I feel like the portability problem is really like a search problem:
Let's say we have N pieces of heterogeneous hardware, and each have K possible configurations. So, to make a program portable, it means one has to search among K^N possible ways to combine them. Fortunately, we have some heuristics about the search problem. But in real world, K can be really large.
BigFish
@solovoy Maybe it's not that simple. Different heterogeneous hardware may cause you to redesign your algorithm totally to make best use of hardware resource. For your approach to work, you need to write efficient algorithm for all cases, which is impossible and too costly.
I feel like the portability problem is really like a search problem: Let's say we have N pieces of heterogeneous hardware, and each have K possible configurations. So, to make a program portable, it means one has to search among K^N possible ways to combine them. Fortunately, we have some heuristics about the search problem. But in real world, K can be really large.
@solovoy Maybe it's not that simple. Different heterogeneous hardware may cause you to redesign your algorithm totally to make best use of hardware resource. For your approach to work, you need to write efficient algorithm for all cases, which is impossible and too costly.