Prev: Opcode Parsing & Invalid Opcodes
Next: "SIE" on a RISC architecture
From: Tim McCaffrey on 8 Jul 2010 14:00
In article
<8c504e90-cc74-4c15-bd14-63f0c1be7fb2(a)d16g2000yqb.googlegroups.com>,
MitchAlsup(a)aol.com says...

>Call Indirect and Jump Indirect Tables with and without hashing are
>generally pretty good given suficient area to contain the footprint
>and a little cleverness of the designers tuning the hash. Here, I
>think, it is much more worthwhile for the software people to choose
>reasonable data structures to contain the indirect tables, and then,
>over time, let the HW guys stumble over things to optimize. So,
>without SW doing anything special, all of a sudden it just gets
>faster. AND it looses none of its obviousness or RAS. Not so with the
>"special creative bent of the moment stuff."
>

I would like to see a prefetch instruction for code, since you (might) know
where that indirect branch is going to go, might as well preload it into the
I-cache.

- Tim

 | 
Pages: 1
Prev: Opcode Parsing & Invalid Opcodes
Next: "SIE" on a RISC architecture