root@fbecc9ba5bb2:/pytorch_knn_cuda# make

Separate cpp shared library that will be loaded to the extern C ffi

nvcc -c -o build/knn_cuda_kernel.so src/knn_cuda_kernel.cu -x cu -Xcompiler -fPIC -shared -Isrc make: nvcc: Command not found Makefile:34: recipe for target 'build/knn_cuda_kernel.so' failed make: *** [build/knn_cuda_kernel.so] Error 127 root@fbecc9ba5bb2:/pytorch_knn_cuda#

did anyone got this error as me, I do not know how to fix it.

Due to some API changes made in pytorch v0.4.1 pytorch_knn_cuda was no longer able to compile (see issue #3).
This PR makes the required changes, it now works with pytorch v.0.4.1

Getting error

gcc -pthread -B /home/bagon/anaconda3/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/bagon/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include -I/home/bagon/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/TH -I/home/bagon/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/THC -I/usr/local/cuda/include -I/home/bagon/anaconda3/pkgs/pytorch_knn_cuda/include -I/home/bagon/anaconda3/include/python3.6m -c /home/bagon/anaconda3/pkgs/pytorch_knn_cuda/src/knn_pytorch.c -o ./home/bagon/anaconda3/pkgs/pytorch_knn_cuda/src/knn_pytorch.o -std=c99
In file included from /home/bagon/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/THC/THCGeneral.h:4:0,
                 from /home/bagon/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/THC/THC.h:4,
                 from /home/bagon/anaconda3/pkgs/pytorch_knn_cuda/src/knn_pytorch.c:1:
/home/bagon/anaconda3/pkgs/pytorch_knn_cuda/src/knn_pytorch.c: In function ‘knn’:
/home/bagon/anaconda3/pkgs/pytorch_knn_cuda/src/knn_pytorch.c:11:24: error: dereferencing pointer to incomplete type ‘THTensor {aka struct THTensor}’
   THArgCheck(ref_tensor->nDimension == 2 , 0, "ref_tensor: 2D Tensor expected");
                        ^
/home/bagon/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/TH/THGeneral.h:111:35: note: in definition of macro ‘THArgCheck’
   _THArgCheck(__FILE__, __LINE__, __VA_ARGS__);

It seems like the definition of struct THTensor was moved from THTensor.h to THTensor.hpp and this causes this issue.

Any way to workaround this?

Tried on 10,000 reference points, 100 query points, and 3 feature dimensions, this crashes with the error:

RuntimeError: cuda runtime error (77) : an illegal memory access was encountered at /pytorch/torch/lib/THC/generic/THCTensorCopy.c:70

The error does not appear when executing the KNN code, but only when trying to assign or access any data on the GPU (including the results) after the code is called. Tested with GeForce GTX 1080 with 8 GB of VRAM.