What do you want to do?

1. peak calling: unless the read numbers are not too different, the peak caller will normalize input and IP track. manual downsampling is also an option
2. generation of bigwig tracks to visualize: read count normalization to RPM, optionally extend reads in read direction by fragment length (MACS can do this for you)
3. there are other normalization methods to generate more "even" tracks that take into account GC content or similar, but they are not in widespread use.
4. Quantitative differences e.g. diffbind have their own normalization methods.

.... I guess there are other use cases and then you would do something else entirely