cudarray icon indicating copy to clipboard operation
cudarray copied to clipboard

Error while installing

Open theasder opened this issue 9 years ago • 8 comments

Trying to install cudarray, however getting error causing installation is being terminated

c++ -bundle -undefined dynamic_lookup -arch x86_64 -arch i386 -Wl,-F. build/temp.macosx-10.10-intel-2.7/./cudarray/wrap/cudart.o -L/usr/local/cuda/lib -lcudart -lcudarray -o build/lib.macosx-10.10-intel-2.7/cudarray/wrap/cudart.so -fPIC
ld: library not found for -lcudarray
clang: error: linker command failed with exit code 1 (use -v to see invocation)
error: command 'c++' failed with exit status 1

theasder avatar Sep 10 '15 18:09 theasder

The linker complains that it cannot find the libcudarray library file. Maybe you can try to figure out where it is located and try to add this location to the library paths in setup.py.

I'm sorry, but my ability to help in this case is limited - I have never tried installing CUDArray using clang.

andersbll avatar Sep 10 '15 18:09 andersbll

Also, where can I get nvcc file?

$ make install
nvcc -gencode arch=compute_20,code=sm_20 -gencode arch=compute_20,code=compute_20 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_30,code=compute_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_35,code=compute_35 -O3 --compiler-options '-O3 -fPIC -Wall -Wfatal-errors' --ftz=true --prec-div=false -prec-sqrt=false --fmad=true -I./include -I/usr/local/cuda/include -c -o src/nnet/pool_b01.o src/nnet/pool_b01.cu
make: nvcc: No such file or directory
make: *** [src/nnet/pool_b01.o] Error 1

theasder avatar Sep 10 '15 19:09 theasder

nvcc is the CUDA compiler. You need to install the CUDA SDK.

andersbll avatar Sep 11 '15 07:09 andersbll

EDIT - never mind the first paragraph, I'm getting make to half-work with minGW, still trying to get it to pick up all the header files. The 2nd paragraph still holds - do you know anyone who has successfully got to run cudarray on Win64? If all fails, could i swap something like PyCUDA in for it maybe?

Hijacking the thread a little bit - I have run into a similar issue when trying to build cudarray for windows. After some fiddling with setuptools, I got to the point where I get the same error, the linker doesn't find cudarray.lib - however, I can't make it find it since at link-time, it is nowhere on my system. Can you provide any hints where in the build process I could look to find out why it doesn't get built?

Nb: Since I'm building on Windows, I am not running "make / make install" - I wonder if I actually could/have to try that with minGW. Do you have any experience/reports of successfully building cudarray on Windows 64-bit?

w0nk0 avatar Sep 11 '15 12:09 w0nk0

@andersbll I have installed CUDA SDK, though terminal can't find nvcc so far.

theasder avatar Oct 28 '15 14:10 theasder

Guys if there is any update on this, i am getting the same error. I have installed CUDA SDK but still getting the error. Am i missing something ?

mhassaan avatar Feb 13 '16 03:02 mhassaan

in the Makefile, this worked for me:

NVCC = /Developer/NVIDIA/CUDA-7.5/bin/nvcc

nresnick avatar Apr 12 '16 20:04 nresnick

I think you didn't run:

make make install

before running:

sudo python setup.py install

adnphllps avatar Jul 05 '16 00:07 adnphllps