Closed pavan4 closed 5 years ago
Hi, I can't remember seeing this error before. Just a guess: are there perhaps any spaces in the paths? Maybe those aren't properly handled. Otherwise, I would try to run VERBOSE=1 make
which should print the exact compiler invocations that are being done. Then one can try to modify the problematic ones to find out what exactly is wrong with them.
It was a bug inside nvcc cmake
Fix: find this file - run_nvcc.cmake
usually something like /usr/share/cmake-3.5/Modules/FindCUDA/run_nvcc.cmake
and change set(nvcc_flags @nvcc_flags@)
to set(nvcc_flags "@nvcc_flags@")
(add double quotes)
Hi @puzzlepaint,
I did notice you mentioned that this compiles with
Ubuntu 14.04
. I am trying to work with16.04
When I compile I get the following issue
nvcc fatal : A single input file is required for a non-link phase when an outputfile is specified
in the generated cmake file for the cuda code specifically for
cuda_unprojection_lookup.cu
andcuda_buffer.cu
Any obvious fixes here?
I tried changing cmake versions from
2.8.12, 3.2.2, 3.5.1
All have the same issue. I am oncuda 8.0
.