[viff-devel] Vedr.: Small VIFF language parser

[Janus Dam Nielsen]

If there are any other ideas for optimizations you would like to see  
in a compiler for Viff then now is the time to come forward.

--
Janus


Den 15/07/2008 kl. 16.49 skrev Martin Geisler:

> Janus Dam Nielsen <fagidiot at daimi.au.dk> writes:
>
>> Hi again,
>>
>> Heres the fruit of half a days work :)
>>
>> Analyzing the expression:
>>     sint n = (a * y + (1 - a) * x);
>>
>> Yields the following results:
>>
>> Final result:
>> ((a * y )+ ((1 - a )* x )) cost: 30
>> (((a * y )+ x )- (x * a )) cost: 30
>> (((a * y )+ x )- (a * x )) cost: 30
>> (((a * y )- (a * x ))+ x ) cost: 30
>> (((y * a )+ x )- (a * x )) cost: 30
>> (((y * a )+ x )- (x * a )) cost: 30
>> ((a * y )+ (x - (x * a ))) cost: 30
>> (((y * a )- (a * x ))+ x ) cost: 30
>> (((y * a )- (x * a ))+ x ) cost: 30
>> (((a * y )- (x * a ))+ x ) cost: 30
>> ((a * (y - x ))+ x ) cost: 20
>>
>> And lo and behold the last line reveals that the expression
>> a * (y - x )+ x
>> has the lowest cost.
>
> That is very cool! :-)
>
>> From the output above it seems that you have taught your analyzer
> about the distributive law, that + and * are commutative.
>
> What about a - a == 0, and that 0 * a == 0? I don't know if those
> rules will help -- they might just blow up the search space... :-)
>
> -- 
> Martin Geisler
> _______________________________________________
> viff-devel mailing list (http://viff.dk/)
> viff-devel at viff.dk
> http://lists.viff.dk/listinfo.cgi/viff-devel-viff.dk