Hi everybody, as discussed with Pekka on the LLVMdev mailing list, I'd like to give it a try to integrate our implementation of "Whole-Function Vectorization" into pocl. The library performs SIMD vectorization of an entire function given in LLVM IR, including all kinds of control flow and splitting of non-vectorizable instructions etc. It is currently integrated into our own OpenCL driver (WFVOpenCL) and can boost performance quite significantly (if you want to get a feeling for this, compare running a kernel with the latest Intel driver with and without the "vec_type_hint" attribute). However, although it is capable of running a significant number of applications, it is really more of a proof-of-concept implementation, and thus is not really robust and the API implementation is nowhere near complete. Since our main focus is on the development of the vectorization technology, I'd like to try using pocl as the basis to get more possibilities for evaluation on real world applications such as the bullet physics engine, clSURF, Luxmark, etc. For you on the other hand, I guess the benefit should not require a lot of explanation, so I hope to get some help on where to start :).
Best regards, Ralf ------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_jan _______________________________________________ pocl-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/pocl-devel
