Re: egcs at FREENIX track of 1999 Usenix conference

[Toon Moene <toon at moene dot indiv dot nluug dot nl>]

Hey John,

That's one for you !

>  It's a technical conference, and papers should reflect
>  this.  The ideal paper would be from an egcs developer
>  and contain low-level hardcore technical information
>  about a specific set of optimizations or other
>  improvements.  For instance, you might describe the alias
>  analysis optimization in loving detail, show situations
>  where it leads to better generated code, and give some
>  performance figures.

If you're in dire need of some performance figures, I doubtlessly  
can dig up some :-)

Perhaps, for those of you tuning in late, it is useful to copy what  
I wrote RMS on the 2nd of August last year - in a pledge to make  
John's work part of gcc - 6 days before I became aware of the  
movement that would result in egcs:

QUOTE

John Carr's alias analysis is the first step to make g77 a real  
_Fortran_ compiler, by allowing it to take advantage of the Fortran  
alias semantics.  Without these semantics, instruction scheduling is  
much more limited in its abilities to arrive at optimal code  
sequences.  In the extreme, without Fortran's alias semantics, it  
would be next to useless writing vectorising compilers, or even  
building vector computers.

UNQUOTE

Monterey, hmmmm ....

Cheers,
Toon.