Re: [patch] Lno branch merge part 12 -- induction variable optimizations

[Mark Mitchell <mark at codesourcery dot com>]

Zdenek Dvorak wrote:

> Hello,
>
>  
>
> > > 5) build_int_cst is rewritten to decide whether to sign extend or to
> > > zero extend depending on the signedness of the type of the constant,
> > > and to mask out bits outside of the precision, which should be what
> > > we want most of the time.  Few places where this is incorrect should
> > > use build_int_csts instead.
> > >      
> > Like Joseph, I think (5) should be completely separated from this patch.
> >
> > You also need to use --param for CONSIDER_ALL_CANDIATES_BOUND and 
> > MAX_CONSIDERED_USES., and probably MAX_RATIO.  The patch to 
> > stor-layout.c is unncessary, and should not be included.  Other people 
> > have already suggested that you avoid the potentially bad compile-time 
> > behavior in remove_statement by relying on DCE.
> >
> > Please resubmit with those changes, and copy me.  I would like someone 
> > else to review this, but I will do it if nobody else does.
> >    
>
> here is the patch that reflects some of the changes above.
> Changes not included:
>
> -- MAX_RATIO is left as a #define.  This constant should basically mean
>   average number of iterations of a loop and it should be replaced
>   by value obtained from profiling as soon as it works reliably enough.
> -- remove_statement is left as it is, since it by far is not the only
>   place where we need stmt_for_bsi.  stmt_for_bsi is also needed on
>   several other places where we need to insert a code that replaces
>   value computed by a statement, and avoiding it in all these instances
>   would be quite complicated.  I have already submitted a patch that makes
>   stmt_for_bsi efficient.
>  
All right.

I will approve this patch, effective September 5th in your local time 
zone, conditional on the fact that nobody else (RTH, Law, Novillo, etc) 
has objected by then.

Thanks,

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com