Before someone dings me on the race condition, imagine I had factored
those repeated @cache calls into a (let [value# @cache] ...)
expression. The other thing I'm missing is a check for whether the
protocol has been changed since the cache entry was created; this
would be done by also maintaining an :expected-protocol field and
comparing its value to protocol-function's protocol.
-Per
On Mon, Mar 22, 2010 at 11:24 PM, Per Vognsen <per.vogn...@gmail.com> wrote:
>>>> lookup (find-protocol-method) over the method invocation itself. Is that
>>>> the
>>>> right way to summarize the performance implications?
>>>
>>> No, since the result is cached. The real every-call overhead is having to
>>> go through the var to detect changes to the protocol, vs the inline-defined
>>> case where the implementation of the interface can't change given the same
>>> instance class.
>>
>> Am I saying something different? The cache requires a hash lookup, right?
>
> I thought the eventual goal was to implement call-site caching in
> bytecode? Inline caching isn't feasible due to the direct and indirect
> cost of regenerating bytecode. But you can still do non-inline
> caching! For the simplest and highest payoff case of monomorphism, it
> would look something like this:
>
> (protocol-function x ...)
>
> =>
>
> (let [cache# <atom created at compile time>]
> (if (= (:expected-type @cache#) (type x))
> ((:expected-function @cache#) x ...) ;; hit
> (let [actual-function <look up protocol-function in (type x) the
> hard way>] ;; miss
> (swap! cache# (constantly {:expected-type (type x),
> :expected-function actual-function}))
> (actual-function x ...))))
>
> It occurs to me that you could even pull off this code transformation
> entirely in-language with CL-style compiler macros while leaving alone
> references to protocol-function in non-operator position.
>
> -Per
>
--
You received this message because you are subscribed to the Google
Groups "Clojure" group.
To post to this group, send email to clojure@googlegroups.com
Note that posts from new members are moderated - please be patient with your
first post.
To unsubscribe from this group, send email to
clojure+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
To unsubscribe from this group, send email to
clojure+unsubscribegooglegroups.com or reply to this email with the words
"REMOVE ME" as the subject.
