Implementing and evaluating an heterogeneous, scalable, tridiagonal linear system solver with OpenCL to target FPGAs, GPUs, and CPUs