[rc5] Re: Supercomputers & Bovine, Q&A
reiter at ruth.rad.tju.edu
Tue Aug 5 14:11:47 EDT 1997
On Tue, 5 Aug 1997, Marc Sissom wrote:
> At 08:57 AM 8/5/97 -0700, you wrote:
> >On Mon, 4 Aug 1997, Murray Stokely wrote:
> >> I always thought PPC chips were hybid as opposed to true RISC like
> >> SPARCS and Alphas. Can anyone with more PPC knowledge shed some light
> >> on this for me?
> >Nope. PPC chips are pure RISC. Always have been.
> Ha! Now things get interesting. Define pure RISC.
> Is it memory access limited to register load-store operations?
> The PPC abides by this - almost. The single precision FP
> load/store instructions don't just load or store. They load
> and convert or convert and store. So here's a CISC instruction
> that's built into a "pure" RISC proc.
> Is it all single cycle instruction execution?
> The PPC does not stick to this(lsw)
> Is it fixed instruction-opcode mapping?
> Nah, refences to r0 often produce the result "0" rather than the
> contents of r0(lhzx).
> Is it large register arrays?
> yep, relatively large anyway.
> Or is it just a Reduced instruction set? Reduced relative to what?
> The Alpha instruction set is even more reduced than the PPC. Is it
> a "purer" RISC?
Even the Alpha is a hybrid these days. The original ones (the 21064s I
belive) could only do word accesses to memory but that is very inefficent
since I would need to fetch 64 bits for a char (a serious concern if I
want to bring in a few chars that all lie next to each other in memory).
So, the DEC people added instructions to access smaller data types. This
is complicating the insturction set, but it made some important code
(the Alpha Linux kernel for example) far more efficent.
A "pure" RISC architecture was tried, and it turns out a hybrid design
is quicker (in most cases). The real question is "how RISC do you want to
be?" and the answer depends very heavily on the intended use of the CPU.
> Marc Sissom | Design Engineer
> DNA Enterprises, Inc. | Phone: 972/644-3301
> 269 W. Renner Parkway | Fax: 972/644-6338
> Richardson, Texas 75080 | http://www.dnaent.com
> To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.
More information about the rc5