On 02/05/2013 01:34 PM, Ralf Karrenberg wrote: > I just managed to integrate libWFV into pocl and got the first few results.
Very good! > There's one other bad thing that I just noticed: These numbers are > kernel times with pocl reusing the compilation results. If I only > measure one run after deleting the temporary files, pocl is *really* > slow (roughly 1.5-3 times slower). This suggests that the implementation > suffers a lot from using scripts, command line tools like opt, and thus > disk I/O. I'm not sure where the compilation time bottleneck really is given pocl is not yet optimized for speed. For now I've assumed the end user executes the kernel more than once to amortize its cost. I doubt it's the scripts that's causing the main slowdown. At least when pocl ends up fully replicating the kernel, the code bloat causes the code generation to take a lot of time, etc. Also, it would be good to improve the program binary format to include the final compilation results in the file to make execution when loading the program from a binary equally fast as the executions after the first kernel launch. Currently it includes only the program in the LLVM bitcode format which should not help much. > Still, the raw kernel performance looks really really good. Yes. I wonder where the performance improvement on BitonicSort for WFVOpenCL comes from. > On a side note: It was really pretty easy to integrate my stuff, I just > run a wrapper pass that invokes WFV on the kernel before all your custom > transformations start, and adjust the loop induction variable increment > of WILoops. It's currently only a hack but shouldn't be hard to make > that code depend on an environment variable or build flag. I think a variable to control the WILoops pass is the cleanest solution here. Then you can setup the increment dynamically based on the vectorization width. -- Pekka ------------------------------------------------------------------------------ Free Next-Gen Firewall Hardware Offer Buy your Sophos next-gen firewall before the end March 2013 and get the hardware for free! Learn more. http://p.sf.net/sfu/sophos-d2d-feb _______________________________________________ pocl-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/pocl-devel
