align the peer mem access check with API document

[changchengx]

In API document, it writes below:
cudaDeviceCanAccessPeer(int* canAccessPeer, int device, int peerDevice )
if device device is capable of directly accessing memory from peerDevice

[AddyLaddy]

I can't see the issue with that code? We're checking for access from the local device cudaDev for each peer i

[changchengx]

I can't see the issue with that code? We're checking for access from the local device cudaDev for each peer i

Yes. It wants to check whether the i could access to the memory allocated from cudaDev. In the API document, it checks whether the device could access to the memory allocated from peerDevice. So, I think the argument i should be assigned to the parameter device.

[kiskra-nvidia]

I'm guessing that in practice the order doesn't really matter but I agree with you @changchengx that the order you suggested is correct.