> On Feb 16, 2025, at 10:39 PM, Chuck Guzis via cctalk <cctalk@classiccmp.org>
> wrote:
>
> On 2/16/25 15:52, Paul Koning via cctalk wrote:
>>
>>
>>> On Feb 16, 2025, at 5:30 PM, paul.kimpel--- via cctalk
>>> <cctalk@classiccmp.org> wrote:
>>>
>>> The question concerned good ALGOL code generation, not the feasibility of
>>> ALGOL code generation.
>>
>> I know that, but just as RISC machines can run very fast no matter what
>> applications you feed them, compilers created with skill can produce
>> excellent code no matter the target machine.
>
> CDC 6000-series turned in better performance benchmarks on COBOL that
> did the high-end IBM S/370 iron. There are huge advantages to fast
> instruction set execution operating on large word sizes. Just ask Don
> Nelson (he who played the bass drum in the Los Trancos Woods marching band).
>
> Two things that worked to CDC's advantage in addition to the simple
> instruction set was that the 6000 was a three-address architecture with
> no condition code register (conditional branches were made on the
> content of a register).
Also multiple functional units, seriously interleaved memory, and a bucket full
of other tricks. The way loads and stores are requested by the programmer
naturally makes them background operations, and the "stunt box" handles that
background process.
Not directly tied to application performance but very nice for the OS is an
amazingly efficient way to switch between processes, the "exchange jump". VAX
almost got this, but the 6000 does it better, taking advantage of the
read/restore cycle to do a read/update and run the context switch at memory
speed. Ignoring initiation time it takes less than 3 microseconds (in 1964!!).
paul
