+- +-
Say hello if visiting :) by Gecko
11 Jan 2023, 07:43:05 pm

Seti is down again by Mike
09 Aug 2017, 10:02:44 am

Some considerations regarding OpenCL MultiBeam app tuning from algorithm view by Raistmer
11 Dec 2016, 06:30:56 am

Loading APU to the limit: performance considerations by Mike
05 Nov 2016, 06:49:26 am

Better sleep on Windows - new round by Raistmer
26 Aug 2016, 02:02:31 pm

Author Topic: OSX Multibeam OpenCL question  (Read 35293 times)

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #30 on: 14 Oct 2014, 02:39:20 pm »
All variants should build OK actually.
I'll check what is regarding NV. MultiBeam OpenCL NV rarely used (for windows and Linux we have CUDA version that faster) so OPENCL_NV define there could be little unmaintained.
Regarding USE_OPENCL_HD5xxx - that's for ATi cards. And should be used together with USE_OPENCL as addition.
What errors do you see for this build ?

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #31 on: 14 Oct 2014, 03:45:38 pm »
Attached is the stderr.txt. Let me know if you need additional info. It was built using the same build flags and preprocessor flags as the intel version that works. The only difference is USE_OPENCL_HD5xxx instead of USE_OPENCL_INTEL.


Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #32 on: 14 Oct 2014, 03:49:06 pm »
Thanks.
-43 is CL_INVALID_BUILD_OPTIONS error. Some compiler option not go it seems.

I'll look closer on weekend probably.

Offline Urs Echternacht

  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 4121
  • ++
Re: OSX Multibeam OpenCL question
« Reply #33 on: 14 Oct 2014, 04:02:49 pm »
These are the buildoptions for OpenCL compiler when HD5xxx is active (which has no specification for __APPLE__ yet) :
#elif USE_OPENCL_HD5xxx
      strcpy(buildoptions,"-w -DRESULT_SIZE=32 -cl-unsafe-math-optimizations -DUSE_OPENCL_HD5xxx -fno-bin-amdil");

Joe, could you try a rebuild using this options in src/GPU_lock.cpp instead :
#elif USE_OPENCL_HD5xxx
#if __APPLE__
      strcpy(buildoptions,"-w -D__APPLE__ -DRESULT_SIZE=32 -cl-unsafe-math-optimizations -DUSE_OPENCL_HD5xxx");
#else
      strcpy(buildoptions,"-w -DRESULT_SIZE=32 -cl-unsafe-math-optimizations -DUSE_OPENCL_HD5xxx -fno-bin-amdil");
#endif
_\|/_
U r s

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #34 on: 15 Oct 2014, 12:11:15 pm »
Hi Urs,

I applied the change that you suggested. I now get error -49 instead of -43. Attached is the stderr.txt

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #35 on: 15 Oct 2014, 02:37:30 pm »
Please check that we on the same build path for HD5 build:

OCL_ZERO_COPY;OCL_CHIRP3;ATI_OS_WIN;SETI7;USE_OPENCL_HD5xxx;USE_OPENCL;USE_SSE;NDEBUG;_WINDOWS;CLIENT;_CONSOLE;USE_I386_OPTIMIZATIONS;USE_I386_XEON;USE_AMD;USE_FFTW

these deifnes (except windows-specific of course) should be set for HD5 build.

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #36 on: 15 Oct 2014, 03:11:25 pm »
The only one that I see which we aren't using us USE_AMD. I will add that and try again.

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #37 on: 15 Oct 2014, 03:22:56 pm »
try to go clFFT_ExecuteInterleaved_mb functionand add OCL_LOG_ERR("smth"); after each kernel argument set.
Like:
         err |= clSetKernelArg(kernelInfo->kernel, 0, sizeof(cl_mem), &memObj[currRead]);OCL_LOG_ERR("arg 1");

that way we will know what argument fails .

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #38 on: 26 Oct 2014, 02:04:17 pm »
So, what was outcome?

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #39 on: 27 Oct 2014, 01:06:36 pm »
Sorry, I have been on the road for the last week and haven't had a chance to follow up. I'll be able to add the debug code and narrow down the troublesome parameter this week.

Thanks
-Joe

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #40 on: 27 Oct 2014, 02:04:34 pm »
Great! Will wait with interest (and I'm not always available too so no problems, lets catch each other :) ).

EDIT: With AP tasks too scarse the need for these builds is great and increases constantly. Share your efforts and issues please, we need to make some breakthrough on this front.
« Last Edit: 30 Oct 2014, 04:47:18 pm by Raistmer »

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #41 on: 05 Nov 2014, 02:23:24 pm »
Understood. I finally have the build modified with the debug info. As soon as I get access to my nvidia and ATI/AMD based machines, I'll run the tests. Should be within the next day or two.

Interestingly, the intel targeted build worked on all three machines, and appeared to be using the GPU based on the time it took for the test to run.

Thanks for the patience.
-Joe

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #42 on: 05 Nov 2014, 03:15:05 pm »
Here is the final output before I get the error. This is running the HD5xxx version on a mac pro with two ATI/AMD GPUs

call 'Setting kernel argument:FindAutoCorrelation_reduce1_kernel_cl' is finished OK in file /Users/joe/projects/seti/sah_v7_opt/AKv8/client/analyzeFuncs.cpp near line 707
INFO: After SetupKernelArgs
INFO: Freeing local variables.
call 'arg pos 115' is finished OK in file /Users/joe/projects/seti/sah_v7_opt/AKv8/client/analyzeFuncs.cpp near line 3693
call 'Setting kernel argument:CalcChirpData_kernel2_cl' is finished OK in file /Users/joe/projects/seti/sah_v7_opt/AKv8/client/analyzeFuncs.cpp near line 3694
call 'Enqueueing kernel:CalcChirpData_kernel2_cl' is finished OK in file /Users/joe/projects/seti/sah_v7_opt/AKv8/client/analyzeFuncs.cpp near line 3709
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.

The full stderr.txt is attached as well.

I will be testing the nvidia one this evening.
Thanks
-Joe

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: OSX Multibeam OpenCL question
« Reply #43 on: 05 Nov 2014, 03:52:18 pm »

-49 is invalid argument index.

Please check arguments for that last call (clSetKernelArg).

Try to update your workspace with latest rev also. Seems Urs can build OpenCL ATi for OS X w/o issues now. Only issue we working on is one missed pulse in special test task, but not any runtime errors...

Offline Joe Fox

  • Volunteer Developer
  • Knight o' The Realm
  • *****
  • Posts: 84
Re: OSX Multibeam OpenCL question
« Reply #44 on: 05 Nov 2014, 04:56:54 pm »
The latest call is

err = clSetKernelArg(CalcChirpData_kernel2_cl,0,sizeof(cl_float2),(void *)&chirprate2)

I will examine what the values of chirprate2 are.

I will also grab the latest code and see if the issue persists.

Thanks

 

Welcome, Guest.
Please login or register.
 
 
 
Forgot your password?
Members
Total Members: 97
Latest: ToeBee
New This Month: 0
New This Week: 0
New Today: 0
Stats
Total Posts: 59559
Total Topics: 1672
Most Online Today: 24
Most Online Ever: 983
(20 Jan 2020, 03:17:55 pm)
Users Online
Members: 0
Guests: 30
Total: 30
Powered by EzPortal