I don't follow. If we had the saturated ADD instruction supported in hardware we wouldn't have to deal with saturation logic in software. Compiler would simply

I don't follow. If we had the saturated ADD instruction supported in hardware we wouldn't have to deal with saturation logic in software. Compiler would simply be able to emit SADD instruction whenever we asked for it to and would be given a chance to spit out more optimized code because that would essentially be a branchless code. Reality though is that we don't have such hardware and we have to deal with it in software, so, instead of a single instruction emitted there will be a bunch of them, and of which there will be certainly branches involved.