On Thu, Nov 07, 2013 at 11:48:27PM +0200, Pekka Jääskeläinen wrote: > Thanks Peter, > > You are probably right that the proprietary libcuda device might > be waste of time. I tried to figure out the licensing terms of > cuda.h and it seems creepy to a non-lawyer, so better not risk it. > > Better invest the time on the Gallium Compute/AMD backend instead to go > towards a proper open source heterogeneous compute stack on a single > OpenCL platform. > > This is the original patch from Tom: > http://people.freedesktop.org/~tstellar/pocl/pocl-gallium-wip.patch > > Looking into it: > - the address space trickery is not needed anymore, there's a > pass (albeit experimental) that fixes the "pocl internal" > address spaces (fake-address-space-map) to the target's. > - built-in lib should be probably reused from libclc, no point > replicating them in pocl tree; detect if it it's installed in > configure and use it as a prerequisite for enabling the > GC/AMD device > - device API to ask for the kernel compiler passes to use (to > e.g. skip WG func generation for SPMD targets), this should be done > after the API calls can replace pocl-workgroup to avoid hassles > with the sh script > > Aside these the remaining code calls the GC APIs from the pocl > device layer. Tom, what is the status of the GC APIs? > Is it now possible for me to install the latest Mesa and a > supported AMD GPU and start experimenting with this? >
I haven't touched the GC APIs since I wrote that patch. I don't have any plans to work on it right now, because almost all the work that needs to be done for the AMD backends is driver specific. One thing I've discussed with some developers from Intel is moving Clover's OpenCL code into core Mesa, which would make it possible for an external OpenCL implementation, like Beignet to be moved into the Mesa tree. This would make it possible to port Pocl's backends to Mesa. Do you have any thoughts about this? -Tom > > On 11/07/2013 10:04 PM, Peter Colberg wrote: > > Dear POCL developers, > > > > I hereby release into the public domain all contributions that I have > > sent to the POCL mailing list up to this point. This includes the > > experimental patches for a libcuda backend. > > > > Regards, > > Peter > > > > ------------------------------------------------------------------------------ > > November Webinars for C, C++, Fortran Developers > > Accelerate application performance with scalable programming models. Explore > > techniques for threading, error checking, porting, and tuning. Get the most > > from the latest Intel processors and coprocessors. See abstracts and > > register > > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > > _______________________________________________ > > pocl-devel mailing list > > [email protected] > > https://lists.sourceforge.net/lists/listinfo/pocl-devel > > > > > -- > --Pekka ------------------------------------------------------------------------------ November Webinars for C, C++, Fortran Developers Accelerate application performance with scalable programming models. Explore techniques for threading, error checking, porting, and tuning. Get the most from the latest Intel processors and coprocessors. See abstracts and register http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk _______________________________________________ pocl-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/pocl-devel
