None of the benchmarks in the SNU OpenCL NAS Parallel Benchmarks are working. They're available from http://aces.snu.ac.kr/Center_for_Manycore_Programming/SNU_NPB_Suite.html I've got a Radeon 5830 and have installed Mesa from git using the instructions at http://dri.freedesktop.org/wiki/GalliumCompute/ Most of the benchmarks fail at Running pass 'AMDGPU DAG->DAG Pattern Instruction Selection' on function The FFT benchmark fails at error: parameter may not be qualified with an address space I do have some other test cases that are working for me, including the ones from http://cgit.freedesktop.org/~tstellar/opencl-example/
I'm having a little trouble figuring out the build system for this suite, could you pick one of the failing examples and give instructions for how to build and execute it? Also, what version of Mesa and LLVM are you using? I recommend using the development versions of both for best results.
This should build and run the EP benchmark: make veryclean make ep CLASS=A cd bin ./run_ocl_gpu.sh ep A If that doesn't work, you might have to edit config/make.def first to make it link to -lOpenCL.
It looks the EP test is failing, because the compiler doesn't support double operations yet. I'm guessing the other tests are failing for the same reason. Adding support for doubles is relatively straight forward. If you are interested in working on this, let me know and I can help give you some pointers.
Nat, does this bug still occur?
(In reply to Vedran Miletić from comment #4) > Nat, does this bug still occur? All NAS benchmarks require cl_khr_fp64, so they won't run on pre GCN hw any time soon.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/462.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.