Type-based aliasing

[Mark Mitchell]

John --

  I'm going to try to summarize the outcome of your attempt at
mode-based alias analysis, i.e., alias analysis following the strict
ANSI C rules for conforming programs.  I'd like you, and others on the
list, to tell me if I've understood the situation correctly.

  You implemented the ANSI aliasing rules.  Unfortunately, they
resulted in misoptimizations of conforming code, because of the fact
that GCC puts some structs/unions in pseudos, and your alias analysis
didn't believe that stores through these pseudos could alias other
storage.  In particular, the MEMs representing stores to these pseudos
did not have MEM_IN_STRUCT_P set.  Is that correct?

  Were there any other problems?

  Thanks,

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com