Welcome to MilkyWay@home

source v0.18

Message boards : Application Code Discussion : source v0.18
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 9566 - Posted: 2 Feb 2009, 20:23:28 UTC
Last modified: 13 Feb 2009, 11:02:27 UTC

Obligatory code release directory link :)

Here's the thread for v0.18 source code stuff. I haven't updated it to the stock app because I'm still testing, but thought it would be a better practice to have more people be able to look at it before i update the stock app.
ID: 9566 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 9577 - Posted: 2 Feb 2009, 22:27:13 UTC - in response to Message 9566.  

just tried to compile it and get this errors:

../astronomy/evaluation_state.c(230): error: class "evaluation_state" has no member "main_integral"
  		fwrite_integral_area(file, es->main_integral);
  		                               ^
../astronomy/evaluation_state.c(231): error: class "evaluation_state" has no member "number_cuts"
  		fprintf(file, "cuts: %d\n", es->number_cuts);
  		                                ^
../astronomy/evaluation_state.c(232): error: class "evaluation_state" has no member "number_cuts"
  		for (i = 0; i < es->number_cuts; i++) {
  		                    ^
../astronomy/evaluation_state.c(233): error: class "evaluation_state" has no member "cuts"
  			fwrite_integral_area(file, es->cuts[i]);
  			                               ^
../astronomy/evaluation_state.c(265): error: class "evaluation_state" has no member "main_integral"
  		fread_integral_area(file, es->main_integral);
  		                              ^
../astronomy/evaluation_state.c(267): error: class "evaluation_state" has no member "number_cuts"
  		fscanf(file, "cuts: %d\n", &es->number_cuts);
  		                                ^
../astronomy/evaluation_state.c(268): error: class "evaluation_state" has no member "number_cuts"
  		for (i = 0; i < es->number_cuts; i++) {
  		                    ^
../astronomy/evaluation_state.c(269): error: class "evaluation_state" has no member "cuts"
  			fread_integral_area(file, es->cuts[i]);
  			                              ^


mic.


ID: 9577 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 9578 - Posted: 2 Feb 2009, 22:49:32 UTC - in response to Message 9577.  

Ohh I think I see what went on. I should have an update later tonight :) The checkpoint functions didn't get updated to the most recent version.
ID: 9578 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10469 - Posted: 13 Feb 2009, 10:59:56 UTC - in response to Message 9578.  

The mw_v0.18b code should fix this problem, and it also has some additional performance improvements.
ID: 10469 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10480 - Posted: 13 Feb 2009, 12:09:06 UTC - in response to Message 10469.  

The mw_v0.18b code should fix this problem, and it also has some additional performance improvements.


another one:
../searches/search_parameters.c(237): error: expected a ")"
  		fprintf(data_file, "%s: %1.2lf\n", BOINC_APP_NAME, BOINC_APP_VERSION);
  		                                                   ^

(the little arrow should be pointing to the B of BOINC_APP_VERSION)

this same line worked fine in the pre 18 versions...

mic.


ID: 10480 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10497 - Posted: 13 Feb 2009, 15:20:33 UTC - in response to Message 10480.  

...solved.

I called the BOINC_APP_VERSION "0.18b", which didn't fit for the "%s: %1.2lf\n"

mic.


ID: 10497 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10499 - Posted: 13 Feb 2009, 15:58:46 UTC - in response to Message 10469.  

Ok, got it compiled. But that's about it - the numbers for fitness are way out of line and wedge 86 errors out with a Floating point exception.

Here are some numbers for wedge 79:

reference: fitness: -2.946683357256020 (from Travis' post)
v18 intel compiled: fitness: -2.993533078315163
v18 gcc compiled: fitness: -2.993533078315124 (without optimisation, used the included make.linux)

Intel compiler and gcc are nearly similar, so I guess it's not the compiler...
mic.


ID: 10499 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 10504 - Posted: 13 Feb 2009, 16:55:41 UTC - in response to Message 10499.  
Last modified: 13 Feb 2009, 17:01:28 UTC

Ok, got it compiled. But that's about it - the numbers for fitness are way out of line and wedge 86 errors out with a Floating point exception.

Here are some numbers for wedge 79:

reference: fitness: -2.946683357256020 (from Travis' post)
v18 intel compiled: fitness: -2.993533078315163
v18 gcc compiled: fitness: -2.993533078315124 (without optimisation, used the included make.linux)

Intel compiler and gcc are nearly similar, so I guess it's not the compiler...

I guess, the culprit lies in the calculate_probabilities function (in evaluation_optimzed.c). In the beginning there are the lines:
        sinb = sin(integral_point[1] / deg);
        sinl = sin(integral_point[0] / deg);
        cosb = cos(integral_point[1] / deg);
        cosl = cos(integral_point[[b]1[/b]] / deg);
The last line should be:
cosl = cos(integral_point[0] / deg);

@ Travis:
This are exactly the lines I was referring to for the code change suggestion in my last PM. It should bring slightly more performance gains than what is done in the 0.18b.
ID: 10504 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10507 - Posted: 13 Feb 2009, 18:49:05 UTC - in response to Message 10504.  
Last modified: 13 Feb 2009, 18:51:02 UTC

I guess, the culprit lies in the calculate_probabilities function (in evaluation_optimzed.c). In the beginning there are the lines:
        sinb = sin(integral_point[1] / deg);
        sinl = sin(integral_point[0] / deg);
        cosb = cos(integral_point[1] / deg);
        cosl = cos(integral_point[[b]1[/b]] / deg);
The last line should be:
cosl = cos(integral_point[0] / deg);

@ Travis:
This are exactly the lines I was referring to for the code change suggestion in my last PM. It should bring slightly more performance gains than what is done in the 0.18b.


Nearly. The fitness is back in line, but wedge 86 still errors out.

[edit]Travis, can you post reference numbers for the wedges 20/21...[/edit]
mic.


ID: 10507 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10521 - Posted: 13 Feb 2009, 20:59:35 UTC - in response to Message 10507.  

I'll see why 86 is giving an error. Should have what stripes 20 and 21 should be spitting out in a bit.
ID: 10521 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10522 - Posted: 13 Feb 2009, 20:59:35 UTC - in response to Message 10507.  
Last modified: 13 Feb 2009, 21:08:16 UTC

I'll see why 86 is giving an error. Should have what stripes 20 and 21 should be spitting out in a bit.

edit: I put the results for stripes 20 and 21 in the stickied "source code and sample workunits" thread. not quite sure what's up with stripe 86 but it shouldn't be too hard to track down.
ID: 10522 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10533 - Posted: 13 Feb 2009, 21:42:50 UTC - in response to Message 10522.  

Ok fixed the issue with stripe 86. The astronomy-parameters-86.txt file was messed up. I put v0.18c in the code releases directory. This also fixes the cosl = cos(integral_point[1] / deg); problem.
ID: 10533 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10535 - Posted: 13 Feb 2009, 21:56:40 UTC - in response to Message 10533.  
Last modified: 13 Feb 2009, 22:03:57 UTC

Ok fixed the issue with stripe 86. The astronomy-parameters-86.txt file was messed up. I put v0.18c in the code releases directory. This also fixes the cosl = cos(integral_point[1] / deg); problem.


Ääähhhmmm... the test-files for 20 21 and 79 seem to be exactly the same...

[edit]would be nice if you could make separate archives for code and test-files, if the code changes often we don't need to dl test-files each time...[/edit]
mic.


ID: 10535 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10537 - Posted: 13 Feb 2009, 22:01:12 UTC - in response to Message 10535.  

wha? really?

oh geez i see the problem. I cut and pasted wrong in the release script ;P

OK v0.18c should be fixed with the right test files for 20 and 21
ID: 10537 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10540 - Posted: 13 Feb 2009, 22:12:39 UTC - in response to Message 10537.  

LOL took me some time to realize why to out was way off...
Already thought that glass of Lagavulin was too much...
mic.


ID: 10540 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10549 - Posted: 13 Feb 2009, 22:34:10 UTC - in response to Message 10540.  

LOL took me some time to realize why to out was way off...
Already thought that glass of Lagavulin was too much...


mm that's some good stuff :D but personally, my favorite is Laphroaig
ID: 10549 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10558 - Posted: 13 Feb 2009, 23:23:13 UTC - in response to Message 10549.  

LOL took me some time to realize why to out was way off...
Already thought that glass of Lagavulin was too much...


mm that's some good stuff :D but personally, my favorite is Laphroaig


It's the Pedro Ximénez Destiller's edition... ;)

I like Laphroaig too - especially the Quarter Cask.

mic.


ID: 10558 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10560 - Posted: 13 Feb 2009, 23:28:34 UTC

back on topic...

you can check out the first v18 results on this host.


mic.


ID: 10560 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 10584 - Posted: 14 Feb 2009, 0:40:29 UTC - in response to Message 10560.  

back on topic...

you can check out the first v18 results on this host.



aww.

I don't suppose you noticed any performance improvement with v0.18?
ID: 10584 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile speedimic
Avatar

Send message
Joined: 22 Feb 08
Posts: 260
Credit: 57,387,048
RAC: 0
Message 10585 - Posted: 14 Feb 2009, 0:43:04 UTC - in response to Message 10584.  

not that much.
about 20-30 sec faster...
mic.


ID: 10585 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · 3 · Next

Message boards : Application Code Discussion : source v0.18

©2024 Astroinformatics Group