Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.arch > #3610

Re: i860 on Bitsavers

Show all headers | View raw

Quadibloc <jsavard@ecn.ab.ca> writes:
>Instructions are 32 bits long. They contain a bit which marks them as
>pipelined. Pipelined instructions execute without any interlocks, so
>you have to wait however many cycles before using their results.

AFAIK, this worked a little differently: Every instruction pushing
into the pipeline would push a result out.  So if you have a pipelined
multiply with pipeline depth three, you have to make another two
muliplies before you get the result out.

That's pretty cool for software pipelining (reduces the register
pressure and the need for software register renaming (called modulo
variable expansion in modulo scheduling)), but ist pretty hard to
program otherwise.

>This I can understand, but I don't understand why a non-pipelined
>instruction discards results of any in-progress pipelined
>instructions;

My guess is that the "non-pipelined" instruction runs through the
pipeline, overwriting the contents of the latches along the way, so
whatever a pipelined instruction put there is gone afterwards.

>A Spartan but high-performance architecture. Why couldn't they have
>kept that instead of going to the Itanium? Presumably, Intel would
>have found it much easier to design new i860 chips embodying the
>improvements we see in the Pentium, thus widening their lead further
>over their competitors - who would no longer have the simplicity of
>RISC as an advantage on their side.

The i860 was not a success (compared to the hype).  New
implementations would have been a hard sell to everyone: Not only
prospective customers, but also Intel management.

Maybe IA-64 flopped for the same reasons as i860: good for HPC, too
hard to program for everything else; but I still think it had more
potential.

- anton
-- 
M. Anton Ertl                    Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

Back to comp.arch | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

i860 on Bitsavers Quadibloc <jsavard@ecn.ab.ca> - 2011-09-10 16:20 -0700
  Re: i860 on Bitsavers anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2011-09-11 07:41 +0000
    Re: i860 on Bitsavers John Levine <johnl@iecc.com> - 2011-09-11 15:58 +0000
      Re: i860 on Bitsavers nmm1@cam.ac.uk - 2011-09-11 16:46 +0100
        Re: i860 on Bitsavers John Levine <johnl@iecc.com> - 2011-09-11 23:37 +0000
    Re: i860 on Bitsavers Quadibloc <jsavard@ecn.ab.ca> - 2011-09-11 12:15 -0700

csiph-web