I am trying to properly setup CUDA in order to take advantage of the GPU in Theano.
After fixing many compilation problems by tuning my .theanorc and nvcc.profile files, I am struggling to fix this linking error:
mod.cu(735): warning: conversion from pointer to smaller integer
mod.cu(1019): warning: statement is unreachable
mod.cu(735): warning: conversion from pointer to smaller integer
mod.cu(1019): warning: statement is unreachable
mod.cu
Creating library C:/Users/niluje/AppData/Local/Theano/compiledir_Windows-8-6.2.9200-Intel64_Family_6_Model_60_Stepping_3_GenuineIntel-2.7.10-64/cuda_ndarray/cuda_ndarray.lib and object C:/Users/niluje/AppData/Local/Theano/compiledir_Windows-8-6.2.9200-Intel64_Family_6_Model_60_Stepping_3_GenuineIntel-2.7.10-64/cuda_ndarray/cuda_ndarray.exp
MSVCRT.lib(atonexit.obj) : error LNK2019: unresolved external symbol __imp_EncodePointer referenced in function __atonexitinit
MSVCRT.lib(crtdll.obj) : error LNK2001: unresolved external symbol __imp_EncodePointer
MSVCRT.lib(atonexit.obj) : error LNK2019: unresolved external symbol __imp_DecodePointer referenced in function _onexit
MSVCRT.lib(crtdll.obj) : error LNK2001: unresolved external symbol __imp_DecodePointer
MSVCRT.lib(gs_report.obj) : error LNK2019: unresolved external symbol __imp_IsDebuggerPresent referenced in function __raise_securityfailure
MSVCRT.lib(gs_report.obj) : error LNK2019: unresolved external symbol IsProcessorFeaturePresent referenced in function __report_gsfailure
MSVCRT.lib(gs_support.obj) : error LNK2019: unresolved external symbol __imp_QueryPerformanceCounter referenced in function __security_init_cookie
MSVCRT.lib(gs_support.obj) : error LNK2019: unresolved external symbol __imp_GetCurrentProcessId referenced in function __security_init_cookie
MSVCRT.lib(gs_support.obj) : error LNK2019: unresolved external symbol __imp_GetCurrentThreadId referenced in function __security_init_cookie
MSVCRT.lib(gs_support.obj) : error LNK2019: unresolved external symbol __imp_GetSystemTimeAsFileTime referenced in function __security_init_cookie
MSVCRT.lib(dllmain.obj) : error LNK2019: unresolved external symbol __imp_DisableThreadLibraryCalls referenced in function DllMain
C:/Users/niluje/AppData/Local/Theano/compiledir_Windows-8-6.2.9200-Intel64_Family_6_Model_60_Stepping_3_GenuineIntel-2.7.10-64/cuda_ndarray/cuda_ndarray.pyd : fatal error LNK1120: 9 unresolved externals
Here's my .theanorcfile:
[global]
device = gpu
floatX = float32
[cuda]
root = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5
[nvcc]
flags = --use-local-env --cl-version=2013 -LC:\Users\niluje\Anaconda\Lib;
compiler_bindir=C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin\amd64
And here is my nvcc.profile file:
TOP = $(_HERE_)/..
NVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice
PATH += $(TOP)/open64/bin;$(TOP)/nvvm/bin;$(_HERE_);$(TOP)/lib;
INCLUDES += "-I$(TOP)/include" "-IC:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\include" "-IC:\Program Files\Microsoft SDKs\Windows\v6.0A\Include" $(_SPACE_)
LIBRARIES =+ $(_SPACE_) "/LIBPATH:$(TOP)/lib/$(_WIN_PLATFORM_)" "/LIBPATH:C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\lib\amd64" "/LIBPATH:C:\Program Files\Microsoft SDKs\Windows\v6.0A\Lib"
CUDAFE_FLAGS +=
PTXAS_FLAGS +=
It seems that this is not an uncommon error, but generally fixes involve changing some setting in the Visual Studio project. However, here I don't have a Visual Studio project. The code is dynamically generated by Theano and compiled at runtime.
Relevant system settings:
Damn it! I figured it out just after posting the question. The solution: slightly different include and library folders:
TOP = $(_HERE_)/..
NVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice
PATH += $(TOP)/open64/bin;$(TOP)/nvvm/bin;$(_HERE_);$(TOP)/lib;
INCLUDES += "-I$(TOP)/include" "-IC:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\include" "-IC:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\Include" $(_SPACE_)
LIBRARIES =+ $(_SPACE_) "/LIBPATH:$(TOP)/lib/$(_WIN_PLATFORM_)" "/LIBPATH:C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\lib\amd64" "/LIBPATH:C:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\Lib\x64"
CUDAFE_FLAGS +=
PTXAS_FLAGS +=
In particular I switched from:
C:\Program Files\Microsoft SDKs\Windows\v6.0A\
to:
C:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A
(I thought that I had to use the Program Files ones because of my 64bits project, but in fact the 64bits files are also included in Program Files (x86))