superlu_dist
superlu_dist copied to clipboard
xiaoyeli
About OMP_NUM_THREADS？
When i set 'unset OMP_NUM_THREADS', running pddrive example with big.rua, FACTOR time is 13s, Mflops value is 0.67.
But set 'export OMP_NUM_THREADS=1', in the same way, FACTOR time is 0.005s, Mflops value is 1780.63.
What could be the reason？ It's not necessarily a problem.I don't know, nobody else has had the same problem.
Before you set any environment variable related to OpenMP, can you do this: $ echo $OMP_NUM_THREADS ? I suspect it's a large number.
Before you set any environment variable related to OpenMP, can you do this: $ echo $OMP_NUM_THREADS ? I suspect it's a large number.
This is cpu info:
When i run 'export OMP_NUM_THREADS=62 && ./pddrive big.rua', the FACTOR time is 10.5s, and Mflops is 0.8. Run 'export OMP_NUM_THREADS=60 && ./pddrive big.rua', the FACTOR time is 0.137s, and Mflops is 61.01.
Before you set any environment variable related to OpenMP, can you do this: $ echo $OMP_NUM_THREADS ? I suspect it's a large number.
Compare data with pardiso:
Before you set any environment variable related to OpenMP, can you do this: $ echo $OMP_NUM_THREADS ? I suspect it's a large number.
Currently, we use superlu_dist-8.1.0.tar.gz package. Could it have been fixed in last version, i will try v8.1.2 tags?
Are you able to resolve the issue? It is very abnormal that when you increase OMP threads from 1 to 2, the runtime is more than doubled.