Closed pnehrer closed 2 months ago
Thanks for bringing this up! Looks like official get_property (and set_property
) function documented takes device_name
as a parameter, so we should probably get rid of our bindings which allow no device name to be passed.
Interestingly, openvino does have 3 non-device specific properties which can be returned successfuly w/o specifying device name (source). In the current state, if you pass either EnableMmap
, CacheDir
, or ForceTbbTerminate
, your test actually does pass.
However, since this is not documented, we should probably remove the option to not pass a device. @abrown what do you think?
Thanks for bringing this up! Looks like official get_property (and
set_property
) function documented takesdevice_name
as a parameter, so we should probably get rid of our bindings which allow no device name to be passed.
We do have Core::get_device_property
, so let's change the parameter there to be an enum that contains only the valid property keys.
Interestingly, openvino does have 3 non-device specific properties which can be returned successfuly w/o specifying device name (source). In the current state, if you pass either
EnableMmap
,CacheDir
, orForceTbbTerminate
, your test actually does pass.
How about this: let's add an enum which contains only the three variants allowed by Core::get_property
and use that type for the parameter.
[edit: @rahulchaphalkar and I talked and none of the above applies; let's wait for his PR]
Retrieving a non-device specific property from core results in a "general error". E.g., the following unit test fails:
Retrieving device-specific properties appear to be working (e.g., for CPU).