Re: Superblock analysis?

"Tommy Thorn" <tommy.thorn@gmail.com>
18 Jul 2006 14:05:38 -0400

          From comp.compilers
Related articles
Superblock analysis? ammalik@cs.uwaterloo.ca (2006-07-18)
Re: Superblock analysis? bill@qswtools.com (Bill Cox) (2006-07-18)
Re: Superblock analysis? tommy.thorn@gmail.com (Tommy Thorn) (2006-07-18)
| List of all articles for this month |
From: "Tommy Thorn" <tommy.thorn@gmail.com>
Newsgroups: comp.compilers
Date: 18 Jul 2006 14:05:38 -0400
Organization: http://groups.google.com
References: 06-07-038
Keywords: analysis
Posted-Date: 18 Jul 2006 14:05:38 EDT

ammalik@cs.uwaterloo.ca wrote:
> 1--Is it possible to have two independent branch instructions in a superblock?


One definition of a superblock have it as "a linear sequence of basic
blocks that has a single entry point only.", thus yes, you can have as
many [independent] branches as you like in a superblock.


> 2--Do compilers perform post register-allocation instruction scheduling using
> superblocks?


That seems like several questions in one. First, you can do register
allocation before or after scheduling (in some approaches even
simultaneously). There are pros and cons to either choice, but
generally, if you have enough registers, you'd do allocation last (to
maximize the scheduling freedom).


You can most definitely schedule using superblocks. Using superblocks
may offer more scheduling opportunities at a lower complexity than
using basic blocks.


You could do a lot worse than reading the seminal dissertation
"Bulldog: a compiler for VLIW architectures" by John R. Ellis
(http://portal.acm.org/citation.cfm?id=912347&dl=GUIDE&coll=&CFID=15151515&CFTOKEN=6184618)


Tommy

Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.