In article ,
Tim Streater writes:
In article , Huge
wrote:

On 2018-03-11, Vir Campestris wrote:
On 10/03/2018 09:59, Huge wrote:
On 2018-03-10, Andrew Gabriel wrote:
Back in the days of X.25 packet switches, I wrote some carefully
crafted assembly code which could switch a packet in 127 instructions.
It allowed for the computer's pipeline, performing base register loads
well enough in advance that there was no stall in the indirect load,
generally interleaving two separate functions in alternate instructions.

"You are not expected to understand this."

Why not?

[Whoosh]

https://thenewstack.io/not-expected-...and-explainer/

Next you'll be saying you don't understand Meltdown and Spectre :P

Except I do.

The comment is ********, of course. And I assume the 127 instructions
for switching a packet will still work even if the internal workings of
the switch changes. E.g. if you are told to buy a more recent, cheaper
model of the switch that has no pipeline and a different memory
architecture.

It was optimised for what was then (1980's) our top range mini-
computer which had a 4-instruction pipeline. It also run on the
slower/cheaper 2- and 0-instruction pipeline systems, but they
were slower for many reasons besides just instruction pipeline
(things like lack of multi-ported memory, and less memory, and
narrower memory bus).
These all ran the same 127 instructions, but only the top range
system gained from specifically hand-crafting those instructions.

--
Andrew Gabriel
[email address is not usable -- followup in the newsgroup]