Thank you guys for the feedback. I believe I've been able to pull a little more speed out of the 64-bit app. It's hard to tell from the quicker test units how much faster it is but the results were valid. I'm going to run the client overnight to see if there's a noticable speed difference and make sure it's all ok. If it checks out I'll probably submit the code over on Simon's developer board tomorrow.

Side note: It is supposed to be "EM64T", sorry about that. Got it mixed with AMD64 =)

Thanks,
Bob Delkhoon (DeNitro)

On the invalid results I got 3 using the 64bit compile, I recall never seeing any errors with the 1.2 version of the Generic SSE2 code, Maybe You should try that as It may be something in the 1.3 version code.

On the EM64T that is Intels fault as they'd rather not be reminded that a better solution to 64bit cpus came from outside Intel. So that's ok as You are still workin on It.

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Thank you guys for the feedback. I believe I've been able to pull a little more speed out of the 64-bit app. It's hard to tell from the quicker test units how much faster it is but the results were valid. I'm going to run the client overnight to see if there's a noticable speed difference and make sure it's all ok. If it checks out I'll probably submit the code over on Simon's developer board tomorrow.

Side note: It is supposed to be "EM64T", sorry about that. Got it mixed with AMD64 =)

Thanks,
Bob Delkhoon (DeNitro)

Bob, I went back to Simons(Chicken) 5.15 SSE2 app version 1.0.0.2 on My PC that uses XP Pro sp2(32bit) which according to the file is version 1.0.0.2 as It didn't make any errors, Hopefully Simon still has the 1.0.0.2 source code, As 1.0.0.3 has made errors and produced invalid results, If Your 64bit app produces any more errors and invalid results on the XP x64 PC It'll be going back to the 32bit 1.0.0.2 app also. Mind You I do like Your 64bit app, Just not the Errors or the Invalid Results.

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Can you think of a function that might benefit from 16 x 4 simd regs (in seti)? Would it keep the register pressure down somewhere? Your most complex simd seti use seems to be your chirp sse3, but compiler seemed to use only registers and not store values in temp memory when I looked at the assembly output.

Had you not brought up my SSE3 chirp, I would have said the same thing--the main loop's data dependencies aren't particularly serial. Given the length of the loop, I can't guarantee how well x86 processors will schedule around instruction latencies, so if I had extra registers, I'd try unrolling it wider to see what effect it had. It's still less certain than on PowerPC, where I'm dealing with an in-order vector unit and can tell what's a win just from examining the generated assembly.

As for the rest of the functions, I doubt too many of them would benefit much, if at all.

[...]
So, EricVonDaniken, have you signed up on Simon's board yet?

Yup, he just did.

Eric, you've got access to the pre-release forum now. Check back on lunatics.at, you will see an extra forum category (as well as an extra Downloads section).

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

[...]
So, EricVonDaniken, have you signed up on Simon's board yet?

Yup, he just did.

Eric, you've got access to the pre-release forum now. Check back on lunatics.at, you will see an extra forum category (as well as an extra Downloads section).

Regards,
Simon.

Hey Simon, Your 1.3 SSE2 Generic has handed out Invalid Results, I went back to 1.2 on one PC and may on the other one that is using the 64 bit app made by Bob If It makes any more Invalid Results and Yours did 4 in a row.

I also tried registering for an account on Your forum and I tried both My email addresses and somehow both don't like Your confirmation emails and so I can read, But not login or post.

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Hi Joker,

interesting that it's returning invalid results - haven't had many reports of such things happening (other than from massive overclocks that went just that little bit too far). Are you sure it's the 1.3 app? The 1.2 and 1.3 only differ in that 1.3 adds CPU MHz detection, not in any other significant way. The function used is part of IPP, and seems to work reliably enough.

I'll check out your registration troubles and manually unlock your account. Sorry it's got to be a bit complicated to sign up - there have already been a few unsavory spam messages posted before, so now there are a couple of measures against that in place, of which the registration confirmation is only one.

Should still work, but sometimes it just doesn't :o)

Regards,
Simon.

<edit>Account is activated. Enjoy!</edit>

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Hi Joker,

interesting that it's returning invalid results - haven't had many reports of such things happening (other than from massive overclocks that went just that little bit too far). Are you sure it's the 1.3 app? The 1.2 and 1.3 only differ in that 1.3 adds CPU MHz detection, not in any other significant way. The function used is part of IPP, and seems to work reliably enough.

I'll check out your registration troubles and manually unlock your account. Sorry it's got to be a bit complicated to sign up - there have already been a few unsavory spam messages posted before, so now there are a couple of measures against that in place, of which the registration confirmation is only one.

Should still work, but sometimes it just doesn't :o)

Regards,
Simon.

<edit>Account is activated. Enjoy!</edit>

I switched back to 1.2 on My PC Joker2(It used to have 1.3 on It under XP Pro sp2 x32), Joker3 has the 64bit app that is based on 1.3 under XP x64.

In any case, Yes, I've had It happen under both 1.3 and the 64bit app.

Here's the 32bit app as I posted earlier, plus there are a couple of pure 64bit ones also.
http://setiathome.berkeley.edu/result.php?resultid=378753838

<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
Optimized Windows SETI@Home Enhanced application
Version info: Windows SSE2 32-bit V5.15 'Chicken Good!' (R-1.3|+freq|xW)
Compiled by Simon Zadra (KWSN - Chicken of Angnor) - Member of the Knights who say Ni! (http://www.kwsn.net)
Download Updates at: http://www.zadra.org/seti_enhanced/

CPU real speed: 2368 MHz

Work Unit Info
True angle range: 0.426463
Optimized Windows SETI@Home Enhanced application
Version info: Windows SSE2 32-bit V5.15 'Chicken Good!' (R-1.3|+freq|xW)
Compiled by Simon Zadra (KWSN - Chicken of Angnor) - Member of the Knights who say Ni! (http://www.kwsn.net)
Download Updates at: http://www.zadra.org/seti_enhanced/

CPU real speed: 2368 MHz

Work Unit Info
True angle range: 0.426463

</stderr_txt>

64bit app that was made from Your 1.3 app.
http://setiathome.berkeley.edu/result.php?resultid=378774946
http://setiathome.berkeley.edu/result.php?resultid=378774917

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Hmm, ever since I installed the 1.3 version of the sse3 app on my computer, not one invalid result. This is on a Dell computer, p4 2.8ghz 512mb RAM with Winxp Home SP2. Crunching like a mad man. Question tho, on CPU-Z, under the instruction set, it lists that EM64T. Do I need to be on a 64 bit operating system to take advantage of that function, or can I do anything with it being on 32 bit?

---

Hmm, ever since I installed the 1.3 version of the sse3 app on my computer, not one invalid result. This is on a Dell computer, p4 2.8ghz 512mb RAM with Winxp Home SP2. Crunching like a mad man. Question tho, on CPU-Z, under the instruction set, it lists that EM64T. Do I need to be on a 64 bit operating system to take advantage of that function, or can I do anything with it being on 32 bit?

You need XP x64 at the very least, No 32bit XP OS will run a 64bit app or program.

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Hmm, ever since I installed the 1.3 version of the sse3 app on my computer, not one invalid result. This is on a Dell computer, p4 2.8ghz 512mb RAM with Winxp Home SP2. Crunching like a mad man. Question tho, on CPU-Z, under the instruction set, it lists that EM64T. Do I need to be on a 64 bit operating system to take advantage of that function, or can I do anything with it being on 32 bit?

You need XP x64 at the very least, No 32bit XP OS will run a 64bit app or program.

Ahh, ok. Thanks. I just wasn't too sure. Don't know enough about all that stuff. Hehe. Thanks again.

---

Hi Joker,

interesting that it's returning invalid results - haven't had many reports of such things happening (other than from massive overclocks that went just that little bit too far). Are you sure it's the 1.3 app? The 1.2 and 1.3 only differ in that 1.3 adds CPU MHz detection, not in any other significant way. The function used is part of IPP, and seems to work reliably enough.

I'll check out your registration troubles and manually unlock your account. Sorry it's got to be a bit complicated to sign up - there have already been a few unsavory spam messages posted before, so now there are a couple of measures against that in place, of which the registration confirmation is only one.

Should still work, but sometimes it just doesn't :o)

Regards,
Simon.

<edit>Account is activated. Enjoy!</edit>

I made a thread in Your forum on this subject, Link.

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

I compiled a second test. It will probably be the last test I put out. I'm going to post the code over on Simon's page and most likely just submit changes to him there to keep things organized. The results seem pretty good so far.

This build is about 80% smaller and 5%-6% faster on the tests I did.
http://tiger.towson.edu/~bdelkh1/setiathome-5.15-DeNitro-em64t_test2.rar

Results may very, I only have 1 x64 machine to test code on so it's hard to tell how it will run on other CPU types.

PLEASE make sure to try this on a test work unit and validate the result before putting this into BOINC!

Hey Simon, Your 1.3 SSE2 Generic has handed out Invalid Results, I went back to 1.2 on one PC and may on the other one that is using the 64 bit app made by Bob If It makes any more Invalid Results and Yours did 4 in a row.

I also tried registering for an account on Your forum and I tried both My email addresses and somehow both don't like Your confirmation emails and so I can read, But not login or post.

This most likely doesn't apply here, but a few months back I had one of my machines that only crunches start randomly giving out errors. I ran Memtest86 on it to test the RAM and sure enough, after running for about 30 minutes I saw errors start appearing. There was only 2 sticks of ram in the system so with trial and error I figured out which was bad and left only the good one in. It's probably a slim chance the RAM is the problem, but it's pretty easy to check with that software.

Thanks,
Bob Delkhoon

I compiled a second test. It will probably be the last test I put out. I'm going to post the code over on Simon's page and most likely just submit changes to him there to keep things organized. The results seem pretty good so far.

This build is about 80% smaller and 5%-6% faster on the tests I did.
http://tiger.towson.edu/~bdelkh1/setiathome-5.15-DeNitro-em64t_test2.rar

Results may very, I only have 1 x64 machine to test code on so it's hard to tell how it will run on other CPU types.

PLEASE make sure to try this on a test work unit and validate the result before putting this into BOINC!

Hey Simon, Your 1.3 SSE2 Generic has handed out Invalid Results, I went back to 1.2 on one PC and may on the other one that is using the 64 bit app made by Bob If It makes any more Invalid Results and Yours did 4 in a row.

I also tried registering for an account on Your forum and I tried both My email addresses and somehow both don't like Your confirmation emails and so I can read, But not login or post.

This most likely doesn't apply here, but a few months back I had one of my machines that only crunches start randomly giving out errors. I ran Memtest86 on it to test the RAM and sure enough, after running for about 30 minutes I saw errors start appearing. There was only 2 sticks of ram in the system so with trial and error I figured out which was bad and left only the good one in. It's probably a slim chance the RAM is the problem, but it's pretty easy to check with that software.

Thanks,
Bob Delkhoon

I got this one when I started up the new app of yours for some reason, It started on a new WU after this one, So all I can do is watch. I know You said Test WUs, But this one seems different. Also checked the ram and It's good, So I'm stumped, But then I'm not a programmer. And I never had a problem with 1.2 code, I'd still ask If He has source code for 1.2 still. Oh and My ram is Registered/ECC PC3200 ram(ECC is on too) as It's a dual opteron PC with a 270 cpu in one socket.

http://setiathome.berkeley.edu/result.php?resultid=378800706
<message>
- exit code -1073741811 (0xc000000d)
</message>
<stderr_txt>
Optimized Windows SETI@Home Enhanced application
Version info: Windows EMT64 64-bit V5.15 Test 1
Modified & Compiled by Bob Delkhoon (DeNitro)
Source downloaded from Simon Zadra (KWSN - Chicken of Angnor)

CPU real speed: 2368 MHz

Work Unit Info
True angle range: 0.421941
Optimized Windows SETI@Home Enhanced application
Version info: Windows EMT64 64-bit V5.15 Test 1
Modified & Compiled by Bob Delkhoon (DeNitro)
Source downloaded from Simon Zadra (KWSN - Chicken of Angnor)

CPU real speed: 2368 MHz

Work Unit Info
True angle range: 0.421941

</stderr_txt>

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Clockspeed is now at 2360MHz(cpu-z 1.36 says 2355MHz) instead of 2368MHz, Done on the fly with Ntune as It's running on My K8N-DL(1008 Bios), Ram is two dimms 1Gb each of SuperTalent D32RB1GW PC3200 ram, One of these days I'll get some Corsair low latency ram for the K8N-DL(Before It's gone). The ram is approved by Supermicro for use with the 2200 chipset(Nforce Pro chipset). Voltages and Timings are done in the Bios, HT Link Multiplier(4x) is adjusted in Ntune along with the Hypertransport Link(944)/Memory Bus(391), Ram is set in the Bios to 166MHz instead of 200MHz and so actual ram speed is 196.3MHz and Ram Timing is at 8-4-4-2.5(set in Bios).

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

I compiled a second test. It will probably be the last test I put out. I'm going to post the code over on Simon's page and most likely just submit changes to him there to keep things organized. The results seem pretty good so far.

This build is about 80% smaller and 5%-6% faster on the tests I did.
http://tiger.towson.edu/~bdelkh1/setiathome-5.15-DeNitro-em64t_test2.rar

Results may very, I only have 1 x64 machine to test code on so it's hard to tell how it will run on other CPU types.

PLEASE make sure to try this on a test work unit and validate the result before putting this into BOINC!

Hey Simon, Your 1.3 SSE2 Generic has handed out Invalid Results, I went back to 1.2 on one PC and may on the other one that is using the 64 bit app made by Bob If It makes any more Invalid Results and Yours did 4 in a row.

I also tried registering for an account on Your forum and I tried both My email addresses and somehow both don't like Your confirmation emails and so I can read, But not login or post.

This most likely doesn't apply here, but a few months back I had one of my machines that only crunches start randomly giving out errors. I ran Memtest86 on it to test the RAM and sure enough, after running for about 30 minutes I saw errors start appearing. There was only 2 sticks of ram in the system so with trial and error I figured out which was bad and left only the good one in. It's probably a slim chance the RAM is the problem, but it's pretty easy to check with that software.

Thanks,
Bob Delkhoon

Ok Bob, I've gone back to 1.2(32bit app) on My XP x64 PC. This is the one errored WU Your Test2 made last night. If You do make another attempt, Please make It from the 1.2 code base.

http://setiathome.berkeley.edu/result.php?resultid=378800747
message>
- exit code -1073741811 (0xc000000d)
</message>
<stderr_txt>
Optimized Windows SETI@Home Enhanced application
Version info: Windows EM64T 64-bit V5.15 Test 2
64-bit Compatibility Edits and Compile by Babak Delkhoon (DeNitro)
Based on Optimized Source Edits by Simon Zadra (KWSN - Chicken of Angnor)

CPU real speed: 2360 MHz

Work Unit Info
True angle range: 0.426464

</stderr_txt>

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

Joker,

again, the only code change from 1.2 to 1.3 was the insertion of CPU MHz detection.

There is no "1.2 Code base" as such :o)

HTH,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Joker,

again, the only code change from 1.2 to 1.3 was the insertion of CPU MHz detection.

There is no "1.2 Code base" as such :o)

HTH,
Simon.

Well in any case I've gone back to 1.2 and I'll stay with It until something better comes out, that is not error prone.

And Yeah I'm beginning to think the extra code which has said My cpu is about 1800MHz and about 5900MHz is at fault and I'd remove It as It's neat, But doesn't help in the crunching to Me.

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

bump

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

I'm using Boincview 1.2.5 to keep track of what WU passess or fails and now both PCs are using (setiathome-5.15-kwsn-sse2.exe) v1.2 as the extra code in v1.3 may be causing the errors and slowing 1.3 down too boot, I've notice that once I switched back to 1.2 that My RAC started to climb and My pending WU's haven't changed really in the mean time. Once the WU known to Boincview as: 9/16/2006 12:42:29 PM 13jn02aa.24708.4769.903410.3.48_1(See link below) passes out of Boincviews buffer, I'll post a screenshot both here and at setiathome for all to see. So Simon(Chicken) could You please remove that cpu speed code as It takes up space and has said My cpu runs as slow as 1808MHz(Ain't possible as It's a 2.0GHz cpu) or as high as 5920MHz(Impossible) and so It's also inacurrate and possibly is whats causing the errors in the results being generated.

http://setiathome.berkeley.edu/result.php?resultid=378800674

---

CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

In a word, no.

I'm sorry you're having trouble with it, but - 10005 Windows app downloads say differently.

I won't remove that code, as I find it pretty useful. In fact, there will be a little more and more specific output in a future release version about your CPU and its capabilities.

Should you need any specific 1.2 release version, I still have the .exe files.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information