OSX Multibeam OpenCL question

[Joe Fox]

Hi all,

I have been doing some work on the OSX OpenCL port of multibeam, and have run in to a roadblock. The code branch that I am using is setisvn sah_v7_opt/AKV8. I have gotten a successful build using xcode 5.0.2 using the readme included in the project, however when I execute the binary against the reference_work_unit.sah I get the following output:

Code: [Select]

Used GPU device parameters are:
Number of compute units: 280
Single buffer allocation size: 64MB
Total device global memory: 1536MB
max WG size: 512
local mem type: Real
period_iterations_num=23
Error in mb oclFFT_2: -49
ERROR: OpenCL kernel/call 'non-strip fft' call failed (-49) in file /Users/joe/projects/seti/sah_v7_opt/AKv8/client/analyzeFuncs.cpp near line 3820.
Waiting 30 sec before restart...


The full stderr.txt is attached if more info is needed.

Earlier versions of this code were working back in early August, but something changed and now I run in to this error -49. Unfortunately I don't know enough about cl to troubleshoot further. Any suggestions on where the problem may lie?

Thanks
-Joe Fox

[Raistmer]

#define CL_INVALID_ARG_INDEX                        -49

Maybe last changes for AP broke MB build. i'll check if it compiles for windows then will see.

[Raistmer]

btw, what device you run on? 280 Cus impressive...

[Joe Fox]

Cool, thanks. This is on my 2013 macbook pro running OSX 10.9.5.

Thanks
-Joe Fox

[Urs Echternacht]

Are you sure that you are using actual code from the svn repository ? From your stderr.txt : "MultiBeam_Kernels_r632.cl" Current svn revision is r2725.

[Joe Fox]

Ah, that r632 is just due to my build environment not setting the release DEFINE properly. The actual svn revision that I am on is r2724

Thanks
-Joe Fox

[Urs Echternacht]

Then could you lookup which svn revision number from august was the last that did work for you ? That would help to identify the possible changes faster.

Could you specify which readme you refering to ?

ps: I have used Xcode 4.5.3 (compiler only) so far, because that was the last version of Xcode that was Lion compatible (needed for AstroPulse v7). For Intel GPUs that is of course not necessary.

[Raistmer]

well, windows build fails with the same error so code tree is broken indeed .
Please wait until I fix it 

[Joe Fox]

Yes, I will go back thru revs to narrow down when it stopped working for me.

As for readme, I used "AKv8/ConfigureLine_AKv8c_OPENCL_SSE3_OSX_MBv7.txt" as my starting point for getting the correct project settings.

Thanks
-Joe

[Joe Fox]

Ah, ok. I'll hold tight until you have the windows side sorted.

Thanks!
-Joe

[Raistmer]

I checked in fix, try to rebuild now.

And please share your findings with team time to time

[Joe Fox]

Unfortunately, it looks like I still get the same error. The only difference I see in the stderr is now local mem type is "Emulated" rather than "Real".

Code: [Select]

Work Unit Info:
...............
Credit multiplier is :  2.85
WU true angle range is :  0.775000
Used GPU device parameters are:
Number of compute units: 280
Single buffer allocation size: 64MB
Total device global memory: 1536MB
max WG size: 512
local mem type: Emulated
period_iterations_num=23
Error in mb oclFFT_2: -49
ERROR: OpenCL kernel/call 'non-strip fft' call failed (-49) in file /Users/joe/projects/seti/sah_v7_opt/AKv8/client/analyzeFuncs.cpp near line 3820.
Waiting 30 sec before restart...

I will continue with the effort to find which revision looks to break things, though it may be next week before I have sufficient time to isolate it.

Thanks
-Joe

[Raistmer]

Please check that fft_execute.cpp file was updated.

[Joe Fox]

Ah, I examined the svn diff, and looks like the code is now wrapped with an ifdef USE_OPENCL_INTEL. When I set that macro in the CFLAGS, then my binary works again. Thank you!

That leads me to a question. What should the USE_OPENCL_??? macros be set at for a wide distribution build, like a beta build? For my local machine, which has an Intel GPU, the USE_OPENCL_INTEL makes sense, but what about for the new Mac Pro desktop workstations? They have an nvidia GPU if I recall correctly. How would a widely distributed binary that was compiled with USE_OPENCL_INTEL work for the nvidia system?

Thanks for the assistance!
-Joe

[Urs Echternacht]

"Emulated" is what i am working at currently.