Yesterday me and @berceanu encountered a problem when trying to build LaserWakefield example from the latest dev using 'laser' (CPU) profile on Hypnos. Today I tried that again and got the same error message:
make[2]: * No rule to make target
/usr/local/lib/libz.so', needed bypicongpu'. Stop.
make[2]: * Waiting for unfinished jobs....
With k80 profile it builds fine.
With @psychocoderHPC we took a look at the output of module list with laser and k80 profiles and it uses different version of hdf5.
For k80 profile output of module list is:
1) gcc/4.9.2 4) cuda/8.0 7) hdf5-parallel/1.8.20
2) cmake/3.10.1 5) openmpi/2.1.2.cuda80 8) libsplash/1.7.0
3) boost/1.62.0 6) pngwriter/0.7.0
and for 'laser' profile it is:
1) gcc/4.9.2 4) psm/3.1 7) pngwriter/0.7.0
2) cmake/3.10.1 5) openmpi/1.8.6 8) hdf5-parallel/1.8.15
3) boost/1.62.0 6) numactl/2.0.7 9) libsplash/1.7.0
@sbastrakov Was there already a solution to this problem?
No there is currently no solution. We are still searching which dependency of picongpu is introducing the bug
I found the issue.
Our sysadmins did build the latest libSplash on a node where a system-wide libz exists in /usr/local/. It does not exist on head nodes or regular nodes... The properly installed libSplash remembers its dependencies at install.
Whatever, I fill a bug report...
Which note was it? I also searched on different node in this lib path.
I have no idea. Afaik, they submit to some node in the default queue for compiles.
Bug is reported as no. 7504 with a long description that they never build libz for any package as I checked but always take some random syslib.
good job!
thx :)
ok, the first toolchain is fixed. Proper rebuilds still pending for the whole software stack to use a non-system zlib.
Proposed documented environment update in #2570
@ax3l @sbastrakov Can this be closed since #2570 was merged and the system was updated?
@sbastrakov works for you as well?
Yes. @PrometheusPi thanks for pointing out. Closing.