Wilco (wilco.dijkstra.delete@this.ntlworld.com) on February 24, 2021 4:37 am wrote:
> Software allows you
> to select whichever one works best, and you can tweak it further, remove bottlenecks etc.

So you need specific software for each CPU, and you need deep knowledge of each CPU, it sounds obvious to me that the best place for this software is inside the CPU.

> However with
> hardware you are stuck with the one in your CPU.

And it is reasonable to imagine that, if someone competent is tasked to implement a memcpy on that CPU that implementation will be optimal, I know, you will say it didn't happen, but we disagre about the reason becuase you keep saying that this isn't possible at all.

> In order for hardware memcpy to work out, it has to
> be as fast as the best software implementation. So far nobody has proven this is feasible.

Depends on CPU, over x86 history rep movsb performance wasn't consistent, which leads to the memcpy function growing to a few kB, but in some of the CPUs rep movsb was the best memcpy available.