BEAST 1.10.4

[Michael Charleston]

Hi all,

Beast has suddenly.... stopped.  I'm running Beast 1 on Linux Ubuntu 20.04 LTS with CUDA and it got to some number of states; computer suspended which may have caused a problem -- but then it's just Seg faulted at

6516000        -46247.6294         -486.7808           -45760.8486         2019.99             9.80159E-4          2.08768             0.4 hours/million states
6517000        -46264.3684         -494.5766           -45769.7918         2019.95             1.07787E-3          2.14288             0.4 hours/million states
6518000        -46271.4236         -492.3047           -45779.1189         2019.97             1.10133E-3          2.51739             0.4 hours/million states
6519000        -46271.7336         -489.6257           -45782.1079         2019.96             1.01288E-3          2.25098             0.4 hours/million states
CUDA error: "Launch failed" (719) from file <GPUInterfaceCUDA.cpp>, line 503.
CUDA error: "Launch failed" (719) from file <GPUInterfaceCUDA.cpp>, line 503.
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x00007fecd45fd9f6, pid=261188, tid=261235
#
# JRE version: Java(TM) SE Runtime Environment (14.0.2+12) (build 14.0.2+12-46)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (14.0.2+12-46, mixed mode, sharing, tiered, compressed oops, g1 gc, linux-amd64)
# Problematic frame:

>

(no, there's nothing printed after "Problematic frame:")

.. Since which running the same command again (with -overwrite)

>beast -beagle_GPU Data-Aligned.xml -overwrite

gives me

OpenCL error: Unknown error from file <GPUInterfaceOpenCL.cpp>, line 115.

I'm not using OpenCL at all -- and AFAIK the beagle-opencl libs are not on the path anyway..

It's weird because this keeps happening: beast 1 runs nicely and then I try to run it again in the same terminal on a different data set, in a different directory, and it gives me this. OpenCL error message.  Even with the command

>beast -version

OpenCL error: Unknown error from file <GPUInterfaceOpenCL.cpp>, line 115.

Any ideas? It's very odd!

[Marc Suchard]

Hi Mike,

I have no great ideas here, as the initial error seems like OS-level trouble.     

On some systems BEAGLE's OpenCL dynamic library gets loaded before the CUDA one and that can cause issues when there's no `libopencl.so` on your library search path.   Since you are not using OpenCL, I'd highly recommend removing `libhmsbeagle-opencl.*.so` from your library search path (probably in /usr/local/lib).  

best, Marc

[Michael Charleston]

Thanks Marc,

I've come to the same conclusions: while I haven't got to the bottom of why it just *stopped*, I'm persevering with optimism -- so far today running bash it's going fine, and I've hunted down all the libhmsbeagle-opencl culprits...  Fingers crossed!

Cheers

Mike