opencv 2.4.8_2 doesn't complete installation -DWITH_CUDA=ON on OS X 10.9.2

[ivano.ras@…]

I'm running MacPorts on a retina MacBookPro with GeForce GT 650M. I've amended the portfile to compile cuda code. All seems to be fine until after half an hour building I get

--->  Building opencv
Error: org.macports.build for port opencv returned: command execution failed
Please see the log file for port opencv for details:
    /opt/local/var/macports/logs/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_graphics_opencv/opencv/main.log
Error: Unable to upgrade port: 1

Log file attached. Thanks.

[ivano.ras@…]

Replying to ivano.ras@…:

I'm running MacPorts on a retina MacBookPro with GeForce GT 650M. I've amended the portfile to compile cuda code. All seems to be fine until after half an hour building I get ---> Building opencv Error: org.macports.build for port opencv returned: command execution failed Please see the log file for port opencv for details:

/opt/local/var/macports/logs/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_graphics_opencv/opencv/main.log

Error: Unable to upgrade port: 1

Log file attached. Thanks.

I've the CUDA Driver Version: 6.0.26 installed, by the way.

[ivano.ras@…]

I've tested it against the normal CUDA Driver Version 5.5.28 and it doesn't work either.

[ivano.ras@…]

And, I've now tested it against the latest CUDA Driver Version 5.5.47. The installation doesn't even start here.

[mf2k (Frank Schima)]

The Priority field is for use by Macports team members' use only. In the future, please use WikiFormatting and Cc the port maintainers (port info --maintainers opencv).

[stromnov (Andrey Stromnov)]

AFAIK, it's a long lasting problem with CUDA binaries and CUDA compiler (ptcc).

You can try to compile with configure.compiler=macports-gcc-4.8 or configure.compiler=macports-clang-3.3 arguments.

[ivano.ras@…]

Many thanks stromnov I'll give that a shot. Regarding the lasting problem, I have to disagree because it was working fine 4 months ago in OS X Mountain Lion and this unfixed problem is really hindering Mac users from using CUDA code... Cheers. Ivano

[stromnov (Andrey Stromnov)]

It works perfectly when you using right compiler toolchain (gcc|cl + nvcc).

Macports use clang by default, nvcc use gcc (​http://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/). So clang++ and nvcc potentially can produce incompatible object files.

[ivano.ras@…]

Yes, I'm well aware of the clang/gcc discrepancy. It's just that if I have compile things myself, what's the point of MacPorts anyway? I'll follow your suggestions though. Cheers. Ivano

[mascguy (Christopher Nielsen)]

Given that we don't support opencv 2.x, can this be closed?

[mascguy (Christopher Nielsen)]

This ticket is no longer relevant, as we don't have an OpenCV 2 port. Can someone close?

[mf2k (Frank Schima)]

Let's call this a duplicate of #41333.