Am 11.03.2015 15:37, schrieb Remi Forax:
[...]
Sometime, yes, if you count the frequency of each branch by example,
here is another code that implements a bimorphic inlining cache that
expand to a dispatch table local to a callsite if necessary, I think you
can adapt this code to implement what Mark want quite easily

https://code.google.com/p/jsr292-cookbook/source/browse/trunk/bimorphic-cache/src/jsr292/cookbook/bicache/RT.java

but obviously the DispatchMap case is not something that will be nice to performance, or not? I am was wondering how performance would drop if I have an array for counting invocations for each part of the cache as well as total number of counts and how that would impact performance. Because obviously on each invocation I would have to call a little method that does the counting as well as an additional guard to cause the reorganization once a certain threshold is reached... And I know every guard will lower performance.

Anyway... based on Marks initial post as well as Remi's idea I would come up with the following... which is mostly an abstraction and refactoring of Remi's code.

class InlineCache<T extends MutableCallSite> {
    int MAX_DEPTH = 5
    int depth
    T callsite
    ...
    InlineCache(T callsite) {...}
    public void installCache()
    public final Object internalFallback(...)
    public abstract MH fallback(T callsite, Object[] args)
    public abstract MH megamorphicFallback(T callsite, Object[] args)
    public final MH insert(MH target, MH... tests)
}

in case of a cache miss and max depth not reached fallback is called and supposed to call insert for the new target, the method will do the integration into the cache, and return a handle you can return from fallback. megamorphicFallback would be called in case of maximum depth is reached and install for example a virtual dispatch or whatever else. installCache would the cache logic as target of the callsite. internalFallback would increase the count as well as calling either fallback or megamorphicFallback. I didn't want to let the cache extend the callsite, since others may need the callsite class for their own purposes. Separation could allow for for example using a different cache type.

Taking your code from https://code.google.com/p/jsr292-cookbook/source/browse/trunk/inlining-cache/src/jsr292/cookbook/icache/RT.java basically line 42 would be in internalFallback 43-45 in megamorphicFallback, 46-47 in internalFallback, 50-57 in fallback, 56-57 would be the test given to insert, 59-63 in internalFallback, and of course we would not link to fallback, but to internalFallback

What do you guys think?

bye Jochen

--
Jochen "blackdrag" Theodorou - Groovy Project Tech Lead
blog: http://blackdragsview.blogspot.com/
german groovy discussion newsgroup: de.comp.lang.misc
For Groovy programming sources visit http://groovy-lang.org

_______________________________________________
mlvm-dev mailing list
mlvm-dev@openjdk.java.net
http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev

Reply via email to