Does it "emulate" cuda?

[Toomoch]

Sorry if this is a really dumb question, but does this allow to run a precompied cuda program on an amb gpu for example?

[psath]

Unfortunately not. You will need CUDA source since it performs Source-to-Source translation.

At one time there was a PTX --> CAL translator (https://ece.northeastern.edu/groups/nucar/GPGPU4/files/dominguez.pdf),
but I am not aware of any current translator that can ingest PTX or CUBIN and output either SPIR or HSAIL, as well as intercept and remap calls to the CUDA Runtime library. (If you find one though, please pass along a pointer as we'd be eager to try it out!)

[engineer1109]

Not support virtual address, cannot run float* d_p = d_q + 100;