[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] extend map-into

From: Andrea Corallo
Subject: Re: [PATCH] extend map-into
Date: Wed, 09 Oct 2019 15:35:00 +0000
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/26.2 (berkeley-unix)

Stefan Monnier <address@hidden> writes:

>> I can tell you what I did just now and you tell me if it makes sense.
>> I've created a function with as body an unrolled loop that calls 10000
>> times (map-into nil 'alist).
>> Once byte compiled I've run it with the current map.el and then with the
>> new one loaded.
> That sounds about right.  Tho, you can also use something like
> `benchmark` for that.
>> On my machine both versions runs in about the same time (~0.04-0.03
>> secs), if there's some difference for this simple test is just noise.
> 0.03 seconds is too small for a meaningful measure, in my experience,
> I recommend you add 2 zeroes to the number of iterations to bring it
> into the realm of "a few seconds".
> But, yes, there's a chance the difference is negligible (more time is
> spent in the funcall/apply machinery itself and/or the hash lookup than
> the specific computation of the "tag" used to index the hash-table,
> which is the part that depends on the set of specializers used on
> a particular argument of a generic function).
>         Stefan

I've repeated the same test going up two magnitude orders (1000000
iterations). Still the execution time (~1.7secs) seems not distinguishable
between the old and the implementation.

For my curiosity I've also tried to run it using
`benchmark-run-compiled' with a very similar execution time as result. I
would have expected to have the unrolled version faster but apparently
funcall is not the bottle-neck here.

As usual when profiling guessing proves to be not a good idea.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]