CUDA 7.5 GPU backend test fail

[davorh]

test fails with GPU backed. CPU backend works fine, latest github version is installed.
Full output attached, error snip:

-- Testing convolution layer with shared param on Mocha.GPUBackend{Float64}...
30-May 10:35:01:INFO:root:Constructing net test-shared-params on Mocha.GPUBackend...
30-May 10:35:01:INFO:root:Topological sorting 5 layers...
30-May 10:35:01:INFO:root:Setup layers...
ERROR: LoadError: LoadError: Bad param
[inlined code] from /home/sageusr/.julia/v0.4/Mocha/src/cuda/cudnn.jl:53
in get_convolution_forward_algorithm at /home/sageusr/.julia/v0.4/Mocha/src/cuda/cudnn.jl:286
in setup_etc at /home/sageusr/.julia/v0.4/Mocha/src/cuda/layers/convolution.jl:38
in call at /home/sageusr/.julia/v0.4/Mocha/src/layers/convolution.jl:120
in setup at /home/sageusr/.julia/v0.4/Mocha/src/layers/convolution.jl:158
in call at /home/sageusr/.julia/v0.4/Mocha/src/net.jl:227
in test_shared_parameters_layers at /home/sageusr/.julia/v0.4/Mocha/test/layers/shared-parameters.jl:25
in test_shared_parameters_layers at /home/sageusr/.julia/v0.4/Mocha/test/layers/shared-parameters.jl:42
in include at ./boot.jl:261
in include_from_node1 at ./loading.jl:320
in anonymous at /home/sageusr/.julia/v0.4/Mocha/test/runtests.jl:26
in map_to! at abstractarray.jl:1286
in map_to! at abstractarray.jl:1295
in map at abstractarray.jl:1308
in test_dir at /home/sageusr/.julia/v0.4/Mocha/test/runtests.jl:25
in include at ./boot.jl:261
in include_from_node1 at ./loading.jl:320
in process_options at ./client.jl:280
in _start at ./client.jl:378
while loading /home/sageusr/.julia/v0.4/Mocha/test/layers/shared-parameters.jl, in expression starting on line 50
while loading /home/sageusr/.julia/v0.4/Mocha/test/runtests.jl, in expression starting on line 85
================================[ ERROR: Mocha ]================================

failed process: Process(/opt/julia/usr/bin/julia --check-bounds=yes --code-coverage=none --color=no /home/sageusr/.julia/v0.4/Mocha/test/runtests.jl, ProcessExited(1)) [1]

Mocha-test.zip

[davidparks21]

Are you using CuDNN (Cuda Neural Network library) version 5? If so, downgrade that to v4 and you should be good to go. The issue with CuDNN v5 is already open in a couple of places (which should probably be consolidated), here's one I posted: #198

Also take a look at this related issue, if Mocha doesn't find the CuDNN v4 library you might need to checkout the latest master from the repo, or just manually apply the simple fix I referenced here: #199

You should be able to continue using the Cuda v7.5 libraries, the problem here only relates to the CuDNN library specifically.

[davorh]

Yes, version 5. Tnx I'll use #199 :)