Elisp native-comp vs. SBCL for inclist-type-hints benchmark (was: [PATCH] Re: Bignum performance)

[Ihor Radchenko <yantar92@posteo.net>]

Emanuel Berg <incal@dataswamp.org> writes:

>> Unfortunately, I cannot use this file because it is referring to
>> ~/public_html/cl/bench/timing.cl, which I do not have.
>
> It is in the same directory
>
>   https://dataswamp.org/~incal/cl/bench/timing.cl

It would be nice if you attached the files to email.
Otherwise, people examining this threads a few years in future may not
be able to access the files.

> So you have to set that manually as for now.

I did, and the results are very different from yours:

$  SBCL_HOME=/usr/lib64/sbcl sbcl --load /tmp/inclist.cl

;; 1.096667 s real time
;; 1.096235 s run time

$ SBCL_HOME=/usr/lib64/sbcl sbcl --load /tmp/inclist-type-hints.cl 
;; 0.55 s real time
;; 0.549992 s run time

(emacs master)
$ perf record ./src/emacs -batch -l /home/yantar92/.emacs.d/straight/repos/elisp-benchmarks/elisp-benchmarks.el --eval '(setq elb-speed 2)' --eval '(elisp-benchmarks-run "inclist")'

* Results

  | test               | non-gc avg (s) | gc avg (s) | gcs avg | tot avg (s) | tot avg err (s) |
  |--------------------+----------------+------------+---------+-------------+-----------------|
  | inclist            |           1.20 |       0.00 |       0 |        1.20 |            0.02 |
  | inclist-type-hints |           1.05 |       0.00 |       0 |        1.05 |            0.02 |
  |--------------------+----------------+------------+---------+-------------+-----------------|
  | total              |           2.26 |       0.00 |       0 |        2.26 |            0.02 |

inclist: 1.1 sec vs. 1.2 sec
inclist-type-hints: 0.55 sec vs. 1.05 sec

with native-comp speed 3, the results are on par for inclist

$ perf record ./src/emacs -batch -l /home/yantar92/.emacs.d/straight/repos/elisp-benchmarks/elisp-benchmarks.el --eval '(setq elb-speed 3)' --eval '(elisp-benchmarks-run "inclist")'

* Results

  | test               | non-gc avg (s) | gc avg (s) | gcs avg | tot avg (s) | tot avg err (s) |
  |--------------------+----------------+------------+---------+-------------+-----------------|
  | inclist            |           1.07 |       0.00 |       0 |        1.07 |            0.02 |
  | inclist-type-hints |           0.99 |       0.00 |       0 |        0.99 |            0.00 |
  |--------------------+----------------+------------+---------+-------------+-----------------|

inclist: 1.1 sec vs. 1.07 sec
inclist-type-hints: 0.55 sec vs. 0.99 sec

There is nothing obvious that is slower in Elisp - most of the time is
spent in the native compiled functions:

    47.83%  emacs     inclist-b6453dcf-34842bf7.eln                  [.] F656c622d696e636c697374_elb_inclist_0
    44.80%  emacs     inclist-type-hints-bb635d76-535ebfb0.eln       [.] F656c622d696e636c6973742d7468_elb_inclist_th_0
     1.45%  emacs     emacs                                          [.] process_mark_stack

So, might be some missing optimization in native-comp itself.
CCing Andrea, in case if he has some insight about further analysis.

-- 
Ihor Radchenko // yantar92,
Org mode contributor,
Learn more about Org mode at <https://orgmode.org/>.
Support Org development at <https://liberapay.com/org-mode>,
or support my work at <https://liberapay.com/yantar92>