[patch] Lno branch merge part 2 -- rtl level invariant motion

[Mark Mitchell]

Zdenek Dvorak wrote:

>Hello,
>
>  
>
>>>On Thu, Jun 10, 2004 at 03:57:43PM +0200, Zdenek Dvorak wrote:
>>>      
>>>
>>>>Lno branch bootstraps on ia64 and x86_64 as well, but I haven't checked
>>>>the produced code there.
>>>>        
>>>>
>>>Hum.  Well, I guess that's going to have to be good enough
>>>for now, until we get enough merged to consider replacing
>>>all of loop.c.
>>>      
>>>
>>thanks to people from IBM, I have some spec numbers for ppc.  They look
>>unexpectedly good to me -- 2.2% improvement on specint, 4.3% on specfp.
>>There are a few regressions (the worst being 6% on wupwise), but overall
>>I was pleasantly surprised.
>>    
>>
>
>sorry for being ambiguous -- these are not the measurements of the
>patch.  The numbers are a comparison of the old loop optimizer with the
>whole new loop optimizer of the lno branch (except for vectorizer), at
>-O2.
>  
>
Those are certainly encouraging numbers!  One of the key tests for me 
with respect to LNO is "is it actually providing an improvement in the 
generated code?" and if those numbers are reproducible, I think the 
answer is unambigously in the affirmative.

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