[Rtk-users] CUDA displaced detector filter failure

[Chao Wu]

Hi all,

I am trying the new cuda displaced detector filter but
got cudaErrorLaunchFailure. This error was returned by cudaFree(dev_geom)
at the end of the code, but since this error type is not registered with
cudaFree(), it would be from previous, asynchronous launches. By
using cudaDeviceSynchronize() I found the error coming after calling
kernel_displaced_weight<<<...>>>(...).

One interesting thing is that when the error occurs depends on the number
of projections processed at once (controlled by the --subset=N option of
rtkfdk.exe for example). If I use subset=1 then it fails right with the
first projection; if subset=2 it fails at the 16th run when processing the
31st and 32nd projections; if subset=4, at the 14th run; if subset=16
(default), at the 6th run; and so on.

The input region of each projection is about 1944x200 (the 2nd dimension
differs a little bit of course) and the output region is then about
3888x200. A GTX580 card is used.

Any suggestions and comments on this? Thanks.

Regards,
Chao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.creatis.insa-lyon.fr/pipermail/rtk-users/attachments/20141010/f898f5fb/attachment.htm>