Add windows install instructions

Currently the INSTALL file says: "I have no idea how to install on Windows.", not very helpful.

For me the following steps were enough to install libgpuarray:

Clone the library from GitHub.
Create build folder, open CMake gui
Change CMAKE_INSTALL_PREFIX to libgpuarray/lib. Perhaps this should be default on windows.
Configure and generate.
Open the .sln file with Visual Studio
Build the INSTALL package (needs PR Remove string definition with no length in gpuarray_buffer_opencl, see #262 #263)
Manually move the dll's to somewhere in the PATH. Perhaps we should automate this on windows.
Build the Cython extension (in the libgpuarray root) with: python setup.py build_ext --compiler=msvc
Install the package with python setup.py install

Now everything works.

Should we add instructions for this somewhere, if so, where?

If you want to update the documentation with these instructions, I will be very open to this. They should go in doc/installation.rst

Hey I'm having trouble building the INSTALL package. Getting the following errors (using Visual Studio 2013):

Error   15  error C2057: expected constant expression   C:\gpuarray\libgpuarray\src\util\integerfactoring.c 830 1   gpuarray
Error   16  error C2466: cannot allocate an array of constant size 0    C:\gpuarray\libgpuarray\src\util\integerfactoring.c 830 1   gpuarray
Error   17  error C2133: 'pInd' : unknown size  C:\gpuarray\libgpuarray\src\util\integerfactoring.c 830 1   gpuarray
Error   32  error C2057: expected constant expression   C:\gpuarray\libgpuarray\src\util\integerfactoring.c 830 1   gpuarray-static
Error   33  error C2466: cannot allocate an array of constant size 0    C:\gpuarray\libgpuarray\src\util\integerfactoring.c 830 1   gpuarray-static
Error   34  error C2133: 'pInd' : unknown size  C:\gpuarray\libgpuarray\src\util\integerfactoring.c 830 1   gpuarray-static
Error   35  error MSB3073: The command "setlocal
"C:\Program Files\CMake\bin\cmake.exe" -DBUILD_TYPE=Debug -P cmake_install.cmake
if %errorlevel% neq 0 goto :cmEnd
:cmEnd
endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
:cmErrorLevel
exit /b %1
:cmDone
if %errorlevel% neq 0 goto :VCEnd
:VCEnd" exited with code 1. C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V120\Microsoft.CppCommon.targets  132 5   INSTALL

Any idea why I am facing this and what I can do to resolve?

Thanks.

There was some unsupported code for MSCV in a recent PR. This has been fixed and the master should work now.

Hey, I tried building again with the fixed version, got the following error:

Error   969 error LNK2019: unresolved external symbol snprintf referenced in function gaIFLsnprintf C:\libgpuarray-master\libgpuarray-master\Build\src\integerfactoring.obj gpuarray
Error   970 error LNK1120: 1 unresolved externals   C:\libgpuarray-master\libgpuarray-master\lib\Debug\gpuarray.dll gpuarray
Error   971 error MSB3073: The command "setlocal
"C:\Program Files\CMake\bin\cmake.exe" -DBUILD_TYPE=Debug -P cmake_install.cmake
if %errorlevel% neq 0 goto :cmEnd
:cmEnd
endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
:cmErrorLevel
exit /b %1
:cmDone
if %errorlevel% neq 0 goto :VCEnd
:VCEnd" exited with code 1. C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V120\Microsoft.CppCommon.targets  132 5   INSTALL

Ok, we need to remove the call to snprintf also.

Okay, please let me know when it is ready to clone. Thanks.

There is this #275.

I don't know when someone will have time to review.

I merged it.

On Wed, Oct 12, 2016 at 11:33 AM, abergeron notifications@github.com
wrote:

There is this #275 #275.

I don't know when someone will have time to review.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#264 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AALC-_QvJ3lMCkwezrd0oLmG5ZhUCxiKks5qzP3hgaJpZM4KKOdR
.

So good news, I was able to build successfully.

I point out some steps to add to windows installation.

The INSTALL package needs to be built with admin privileges in release mode. So this needs to be built in the Visual Studio Developer Command Prompt with "msbuild libgpuarray.sln /p:Configuration=Release"
Restart is necessary after installing pygpu

I should point out that I tried running theano after this, but the kernel crashed. However, when I changed the device = gpu0 in the .theanorc.txt file from device = cuda, it worked just fine.

Thank you for all your help. You guys are great!

Great, I'll try to get to writing these instructions sometime soon.

device=gpu use the old backend. So the crash still need to be fixed.

Le 13 oct. 2016 02:44, "Jonas Adler" notifications@github.com a écrit :

Great, I'll try to get to writing these instructions sometime soon.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#264 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AALC-2JWqRe7FzjxfcaDWX0Wpe8ONFvjks5qzdNpgaJpZM4KKOdR
.

Just to note that we are working on conda packages. This will probably be the only supported way to install it for user. We can't support all combination of compiler on all platform we have.

So mostly, this issue is about adding install instruction with conda on windows. Not the full installation.

Any update on the conda packages?

Hello ! Do you use the same Python environment in both Spyder and command line ? Also, which version of GCC do you use ? We recommend the version provided by conda package m2w64-toolchain: http://deeplearning.net/software/theano/install_windows.html#install-requirements-and-optional-packages .

@michaelklachko

@michaelklachko fastmath is always disabled because we often had problem with wrong results due to the reduced precision. In addition to that the NVIDIA docs don't recommend that it be enabled anymore.

@notoraptor

@notoraptor Sorry for the late response. In Spyder I was using the iPython interpreter. The command line uses the default console python interpreter.

@iceman121

@iceman121 What is the version of your console Python interpreter (result of command python -V), and what is the version of your Spyder's IPython ? Also, do you use a conda environment either in console or Spyder ?

@notoraptor

@notoraptor

Python version= 3.5.3
IPython = 6.0.0

Use of Conda environment= No.

I just want to note that we only support conda packages on windows for pygpu/libgpuarray.

@iceman121

Hello, @iceman121 . Sorry for late response, I was not working on Windows these las days.

But your problem is quite strange. Could you give the full output of following commands ?

In command line:

where python
python -V
gcc -v
conda info --envs

In Spyder IPython:

import sys
print(sys.version)
print(sys.prefix)

The goal is to check if the Python interpreter in your command line is the same as the one in Spyder (do you have many Python installations in your system ?).

@notoraptor

Hi @notoraptor
Please find below-

In command line:

where python
C:\Continuum\python.exe

python -V
Python 3.5.3 :: Anaconda custom (64-bit)

gcc -v
Using built-in specs.
COLLECT_GCC=C:\Continuum\Scripts\gcc.bat\..\..\MinGW\bin\gcc.exe
COLLECT_LTO_WRAPPER=c:/continuum/mingw/bin/../libexec/gcc/x86_64-w64-mingw32/4.7.0/lto-wrapper.exe
Target: x86_64-w64-mingw32
Configured with: ../../../build/gcc/src/configure --target=x86_64-w64-mingw32 --prefix=/c/bb/vista64-mingw32/mingw-x86-x86_64/build/build/root --with-sysroot=/c
/bb/vista64-mingw32/mingw-x86-x86_64/build/build/root --enable-languages=all,obj-c++ --enable-fully-dynamic-string --disable-multilib
Thread model: win32
gcc version 4.7.0 20111220 (experimental) (GCC)

conda info --envs
# conda environments:
#
root                  *  C:\Continuum

For Spyder IPython:

print(sys.version)
3.5.3 |Anaconda custom (64-bit)| (default, May 15 2017, 10:43:23) [MSC v.1900 64 bit (AMD64)]

print(sys.prefix)
C:\Continuum

As far as I remember I have only one interpreter on my system. I think this validates it.

@iceman121

Hello, @iceman121 , indeed, it seems there is only 1 Python installation. Quite strange that theano GPU test works in Spyder but not in console.

The last thing I could suggest is (in command line):

Check that you use m2w64-toolchain as gcc package (conda install m2w64-toolchain). If this package is installed, maybe you could find a gcc program in C:\Continuum\Library\mingw-w64\bin\ ?
Make sure default conda environment (root) is activated (activate root).
Once root env is activated, then try theano test again. Also run where gcc to check if conda gcc package is indeed used in conda env.

@notoraptor

@notoraptor Results below:

Found the gcc.exe in the Library location.
Done.
Tried, failed. where gcc responds as the same location.

Ok.

It seems I am encountering a similar problem as you, but with Python 3.5, not with Python 2.7.

Could you try to create a conda environment with Python 2.7 (e.g. conda create -n python2 python=2, then activate python2), install everything needed in that env ( http://www.deeplearning.net/software/theano_versions/dev/install_windows.html ) and check if theano GPU test works on command line with this environment ?

I'll check this issue for Python 3.5.

I have installed m2w64-toolchain and still getting the following error with libgpuarray.

Can not use cuDNN on context None: cannot compile with cuDNN. We got this error:
b'C:/Users/ankushbabbar/Miniconda2/envs/dlwin36/Library/mingw-w64/bin/../lib/gcc/x86_64-w64-mingw32/5.3.0/../../../../x86_64-w64-mingw32/bin/ld.exe: cannot find -lcudnn\r\ncollect2.exe: error: ld returned 1 exit status\r\n'

I'm using the follwing THEANO_FLAGS
THEANO_FLAGS=floatX=float32,device=cuda0,optimizer_including=cudnn,gpuarray.preallocate=0.8,dnn.conv.algo_bwd_filter=deterministic,dnn.conv.algo_bwd_data=deterministic,dnn.include_path=F:\Softwares\fast_ai\CUDA\v8.0\include,dnn.library_path=F:\Softwares\fast_ai\CUDA\v8.0\lib\x64

The problem seems to be that you don't specify a bin_path for cudnn (which is important on windows).

However, instead of specifying all the individual paths by hand try using the following flag instead of dnn.include_path and dnn.library_path: cuda.root=F:\Softwares\fast_ai\CUDA\v8.0 This should set all the other dnn flags automatically.

Navigation Menu

Repository navigation

Add windows install instructions #264

28 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Add windows install instructions #264

Description

Activity

abergeron commented on Sep 29, 2016

iceman121 commented on Oct 10, 2016

abergeron commented on Oct 11, 2016

iceman121 commented on Oct 11, 2016

abergeron commented on Oct 11, 2016

iceman121 commented on Oct 12, 2016

abergeron commented on Oct 12, 2016

nouiz commented on Oct 12, 2016

iceman121 commented on Oct 12, 2016

adler-j commented on Oct 13, 2016

nouiz commented on Oct 14, 2016

nouiz commented on Jan 12, 2017

nouiz commented on Jan 12, 2017

adler-j commented on Feb 13, 2017

28 remaining items

notoraptor commented on Jun 6, 2017

abergeron commented on Jun 6, 2017

iceman121 commented on Jun 11, 2017

notoraptor commented on Jun 12, 2017

iceman121 commented on Jun 12, 2017

abergeron commented on Jun 12, 2017

notoraptor commented on Jun 15, 2017

iceman121 commented on Jun 15, 2017

notoraptor commented on Jun 16, 2017

iceman121 commented on Jun 16, 2017

notoraptor commented on Jun 16, 2017

ankushbbbr commented on Sep 23, 2017

abergeron commented on Sep 25, 2017

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions