As far as I know what you see when asking NVCC to keep all intermediate files is PTX code. This is not the code the device executes.

I haven't heard of anyone who reverse engineered the toolchain but there's an awesome tool that helps you reverse engineer your own binaries: http://www.cs.rug.nl/~wladimir/decuda/ This is relevant because the compiler creates device specific binaries that you can't get the assembler code for. So if you want to know exactly what your kernel is doing disassemble it with decuda. Unfortunately the tool is a bit outdated but it still might be useful to some.