News for hackers. Stuff that matters.

Auto-parallelizer: #pragma loop

I just discovered that MSVC++ can automatically run each iteration of a loop in parallel in separate threads. Add "#pragma loop(hint_parallel(8))" and "#pragma loop(ivdep)" before the loop, and compile with the /Qpar option. This simple change sped up my cryoablation simulation code by 4x.
( 3 comments | 2 points | on HN )
Re:Auto-parallelizer: #pragma loop by boas on Wednesday April 15, 2015 @03:58PM (#9383640)
A lot of the radiology software that I write involves iterating over a 3D grid. The auto-parallelizer lets you easily process each 2D slice on a different processor core. My laptop has 2 cores x 2 hyperthreads per core = 4 virtual cores, so 4x speedup.
Re:Auto-parallelizer: #pragma loop by dalke on Wednesday April 15, 2015 @03:57PM (#9383633)
You might also be interested in OpenMP, which does similar things for more compilers than MSVC++.
Re:Auto-parallelizer: #pragma loop by boas on Wednesday April 15, 2015 @04:18PM (#9383790)
Thanks! I'll look into this. I also use g++ on Linux, so a cross-platform solution would be nice.