On 7/7/11 15:04 , Oliver Bock wrote:
> On 4/12/11 12:36 , Oliver Bock wrote:
>> On 4/12/11 12:23 , Bernd Machenschalk wrote:
>>> I propose that the client should actually keep the "canonical" list of
>>> OpenCL devices, and a BOINC API function (e.g.
>>> boinc_get_opencl_device_id(int,...)) should be called in the application
>>> to map the --device argument to the actual OpenCL device id.
>>
>> That should solve the "how to pass the non-integer OpenCL device ID to
>> the science app" question. Now we "just" need to figure out how to get
>> to the consistent mapping across frameworks. Maybe we should post a
>> feature request for CUDA like Rom did for OpenCL. Both frameworks should
>> return a unique vendor-independent device ordinal, e.g. the PCI ID...
> 
> FYI, the CUDA 4.0 runtime API now returns three new device properties
> (pciBusID, pciDeviceID, pciDomainID) via cudaGetDeviceProperties() that
> might help...

This also possible if you use the 4.0 driver API since
CU_DEVICE_ATTRIBUTE_PCI_BUS_ID, CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID,
CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID have been added to cuDeviceGetAttribute()


Oliver


_______________________________________________
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.

Reply via email to