Zoom Logo

2021 Code Performance Series - Shared screen with speaker view
Tobias Weinzierl
01:07:21
Lack of vectorisation
Tuomas Koskela
01:14:40
0
Tuomas Koskela
01:14:54
We only access every element once
Tuomas Koskela
01:15:17
Ah, right
Tuomas Koskela
01:15:57
50%?
George Bisbas
01:22:56
Sorry, it was raised up from the roofline question
Tobias Weinzierl
01:23:20
Break till 10:15
Jamie Quinn
01:47:47
How well does likwid pin play with other wrappers e.g. scorep?
Jamie Quinn
01:49:27
Grand, thanks!
Tobias Weinzierl
01:49:27
We have the afternoon ;-)
Brian Wylie (JSC)
01:54:57
scalasca/scan does no bindings so should be orthogonal to likwid-pin: you'll need to indicate that likwid-pin is an "imposer" to be ignored when scan parses the commandline.
Jon McCullough
02:15:40
Is this most useful if you are only running on a single machine?
Tobias Weinzierl
02:16:28
We do the questions now and will start the third session at 10 past
Jon McCullough
02:18:23
Yes, that would be useful - my code is MPI only.
Jon McCullough
02:18:36
Thanks!
Tuomas Koskela
02:32:32
What do distances 10 and 32 mean?
George Bisbas
03:02:58
What is the analogous when using dynamic scheduling?
Lawrence Mitchell
03:03:55
TBF I was teaching some of this earlier in the year
Tobias Weinzierl
03:04:11
So we should have asked your students
Lawrence Mitchell
03:04:21
I think they are not here...
Lawrence Mitchell
03:04:38
But Connor might have recalled from last year
Connor Ward
03:05:56
Yeah I remember now
George Bisbas
03:07:12
some temporal blocking :-) as well
Jamie Quinn
03:12:13
FYI - capitaisation is important for tiny.cc url
Lawrence Mitchell
03:13:26
It points here https://rrze-hpc.github.io/layer-condition/
Ilektra Christidi
03:22:03
-m
Tobias Weinzierl
03:27:14
vector efficiency will go down
George Bisbas
03:27:14
register spilling?
Jamie Quinn
03:27:22
More significant boundary effects?