@TBar

I suppose you can't use the same AVX2 FFTW library with a SSE41 CPU although FFTW hints that you can;

If the CPU does not support AVX SIMD extensions you must compile fftw-3.3.6-bl1 only with --enable-sse2

So, on your machine there isn't any difference with run-times between AVX & AVX2?

I confirm that there are no signifcant difference in execution time when use AVX2 extension.

It would be interesting to see if the same happens using new Skylake and Kabylake i7 processors.

Gianfranco

---

I don't want to believe, I want to know!

Same test on a Skylake....

KWSN-Darwin-MBbench v2.1.07
Running on domisses-MacBook-Pro.local at Mon Jan 23 10:17:12 2017
---------------------------------------------------
Starting benchmark run...
---------------------------------------------------
Listing wu-file(s) in /testWUs :
18dc09ah.26284.16432.6.33.125.wu reference_work_unit_r3215.wu

Listing executable(s) in /APPS :
MBv8_8.22r3588_avx2_x86_64-apple-darwin

Listing executable in /REF_APPs :
MBv8_8.06r3366_avx_x86_64-apple-darwin
---------------------------------------------------
Current WU: 18dc09ah.26284.16432.6.33.125.wu
---------------------------------------------------
Running default app with command : MBv8_8.06r3366_avx_x86_64-apple-darwin
1466.91 real 1462.82 user 1.52 sys
Elapsed Time: ………………………………… 1467 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3588_avx2_x86_64-apple-darwin
1219.52 real 1215.32 user 1.72 sys
Elapsed Time : ……………………………… 1220 seconds
Speed compared to default : 120 %
-----------------
Comparing results
------------- R1:R2 ------------ ------------- R2:R1 ------------
Exact Super Tight Good Bad Exact Super Tight Good Bad
Spike 0 0 0 0 0 0 0 0 0 0
Autocorr 0 0 0 0 0 0 0 0 0 0
Gaussian 0 0 0 0 0 0 0 0 0 0
Pulse 0 0 0 0 1 0 0 0 0 0
Triplet 0 1 1 1 2 0 1 1 1 6
Best Spike 0 0 0 0 1 0 0 0 0 1
Best Autocorr 0 0 0 0 1 0 0 0 0 1
Best Gaussian 1 1 1 1 0 1 1 1 1 0
Best Pulse 0 0 0 0 1 0 0 0 0 1
Best Triplet 0 0 0 0 1 0 0 0 0 1
---- ---- ---- ---- ---- ---- ---- ---- ---- ----
1 2 2 2 7 1 2 2 2 10

Unmatched signal(s) in R1 at line(s) 359 376 393 419 435 473 498
Unmatched signal(s) in R2 at line(s) 359 376 393 410 427 444 461 477 515 539
For R1:R2 matched signals only, Q= 100.0%
Result : Different.
---------------------------------------------------
Done with 18dc09ah.26284.16432.6.33.125.wu.
Current WU: reference_work_unit_r3215.wu
---------------------------------------------------
Running default app with command : MBv8_8.06r3366_avx_x86_64-apple-darwin
1117.35 real 1113.86 user 1.11 sys
Elapsed Time: ………………………………… 1117 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3588_avx2_x86_64-apple-darwin
986.61 real 983.01 user 1.18 sys
Elapsed Time : ……………………………… 987 seconds
Speed compared to default : 113 %
-----------------
Comparing results
------------- R1:R2 ------------ ------------- R2:R1 ------------
Exact Super Tight Good Bad Exact Super Tight Good Bad
Spike 0 0 0 0 13 0 0 0 0 0
Autocorr 0 0 0 0 1 0 0 0 0 0
Gaussian 0 0 0 0 6 0 0 0 0 0
Pulse 0 0 0 0 0 0 0 0 0 0
Triplet 0 0 0 0 2 0 0 0 0 0
Best Spike 0 0 0 0 1 0 0 0 0 1
Best Autocorr 0 0 0 0 1 0 0 0 0 1
Best Gaussian 0 0 0 0 1 0 0 0 0 1
Best Pulse 0 0 0 0 1 0 0 0 0 1
Best Triplet 0 0 0 0 1 0 0 0 0 0
---- ---- ---- ---- ---- ---- ---- ---- ---- ----
0 0 0 0 27 0 0 0 0 4

Unmatched signal(s) in R1 at line(s) 240 257 273 289 305 321 337 353 369 385 401 417 433 449 466 482 499 526 553 580 607 634 661 677 694 720 745
Unmatched signal(s) in R2 at line(s) 240 256 273 299
Result : Different.
---------------------------------------------------
Done with reference_work_unit_r3215.wu.

---

Same test on a Skylake....

Interesting. The AVX2 CPU App you tested was the one in the folder named AVX2_Test2.zip? The AVX2_Test2 was compiled on the 21st and used the AVX2 FFTW & AVX2 APP.
Strange it produced different results than the two previous AVX CPU Apps.

While trying different compiles, I decided to try the Intel iGPU App compiled to use AVX again. I tried it last year and it wasn't any better than the iGPU App compiled to use ssse3.
Try it again and see if it's any better than the last Intel iGPU App which was MBv8_8.22r3586_Intel_ssse3_x86_64-apple-darwin; Intel-AVX.zip

It maybe better TBar to only test what your rig is actually capable of, trying to do things that it isn't really capable of IMHO isn't helping matters at all and could very likely be counter productive. ;-)

Remember the KISS principle.

Cheers.

It maybe better TBar to only test what your rig is actually capable of, trying to do things that it isn't really capable of IMHO isn't helping matters at all and could very likely be counter productive. ;-)

Remember the KISS principle.

Cheers.

So....Who do you suggest trying to compile the Mac Intel iGPU App?
You are aware the Mac CPU AVX App works Extremely well aren't you? Who do you suggest compile that App that wouldn't exist if I hadn't of compiled it?

Same test on a Skylake....

Interesting. The AVX2 CPU App you tested was the one in the folder named AVX2_Test2.zip? The AVX2_Test2 was compiled on the 21st and used the AVX2 FFTW & AVX2 APP.
Strange it produced different results than the two previous AVX CPU Apps.

While trying different compiles, I decided to try the Intel iGPU App compiled to use AVX again. I tried it last year and it wasn't any better than the iGPU App compiled to use ssse3.
Try it again and see if it's any better than the last Intel iGPU App which was MBv8_8.22r3586_Intel_ssse3_x86_64-apple-darwin; Intel-AVX.zip

KWSN-Darwin-MBbench v2.1.07
Running on domisses-MacBook-Pro.local at Mon Jan 23 13:16:24 2017
---------------------------------------------------
Starting benchmark run...
---------------------------------------------------
Listing wu-file(s) in /testWUs :
18dc09ah.26284.16432.6.33.125.wu reference_work_unit_r3215.wu

Listing executable(s) in /APPS :
MBv8_8.22r3589_Intel_avx_x86_64-apple-darwin

Listing executable in /REF_APPs :
MBv8_8.06r3366_avx_x86_64-apple-darwin
---------------------------------------------------
Current WU: 18dc09ah.26284.16432.6.33.125.wu
---------------------------------------------------
Running default app with command : MBv8_8.06r3366_avx_x86_64-apple-darwin
1647.26 real 1642.98 user 1.68 sys
Elapsed Time: ………………………………… 1647 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3589_Intel_avx_x86_64-apple-darwin
0.12 real 0.08 user 0.02 sys
./benchmark: line 145: 164700 / 0: division by 0 (error token is "0")

---

You are aware the Mac CPU AVX App works Extremely well aren't you? Who do you suggest compile that App that wouldn't exist if I hadn't of compiled it?

Your work on apps building and dedication are worthwhile, absolutely no doubts.
Would be good to have more capable hardware to test but that's life - not all wishes fulfill. From bright side there is opportunity for someone with capable hardware but w/o compilers knowledge to take his part in collective effort.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

From bright side there is opportunity for someone with capable hardware but w/o compilers knowledge to take his part in collective effort.

The OSX compiler situation is a minefield. Everything above Darwin 13.x has problems with certain SETI Apps. In Darwin 14.x there are problems with the OS and 10.10 SDK that prevents compiling an OpenCL App. In every xCode above 6.2 there are problems with Petri's CUDA code. The Best environment for compiling is either Darwin 12.x or 13.x with xCode 6.1.1 or 6.2. Of course to boot to any OS lower than Darwin 14.x I have to REMOVE the 3 nVidia Cards and install the old GTS 250 to even boot the OS. It's quite a PITA. The last two Apps tested here were compiled with xCode 7.3.1 in Darwin 15.6...it doesn't seem to be working very well, does it. I've removed xCode 7.3.1 and gone back to 6.2, which actually works. I'll try another iGPU App with xCode 6.2 in Darwin 15.6, if that doesn't work we'll just have to wait until the next time I need to REMOVE the 3 cards and boot into a lower OS version. That will probably be when nVidia releases a New Web driver that supports the latest update to Darwin 15.6, which should be in a day or two.

New Intel App, Intel_iGPU_1-23-17.zip. http://www.arkayn.us/forum/index.php?topic=191.msg4502#msg4502

...Expand the MB Bench folder in your Home folder. Place MBv8_8.22r3590_Intel_avx_x86_64-apple-darwin in the APPS folder, MultiBeam_Kernels_r3590.cl in the KWSN-OSX-bench-MB_v2.1.07 folder, the 2 WUs in the testWUs folder, and the CPU App MBv8_8.06r3366_avx_x86_64-apple-darwin into the REF_APPs folder. Open a Terminal window, cd to the KWSN-OSX-bench-MB_v2.1.07 folder, and run ./benchmark.

That will probably be when nVidia releases a New Web driver that supports the latest update to Darwin 15.6, which should be in a day or two.

TBar here is a link where you can download the latest Nvidia web driver build 15G1217 compatible with Darwin 15.6.0

https://images.nvidia.com/mac/pkg/346/WebDriver-346.03.15f06.pkg


Gianfranco

---

I don't want to believe, I want to know!

Thanks for the heads up. It's still not showing up in the driver search, and since I could't boot to that system I had no other way of checking. Back to the Main system on the SSD now.
Urs provided another Intel mod last night so I compiled a few Apps in Mavericks. This time there is a change, the iGPU App no longer works correctly on my nVidia cards. Now to see if that's a good thing or not.
I also tried the avx2 CPU App using the fftw libraries from the other day, hopefully it will work this time. It's in the same location; http://www.arkayn.us/forum/index.php?topic=191.msg4502#msg4502

New Intel App, Intel_iGPU_1-23-17.zip. http://www.arkayn.us/forum/index.php?topic=191.msg4502#msg4502

...Expand the MB Bench folder in your Home folder. Place MBv8_8.22r3590_Intel_avx_x86_64-apple-darwin in the APPS folder, MultiBeam_Kernels_r3590.cl in the KWSN-OSX-bench-MB_v2.1.07 folder, the 2 WUs in the testWUs folder, and the CPU App MBv8_8.06r3366_avx_x86_64-apple-darwin into the REF_APPs folder. Open a Terminal window, cd to the KWSN-OSX-bench-MB_v2.1.07 folder, and run ./benchmark.

KWSN-Darwin-MBbench v2.1.07
Running on domisses-MacBook-Pro.local at Tue Jan 24 13:04:05 2017
---------------------------------------------------
Starting benchmark run...
---------------------------------------------------
Listing wu-file(s) in /testWUs :
18dc09ah.26284.16432.6.33.125.wu reference_work_unit_r3215.wu

Listing executable(s) in /APPS :
MBv8_8.22r3606_Intel-mod_ssse3_x86_64-apple-darwin

Listing executable in /REF_APPs :
MBv8_8.06r3366_avx_x86_64-apple-darwin
---------------------------------------------------
Current WU: 18dc09ah.26284.16432.6.33.125.wu
---------------------------------------------------
Running default app with command : MBv8_8.06r3366_avx_x86_64-apple-darwin
1586.11 real 1581.69 user 1.80 sys
Elapsed Time: ………………………………… 1586 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3606_Intel-mod_ssse3_x86_64-apple-darwin
1667.80 real 53.55 user 58.78 sys
Elapsed Time : ……………………………… 1668 seconds
Speed compared to default : 95 %
-----------------
Comparing results
------------- R1:R2 ------------ ------------- R2:R1 ------------
Exact Super Tight Good Bad Exact Super Tight Good Bad
Spike 0 0 0 0 0 0 0 0 0 0
Autocorr 0 0 0 0 0 0 0 0 0 1
Gaussian 0 0 0 0 0 0 0 0 0 0
Pulse 0 0 0 0 1 0 0 0 0 1
Triplet 0 2 2 3 0 0 2 2 3 0
Best Spike 0 0 1 1 0 0 0 1 1 0
Best Autocorr 0 0 0 1 0 0 0 0 1 0
Best Gaussian 1 1 1 1 0 1 1 1 1 0
Best Pulse 0 0 0 0 1 0 0 0 0 1
Best Triplet 0 0 0 1 0 0 0 0 1 0
---- ---- ---- ---- ---- ---- ---- ---- ---- ----
1 3 4 7 2 1 3 4 7 3

Unmatched signal(s) in R1 at line(s) 393 473
Unmatched signal(s) in R2 at line(s) 359 410 490
For R1:R2 matched signals only, Q= ????
Result : Weakly similar.
---------------------------------------------------
Done with 18dc09ah.26284.16432.6.33.125.wu.
Current WU: reference_work_unit_r3215.wu
---------------------------------------------------
Running default app with command : MBv8_8.06r3366_avx_x86_64-apple-darwin
1107.28 real 1103.85 user 1.07 sys
Elapsed Time: ………………………………… 1107 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3606_Intel-mod_ssse3_x86_64-apple-darwin
1081.87 real 208.19 user 34.19 sys
Elapsed Time : ……………………………… 1082 seconds
Speed compared to default : 102 %
-----------------
Comparing results
------------- R1:R2 ------------ ------------- R2:R1 ------------
Exact Super Tight Good Bad Exact Super Tight Good Bad
Spike 0 8 10 13 0 0 8 10 13 0
Autocorr 0 1 1 1 0 0 1 1 1 0
Gaussian 0 0 0 2 4 0 0 0 2 3
Pulse 0 0 0 0 0 0 0 0 0 0
Triplet 0 0 1 2 0 0 0 1 2 0
Best Spike 0 1 1 1 0 0 1 1 1 0
Best Autocorr 0 1 1 1 0 0 1 1 1 0
Best Gaussian 0 0 0 1 0 0 0 0 1 0
Best Pulse 0 0 0 0 1 0 0 0 0 1
Best Triplet 0 0 1 1 0 0 0 1 1 0
---- ---- ---- ---- ---- ---- ---- ---- ---- ----
0 11 15 22 5 0 11 15 22 4

Unmatched signal(s) in R1 at line(s) 499 526 553 634 720
Unmatched signal(s) in R2 at line(s) 499 526 607 693
For R1:R2 matched signals only, Q= 17.34%
Result : Weakly similar.
---------------------------------------------------
Done with reference_work_unit_r3215.wu.

---

...Urs provided another Intel mod last night so I compiled a few Apps in Mavericks. This time there is a change, the iGPU App no longer works correctly on my nVidia cards. Now to see if that's a good thing or not....

Its not enough, but at least it seems to do something, even if not that what we want.

---

_\|/_
U r s

Its not enough, but at least it seems to do something, even if not that what we want.

Seems another mod or two is needed on the iGPU App.

Laurent, please look at the Activity Monitor and post the Idle Wake Ups for the current ATI App 3587, it should be around 300 to 1000. Then try the New ATI App 3604 and see if the numbers are any better. Also, does the CPU App work now?

I compiled a version for iGPU using the new fftw-3.3.6.-pl1 with AVX & AVX2 SIMD extensions with the following results

KWSN-Darwin-MBbench v2.1.07
Running on Andromeda at Mar 24 Gen 16:43:50 2017
---------------------------------------------------
Starting benchmark run...
---------------------------------------------------
Listing wu-file(s) in /testWUs :
reference_work_unit_r3215.wu

Listing executable(s) in /APPS :
MBv8_8.22r3609_ssse3_Intel_x86_64-apple-darwin

Listing executable in /REF_APPs :
MBv8_8.22r3603_avx_x86_64-apple-darwin
---------------------------------------------------
Current WU: reference_work_unit_r3215.wu
---------------------------------------------------
Skipping default app MBv8_8.22r3603_avx_x86_64-apple-darwin, displaying saved result(s)
Elapsed Time: ………………………………… 938 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3609_ssse3_Intel_x86_64-apple-darwin
Elapsed Time : ……………………………… 271 seconds
Speed compared to default : 346 %
-----------------
Comparing results
Result : Strongly similar, Q= 99.51%
---------------------------------------------------
Done with reference_work_unit_r3215.wu.

---

I don't want to believe, I want to know!

It's been like that for a while now. The Un-Modded Intel App runs fine on a nVidia GPU, but fails on an Intel iGPU. The Current Stock Mac nVidia OpenCL App is really the Intel App configured to run on nVidia GPUs because the Regular nVidia OpenCL builds take longer to run on a nVidia GPU than it does on a CPU. Plus the Idle Wake Ups on the nVidia SoG App is in the Tens of Thousands whereas the Intel App run on a nVidia GPU is Normal. What do you get on your nVidia card when you compile an App using the nVidia SoG path?

This is the result from last nights SoG build;

Current WU: 18dc09ah.26284.16432.6.33.125.wu
---------------------------------------------------
Skipping default app MBv8_8.05r3344_sse41_x86_64-apple-darwin, displaying saved result(s)
Elapsed Time: ………………………………… 3517 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3609_NV-SoG_ssse3_x86_64-apple-darwin
     4069.81 real      3676.38 user       384.25 sys
Elapsed Time : ……………………………… 4070 seconds
Speed compared to default : 86 %
-----------------
Comparing results
Result      : Strongly similar,  Q= 99.49%

The Idle Wake Ups reached around 25k and the CPU use up to 106%
On the Other hand, the Un-Modded Intel build works fine on that nVidia GPU.

What do you get on your nVidia card when you compile an App using the nVidia SoG path?

TBar,
these are the results obtained with my Nvidia GPU and the SoG App

KWSN-Darwin-MBbench v2.1.07
Running on Andromeda at Mer 25 Gen 07:41:57 2017
---------------------------------------------------
Starting benchmark run...
---------------------------------------------------
Listing wu-file(s) in /testWUs :
reference_work_unit_r3215.wu

Listing executable(s) in /APPS :
MBv8_8.22r3609_ssse3_NV_SoG_x86_64-apple-darwin

Listing executable in /REF_APPs :
MBv8_8.22r3603_avx_x86_64-apple-darwin
---------------------------------------------------
Current WU: reference_work_unit_r3215.wu
---------------------------------------------------
Running default app with command : MBv8_8.22r3603_avx_x86_64-apple-darwin
940.89 real 938.01 user 0.77 sys
Elapsed Time: ………………………………… 941 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3609_ssse3_NV_SoG_x86_64-apple-darwin
463.46 real 154.04 user 94.91 sys
Elapsed Time : ……………………………… 463 seconds
Speed compared to default : 203 %
-----------------
Comparing results
Result : Strongly similar, Q= 99.50%
---------------------------------------------------
Done with reference_work_unit_r3215.wu.


I hope that these results will be helpful!

---

I don't want to believe, I want to know!

The NV SoG App takes about twice as long to be performed respect the iGPU.

---

I don't want to believe, I want to know!

That's when using the Modified reference_work_unit_r3215.wu. I get better performance when using that Modded Work Unit as well, but, the Idle Wake Ups will spike to 40k. Try it with a Normal WU. I usually get crashes with the NV-SoG App when running a BLC task, but, it usually works with the Arecibo tasks....slowly. That's why SETI is using the Intel Build for the Mac nVidia OpenCL App.

Current WU: reference_work_unit_r3215.wu
---------------------------------------------------
Skipping default app MBv8_8.05r3344_sse41_x86_64-apple-darwin, displaying saved result(s)
Elapsed Time: ………………………………… 2521 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3609_NV-SoG_ssse3_x86_64-apple-darwin
      946.76 real       613.77 user       156.33 sys
Elapsed Time : ……………………………… 947 seconds
Speed compared to default : 266 %
-----------------
Comparing results
Result      : Strongly similar,  Q= 99.51%

Too bad the Current Intel Build doesn't work correctly on a large number of Actual Intel iGPUs.

I compiled a version for iGPU using the new fftw-3.3.6.-pl1 with AVX & AVX2 SIMD extensions with the following results

KWSN-Darwin-MBbench v2.1.07
Running on Andromeda at Mar 24 Gen 16:43:50 2017
---------------------------------------------------
Starting benchmark run...
---------------------------------------------------
Listing wu-file(s) in /testWUs :
reference_work_unit_r3215.wu

Listing executable(s) in /APPS :
MBv8_8.22r3609_ssse3_Intel_x86_64-apple-darwin

Listing executable in /REF_APPs :
MBv8_8.22r3603_avx_x86_64-apple-darwin
---------------------------------------------------
Current WU: reference_work_unit_r3215.wu
---------------------------------------------------
Skipping default app MBv8_8.22r3603_avx_x86_64-apple-darwin, displaying saved result(s)
Elapsed Time: ………………………………… 938 seconds
---------------------------------------------------
Running app with command : MBv8_8.22r3609_ssse3_Intel_x86_64-apple-darwin
Elapsed Time : ……………………………… 271 seconds
Speed compared to default : 346 %
-----------------
Comparing results
Result : Strongly similar, Q= 99.51%
---------------------------------------------------
Done with reference_work_unit_r3215.wu.

Could you post stderr of results. Also, is it iGPU indeed or Intel CPU AVX2 instead? That iGPU thing looks confusing...

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.