Hi,
I've been processing a large number (say thousands) of arrays, each with the same, fixed, but small size (like 20 in one dimension). I believe this pertains to a scenario well-suited with 'new array execute functions' rather than the multi-threaded fftw. Still, the job was divided by arrays and sent to several threads, all of which are sharing the same one 'plan' (maintained by the master thread). I wonder if there's any bottleneck to be expected due to the overhead between the threads in accessing the 'plan'? Because I did notice the overall speed seemed to flat once >5/6 threads were used.
Hi, I've been processing a large number (say thousands) of arrays, each with the same, fixed, but small size (like 20 in one dimension). I believe this pertains to a scenario well-suited with 'new array execute functions' rather than the multi-threaded fftw. Still, the job was divided by arrays and sent to several threads, all of which are sharing the same one 'plan' (maintained by the master thread). I wonder if there's any bottleneck to be expected due to the overhead between the threads in accessing the 'plan'? Because I did notice the overall speed seemed to flat once >5/6 threads were used.
thanks, gz15028