auto vectorized and auto parallelized only interested loop.

[aketh_t_]

Hi all,

It turn out one of the loop I was interested in offloading to MIC as per optimization report suggested it was both auto vectorized and auto parallelized.

What I would want is only auto parallelize only this loop.

The -vec-report vectorised all loops and paralelized them.

This is important to for me as I would want to check percentage reduction in time taken on my application just by reducing the time taken in this loop.

How do I do this? Is there a way to do this?

[TimP]

Do you mean is it ok to set pragma no vector on a test loop?  Remember Mic performance depends strongly on vectorization.

[aketh_t_]

No,

I want vectorization and parallelisation(thread level) only for the one loop I am interested in. Default options for the rest of the application.

Thats because I want to know, how much faster my application was after optimization of only the loop I was interested in.  

[TimP]

Default options involve fairly aggressive vectorization.  You might try adding no-vec option with pragma simd on the test loop.  For selective parallelism openmp is likely answer.