Optimized Einstein@Home App

Michael Hou
Michael Hou
Joined: 15 Oct 06
Posts: 2
Credit: 1060
RAC: 0
Topic 191968

Hi everyone, I have 3 question here.

1. Is the App version on the "Faster all-sky pulsar search" already optimized to those cpu support SSE3?

2. If the App remove the graphic display code, will it run faster?

3. If the graphic is removed, is there a version that is "no graphic" and cpu optimized? (Just like the SETI@Home)

Thank you for answering my questions.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 89

Optimized Einstein@Home App

Bernd says he was unable to gain any speedup using SSE3 over the base SSE code. Since Akos was able to get a benefit with the s4 code and was hired on as a consultant for S5 I'm not sure why they weren't able to use SSE3 rounding to benefit the current app. One possibility is that the faster rounding came at a slight increase in inaccuracy, and with the longer integration time it builds up to a nontrivial level. That speculation is purely a SWAG on my part however.

There's not a no gfx version of the app, and since akos never chopped it out when he was hacking s4 to optimize it I'd assume there's no penalty to having it disabled.

Michael Hou
Michael Hou
Joined: 15 Oct 06
Posts: 2
Credit: 1060
RAC: 0

I see, thank you for

I see, thank you for answering my questions.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

Hello, RE: Bernd says he

Message 48533 in response to message 48531

Hello,

Quote:
Bernd says he was unable to gain any speedup using SSE3 over the base SSE code. Since Akos was able to get a benefit with the S4 code and was hired on as a consultant for S5 I'm not sure why they weren't able to use SSE3 rounding to benefit the current app. One possibility is that the faster rounding came at a slight increase in inaccuracy, and with the longer integration time it builds up to a nontrivial level. That speculation is purely a SWAG on my part however.


The SSE3 trunctaion is absolutely perfect, so there aren't any differences between the results. The SSE3 optimised code is about 15-20% faster.

Quote:
There's not a no gfx version of the app, and since akos never chopped it out when he was hacking s4 to optimize it I'd assume there's no penalty to having it disabled.


You are right.

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4273
Credit: 245297764
RAC: 11812

RE: The SSE3 trunctaion is

Message 48534 in response to message 48533

Quote:
The SSE3 trunctaion is absolutely perfect, so there aren't any differences between the results. The SSE3 optimised code is about 15-20% faster.


Akos, do you refer to the current code or an older one (the one we started S5R1 with or even the 4.37 from S4)? I thougt that in the current code FISTTP is not of much use, and definitely not speeds up the overall computation by 20 or even 15%.

BM

BM

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

Hello

Message 48535 in response to message 48534

Hello Bernd!

Quote:
Quote:
The SSE3 trunctaion is absolutely perfect, so there aren't any differences between the results. The SSE3 optimised code is about 15-20% faster.

Akos, do you refer to the current code or an older one (the one we started S5R1 with or even the 4.37 from S4)? I thougt that in the current code FISTTP is not of much use, and definitely not speeds up the overall computation by 20 or even 15%.


I have found that the current official 4.24 windows application didn't use any SSE3 instructions. I didn't understand the reason of it, but i though that you didn't want to release it because of compatibility or something else. So, i tried to implemet the SSE3 code snipetts that you also have and i found that it is a bit faster. I'm terrible sorry my inactivity, but i don't have any time.

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4273
Credit: 245297764
RAC: 11812

RE: I'm terrible sorry my

Message 48536 in response to message 48535

Quote:
I'm terrible sorry my inactivity, but i don't have any time.

We're quite busy, too, getting the new hierarchical search code running which we intend to use for the next run. I doubt that I will have any time left for any more work on the current code.

BM

BM

daniel
daniel
Joined: 11 Oct 06
Posts: 11
Credit: 4990
RAC: 0

where can i find these aps

Message 48537 in response to message 48536

where can i find these aps and which one is the right one for me

Richard M
Richard M
Joined: 11 Nov 04
Posts: 78
Credit: 249429146
RAC: 935242

RE: where can i find these

Message 48538 in response to message 48537

Quote:
where can i find these aps and which one is the right one for me

You are running it now.

;-)

TTYL
Richard

daniel
daniel
Joined: 11 Oct 06
Posts: 11
Credit: 4990
RAC: 0

o its automatic

Message 48539 in response to message 48538

o its automatic

Richard M
Richard M
Joined: 11 Nov 04
Posts: 78
Credit: 249429146
RAC: 935242

RE: o its automatic Once

Message 48540 in response to message 48539

Quote:
o its automatic

Once the science app is offical it's downloaded automatically.
New science apps that are improved (for whatever reason) are beta tested prior to their release. There is usually a notice of new apps in this forum when they become available and you can help test them on your pc then. However, from what I read in this thread, there doesn't appear to be much chance of that happening in this science run.

TTYL
Richard

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.