require('cutorch') does not work

[tuming1990]

Hi,

My system is MAC OS 10.10.4

I installed cutorch and cunn using "luarocks install cutorch/cunn". There was no error during this process and at the end it showed that cutorch/cunn is installed in ....

But when I call require('cutorch/cunn'), it gives the error:
/usr/local/share/lua/5.1/trepl/init.lua:363: /usr/local/share/lua/5.1/trepl/init.lua:363: /usr/local/share/lua/5.1/cutorch/init.lua:2: cannot load '/usr/local/lib/lua/5.1/libcutorch.so'
stack traceback:
[C]: in function 'error'
/usr/local/share/lua/5.1/trepl/init.lua:363: in function 'require'
[string "_RESULT={require('cunn')}"]:1: in main chunk
[C]: in function 'xpcall'
/usr/local/share/lua/5.1/trepl/init.lua:630: in function 'repl'
/usr/local/lib/luarocks/rocks/trepl/scm-1/bin/th:185: in main chunk
[C]: at 0x0107d8a530

require('nn') just works fine, only 'cutorch' and 'cunn'.

I installed CUDA 6.5 driver and samples could be compiled correctly.

Anyone know how to solve this problem? Thanks

[soumith]

You've done a non-standard install (did not follow our instructions on torch.ch ), i.e. you installed torch global to the system and not local to ~/ .
While that is not usually a problem, it's really now up to you to debug. There could be so many things weird going on. First step is to see the install log and see why /usr/local/lib/lua/5.1/libcutorch.so does not exist.

[tuming1990]

I checked. Both libcutorch.so and libcunn.so is in folder usr/local/lib/lua/5.1. I will reinstall torch again. Thanks.

[soumith]

if you are reinstalling torch, please use https://github.com/torch/ezinstall/blob/master/clean-old.sh to clean up your old installation.

[tuming1990]

Still the sam problem...

/Users/mingtu/torch/install/share/lua/5.1/trepl/init.lua:363: /Users/mingtu/torch/install/share/lua/5.1/cutorch/init.lua:2: cannot load '/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so'
stack traceback:
[C]: in function 'error'
/Users/mingtu/torch/install/share/lua/5.1/trepl/init.lua:363: in function 'require'
[string "_RESULT={require('cutorch')}"]:1: in main chunk
[C]: in function 'xpcall'
/Users/mingtu/torch/install/share/lua/5.1/trepl/init.lua:630: in function 'repl'
...ngtu/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:185: in main chunk
[C]: at 0x01000431c0

I can still find libcutorch in folder /Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so

[soumith]

what's the output for the command:

luajit -l libcutorch

[soumith]

also for the command:

luajit -l torch -l libcutorch

[tuming1990]

luajit -l libcutorch
luajit: error loading module 'libcutorch' from file '/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so':
dlopen(/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find:
/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #20 length (0) too small in /Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so
stack traceback:
[C]: at 0x010fe58dd0
[C]: at 0x010fe583f0
[C]: at 0x010fe2a1c0

luajit -l torch -l libcutorch
luajit: error loading module 'libcutorch' from file '/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so':
dlopen(/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find:
/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #20 length (0) too small in /Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so
stack traceback:
[C]: at 0x01062dadd0
[C]: at 0x01062da3f0
[C]: at 0x01062ac1c0

[tuming1990]

I'm trying this:
#126

[soumith]

great, that might fix it.

[tuming1990]

Still doesn't work. After updating cmake, do I need to reinstall torch?

[soumith]

after updating cmake, you'll have to reinstall cutorch / cunn:

luarocks install cutorch
luarocks install cunn

[tuming1990]

Yes. I did that. But same error occurs...

[soumith]

what CUDA version are you running? can you upgrade to CUDA 7.0 (or even 7.5), that fixes these OSX compile issues to a large extent. CUDA on OSX used to be pretty buggy on the compiler chain till 6.5

[tuming1990]

luajit -l libcutorch
luajit: error loading module 'libcutorch' from file '/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so':
dlopen(/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find:
/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #19 length (0) too small in /Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so
stack traceback:
[C]: at 0x010fe58dd0
[C]: at 0x010fe583f0
[C]: at 0x010fe2a1c0

luajit -l torch -l libcutorch
luajit: error loading module 'libcutorch' from file '/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so':
dlopen(/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find:
/Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #19 length (0) too small in /Users/mingtu/torch/install/lib/lua/5.1/libcutorch.so
stack traceback:
[C]: at 0x01062dadd0
[C]: at 0x01062da3f0
[C]: at 0x01062ac1c0

Almost the same but "load command #20" changes to "#19"

[tuming1990]

I use cuda 6.5. I will try to upgrade to 7.0.

[tuming1990]

Thanks. It works now 👍

[ghost]

Hi I have updated to CUDA 7.5 and I am getting same error when do
require 'cunn'

I am using ubuntu 14.04

luajit -l torch -l libcutorch
luajit: error loading module 'libcutorch' from file '/usr/local/torch/install/lib/lua/5.1/libcutorch.so':
libcudart.so.7.0: cannot open shared object file: No such file or directory
stack traceback:
[C]: at 0x0047a410
[C]: at 0x0046ab90
[C]: at 0x00406670

[ghost]

I've got the same problem as @jbansal .

[ghost]

Hi nmeln,
Please check your LD_LIBRARY_PATH setting. I had this problem as it was pointing to old lua folder and it is installed in new folder. So, once I set this LD_LIBRARY_PATH, it solved my problem.

Regards,
Jitendra Bansal

[ghost]

@jbansal Thank you for the answer.
I think the reason why there was a problem is that I installed CUDA 7.5 after installing torch.
And it turned out that cmake in my system was outdated (2.x) so I updated it.
Then I cleaned-up (using the official script) and reinstalled torch.
Everything works now!

[zdb26]

@nmeln Unfortunately, I install torch after CUDA 7.5, with the cmake version 3.3.2, but I'v also met the exactly same problem.......

[zdb26]

And: My env is: OSX10.10.5, Xcode6.1.1, CUDA7.5, cmake 3.3.2

localhost:torch zhoudongbin$ luajit -l torch -l libcutorch
luajit: error loading module 'libcutorch' from file '/Users/zhoudongbin/torch/install/lib/lua/5.1/libcutorch.so':
dlopen(/Users/zhoudongbin/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find:
/Users/zhoudongbin/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #19 length (0) too small in /Users/zhoudongbin/torch/install/lib/lua/5.1/libcutorch.so
stack traceback:
[C]: at 0x010e2c06d0
[C]: at 0x010e2bfcf0
[C]: at 0x010e291770

[roeeaharoni]

Running OSX 10.11.2, CUDA7.5, cmake 3.4.3 and getting:

luajit -l torch -l libcutorch
luajit: error loading module 'libcutorch' from file '/Users/roeeaharoni/torch/install/lib/lua/5.1/libcutorch.so':
dlopen(/Users/roeeaharoni/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find:
/Users/roeeaharoni/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #19 length (0) too small in /Users/roeeaharoni/torch/install/lib/lua/5.1/libcutorch.so
stack traceback:
[C]: at 0x010b437260
[C]: at 0x010b436880
[C]: at 0x010b4081d0

[mattyd2]

@roeeaharoni have you been able to resolve this issue? I'm running into the same issue.

Environment:

• OSX10.11
• Xcode 6.2 to be compatible with CUDA 7.5.20
• CUDA 7.5.20
• cmake 3.5.0

When I run luajit -l torch -l libcutorch - I've removed my username for brevity.

luajit: error loading module 'libcutorch' from file '~/torch/install/lib/lua/5.1/libcutorch.so': dlopen(~/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find: ~/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #19 length (0) too small in ~/torch/install/lib/lua/5.1/libcutorch.so stack traceback: [C]: at 0x0104946260 [C]: at 0x0104945880 [C]: at 0x01049171d0

When I run $ luajit -l libcutorch:

luajit: error loading module 'libcutorch' from file '/Users/matthewdunn/torch/install/lib/lua/5.1/libcutorch.so': dlopen(/Users/matthewdunn/torch/install/lib/lua/5.1/libcutorch.so, 6): no suitable image found. Did find: /Users/matthewdunn/torch/install/lib/lua/5.1/libcutorch.so: malformed mach-o image: load command #19 length (0) too small in /Users/matthewdunn/torch/install/lib/lua/5.1/libcutorch.so stack traceback: [C]: at 0x0109aae260 [C]: at 0x0109aad880 [C]: at 0x0109a7f1d0

I pretty sure I've done a completely standard install as I did it on a clean install of OSX 10.11

Any help will be greatly appreciated.

[Shaahinan]

Hi Guys, I am trying to run a .lua code with Torch7, I came across these errors, I would be really appreciative if someone can help me:
/home/angizi/torch/install/bin/luajit: /home/angizi/torch/install/share/lua/5.1/trepl/init.lua:389: module 'cutorch' not found:No LuaRocks module found for cutorch
no field package.preload['cutorch']
no file '/home/angizi/.luarocks/share/lua/5.1/cutorch.lua'
no file '/home/angizi/.luarocks/share/lua/5.1/cutorch/init.lua'
no file '/home/angizi/torch/install/share/lua/5.1/cutorch.lua'
no file '/home/angizi/torch/install/share/lua/5.1/cutorch/init.lua'
no file '/home/angizi/.luarocks/share/lua/5.2/cutorch.lua'
no file '/home/angizi/.luarocks/share/lua/5.2/cutorch/init.lua'
no file '/home/angizi/torch/install/share/lua/5.2/cutorch.lua'
no file '/home/angizi/torch/install/share/lua/5.2/cutorch/init.lua'
no file './cutorch.lua'
no file '/home/angizi/torch/install/share/luajit-2.1.0-beta1/cutorch.lua'
no file '/usr/local/share/lua/5.1/cutorch.lua'
no file '/usr/local/share/lua/5.1/cutorch/init.lua'
no file '/home/angizi/.luarocks/lib/lua/5.1/cutorch.so'
no file '/home/angizi/torch/install/lib/lua/5.1/cutorch.so'
no file '/home/angizi/torch/install/lib/cutorch.so'
no file '/home/angizi/.luarocks/lib/lua/5.2/cutorch.so'
no file '/home/angizi/torch/install/lib/lua/5.2/cutorch.so'
no file './cutorch.so'
no file '/usr/local/lib/lua/5.1/cutorch.so'
no file '/usr/local/lib/lua/5.1/loadall.so'
stack traceback:
[C]: in function 'error'
/home/angizi/torch/install/share/lua/5.1/trepl/init.lua:389: in function 'require'
main.lua:11: in main chunk
[C]: in function 'dofile'
...gizi/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
[C]: at 0x00406670

Please help me

[chaopengshen]

I'm getting the same error. The machine in question has an existing global torch installation. I have no root authority so I have to install luarocks packages locally. I find the code to be calling the cutorch.so in the global torch directory, instead of the local directory I installed it too. I suspect other packages I installed has conflicts with global cutorch. I could have installed the entire torch myself but this system has RHEL 6.0 which does not support "bash install-deps". These get me stuck!

[albanD]

@chaopengshen the install-deps script just install dependencies for torch, if torch is already installed somewhere else, these dependencies are already installed, you can skip this step and just run the ./install.sh script.