Computation Error With Optimized Apps

[Lysia]

Lysius can you also run the two clients attached below?. They are the ones we've tested before release.

Running for about 50minutes now, no pulse has been found so far. This SSE2-App seems better, but final results will take something like another 24 hours.

[sunu]

Ok. I'm still running the two apps I got yesterday from crunch3r's page. Next I'll run the two apps I attached above and my 64bit sse3 app. With this I've run more than a 100 WUs without a single problem, not even a third wingman-arbiter was needed in any of them AFAIK. This will be our reference.

[Lysia]

I have used the 32-bit SSE3-App from crunch3r's Page (sha1sum: 2551c1a0553aa696327bd26d0bd4866446bb9018) on my Desktop PC (E6750) with about 15 WUs, 5 left for validation, the rest validated without a problem. So this app seems to be OK, too (at least much better than the SSE2).

[Lysia]

I have recognized, that Indices.txt (that is generated while running AP; is there any information on what this file, fold.dat and wisdom.dat do?) is the same for both SSE3-apps and the "correct" SSE2-app, but differs for the broken SSE2-app. I have attached both versions.

[attachment deleted by admin]

[Urs Echternacht]

Lysius, sunu, i can see the same. The linux 32bit SSE2 release version is definitly broken. My tests are still running, but this far into the wu it is obvious. After i saw sunu's post about the last prerelease version earlier today, i looked that one up and started parallel another test. After some more hours the intermediate pulse.out files are too different to compare to each other.
Results should be ready by tomorrow.

[sunu]

Urs, it's not only broken, it's more than that. 
It gives results that they are not only wrong but inconsistent. My results are weakly similar or different compared to Lysius's.
My sse3 (from crunch3r's page) results are strongly similar to Lysius's results.

Now I've started the two apps I've posted above (fortunately I keep everything from my tests) and then the 64bit sse3 app. We've tested these apps, but no harm doing them again. By tomorrow evening I think we'll finish.

Lysius I don't know what the other files do exactly, maybe Joe Segur will enlighten us. Nevertheless pulse.out is the one returned to berkeley with the results so it is the file that matters.

[Raistmer]

fold.dat is part of checkpoint set, wisdom.dat is FFTW library wisdom file (FFTW try to keep best FFT algorithm between its runs), indices.txt - some info file about what areas were blanked in particular task.
Result file is pulse.out indeed.

[sunu]

Thanks raistmer.

[arkayn]

I have pulled my SSE2 package until you get to the root cause of the problem.

[sunu]

Arkayn it's possible that your SSE2 package has already a good AP client. What is the date of the AP client in your package? Can you attach it here?

[Urs Echternacht]

My testruns with the different SSE2 AstroPulse applications (prerelease and release) just finished. The results are different.

add: Send Crunch3r a pm about the testresults.

[attachment deleted by admin]

[Lysia]

The release apps are much faster than the prerelease (both SSE2 and SSE3). Both prerelease apps are at 1550m CPU-Time now, which is more than the release-apps took and they still did not find the pulses at <dm>-14010</dm> yet.

Because the results of the SSE2-app are quite random, I would guess at least some memory initialization has been "optimized" away.

[PatrickV2]

I guess I'm a bit late to the party here, but it's only recently that I noticed that a Linux machine of mine was _very_ quickly processing away the astropulse units it got (the machine is mostly unattended, so I didn't notice this immediately).

All of them finished within about 15 seconds max. The log from one of these:

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<message>
process exited with code 193 (0xc1, -63)
</message>
<stderr_txt>
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
AstroPulse v. 5.03
Linux 32 bit
FFTW   USE_CONVERSION_OPT   USE_SSE2->SSE3_EMU   
SIGILL: illegal instruction
Stack trace (7 frames):
[0x80797d4]
[0xb7f62400]
[0x8059020]
[0x8049e14]
[0x8048d58]
[0x8217824]
[0x8048131]

Exiting...

</stderr_txt>
]]>


(Link: http://setiathome.berkeley.edu/result.php?resultid=1192493744)

I'm quite (but not 100%) sure I installed the SSE2 apps. I have now installed the SSE apps, but have to wait until an ap unit is going to be crunched, to see if this fixed it...

Regards, Patrick.

[Urs Echternacht]

[size=0]I guess I'm a bit late to the party here, but it's only recently that I noticed that a Linux machine of mine was _very_ quickly processing away the astropulse units it got (the machine is mostly unattended, so I didn't notice this immediately).

All of them finished within about 15 seconds max. The log from one of these:

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<message>
process exited with code 193 (0xc1, -63)
</message>
<stderr_txt>
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
AstroPulse v. 5.03
Linux 32 bit
FFTW   USE_CONVERSION_OPT    USE_SSE2->SSE3_EMU
SIGILL: illegal instruction
Stack trace (7 frames):
[0x80797d4]
[0xb7f62400]
[0x8059020]
[0x8049e14]
[0x8048d58]
[0x8217824]
[0x8048131]

Exiting...

</stderr_txt>
]]>


(Link: http://setiathome.berkeley.edu/result.php?resultid=1192493744)

I'm quite (but not 100%) sure I installed the SSE2 apps. I have now installed the SSE apps, but have to wait until an ap unit is going to be crunched, to see if this fixed it...

Regards, Patrick.[/size]

Patrick, thanks for reporting your problem. I colorized the part in the quoted output of your post that identifies the AstroPulse SSE2 application version in question. This is the bad one definitely. To use the SSE version until this problem gets resolved is recommended.

[Claggy]

[size=0]I guess I'm a bit late to the party here, but it's only recently that I noticed that a Linux machine of mine was _very_ quickly processing away the astropulse units it got (the machine is mostly unattended, so I didn't notice this immediately).

All of them finished within about 15 seconds max. The log from one of these:

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<message>
process exited with code 193 (0xc1, -63)
</message>
<stderr_txt>
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
AstroPulse v. 5.03
Linux 32 bit
FFTW   USE_CONVERSION_OPT    USE_SSE2->SSE3_EMU
SIGILL: illegal instruction
Stack trace (7 frames):
[0x80797d4]
[0xb7f62400]
[0x8059020]
[0x8049e14]
[0x8048d58]
[0x8217824]
[0x8048131]

Exiting...

</stderr_txt>
]]>


(Link: http://setiathome.berkeley.edu/result.php?resultid=1192493744)

I'm quite (but not 100%) sure I installed the SSE2 apps. I have now installed the SSE apps, but have to wait until an ap unit is going to be crunched, to see if this fixed it...

Regards, Patrick.[/size]

Patrick, thanks for reporting your problem. I colorized the part in the quoted output of your post that identifies the AstroPulse SSE2 application version in question. This is the bad one definitely. To use the SSE version until this problem gets resolved is recommended.

Speedimic and Officedude both had this error about 15 days ago on hosts 4462906 and 3370265, Speedimic tried to redownload the same SSE2 app,
but the link was Broken by then, suspect that might have been from Lunatics, Both have swapped to the SSE app.

Claggy