I’m sorry if this is a dumb question or if it has been asked hundreds of times already, but I can’t seem to find any documentation. I have a .cu file which I can compile into an executable file. However, I want to alter some of the generated PTX from that file, so I’m using -keep or --ptx to get that ptx file. From there, I can’t figure out how to build from the modified ptx code and reintegrate with the host code in the .cu file. I’ve found some posts mentioning the .cubin file format, and I can convert .ptx to .cubin as well, but I’m stuck there too. The best I’ve come up with so far is using CUTIL functions like cuLoadModule and cuLaunchGridAsync, but it seems to me there must be a better option than that since nvcc itself generates the PTX and doesn’t use CUTIL.