Compiling error with GPU

[qazmichaelgw]

I use the kinetic-devel branch. When I compile it in Ubuntu 16.04, there are some compiling errors with respect to CUDA in the autorally_control package.

I got the following error:

/home/gaowei/catkin_ws/src/autorally/autorally_control/include/autorally_control/path_integral/generalized_linear.cut(98): error: a pointer to a bound function may only be used to call the function
/home/gaowei/catkin_ws/src/autorally/autorally_control/include/autorally_control/path_integral/neural_net_model.cut(79): error: a pointer to a bound function may only be used to call the function
/home/gaowei/catkin_ws/src/autorally/autorally_control/include/autorally_control/path_integral/neural_net_model.cut(80): error: a pointer to a bound function may only be used to call the function
/home/gaowei/catkin_ws/src/autorally/autorally_control/include/autorally_control/path_integral/generalized_linear.cut(98): error: a pointer to a bound function may only be used to call the function

Does anyone have any hint on this? Thanks a lot!

[gradyrw]

This is actually just a normal c++ error, not a CUDA error. It looks like the cnpy api (https://github.com/rogersce/cnpy), which is used to load python numpy arrays, changed recently. You're likely using the new version where "data" is now a member function instead of a data member. This is what's causing the error.

If you change: double foo = reinterpret_cast<double>(foo_raw.data);

to instead be: double* foo = foo_raw.data();

in generalized_linear and neural_net_model then it should fix the problem. We'll need to figure out a way of preventing this type of problem in the future.

[qazmichaelgw]

Thanks a lot. I got it work by changing it to:

double* theta_ptr = theta_raw.data<double>();

[JasonGibson274]

Closing issue since no longer relevant with the write that happened in between then and now