Re: On elisp running native - Update 9

[Andrea Corallo <akrl@sdf.org>]

Gong-Yi Liao 廖宮毅 <gongyi.liao@gmail.com> writes:

> Andrea,
>
> Question:
> What's the difference between Update 1's and   Update 9's benchmark
> table's byte code column?
>
> The byte code column's  run time results have  decreased roughly 78%
> from Update 1's 507.23s to Update 9's 114.91, that seems unbelievable
> if all the Update 9's byte code result are purely based on byte code
> without any natively compiled code.
>
> Are Update 9's byte code result based on natively compiled built-in
> function and byte code compiled benchmark functions or they are
> purely all byte code compiled?
>
> Or there are drastic changes in benchmark environment settings
> (hardware, number of repetitions) betweens these two updates?
>
> Thanks,
> Gong-Yi. 

Hi Gong-Yi,

The first measures were based on a collection of benchmarks that later
became what we have in ELPA now.

In the transition a number of things changed: the set slightly changed,
IIRC also the durations have been normalized to be more homogeneous and
the result is given in avg per iteration instead of being cumulated.

Lastly the two tests has been run on two different machines.

Probably would be interesting to re-add the u-benchmark for type-hints
in the future.

Regards

  Andrea

-- 
akrl@sdf.org