Quote:

I doesn't speed up a single thread at all, so if the program is completely linear and isn't broken down into modules, neither helps.

But you can write a "program" that benefits from multiple threads if it can be broken down into separate processes and there are modules that can be run concurrently.

This idea, for me, dates back to the 6809 (from there to the 68K and now PPCs).

Also, code modularity just makes sense from the perspective of system resource use and it helps if you are implementing a virtual memory system.

At the extreme, loading and unloading these modules slows things down, but it also can help maximize the utility of your available memory.

You have to remember that until we had MMUs, we were limited to 64K on the 6809 (and in some ways still were after as that was the limit for direct cpu addressing).
For our 68K systems, generally 16MB (the 68020 system could address more, but at the time memory was expensive).

The current limit (a legacy issue) of 2GB seems almost expansive in comparison.

[ Edited by Jim 02.06.2018 - 14:53 ]