Computation Error With Optimized Apps

[Lysia]

Done. Results of the prerelease apps are exactly identical. The results are very similar to the ones of the released SSE3-app (peak power differs slightly, and one of the time_series differs in one digit). I have attached all results (including those from the released apps) and a diff between the nearly identical results.
My result is identical to Urs' for the prerelease.

All "good" apps have the same indices.txt that is produced right at the beginning of the calculation and it differs significantly from the broken app. Although it is of no use after the calculation, it might be the cause of the miscalculation (or the difference has the same origin), so i would start there when searching for the bug.

EDIT: For the comparison of the performance
prerelease SSE2:

Code: [Select]

real    1847m42.197s
user    1703m37.728s
sys     93m4.481sprerelease SSE3:

Code: [Select]

real    1777m46.620s
user    1631m52.455s
sys     93m21.558srelease SSE2:

Code: [Select]

real    1429m15.220s
user    1317m58.278s
sys     86m18.336srelease SSE3:

Code: [Select]

real    1475m25.372s
user    1363m40.233s
sys     86m3.655s

[attachment deleted by admin]

[sunu]

I got all the results from Lysius, Urs and mine and compared each other and I agree with what you've all said above. So to recap:

SSE2 app from Crunch3r's page is not good.

SSE3 app from Crunch3r's page and pre-release apps I've posted above are good.

So for the time being, till Crunc3r makes a new one, everyone who wants a SSE2 app you can use the one attached.

[Urs Echternacht]

Checking results :
at first the ones that are supposed to be ok :

Code: [Select]

.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ls
pulse_503pr_sse2.out  pulse.out_pre  pulse.out_rel_SSE3  pulse.out_SSE3  rescmpAP2
.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ./rescmpAP2 pulse_503pr_sse2.out pulse.out_pre
                \: <ap_signal>14,<pulses>4,<best_pulses>10
                    : <ap_signal>14,<pulses>4,<best_pulses>10
             All Signals: Checked  14, 14 , Strongly Similar
                  Pulses: Checked   4,  4 , Strongly Similar
             Best Pulses: Checked  10, 10 , Strongly Similar
.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ./rescmpAP2 pulse_503pr_sse2.out pulse.out_rel_SSE3
                \: <ap_signal>14,<pulses>4,<best_pulses>10
                    : <ap_signal>14,<pulses>4,<best_pulses>10
             All Signals: Checked  14, 14 , Strongly Similar
                  Pulses: Checked   4,  4 , Strongly Similar
             Best Pulses: Checked  10, 10 , Strongly Similar
.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ./rescmpAP2 pulse_503pr_sse2.out pulse.out_SSE3
                \: <ap_signal>14,<pulses>4,<best_pulses>10
                    : <ap_signal>14,<pulses>4,<best_pulses>10
             All Signals: Checked  14, 14 , Strongly Similar
                  Pulses: Checked   4,  4 , Strongly Similar
             Best Pulses: Checked  10, 10 , Strongly SimilarThey 'are strongly similar.

Now against the ones from that bad release version :

Code: [Select]

.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ls
pulse_503pr_sse2.out  pulse_503rel_sse2.out  pulse.out_pre  pulse.out_rel_SSE2  pulse.out_rel_SSE3  pulse.out_SSE2  pulse.out_SSE3  rescmpAP2
.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ./rescmpAP2 pulse_503pr_sse2.out pulse_503rel_sse2.out
                \: <ap_signal>14,<pulses>4,<best_pulses>10
                    : <ap_signal>40,<pulses>30,<best_pulses>10
             All Signals: Weakly similar or Different.
                  Pulses: Weakly similar or Different.
             Best Pulses: Weakly similar or Different.
.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ./rescmpAP2 pulse_503pr_sse2.out pulse.out_rel_SSE2
                \: <ap_signal>14,<pulses>4,<best_pulses>10
                    : <ap_signal>40,<pulses>30,<best_pulses>10
             All Signals: Weakly similar or Different.
                  Pulses: Weakly similar or Different.
             Best Pulses: Weakly similar or Different.
.../ap_5.03/linux/ap_5.03_sse2_lx32_r03/compare> ./rescmpAP2 pulse_503pr_sse2.out pulse.out_SSE2
                \: <ap_signal>14,<pulses>4,<best_pulses>10
                    : <ap_signal>40,<pulses>30,<best_pulses>10
             All Signals: Weakly similar or Different.
                  Pulses: Weakly similar or Different.
             Best Pulses: Weakly similar or Different.They are obviously (no. of signals) different.

[Lysia]

So for the time being, till Crunc3r makes a new one, everyone who wants a SSE2 app you can use the one attached.

I think I will do some performance comparison tomorrow and run the SSE-version. Might be that the SSE-version of the released app is faster than the SSE2-version of the prerelease. Can you also attach the SSE-version of the prerelease?

[sunu]

The sse app in crunch3r's page is the same we had for testing, it didn't change. And it is slower than the sse2 app.

[PatrickV2]

I got all the results from Lysius, Urs and mine and compared each other and I agree with what you've all said above. So to recap:

SSE2 app from Crunch3r's page is not good.

SSE3 app from Crunch3r's page and pre-release apps I've posted above are good.

So for the time being, till Crunc3r makes a new one, everyone who wants a SSE2 app you can use the one attached.

Okidoki, I installed the app you attached. However, until my BOINC install begins with ap units (none in sight yet), I cannot comment if it works. I do hope I'm not going to mess up more stuff for my wingmen...

Regards, Patrick.

[arkayn]

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?

This is the version that is in the SSE2 combined package. I downloaded it from this site when it was made available.

[attachment deleted by admin]

[sunu]

Patrick it's good, there is no concern.

Arkayn this client is different from both the bad one and the pre-release one. I've started testing it. Tomorrow I'll have the results.

[arkayn]

I wonder if it is the SSE version then.

We really need to rename the files so we can tell what they are without having to run them or look at the code.

[Raistmer]

I wonder if it is the SSE version then.

We really need to rename the files so we can tell what they are without having to run them or look at the code.

In original sources USE_SSE* in stderr defines what SSE level should be used on targed hosts. So you can rename apps based on this value

[sunu]

Arkayn your client is sse2 all right. In a few hours it should finish.

[arkayn]

I started with the OS X versions and renamed them with the revision number so anybody can tell what version they have.

After this is cleaned up, I hope to get the Linux apps renamed as well.

[sunu]

Arkayn your client is good. It's also faster than the pre-release app I've posted above.

Anyone who wants an SSE2 app is advised to get arkayn's file from post #51 above.

[PatrickV2]

Arkayn your client is good. It's also faster than the pre-release app I've posted above.

Anyone who wants an SSE2 app is advised to get arkayn's file from post #51 above.

So that means I should forget about the app you attached? Perhaps you can also edit that post of yours to 'lessen' the confusion?

Regards, Patrick.

[sunu]

I forgot, done!