Re: [ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-29 Thread Philippe Tillet
Hi, Just to let you know that the problem was indeed fixed using: opencl/kernels/vector.hpp::avbv: opencl_source(template,statements, binding_policy) in opencl/vector_operations.hpp::avbv enqueue(template, program, statements, binding_policy) This is not the safest thing ever, since template, s

Re: [ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-28 Thread Philippe Tillet
Hey, There is no such generator objects. Actually, the generation of the source for each kernel is independent. The interface of the generator is now much clearer and really only requires generate_opencl_source(template, list of statements corresponding to the template, throws an exception if the

Re: [ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-28 Thread Karl Rupp
Hey, > There is one program of around 1000 kernels. For comparison, the current > vector operations program take 2seconds to compile on my laptop. Since > the 1.6seconds are including all the flip, reciprocal, but not the x = > a*x + b*y -like operations, I suspect that this would be better in eve

Re: [ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-28 Thread Philippe Tillet
Hi, 2014-05-28 10:25 GMT+02:00 Karl Rupp : > Hi, > > > > An additional information. Now that the test passes, i can safely > >> compare the JIT compilation times. >> Without the optimized kernels : 1.6seconds >> With the optimized kernels : 6.5 seconds >> I fear that this might get untractable w

Re: [ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-28 Thread Karl Rupp
Hi, > An additional information. Now that the test passes, i can safely > compare the JIT compilation times. > Without the optimized kernels : 1.6seconds > With the optimized kernels : 6.5 seconds > I fear that this might get untractable when further operations are > added. I should probably find

Re: [ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-27 Thread Philippe Tillet
An additional information. Now that the test passes, i can safely compare the JIT compilation times. Without the optimized kernels : 1.6seconds With the optimized kernels : 6.5 seconds I fear that this might get untractable when further operations are added. I should probably find a way to disable

[ViennaCL-devel] Influence of the number of kernels on the behavior of a program

2014-05-27 Thread Philippe Tillet
Hello, The integration of the kernel generator has been a nightmare! Anyway, I've realized that thousands of kernels per scalartype are required, in order to obtain optimal performance. Why so much? - flip_a, reciprocal_a, flip_b, reciprocal_b requiring their own kernel - The generator interprets