, assuming that they work on your OS. If the tests succeed, there would be little reason for users of HDF to bother with fixing up the build scripts, and it would suffice to notify the HDF developers of the build problems.

[aazue]

Hi
I have make test this lib GNU an ICC
All test for two passed and perfectly working
I am with Debian O/S but probably must also result ok for your O/S almost based BSD

Change your processors type (march mtune) this flags are important.

make distclean
LANG=C; export LANG
CC="/opt/intel/Compiler/11.1/072/bin/intel64/icc"; export CC
LD_LIBRARY_PATH=/opt/intel/Compiler/11.1/072/lib/intel64;export LD_LIBRARY_PATH;export LD_LIBRARY_PATH
AR="/opt/intel/Compiler/11.1/072/bin/intel64/xiar";export AR
CFLAGS="-O2 -m64 -mieee-fp -ipo -override-limits -march=core2 -mtune=core2 -Wpointer-arith -fno-strict-aliasing "; export CFLAGS

./configure
make --jobs=4
make check

Warning: do not execute (make check)
if you having users linked shared working to your machine in same time
all resource machine is used full with some test.

I have made some tests (side btree...) (with personal source origin)
very very nice performance for Icc.
as to inverted some times i am disappointed and furious ,
like that result i like ICC and i am happy...

Remember before to evaluate performance to flush machine cache to default
(sync; echo 3 > proc/sys/vm/drop_caches)
Regards

[koppenhoefer]

The answer is that Xcode 3.2.2 and ifort are not compatible and can cause incorrect runtime results or segmentation faults.

Read more here: http://software.intel.com/en-us/forums/showthread.php?t=73942&o=d&s=lr

In short, to get a successful build / make check of hdf5 1.8.5 add '-use-asm' to the CFLAGS.

[aazue]

Hi
I think that you merge 2/3 problems
package that you have give result wrong build work perfectly all test passed...
Better that you dissociate threads subject with exactly as in relation with each step problem.
and that you not incriminate the wrong object, in your initial request .....
Regards