Re: [Cocci] [PATCH] Coccinelle: null: Optimise disjunctions in SmPL script “eno.cocci”

Re: [Cocci] [PATCH] Coccinelle: null: Optimise disjunctions in SmPL script “eno.cocci”

[Julia Lawall]

[-- Attachment #1: Type: text/plain, Size: 2983 bytes --]



On Sun, 25 Oct 2020, Markus Elfring wrote:

> From: Markus Elfring <elfring@users.sourceforge.net>
> Date: Sun, 25 Oct 2020 18:54:36 +0100
>
> A disjunction is applied by this script for the semantic patch language.
> This construct uses short-circuit evaluation. It has got the consequence
> that the last element of the specified condition will only be checked
> if all previous parts did not match. Such a technical detail leads to
> a recommended ordering of condition parts if you would like to care for
> optimal run time characteristics of SmPL code.
>
> An usage incidence was determined for the specified identifiers in source
> files from the software “Linux next-20201023” by another SmPL script.
>
> See also:
> Determination of an usage statistic for memory allocation calls
> https://lore.kernel.org/cocci/2774601.u91sIFNy1E@sonne/
>
> This analysis result indicates a clear ranking for such function calls.
> Thus reorder the SmPL disjunction items according to their usage incidence.

Did you actually test this before and after the change and see a
difference in performance?  On my laptop, I see absolutely no difference,
for the patch mode and for the context mode.  I didn't try the other
cases.

Before the change:

patch: 440.182u 2.049s 1:53.12 390.9%	0+0k 0+232io 4pf+0w
context: 392.931u 2.164s 1:41.53 389.1%	0+0k 432+360io 5pf+0w

After the change:

patch: 442.182u 2.090s 1:54.13 389.2%	0+0k 0+240io 4pf+0w
context: 392.742u 2.035s 1:41.37 389.4%	0+0k 0+360io 4pf+0w

julia

>
> Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
> ---
>  scripts/coccinelle/null/eno.cocci | 15 +++++++++++----
>  1 file changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/scripts/coccinelle/null/eno.cocci b/scripts/coccinelle/null/eno.cocci
> index 81584ff87956..969cab5116a9 100644
> --- a/scripts/coccinelle/null/eno.cocci
> +++ b/scripts/coccinelle/null/eno.cocci
> @@ -17,8 +17,16 @@ virtual report
>  @depends on patch@
>  expression x,E;
>  @@
> -
> -x = \(kmalloc\|kzalloc\|kcalloc\|kmem_cache_alloc\|kmem_cache_zalloc\|kmem_cache_alloc_node\|kmalloc_node\|kzalloc_node\)(...)
> + x =
> +(kzalloc
> +|kmalloc
> +|kcalloc
> +|kmem_cache_alloc
> +|kmem_cache_zalloc
> +|kzalloc_node
> +|kmalloc_node
> +|kmem_cache_alloc_node
> +)(...)
>  ... when != x = E
>  - IS_ERR(x)
>  + !x
> @@ -27,8 +35,7 @@ x = \(kmalloc\|kzalloc\|kcalloc\|kmem_cache_alloc\|kmem_cache_zalloc\|kmem_cache
>  expression x,E;
>  position p1,p2;
>  @@
> -
> -*x = \(kmalloc@p1\|kzalloc@p1\|kcalloc@p1\|kmem_cache_alloc@p1\|kmem_cache_zalloc@p1\|kmem_cache_alloc_node@p1\|kmalloc_node@p1\|kzalloc_node@p1\)(...)
> +*x = \(kzalloc@p1\|kmalloc@p1\|kcalloc@p1\|kmem_cache_alloc@p1\|kmem_cache_zalloc@p1\|kzalloc_node@p1\|kmalloc_node@p1\|kmem_cache_alloc_node@p1\)(...)
>  ... when != x = E
>  * IS_ERR@p2(x)
>
> --
> 2.29.1
>
> _______________________________________________
> Cocci mailing list
> Cocci@systeme.lip6.fr
> https://systeme.lip6.fr/mailman/listinfo/cocci
>

Re: [Cocci] Coccinelle: null: Optimise disjunctions in SmPL script “eno.cocci”

[Julia Lawall]

On Sun, 25 Oct 2020, Markus Elfring wrote:

> >> This analysis result indicates a clear ranking for such function calls.
> >> Thus reorder the SmPL disjunction items according to their usage incidence.
> >
> > Did you actually test this before and after the change and see a
> > difference in performance?
>
> Would you become interested to configure a representative test environment
> for safe comparisons of corresponding run time characteristics
> of the affected software?

In what sense could the comparison possibly be unsafe?

Just use time and run spatch on whatever machine you want.  It's not that
complicated.

> > On my laptop, I see absolutely no difference,
> > for the patch mode and for the context mode.
>
> Does such information trigger a desire to clarify involved aspects in more detail?

Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz

julia

Re: [Cocci] Coccinelle: null: Optimise disjunctions in SmPL script “eno.cocci”

[Julia Lawall]

[-- Attachment #1: Type: text/plain, Size: 3566 bytes --]



On Sun, 25 Oct 2020, Markus Elfring wrote:

> >> Would you become interested to configure a representative test environment
> >> for safe comparisons of corresponding run time characteristics
> >> of the affected software?
> >
> > In what sense could the comparison possibly be unsafe?
>
> * Our test systems are obviously different.
>   Thus concerns can be considered for reproducibility of test results
>   on other possible configurations.
>
> * We share only a tiny fraction of technical information which would probably
>   be needed for “benchmarks”.
>
>
> > Just use time and run spatch on whatever machine you want.
>
> fring@Sonne:~/Projekte/Linux/next-patched>
> elfring@Sonne:~/Projekte/Linux/next-patched> git checkout next-20201023 && XX=$(date) && time spatch -D patch --timeout 9 --jobs 4 --chunksize 1 --include-headers --no-includes --dir . scripts/coccinelle/null/eno.cocci > ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno1.diff 2> ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno1-errors.txt; YY=$(date) && echo "$XX | $YY"
> …
> real	2m54,266s
> user	10m15,553s
> sys	0m4,004s
> So 25. Okt 20:53:56 CET 2020 | So 25. Okt 20:56:51 CET 2020
> elfring@Sonne:~/Projekte/Linux/next-patched> git checkout next-20201023 && XX=$(date) && time spatch -D context --timeout 9 --jobs 4 --chunksize 1 --include-headers --no-includes --dir . scripts/coccinelle/null/eno.cocci > ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno2.txt 2> ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno2-errors.txt; YY=$(date) && echo "$XX | $YY"
> …
> real	2m38,494s
> user	9m39,364s
> sys	0m4,094s
> So 25. Okt 20:58:05 CET 2020 | So 25. Okt 21:00:44 CET 2020
> elfring@Sonne:~/Projekte/Linux/next-patched> git checkout optimise_disjunction_in_eno.cocci-1 && XX=$(date) && time spatch -D patch --timeout 9 --jobs 4 --chunksize 1 --include-headers --no-includes --dir . scripts/coccinelle/null/eno.cocci > ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno3.diff 2> ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno3-errors.txt; YY=$(date) && echo "$XX | $YY"
> …
> real	2m46,097s
> user	10m15,467s
> sys	0m3,984s
> So 25. Okt 21:00:56 CET 2020 | So 25. Okt 21:03:42 CET 2020
> elfring@Sonne:~/Projekte/Linux/next-patched> git checkout optimise_disjunction_in_eno.cocci-1 && XX=$(date) && time spatch -D context --timeout 9 --jobs 4 --chunksize 1 --include-headers --no-includes --dir . scripts/coccinelle/null/eno.cocci > ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno4.txt 2> ~/Projekte/Bau/Linux/scripts/Coccinelle/call_checks/20201023/eno4-errors.txt; YY=$(date) && echo "$XX | $YY"
> …
> real	2m41,472s
> user	9m37,492s
> sys	0m3,834s

In the patch case, the user and system time are essentially identical.  In
the context case, the difference in user time is 2 seconds out of 9.5
minutes, 0.3%.

In the patch case, the real time is a bit slower.  But I believe that this
is due to the time to read in the data from the file system.  I also had a
number like that, but the difference disappeared when I reran it
afterwards, which meant running that case in the same conditions as the
others.

In the context case, the real time is slightly slower with your patch.

So I see no compelling evidence for making the change.

julia

> So 25. Okt 21:03:56 CET 2020 | So 25. Okt 21:06:37 CET 2020
>
>
> > Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz
>
> AMD Phenom(tm) II X4 850 Processor
>
> Will any other aspects become relevant?
>
> Regards,
> Markus
>