[Info-vax] Reading Gordon Bell's VAX strategy document

[gah4]

On Sunday, September 24, 2023 at 11:48:43 PM UTC-7, John Dallman wrote:
> In article <6c594b48-68c2-4400... at googlegroups.com>,

(I wrote)
 
> > The most important thing you want, when you start reading an 
> > instruction, is to know where the next one is. For IBM S/360, 
> > you always know from the first byte where the next instruction 
> > is. Even with the above change, it isn't easy for VAX, but 
> > would be close.

> I wonder if that was deliberate for S/360? The original paper on the 
> architecture does not mention anything about the encoding; a really old 
> copy of the "Principles of Operation" would be interesting. 
 
Some parts were amazingly lucky in that virtual machines work well.

One interesting one is that when S/360 machines have nothing else to
do, they stop executing instructions. There is no idle loop.

The reason for that, is that for rented machines they charged based on
actual usage.  There is a meter that measures how often it is not in
a wait state.

But S/360 came not so long after Stretch, designed to be fast and where
many pipelined processor ideas started. Early on, there was the 360/92,
later replaced by the model 91, using pipelining techniques. 

One not talked about much, but that I have known for a long time, is
that hexadecimal floating point is more convenient for pipelining.

S/360 has three instruction lengths, which you know from the first
two bits of the opcode, Even if you are not designing a fancy processor,
it is nice to know where the next instruction is.  While S/360 is not RISC,
compared to VAX, it looks very RISCy.  

The book by Blaauw and Brooks, "Computer Architecture" describes
many architectures, but includes some details that actually went
into S/360, because they designed it. 

The one I am remembering now is the big endian choice, which they
believe is the right choice.  The VMS DUMP program shows why.

It might be in there.