Pekka, Thank you for the writing of the POCL designs. It is very helpful.
A few questions: - generate the workgroup function: . a total of N copies of the original kernel (work-item) function code is stashed into generating the workgroup function through either WorkitemDuplicate or WorkitemLoop. Who decides what N is? It is the workgroup dimension parameter(s) available to/after the clEnqueueNDRangeKernel() call? - address-space conversion for the workgroup function: There is code in the workgroup launcher (kernel_workgroup) function that analyze/converts among different address spaces before calling the workgroup function. Could you leverage a bit more on what is happening here? In case the host and the device have separated memory space, how is the address mapping actually done? (it is being heavily discussed over the LLVM-Dev list) - Workgroup auto vectorization: For current LLVM-0.8 release, it mentions that it is using an adopted LLVM LoopVectorizer. What new/improved feature(s) have been added so far? Thank you very much Chuck On 8/9/2013 8:04 AM, Pekka Jääskeläinen wrote: > Hi Chuck Zhao, > > On 07/25/2013 08:45 PM, Chuck Zhao wrote: >> I am asking the list if there is any written document describing the >> kernel compiler? > > I now added some preliminary documentation for the kernel compiler: > http://pocl.sourceforge.net/docs/html/design.html > > I hope this helps a bit. I can add more later for the parts causing > the most confusion. > ------------------------------------------------------------------------------ Get 100% visibility into Java/.NET code with AppDynamics Lite! It's a free troubleshooting tool designed for production. Get down to code-level detail for bottlenecks, with <2% overhead. Download for free and get started troubleshooting in minutes. http://pubads.g.doubleclick.net/gampad/clk?id=48897031&iu=/4140/ostg.clktrk _______________________________________________ pocl-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/pocl-devel
