Re: Cost of handling misaligned access

Michael S wrote:

On Fri, 7 Feb 2025 11:06:43 +0100
Terje Mathisen <terje.mathisen@tmsw.no> wrote:
 

Michael S wrote:

On Thu, 6 Feb 2025 21:36:38 +0100
Terje Mathisen <terje.mathisen@tmsw.no> wrote:

BTW, when I timed 1000 calls to that 5-6 us program, to get around
teh 100 ns timer resolution, each iteration ran in 5.23 us.

>
That measurement could be good enough on desktop. Or not.
It certainly not good enough on laptop and even less so on server.
On laptop I wouldn't be sutisfied before I lok my program to
particualr core, then do something like 21 measurements with 100K
calls in each measurement (~10 sec total) and report median of 21.

>
Each measurement did 1000 calls, then I ran 100 such measurements.
The 5.23 us value was the lowest seen among the 100, with average a
bit more:
>
>
Slowest: 9205200 ns
Fastest: 5247500 ns
Average: 5672529 ns/iter
Part1 = 3338
>
My own (old, but somewhat kept up to date) cputype program reported
that it is a "13th Gen Intel(R) Core(TM) i7-1365U" according to CPUID.
>
Is that sufficient to judge the performance?
>
Terje
>

 Not really.
i7-1365U is a complicated beast. 2 "big" cores, 8 "medium" cores.
Frequency varies ALOT, 1.8 to 5.2 GHz on "big", 1.3 to 3.9 GHz on
"medium".

OK. It seems like the big cores are similar to what I've had previously, i.e. each core supports hyperthreading, while the medium ones don't. This results in 12 HW threads.

As I said above, on such CPU I wouldn't believe the numbers before
total duration of test is 10 seconds and the test run is locked to
particular core. As to 5 msec per measurement, that's enough, but why
not do longer measurements if you have to run for 10 sec anyway?

The Advent of Code task required exactly 250 keys and 250 locks to be tested, this of course fits easily in a corner of $L1 (2000 bytes).
The input file to be parsed was 43*500 = 21500 bytes long, so this should also fit in $L1 when I run repeated tests.
Under Windows I can set thread affinity to lock a process to a given core, but how do I know which are "Big" and "Medium"?
Terje
--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"