clBuildProgram confusion

Accelerated Computing CUDA CUDA Programming and Performance
1

I have an OpenCL program that will compile and run just fine on an nVidia GeForce GTX 1050, but won’t compile at all on a Lenovo P50 with an nVidia Quadro M2000M. I get the following error from clGetProgramBuildInfo

fcl build 1 succeeded.

error: undefined reference to `jmp()’
error: backend compiler failed build.

Obviously, I have no clue how that happened. So I thought maybe somehow my program isn’t compiling behind the scenes and there’s some issue with my code that a difference in driver details between the two GPUs makes manifest. To take a deeper dive, I thought I’d pass the parameter to clBuildProgram that as -cl-nv-verbose, and get more details. However, I get nothing.

Any ideas?

Other, simpler CL programs compile and run just fine in the same application, so I don’t believe this to be a configuration issue.

2

what are the GPU driver versions on the machine with the GTX 1050 and the machine with the M2000M ?

3

The driver on the M2000M is 388.19, which, I updated when I did the CUDA Computing Kit installation.

On the GTX, I don’t have the number handy but I also updated it to the latest CUDA Computing Kit and after dillying with an INCLUDE path everything worked. So, its something about the M2000M and driver combination that sinks me.

4

Confirming that the drivers on both are 388.19.