Does cub::DeviceRadixSort support that the input and the output is the same pointer?

[desert0616]

Hi,
I was wondering whether it is ok if I set the input and the output to the same pointer? It is usual to sort an array and output the sorted results to the original location.

Thanks.

[dumerrill]

Not yet. We will soon.

[desert0616]

Does "not yet" means any overlapped input and output used for cub device functions will get an incorrect result?

Thanks

[dumerrill]

Yeah, there is potential for incorrect results (depending on the number of
internal sorting passes, e.g., if it is an odd number, the results would
end up in the temporary buffer.)

On Sat, Oct 8, 2016 at 1:19 AM, Mo Sha notifications@github.com wrote:

Does "not yet" means any overlapped input and output used for cub device
functions will get an incorrect result?

Thanks

—
You are receiving this because you modified the open/close state.
Reply to this email directly, view it on GitHub
https://github.com/NVlabs/cub/issues/60#issuecomment-252403849, or mute
the thread
https://github.com/notifications/unsubscribe-auth/ABaFwFX54vpixDtoe0UgGOhPw3BRfNg2ks5qxyfugaJpZM4KRmzl
.

[desert0616]

Thank you so much for the reply .Dose the DoubleBuffer version api can avoid this issue?