undefined symbol: cudaSetupArgument

[WayneChen-cloud]

hi, after I run Make, I still meet this error when I run the demo.ipynb: ModuleNotFoundError: No module named 'nms.cpu_nms' How can i deal with it?

[WayneChen-cloud]

I try to add import pyximport pyximport.install() in nums.py, and it can find 'nms.cpu_nums', but there is another problem:

fatal error: numpy/arrayobject.h: 没有那个文件或目录

include "numpy/arrayobject.h"

      ^~~~~~~~~~~~~~~~~~~~~

compilation terminated. DistutilsExecError Traceback (most recent call last) ~/anaconda3/envs/transpose/lib/python3.9/distutils/unixccompiler.py in _compile(self, obj, src, ext, cc_args, extra_postargs, pp_opts) 116 try: --> 117 self.spawn(compiler_so + cc_args + [src, '-o', obj] + 118 extra_postargs)

~/anaconda3/envs/transpose/lib/python3.9/distutils/ccompiler.py in spawn(self, cmd) 909 def spawn(self, cmd): --> 910 spawn(cmd, dry_run=self.dry_run) 911

~/anaconda3/envs/transpose/lib/python3.9/distutils/spawn.py in spawn(cmd, search_path, verbose, dry_run) 90 cmd = cmd[0] ---> 91 raise DistutilsExecError( 92 "command %r failed with exit code %s" % (cmd, exitcode))

DistutilsExecError: command '/usr/bin/gcc' failed with exit code 1

[WayneChen-cloud]

After I "export CFLAGS=-I/nvme/usr/anaconda3/envs/transpose/lib/python3.9/site-packages/numpy/core/include" in the shell, the error is changed. The new error is as follows:

/nvme/usr/.pyxbld/temp.linux-x86_64-3.9/pyrex/nms/gpu_nms.c:628:10: fatal error: gpu_nms.hpp: 没有那个文件或目录

include "gpu_nms.hpp"

      ^~~~~~~~~~~~~

compilation terminated. how can I solve it?

[WayneChen-cloud]

hi, I run the 'make clean', and make it again. the error above disappare. But a new error occurs: ImportError: /nvme/usr/TransPose-main/lib/nms/gpu_nms.cpython-39-x86_64-linux-gnu.so: undefined symbol: cudaSetupArgument What does it mean? And how can i fix it?

[yangsenius]

Did you make the files in the correct directory? Please run the make command under the lib directory

[WayneChen-cloud]

hi, thanks for your reply! It seems that the function named 'locate_cuda()' didn't return the right results in my machine. So i just give it the right path of cuda and nvcc manually, and it work for me.