OMPi - Processing a large/long vector of floats
Posted: Thu Sep 21, 2017 7:06 pmHi
I have struggled with processing a large(ish) vector of single precision floats (~ 6MB) on the Epiphany, using OMPi. Since the data is way too big for the local (32kb)
memory of a single epiphany core, I need a way to efficiently pipe this vector to any number of epiphany cores (1-16) in manageable chunks. Then each core can do
a relatively simple calculation, a moving average, and write the result vector back to shared RAM. Each core should do the moving average calculation
with a different window length parameter, so that I can calculate up to 16 different of these averages simultaneously.
What I have failed to do is piping the input vector and the 16 ouput vectors efficiently back and forth between the epiphany and the host. I guess my C programming
and understanding of memory address space is a bit rusty :!
I would really appreciate if anyone here was able to give me an indication on how to do this best
Cheers
S.
I have struggled with processing a large(ish) vector of single precision floats (~ 6MB) on the Epiphany, using OMPi. Since the data is way too big for the local (32kb)
memory of a single epiphany core, I need a way to efficiently pipe this vector to any number of epiphany cores (1-16) in manageable chunks. Then each core can do
a relatively simple calculation, a moving average, and write the result vector back to shared RAM. Each core should do the moving average calculation
with a different window length parameter, so that I can calculate up to 16 different of these averages simultaneously.
What I have failed to do is piping the input vector and the 16 ouput vectors efficiently back and forth between the epiphany and the host. I guess my C programming
and understanding of memory address space is a bit rusty :!
I would really appreciate if anyone here was able to give me an indication on how to do this best
Cheers
S.