Hi,
Does TI's compiler support environment variable for target devices?
For instance, export OMP_NUM_THREADS=4
#pragma omp target
#pragma omp parallel for
for(...)
Will four threads running on the DSP? I have no clue because the current compiler version does not support print out to console from other cores except core 0.
But from the performance angle, it seems not...??
If it supports parsing environment variable, for the case that I have two parallel regions running on the host and device, respectively, e.g.,
//Running on the ARM
#pragma omp parallel
{
...
}
//Running on the DSP
#pragma omp target
#pragma omp parallel
{
...
}
IF I set OMP_NUM_THREADS=4, both parallel regions will launch 4 threads, right? Does there any way to control via environment variable, a.k.a different threads? I think it is not hard to define extensions such as OMP_NUM_THREADS_HOST, OMP_NUM_THREADS_DEVICE in the omp runtime??
