by passing the opencl id as a cmdline arg
On 14-Mar-2011 5:10 PM, Richard Haselgrove wrote: > And how is BOINC going to signal to an application > > "When you start up, use device #CL_DEVICE_VENDOR_ID (which is, by the way, a > cl_uint, not the hardware device # passed by legacy versions of BOINC)" ? > > From: "David Anderson" <[email protected]> > > >> clGetDeviceInfo() returns the following; >> I'm hoping this is the same ID as used by the vendor-specific APIs. >> -- David >> >> CL_DEVICE_VENDOR_ID Return type: cl_uint >> A unique device vendor identifier. An example of a unique device identifier >> could be the PCIe ID. >> >> On 14-Mar-2011 4:30 PM, Richard Haselgrove wrote: >>> I'm still wondering who is responsible for establishing the mapping between >>> logical and physical device numbering, when the hardware and software >>> enumerators return different answers. >>> >>>> On 14-Mar-2011 2:37 AM, Richard Haselgrove wrote: >>>>> >>>>> The issue is not really one of plan_classes, but of whether the construct >>>>> >>>>> <coproc> >>>>> <type>OpenCL</type> >>>>> </coproc> >>>>> >>>>> is recognised by the client: no, it isn't, according to coproc_detect.cpp >>>> >>>> Nor should it be; OpenCL is a language system, not a GPU type. >>>> >>>> -- David >>>> _______________________________________________ >>>> boinc_dev mailing list >>>> [email protected] >>>> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev >>>> To unsubscribe, visit the above URL and >>>> (near bottom of page) enter your email address. >>>> >>> >>> >> > > _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
