Is malloc() correct in stream.c?

[greensquirrel]

Dear NUMA-STREAM developers!

If I understand it correctly, the latest stream.c @ e5aa9ca should be NUMA aware.
The OPENMP threads are spread across the CPUs but all these threads seem to do their work
on the same memory chunks that have been allocated via malloc(). Therefore, I think that
stream.c is not completely NUMA aware. I expect NUMA aware memory allocations like e.g.
numa_alloc_onnode().

Is stream-pthread the recommended NUMA aware benchmark? At the first glance, it allocates memory for each thread individually via numa_alloc_onnnode() and the code seems to be
be NUMA aware.

Which version of stream have you used in your paper "Measuring NUMA effects with the STREAM benchmark"?

Thank you in advance!