Search code examples
cmakecudanvcc

CUDA compilation with relocatable code: "Could not find fatbin in ..."

As part of a larger CMake project, I am adding a CUDA library. The rest of the project is C++, compiled with clang.

To test that the library works correctly, I'm creating a small executable and linking the CUDA library to it:

add_library(kernels STATIC
    kernels.cu
)
set_target_properties(kernels PROPERTIES CUDA_SEPARABLE_COMPILATION ON)

# --- Test executable
add_executable(main
    main.cpp
)
target_link_libraries(main PRIVATE kernels)

The library compiles fine, but I get the following error when nvcc is invoked to do the device linking part of the process on my executable (target main):

nvlink fatal   : Could not find fatbin in '[some long path]/main.cpp.o'
nvlink fatal   : elfLink internal error

What is preventing this step from working?

Solution

  • I couldn't reproduce this issue in a fresh, tiny CMake project, so I eventually figured out that some flag from my larger project wasn't playing along.

    It turns out that Thin LTO, which was enabled in CMAKE_CXX_FLAGS is causing this issue. I disabled it for this particular target with:

    target_compile_options(main PRIVATE "-fno-lto")