OpenBLAS multi-thread using on NVIDIA tegra k1

[alex-xia-xia]

The program called openblas_set_num_threads(4) before calling dgemm function. Meanwhile, I use top command to monitor CPU usage, I found only one core is running.

Otherwise, I call openblas_set_num_threads(1), the time of DGEMM elapsed is equal to openblas_set_num_threads(4);

[martin-frbg]

What is the size of the matrix ? There has been some work lately to avoid creating multiple threads when the dimensions are so small that the overhead from creating multiple threads would outweigh the benefits of parallelization.

[alex-xia-xia]

matrix size: 200 If the program link dynamic library, only one core is running.

When the program link static library, multi-threads is working. For example, after called openblas_set_num_threads(4), I use top to monitor CPU usage, four cores are running.

[joeking11829]

I have the same issue on TK1.

How to use OpenBLAS as a static library..? BTW, I use dlib with OpenBLAS....

[dtmoodie]

I'm having the same issue without the possibility of statically linking to openblas. Has anyone found a solution to link dynamically?_