CUDA: Invalid "7.5+PTX" GPU architecture is defined as one of the "common" architectures
I have a project with uses NVRTC, the CUDA run-time compilation library. This is its main CMakeLists.txt
, with the key lines being:
PROJECT(cuda-api-wrappers LANGUAGES CUDA CXX)
find_package(CUDAToolkit REQUIRED)
target_link_libraries(nvrtc INTERFACE CUDA::nvrtc)
and in a subdirectory I have:
target_link_libraries(vectorAdd_nvrtc nvrtc)
Now, I'm using Travis CI to build that project. I've upped the CMake version in there to 3.21.0-rc2 using the .travis.yml to make sure I'm filing this issue against the latest and greatest (but am also seeing this with 3.17.5 for example).
When the Travis machine comes up, this is what happens; and the key lines are:
-- The CUDA compiler identification is NVIDIA 10.2.89
...snip...
-- Found CUDAToolkit: /usr/local/cuda-10.2/include (found version "10.2.89")
...snip...
-- Automatic GPU detection failed. Building for common architectures.
-- Autodetected CUDA architecture(s): 3.5;5.0;5.3;6.0;6.1;7.0;7.5;7.5+PTX
There is no such architecture as "7.5+PTX". Is this a bug?
Edited by Brad King