Re: Help requested: Interpreting results from bench.pl

On 3/12/26 10:40, Paul "LeoNerd" Evans wrote:
> Hi all,
> 
> I've been trying to measure performance of a set of changes on a
> branch. To do this I've arranged to run Porting/bench.pl at several
> stages along the branch, so I can compare a bunch of numbers between
> the runs. What I end up with looks like this:
> 
> AVERAGE
> 
>          blead      p1    p1+1     p1+4    p1+5     p1+6    p1+7     p1+8
>         ------ ------- ------- -------- ------- -------- ------- --------
>      Ir 100.00   99.98   99.98    99.98   99.98    99.98   99.98    99.98
>      Dr 100.00   99.98   99.98    99.98   99.98    99.98   99.98    99.98
>      Dw 100.00   99.99   99.99    99.99   99.99    99.99   99.99    99.99
>    COND 100.00   99.97   99.97    99.97   99.97    99.97   99.97    99.97
>     IND 100.00  100.00  100.00   100.00  100.00   100.00  100.00   100.00
> 
> COND_m 100.00  112.64  102.73    96.57   84.61   107.92  113.36   113.77
>   IND_m 100.00  100.00   99.24    99.24  100.00   100.00  100.00    99.99
> 
>   Ir_m1 100.00   66.28   96.14    91.83   47.18    66.57   66.12    54.55
>   Dr_m1 100.00  100.26  100.29   100.13  100.16   100.34  100.27   100.24
>   Dw_m1 100.00   99.40   99.40    99.40   99.40    99.20   99.80    99.80
> 
>   Ir_mm 100.00  100.00  100.00   100.00  100.00   100.00  100.00   100.00
>   Dr_mm 100.00  100.00  100.00   100.00  100.00   100.00  100.00   100.00
>   Dw_mm 100.00  100.00  100.00   100.00  100.00   100.00  100.00   100.00
> 
> (ignore the missing +2 and +3; I deleted those columns having
> determined that those changes make no difference at all to the numbers
> here).
> 
> I get that "bigger numbers are faster", so I can see that various of
> these commits in some sense make the perl "slower", and occasional
> spikes above 100 that might apparently make some things faster.
> 
> But beyond that I don't really have a feel for how to interpret all of
> this. For instance, the commits between +4 and +5 have made the Ir_m1
> score drop from 91.83 all the way down to 47.18 - what does that
> actually mean? If I want to look in more detail into the code and try
> to work out what specifically slowed it down and how to improve it so
> it isn't as bad, what should I be looking for?
> 

I've been told, and it may even be documented that the rows below IND_m 
are not significant; so don't even bother looking at them.  And it might 
be rows below plain 'IND' that aren't important.

There are people who say that this whole cachegrind output is not 
relevant to modern processors.  I plead ignorance.