[Info-vax] Assembly languages, was: Re: OT: PDP-11 history in arstechnica

[Johnny Billquist]

On 2022-03-19 02:21, gah4 wrote:
> On Friday, March 18, 2022 at 12:22:10 PM UTC-7, Simon Clubley wrote:
> 
> (snip)
> 
>> Yes, but that doesn't change my point that with every architecture
>> I know (both DEC and non-DEC) that uses a MOV/MOVE mnemonic variant,
>> then MOV/MOVE is actually a copy to destination instead as the source
>> is not destroyed during the copy.
> 
> In the case of overlapping operands, the source might get,
> at least partially, destroyed.

That is just because the source would then also be the destination. And 
it's being destroyed for being destination, so that don't count.

> In the days of magnetic core memory, which has a destructive read
> cycle, one could have designed processors that did clear the source.

The PDP-8 do. But then again, the instruction isn't called MOVE either. 
And it only applies to the accumulator, which isn't in core.

But machines like the PDP-11 actually do take advantage of the 
destructive behavior of core memory. Instructions which modifies the 
destination will not refresh the old data back after it have been read 
out, since a new value will be written soon anyway. (Think stuff like 
ADD...)
If someone knows Unibus transactions, that's what the DATIP transaction 
is there for.

   Johnny