JS speed - Konq vs. Mozilla

[Dirk Mueller]

On Sam, 07 Sep 2002, Koos Vriezen wrote:

> Still, I'm not sure why its faster. Maybe its the 'Collector::collect()'
> that's faster in the next run (less blocks).

Profilling the collect() step, it seems the biggest problem is the massive 
amount of cache misses it produces, that is over 78% of all (of testkhtml). 

It might be better to put the ValueImp flags into a separate, continuous 
memory region, because the current pointer following causes it to refetch 
all allocated memory regions into the CPU cache, which is extremely slow. 
Also, as the allocated regions are a multiple of the cache association size, 
this currently screams for caching problems. 

Do you want to try that (allocating a separate bitarray for those flags) ?

Might have a nice gain in speed. 


-- 
Dirk (received 526 mails today)