🚀 go-pugleaf

On 2018-08-13 19:07, John Levine wrote:
> In article <44627835.555851238.7102...@news.eternal-september.org>,
> Peter Flass  <peter...@yahoo.com> wrote:
>>>> And the epitome of the CISC processor, the VAX-11.
>>>
>>> Sixteen addressing modes, single instructions to calculate CRCs and polynomials,
>>> string translation instructions... the VAX had just about everything except an
>>> opcode for "phase of the moon."  Too many bells and whistles to be considered
>>> elegant, IMO.
> 
> It suffered from a surprising degree of technical myopia.  The VAX was
> designed to be implemented in serial microcode, and was quite painful
> to overlap or pipline.  The instruction format was an opcode and a series
> of variable length operand descriptors which had to be decoded in order since
> one decriptor could modify the register used in a subsequent one, e.g.
> 
>   ADDW2 (R3)+,(R3)

Decoding is not the problem, and can be done without anything blocking 
anything else.
Evaluation is where you get into "problems", as there can be 
interdependencies. But that has no relevance for the decoding as such.

> It can't fetch the second operand until it's decoded the first one,
> since the first one modifies R3.  There were also a lot of absurdly
> complex instructions like evaluate polynomial which would have made
> a lot more sense as subroutines, at least if they had a lighter
> weight subroutine call instruction.

The VAX do have more light weight subroutine calls than the generic 
CALLS/CALLG. I don't know why people never seem to know this. There is 
the JSB/RSB pair, which is just the simple push the return address and 
jump and nothing more opcodes (and the return is obviously just a pop 
the return address).

> I gather the later Vaxen did complex pipelines with interlocks but it
> was a lot of effort to solve a problem that they needn't have created
> in the first place.  Compare it to the 360/370 which is carefully
> designed so that as soon as you've looked at the opcode you can
> evaluate all of the operands in parallel.

The pipelines VAX implementation happened already in 1984 with the VAX 
8600. Yes, it did take some effort, but it was not as complicated as you 
seem to imply. And all later VAXen was pipelined.

   Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: b...@softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol