c - Reasons for omp_set_num_threads(1) slower than no openmp -
i believe agree title of post. can point me reason ? reference book etc ? have tried find no luck.
i believe reason openmp has synchronization overhead no openmp project doesn't have.
hope can expand reason more?
thanks
while there overhead @ runtime using openmp 1 thread, more important issue code transformations compiler has perform generate openmp code (in particular outlining parallel region code separate functions [done gcc , icc; pgi different...]) affecting other code optimizations (like vectorization). information compiler has in single function allows optimizations potentially gets lost when parts of code executed in outlined functions, generated code may worse.
Comments
Post a Comment