On modern x86 (e.g. netburst ) INC may be slower than ADD - because it doesn't modify carry flag, and if the next op is some ( not any ) conditional branch, cpu will flush the pipeline (so called partial flags stall)
Actually any conclusions you draw about C++ language structure based on the assembly (machine code) that you end up with apply only for the specific platform/compiler combination.
Any other compiler may produce different machine code (especially if it has a different optimiser) and any other platform (processor) will have a different instruction set with different characteristics .
If the performance difference between ++i and i++ matters for a project, I am guessing that anyone even remotely competent is going to run a few tests to measure it. It's true that ++i tends to generally be faster on x86 based systems, but you can't really assume that unless you test it on every single x86 CPU out there.
With that said, using either operators will have no notable impact unless all of this is true:
1) 99% or more CPU usage. Not 60%, 80% or 90%.
2) increment is done in a performance critical section multiple times
3) increment cannot be optimized or the compiler will not optimize it
Comment