[Starpu-devel] Issue with lws

[Mathieu Faverge <mathieu.faverge@inria.fr>]

Hello,

Just to have the info somewhere. I think there is an issue with the work 
stealing in the lws scheduler.
I ran PaStiX more than 100 times in a row on top of dmdas and I had no 
error. On top of lws, I have undefined behavior with approximately 1 run 
over 5 failing depending on the size of the pb.

When I'm lucky, I got this assert:
/cm/shared/dev/apps/runtime/starpu/1.2.3/gcc@6.3.0-openmpi@3.0.0-hwloc@1.11.9/lib/libstarpu-1.2.so.3(_starpu_release_data_enforce_sequential_consistency+0x235)[0x7f70979983f5]
/cm/shared/dev/apps/runtime/starpu/1.2.3/gcc@6.3.0-openmpi@3.0.0-hwloc@1.11.9/lib/libstarpu-1.2.so.3(_starpu_release_task_enforce_sequential_consistency+0x83)[0x7f7097998583]
/cm/shared/dev/apps/runtime/starpu/1.2.3/gcc@6.3.0-openmpi@3.0.0-hwloc@1.11.9/lib/libstarpu-1.2.so.3(_starpu_handle_job_termination+0x169)[0x7f7097988a29]
/cm/shared/dev/apps/runtime/starpu/1.2.3/gcc@6.3.0-openmpi@3.0.0-hwloc@1.11.9/lib/libstarpu-1.2.so.3(_starpu_cpu_driver_run_once+0x160)[0x7f7097a00bd0]
/cm/shared/dev/apps/runtime/starpu/1.2.3/gcc@6.3.0-openmpi@3.0.0-hwloc@1.11.9/lib/libstarpu-1.2.so.3(_starpu_cpu_worker+0x3d)[0x7f7097a014cd]
/lib64/libpthread.so.0(+0x7e25)[0x7f709be01e25]
/lib64/libc.so.6(clone+0x6d)[0x7f709769934d]
simple: core/dependencies/implicit_data_deps.c:446: 
_starpu_release_data_enforce_sequential_consistency: Assertion 
`task_dependency_slot->task == task' failed.
Abandon

Otherwise, I have segfault in other parts of the code, or deadlock. If 
someone can point me out where to look, or better give me the solution :)

Mathieu






--
--
Mathieu Faverge
Maitre de conférence / Associate Professor
Institut Polytechnique de Bordeaux - ENSEIRB-Matmeca
INRIA Bordeaux - Sud-Ouest, HiePACS Team
200 avenue de la vielle tour
33405 Talence Cedex
Phone: (+33) 5 24 57 40 73