Your "main" part has no end.
End it with
loop: sjmp loop
or something more elegant. Now the CPU simply runs on like a "chicken without a head", as we say in the Netherlands.

Try it on a real chip, don't believe simulators too much. They often have problems simulating I/O operations. I recommend using simulators only if you want to test program flow.

Why do you trust the flashing LED routine so much you left it out? Maybe the problem is there!