cunn tests fail with cuda 8.0rc on GTX1070

[htoyryla]

I made a fresh install of Ubuntu 14.04 with GTX1070, cuda 8.0rc, torch + cutorch + cunn. Nvidia driver version 367.35

Testing cunn fails as reported here https://gist.github.com/htoyryla/6ab33ffc9794d3d002ce066553142cd6

During installation of cutorch and cunn, the architecture 6.1. is correctly detected. Install log for cunn here https://gist.github.com/htoyryla/aa77379e493bf155de3eac709a2cb41f

Running programs like neural-style fail with cunn, but work with cudnn. As far as I can see, torch works with cudnn without problems.

Error output from neural-style when running without cudnn (same error is also reported by cunn.test(), see gist above).

/home/hannu/torch/install/bin/luajit: /home/hannu/torch/install/share/lua/5.1/nn/Container.lua:67:
In 7 module of nn.Sequential:
/home/hannu/torch/install/share/lua/5.1/nn/THNN.lua:109: wrong number of arguments for function call
stack traceback:
    [C]: in function 'v'
    /home/hannu/torch/install/share/lua/5.1/nn/THNN.lua:109: in function 'SpatialMaxPooling_updateOutput'
    ...nnu/torch/install/share/lua/5.1/nn/SpatialMaxPooling.lua:42: in function <...nnu/torch/install/share/lua/5.1/nn/SpatialMaxPooling.lua:31>
    [C]: in function 'xpcall'
    /home/hannu/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
    /home/hannu/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
    neural_style.lua:204: in function 'main'
    neural_style.lua:515: in main chunk
    [C]: in function 'dofile'
    ...annu/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
    [C]: at 0x004065d0

[soumith]

Update nn and cutorch and cunn

luarocks install nn

luarocks install cutorch

luarocks install cunn

[htoyryla]

OK, that helped. Reinstalling nn was the trick, the others I had already reinstalled a few times. Silly of me not going back to nn immediately.

There was still this:

VolumetricDilatedConvolution 
 Function call failed
/home/hannu/torch/install/share/lua/5.1/nn/THNN.lua:109: $ Torch: not enough memory: you tried to allocate 21GB. Buy new RAM! at /tmp/luarocks_torch-scm-1-8091/torch7/lib/TH/THGeneral.c:226
stack traceback:
    [C]: in function 'v'
    /home/hannu/torch/install/share/lua/5.1/nn/THNN.lua:109: in function 'VolumetricDilatedConvolution_updateOutput'
    ...nstall/share/lua/5.1/nn/VolumetricDilatedConvolution.lua:32: in function 'forward'
    /home/hannu/torch/install/share/lua/5.1/cunn/test.lua:5346: in function 'v'
    /home/hannu/torch/install/share/lua/5.1/cunn/test.lua:6029: in function </home/hannu/torch/install/share/lua/5.1/cunn/test.lua:6027>
    [C]: in function 'xpcall'
    /home/hannu/torch/install/share/lua/5.1/torch/Tester.lua:477: in function '_pcall'
    /home/hannu/torch/install/share/lua/5.1/torch/Tester.lua:436: in function '_run'
    /home/hannu/torch/install/share/lua/5.1/torch/Tester.lua:355: in function 'run'
    /home/hannu/torch/install/share/lua/5.1/cunn/test.lua:6049: in function 'test'
    (command line):1: in main chunk
    [C]: at 0x004065d0

--------------------------------------------------------------------------------
luajit: /home/hannu/torch/install/share/lua/5.1/torch/Tester.lua:363: An error was found while running tests!
stack traceback:
    [C]: in function 'assert'
    /home/hannu/torch/install/share/lua/5.1/torch/Tester.lua:363: in function 'run'
    /home/hannu/torch/install/share/lua/5.1/cunn/test.lua:6049: in function 'test'
    (command line):1: in main chunk
    [C]: at 0x004065d0

but even that disappeared when I ran the test a second time (after installing cudnn too).

[soumith]

That looks like a bad test, I'll fix it. Don't treat it as a failure for now

[IEWbgfnYDwHRoRRSKtkdyMDUzgdwuBYgDKtDJWd]

@htoyryla Would you be able to say which drivers/steps you took to install drivers/cuda on 14.04? Im having LOTS of issues, could even tip you btc for the help. my original issue opened is here jcjohnson/neural-style#460 if you might have the time to look. thanks!