I'm sure Philippe wanted to send this to viennacl-devel ;-)
-------- Original Message -------- Subject: Re: [ViennaCL-devel] How to use multiple cores/CPUs Date: Sun, 15 Dec 2013 16:58:14 +0800 From: Philippe Tillet <phil.til...@gmail.com> To: Karl Rupp <r...@iue.tuwien.ac.at> Hello, This is still experimental, but you can pull the feature in the branch "feature-blas_linking", Albert. The tests all pass so it should be stable enough for basic usages. You just need to define VIENNACL_WITH_CBLAS, and link your program with openblas. On my laptop, I observed 300x speed-ups on GEMM. I'm running on a core i5 2410m @ 2.3GHz, so the peak is : 2.3 (Freq) * 2 (Cores) * 8 (AVX) * 2 (Mul-add) = 72GFlop/s,and ViennaCL with OpenBlas linking gives me 69.5Gflop/s . Pretty impressive. (Comparatively, the basic single-threaded or strided implementation leads to 0.2Gflop/s) Best regards, Philippe 2013/12/14 Karl Rupp <r...@iue.tuwien.ac.at <mailto:r...@iue.tuwien.ac.at>> Hi, > It seems like there is another big problem with the use of SSE, it cannot account for strides and offset. So probably it should only be used when stride=1 and start=0. Since we would provide blas-linking, I don't know if checking the strides and the starts accordingly is worth the code bloat it would induce... Strictly speaking, it also requires aligned allocations (which we don't check for), internal row/column length being a multiple of 2 or 4 (which we usually have, but not always), and sizes being multiples of 2 or 4. Lots of constraints. The code still has some value, so I don't want to throw it away. Instead, I'm hoping for the day where somebody gets motivated extending and maintaining this. :-) However, I think that at the moment there are more urgent things to do, e.g. linking with other BLAS libraries. Best regards, Karli ------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk _______________________________________________ ViennaCL-devel mailing list ViennaCL-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/viennacl-devel
