gdalwarp & OpenCL, linux and windows, nvidia: I'm doing something wrong. what it is?

المشرف العام

Administrator
طاقم الإدارة
gdalwarp & OpenCL, linux and windows, nvidia: I'm doing something wrong. what it is?

after installing - linux (ubuntu 15.04): the nvidia propretary drivers and the related OpenCL packages , plus the "build-dep" set for the gdal package - windows 7 : the nvidia cuda kit in its entirety, having visual studio 2010 already installed, having already compiled the proj.4 dll

I've downloaded the current gdal sources (2.0), and built it both with and without OpenCL support (linux: the configure switch; windows: editing alg/makefile.vc to add -DHAVE_OPENCL)

the resulting binaries differ as expected, and actually work.

the problem is that regardless if I launch gdalwarp on a 35MB geotiff .... -wo "USE_OPENCL=TRUE" (-r lanczos) or FALSE, the execution time stays exactly the same, so I suspect that opencl, or the Nvidia version of OpenCL, doesn't actually kick in.

Unfortunately, afaik, I've no way to monitor gpu usage (mid to low-end cards, not teslas; nvidia-smi under linux says explicitly "not supported" when I query for gpu usage).

what I'm not doing correctly here?

How can I raise enough the verbosity to check if the Nvidia OpenCL is actually in use at runtime or not?

Thank you in advance



أكثر...
 
أعلى