I am currently writing some code using OpenCL on CUDA devices and came across this dilemma that generated binaries for the program using OpenCL on a cluster node with multiple T10 devices, couldn’t be used to re-create the program using clCreateProgramWithBinary(). After checking the binaries, for AMD gpus, the binaries are in ELF, while in PTX for Nvidia gpus.
Does anyone have a suggestion/solution for such problem?
Your advice is greatly appreciated! Thank you in advance.