GPU-ready (with free driver) buildd/CI/porterbox?

2019-11-18 Thread Mo Zhou
Hi folks, Just a bold thought. Two Debian developers intended to package basic libraries of the AMD ROCm (freesoftware-licensed) [1], but both of them don't have access to such hardware. Given that there are also a number of packages in debian with OpenCL enabled, I'm wondering wether a shared b

Re: GPU-ready (with free driver) buildd/CI/porterbox?

2019-11-19 Thread Sam Hartman
> "Mo" == Mo Zhou writes: Mo> Hi folks, Just a bold thought. Mo> Two Debian developers intended to package basic libraries of the Mo> AMD ROCm (freesoftware-licensed) [1], but both of them don't Mo> have access to such hardware. Debian has from time to time funded hardware

Re: GPU-ready (with free driver) buildd/CI/porterbox?

2019-11-19 Thread Sam Hartman
> "PICCA" == PICCA Frederic-Emmanuel > writes: PICCA> Hello >> Debian has from time to time funded hardware for people doing >> important work. I'd definitely be happy to receive a >> reimbursement request for such hardware from Debian developrs. >> For non DDs, I wo

Re: GPU-ready (with free driver) buildd/CI/porterbox?

2019-11-19 Thread Christian Kastner
On 19.11.19 05:59, Mo Zhou wrote: > Given that there are also a number of packages in debian with OpenCL > enabled, I'm wondering wether a shared buildd/CI/porterbox with GPU[2] > could be helpful? I would also be interested in this.

Re: GPU-ready (with free driver) buildd/CI/porterbox?

2019-11-19 Thread Simon Richter
Hi, On Tue, Nov 19, 2019 at 09:21:16AM -0500, Sam Hartman wrote: > I'm hoping there's no need for a GPU on a buildd. > That would mean that the software required an active GPU to *build*, > which seems problematic on a number of fronts. As far as I've understood, that makes sense for some neura

Re: GPU-ready (with free driver) buildd/CI/porterbox?

2019-11-19 Thread Simon Richter
Hi, On Tue, Nov 19, 2019 at 02:48:48PM +, PICCA Frederic-Emmanuel wrote: > It would be nice also to be able to test the OpenCL icd implementations and > work with real hardware. That is mostly upstream's job -- ICD packagers should just verify that the package still runs "Hello World" on th