Message boards :
Number crunching :
Linux (ARM processor) app and alternatives
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8
Author | Message |
---|---|
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Does the ARM64 app need rebuilding now the vfp and neon chirps are fixed? Or aren't they used on that app, as i see no function choices in their stderr.txt's: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=26799100 Claggy |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
|
Tom Rinehart Send message Joined: 12 Dec 01 Posts: 113 Credit: 13,255,975 RAC: 6 |
I can't get it to show any of the tests, and I do run it with verbose output. The ARM64 code doesn't use vfp and neon functions. The only neon code it is using is through FFTW. The vfp and neon functions need to be ported to AARCH64. There is a porting guide: https://community.arm.com/processors/b/blog/posts/porting-to-arm-64-bit |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
We have on main: Android (ARM64 processor) 8.00 (arm64-neon) 22 Jan 2016, 0:38:52 UTC 231 GigaFLOPS Android (ARM64 processor) 8.00 (arm64-vfpv4) 22 Jan 2016, 0:38:52 UTC 224 GigaFLOPS Android (ARM64 processor) 8.01 4 Jan 2017, 3:33:29 UTC 111 GigaFLOPS What those neon and vfp plan classes mean then? EDIT: for beta list even wider: Android (ARM64 processor) 8.00 4 Nov 2016, 19:09:50 UTC 1 GigaFLOPS Android (ARM64 processor) 8.00 (api24) 23 Feb 2017, 20:28:26 UTC 0 GigaFLOPS Android (ARM64 processor) 8.01 (arm64-neon) 5 Jan 2016, 23:44:30 UTC 6 GigaFLOPS Android (ARM64 processor) 8.01 (arm64-vfpv4) 5 Jan 2016, 23:44:30 UTC 5 GigaFLOPS Linux (ARM64 processor) 8.01 10 Feb 2017, 21:46:51 UTC 6 GigaFLOPS As I understand you both speaking about last one, Linux one. But what about Android-based? Do they show bench results? Android comletely on Eric as I understand. Who built Linux A64? SETI apps news We're not gonna fight them. We're gonna transcend them. |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
I can't get it to show any of the tests, and I do run it with verbose output. The ARM64 code doesn't use vfp and neon functions. The only neon code it is using is through FFTW. The vfp and neon functions need to be ported to AARCH64. There is a porting guide: https://community.arm.com/processors/b/blog/posts/porting-to-arm-64-bit I think that is exactly the same as what i got when building at rev 1924 (Before all the Arm function choices were added), so a generic code path only. Claggy |
Tom Rinehart Send message Joined: 12 Dec 01 Posts: 113 Credit: 13,255,975 RAC: 6 |
Who built Linux A64? I built the Linux ARM64 app. The Android 64 ones with VFP and NEON are the 32-bit apps just cast to run on Android 64. See Eric's note on beta testing the Android 64 app - https://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2347#60010 The ARM64 processors will run 32 bit apps. The Linux ARM64 8.01 app is about 80% faster than the Linux ARM 8.03 app. I didn't compare against 8.06. - Tom |
Tom Rinehart Send message Joined: 12 Dec 01 Posts: 113 Credit: 13,255,975 RAC: 6 |
The Linux ARMHF 8.06 app is now available on the main SETI project. Also the Linux ARM64 8.02 app (which I think is 8.01 from Beta) is also available. |
Tom Rinehart Send message Joined: 12 Dec 01 Posts: 113 Credit: 13,255,975 RAC: 6 |
The Linux ARMHF 8.06 app is now available on the main SETI project. Also the Linux ARM64 8.02 app (which I think is 8.01 from Beta) is also available. Looking at the the Applications page (https://setiathome.berkeley.edu/apps.php), the Linux ARM 8.06 app is now processing at about 1750 GigaFLOPS. This is more than all the Android ARM apps combined. Thanks to all the people that helped improve the Linux ARM app - particularly Raistmer! There is only one optimization that is left to fix - opt VFP folding. I have five different types of ARM computers running Linux and the increase in processing throughput is impressive on each of them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
There is only one optimization that is left to fix - opt VFP folding. yep, it appeared more complex debug needed (or I missed something obvious). Unfortunately, no more time for it currently, maybe some later... On bright side - rebuilt x64 for Windows shows good speedup over already released one in Mike's tests. Will send them for beta soon. SETI apps news We're not gonna fight them. We're gonna transcend them. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.