OS X failures
log in

Advanced search

Message boards : OS X and macOS : OS X failures

Author Message
Steve Hawker*
Send message
Joined: 11 Nov 16
Posts: 13
Credit: 44,799
RAC: 0
Message 24 - Posted: 17 Nov 2016, 20:47:12 UTC
Last modified: 17 Nov 2016, 20:48:53 UTC

All fails for mt_mac (OpenMP-multithreaded for Mac) are for the same host. All fails for opencl_ati_102_mac are also for the same host. Maybe, there are problems with these app versions but better statistics is needed.

Yeah, that's mine.

FWIW this host runs Primegrid, SETI and Einstein on CPU and GPU without any issues.

After my experience with the ATI on Win 7, I tried suspending CPU operation but it made no difference to the GPU results.

I'm always in need of GPU apps for my MacBook so I really appreciate you supporting this platform.

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 26 - Posted: 17 Nov 2016, 21:25:47 UTC - in response to Message 24.

The problem with mt_mac is localized already. I built the binary with gcc 4.8 (because clang does not support OpenMP) and forget to specify the -static-libgcc and -static-libstdc++ flags when linking. I hope that static linking will solve the problem. The new binary will be provided tomorrow. The problem with opencl_ati_102_mac needs further investigation.

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 27 - Posted: 17 Nov 2016, 22:18:20 UTC - in response to Message 24.

The new statically linked version of mt_mac is provided already, however, I didn't test it on Mac that does not have macports installed, so I am not sure that the problem is solved. Can you try it?

Steve Hawker*
Send message
Joined: 11 Nov 16
Posts: 13
Credit: 44,799
RAC: 0
Message 28 - Posted: 18 Nov 2016, 8:33:24 UTC - in response to Message 27.

The new statically linked version of mt_mac is provided already, however, I didn't test it on Mac that does not have macports installed, so I am not sure that the problem is solved. Can you try it?

Success!!

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 29 - Posted: 18 Nov 2016, 14:18:40 UTC - in response to Message 28.

The new statically linked version of mt_mac is provided already, however, I didn't test it on Mac that does not have macports installed, so I am not sure that the problem is solved. Can you try it?

Success!!

That's good!

Can I ask you to help me with testing the opencl_ati_102_mac version? It should not take much time. Here is a link to the archive containing the app and the data for simple NaCl crystallite.
To perform testing, start the Terminal in the folder where you have extracted the archive and run:

./XaNSoNS_OCL_mac_1.0 NaCl1.xml > stdout1

and after that, run:

./XaNSoNS_OCL_mac_1.0 NaCl2.xml > stdout2

The estimated computational time for the first run is a few seconds and less than a second for the second run. If it takes much longer, then something is wrong. If the calculations complete successfully, please upload the files stdout1, stdout2, NaCl_1_xray_1D.txt and NaCl_2_xray_1D.txt to somewhere and provide me a link.
Thanks in advance!

Steve Hawker*
Send message
Joined: 11 Nov 16
Posts: 13
Credit: 44,799
RAC: 0
Message 30 - Posted: 18 Nov 2016, 18:06:29 UTC - in response to Message 29.

Sadly no go. For some reason it couldn't find an OpenCL device, which is most odd.

Here's the stdout1 file:

Parsing calculation parameters...
Calculation-->source is set to Xray.
Calculation-->scenario is set to Debye.
Calculation-->PrintAtoms is set to No.
Calculation-->PolarFactor is set to No.
q-->N is set to 1024.
Calculation-->wavelength is set to default.
Calculation-->PartialIntensity is set to No.

Parsing Block 0...
Block-->centered is set to No.
Block-->centeredAtoms is set to No.
Block-->mol_rotation is set to No.
Block-->mol_Uiso is set to 0.
Atoms-->filename is set to default value.
Atom-->occ is set to 1.0.
Atom-->Uiso is set to 0.
Atom-->occ is set to 1.0.
Atom-->Uiso is set to 0.

All blocks have been parsed.

The following OpenCL platforms are found:
Platform 0: Apple
Platform contains the following OpenCL devices:
Device 0:
GPU: ATI Radeon HD 6750M
Number of compute units: 6
GPU clock rate: 150 MHz
Theoretical peak performance: 0 GFLOPs

Error: No OpenCL devices found.

stdout2 was identical

In case you're wondering, here's the appropriate entry in the BOINCmgr log for this host:

Fri Nov 18 06:21:37 2016 | | OpenCL: AMD/ATI GPU 0: ATI Radeon HD 6750M (driver version 1.2 (Nov 18 2015 20:53:06), device version OpenCL 1.2, 1024MB, 1024MB available, 72 GFLOPS peak)
Fri Nov 18 06:21:37 2016 | | OpenCL CPU: Intel(R) Core(TM) i7-2720QM CPU @ 2.20GHz (OpenCL driver vendor: Apple, driver version 1.1, device version OpenCL 1.2)


Happy to try anything to get it to work.

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 31 - Posted: 18 Nov 2016, 19:42:02 UTC - in response to Message 30.


Happy to try anything to get it to work.

Thank you! Can you replace the old executable with the new one from this archive and try again?

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 34 - Posted: 19 Nov 2016, 11:03:29 UTC - in response to Message 30.


Happy to try anything to get it to work.

I've got your message, thank you! The bug was not platform-related, but GPU-related. I think that all old AMD GPUs were affected. I will update the OpenCL apps today. I'll let you know when it's done.

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 35 - Posted: 19 Nov 2016, 17:28:47 UTC - in response to Message 34.


I will update the OpenCL apps today. I'll let you know when it's done.

It's done.

Steve Hawker*
Send message
Joined: 11 Nov 16
Posts: 13
Credit: 44,799
RAC: 0
Message 46 - Posted: 21 Nov 2016, 14:40:11 UTC - in response to Message 35.


I will update the OpenCL apps today. I'll let you know when it's done.

It's done.

100% success rate so far (147/147)
Thank you - especially because so few projects take the time to develop apps for OS X and even fewer have a GPU app that works. So as long as you generate tasks, I'll be gobbling them up for my GPU.

Message boards : OS X and macOS : OS X failures


Main page · Your account · Message boards


© 2021 Vladislav Neverov (NRC 'Kurchatov institute'), Nikolay Khrapov (Institute for Information Transmission Problems of RAS)