Is it correct?

[kznsq]

Hello!
I think this straightforward port to parallel implementation is not correct.
Because (x, y) patch must be computed from (x+dir, y+dir) patches that already processed.
But in your code I cannot find any synchronizations.

Thanks,
Andrey.

[abiusx]

Hi, Yes, that is intended. The kernels run in order, so usually x+dir and y+dir are computed before x and y, especially since the kernels do not have any synchronization and are not complicated. Some kernels definitely do fall outside of order, but that is not vital to the output of the algorithm. Regards -A

…

On Dec 28, 2016, at 7:56 AM, kznsq ***@***.***> wrote: Hello! I think this straightforward port to parallel implementation is not correct. Because (x, y) patch must be computed from (x+dir, y+dir) patches that already processed. But in your code I cannot find any synchronizations. Thanks, Andrey. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABVjW5PYgIDbvdMSPkqoUHtbiKDkKCPIks5rMlyWgaJpZM4LW5pq>.