[rescue] SGI, MIPS, and IA-64

[dave at cca.org]

jdboyd at cs.millersville.edu writes:

>Well, so far the ia-64 certainly isn't entering the mainstream.  What do you 
>think of this chip anyway?  It appears to be decent.

Horrific monstrosity.

>Unlike the i860 or the 3210, it looks like the ia-64 should be easy to have
>excelent optimizing compilers for.

Errr... I think the other way around. IA64 is going to be difficult to
get performance out of, and will require more sophisticated compilers
than currently exist. (That's one of the two main drawbacks of VLIW,
and IA64 is much worse that VLIW. (The other traditional VLIW problem is
NOPs wasting your bandwidth - which IA64 seems to get around.))

>I recently noticed that it does register windowing, which I never new before.
>That does start to make 128general registers look kinda small though.  
>I think the bundles are cool (a bundle has 3 41bit instructions in them, which
>are guaranteed to execute simultaneously, to my understanding), plus for the
>most part it looks like IA-64 can execute 2 bundles at once (meaning 6 parallel
>instructions).

They should have just done a nice clean simple VLIW. They've added so
much complexity that implimentation becomes absolute hell (and sucks
up way too much of your transistor budget, and drives your compiler
team to suicide).

------ David Fischer ------- dave at cca.org ------- http://www.cca.org ------
----------------- Live each day like the Triffids are coming. -------------