[millstone]

What is a "concurrent program?" If concurrency is about program structure, does that structure have to be explicit? If I rewrite a loop to support OOE is it thereby concurrent?

A uniprocessor may be pipelined, superscalar, support SIMD, etc which would seem to satisfy the definition of parallelism. So isn't parallelism achievable on a uniprocessor?

The more I think about this stuff the less sense the distinction makes.

[elteto]

I would define a concurrent program as one with multiple independent flows of execution. Whether they are scheduled on multiple CPUs at the same time or just one is what makes them parallel or not.

If you look at a what happens inside a CPU then you probably could say a uniprocessor has some degree of parallelism, but this is not a common point of view AFAIK, it is too low level. Also, the execution contexts inside a pipelined CPU are not entirely independent, you can have pipeline stalls due to dependencies. This is not the case with, say, two application threads running independent tasks.

[gpderetta]

> So isn't parallelism achievable on a uniprocessor?

Of course! Normally unqualified parallelism refers to multiprocessor parallelism, but instruction level parallelism is also a thing. So is overlapping io or memory access with computation. Even memory accesses themselves can be parallelized (caches are multi ported, CPUs allow multiple outstanding cache misses), or IO (multiple disks or network cards).

[agumonkey]

From what I recall, Quake existed because of instruction level parallelism between apu and fpu. This indeed has importance.