I have succeeded to get my program running on Nvidia using ComputeCpp 1.2.0 (see my comment in yesterday’s thread about the regression bug in 1.3.0).
Unfortunately, I had to comment out lines using mathematical functions from cl::sycl (cos, sin, fabs, rsqrt). The comment on the “Targeting NVIDIA PTX” seems to indicate that this is a known problem. Unfortunately, there is only a cryptic explanation about mapping OpenCl functions to their PTX counterparts. Is there some kind of a workaround that I could use to get my code running?