Build error: nvcc fatal: A single input file is required for a non-link phase when an outputfile is specified

Question

Build error: nvcc fatal: A single input file is required for a non-link phase when an outputfile is specified

yjdqk opened this issue 2 years ago · comments

yjdqk commented 2 years ago

nvcc fatal:unKnown option 'fPIC'

leslie_ds · Answer 1 · Tue Jan 18 2022 15:15:53 GMT+0800 (China Standard Time)

did you solve it

Mark · Answer 2 · Wed Jan 19 2022 00:54:58 GMT+0800 (China Standard Time)

I had problems with nvcc too. They went away when I upgraded to the latest version (11.6). Check your version with nvcc --version

mQvQ · Answer 3 · Thu Jan 20 2022 19:10:41 GMT+0800 (China Standard Time)

I had the problem too. Add ' -XCompiler' can solve it.
https://www.coder.work/article/3276225

Thomas Germer · Answer 4 · Fri Jan 21 2022 05:42:54 GMT+0800 (China Standard Time)

I also had the error message nvcc fatal : Unknown option 'fPIC -Xcompiler' which could be solved by adding a space before -fPIC in CMakeLists.txt

After that, I got the error nvcc fatal : Unknown option '-extended-lambda' which could again be solved by adding a space before that option in this file and also in dependencies/tiny-cuda-nn/CMakeLists.txt.

Now I am stuck at nvcc fatal : Unsupported gpu architecture 'compute_75+PTX'. I have an RTX 3060, so it should be supported according to the README, but maybe the nvcc version which NVIDIA ships by default on Ubuntu 20.04 is too old. I am reluctant to upgrade since I tried that a few times in the past and it never went well.

EDIT: ~~I upgraded nvcc to 11.6 and for the first time in history, it did not fry my entire system.~~
EDIT2: I spoke too early. First reboot worked, but after a second reboot this morning, the screen stayed black. Surprisingly, it came back on after a third reboot, but after a while, the process nvidia-powered appeared, which is running at 100% CPU utilization.
EDIT3: The system now randomly fails to boot.
EDIT4: The installation page for older cuda versions has incorrect instructions for deb (network). It will install the latest (broken) version instead.

New issues so far:

The installation instructions on https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_local are missing reboot
nvcc version 11.6 is now shadowed by the default nvcc, which is still 10.1. Fortunately, the fix is partially covered in the README (I replaced 11.4 with 11.6)

export PATH="/usr/local/cuda-11.6/bin:$PATH"
export LD_LIBRARY_PATH="/usr/local/cuda-11.6/lib64:$LD_LIBRARY_PATH"

Next, I got:

CMake Error in instant-ngp/build/CMakeFiles/CMakeTmp/CMakeLists.txt:
  CUDA_ARCHITECTURES is empty for target "cmTC_97a7d".

This could be fixed by defining CMAKE_CUDA_COMPILER as descriped in #25

cmake . -B build -DCMAKE_CUDA_COMPILER=/usr/local/cuda-11.6/bin/nvcc

After compiling for maybe 15 minutes, I get the next error:

/usr/bin/ld: libngp.a(testbed_image.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::Trainer<float, __half, __half>::training_step(CUstream_st*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, float*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const*)::{lambda()#1}>(CUstream_st*, bool, tcnn::Trainer<float, __half, __half>::training_step(CUstream_st*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, float*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const*)::{lambda()#1})':
tmpxft_00001340_00000000-6_testbed_image.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_7TrainerIf6__halfS3_E13training_stepEP11CUstream_stRKNS_9GPUMatrixIfLNS_12MatrixLayoutE1EEESB_PfPSA_EUlvE_EEvS6_bT_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_7TrainerIf6__halfS3_E13training_stepEP11CUstream_stRKNS_9GPUMatrixIfLNS_12MatrixLayoutE1EEESB_PfPSA_EUlvE_EEvS6_bT_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `tcnn::ShampooOptimizer<float>::step(CUstream_st*, float, float*, float*, float const*)::{lambda()#1}::operator()() const':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv]+0xacf): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv]+0xebb): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv]+0x12c1): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `tcnn::ShampooOptimizer<__half>::step(CUstream_st*, float, float*, __half*, __half const*)::{lambda()#1}::operator()() const':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv]+0xaec): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv]+0xed8): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o):tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv]+0x12de): more undefined references to `cublasSetWorkspace_v2' follow
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<float>::step(CUstream_st*, float, float*, float*, float const*)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<float>::step(CUstream_st*, float, float*, float*, float const*)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE4stepEP11CUstream_stfPfS6_PKfEUlvE_EEvS5_bT_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE4stepEP11CUstream_stfPfS6_PKfEUlvE_EEvS5_bT_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS6_bS9_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS6_bS9_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#2}>(CUstream_st*, bool, tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#2})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS6_bS9_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS6_bS9_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<__half>::step(CUstream_st*, float, float*, __half*, __half const*)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<__half>::step(CUstream_st*, float, float*, __half*, __half const*)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS3_PKS3_EUlvE_EEvS6_bT_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS3_PKS3_EUlvE_EEvS6_bT_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<__half>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<__half>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS7_bSA_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS7_bSA_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o):tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS7_bSA_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS7_bSA_]+0x406): more undefined references to `cudaGraphExecUpdate' follow
collect2: error: ld returned 1 exit status

Another issue mentions that a similar error could be fixed by defining CMAKE_CUDA_COMPILER, which I already do, so I am stuck again.

yjdqk · Answer 5 · Fri Jan 21 2022 11:57:24 GMT+0800 (China Standard Time)

I also had the error message nvcc fatal : Unknown option 'fPIC -Xcompiler' which could be solved by adding a space before -fPIC in CMakeLists.txt

After that, I got the error nvcc fatal : Unknown option '-extended-lambda' which could again be solved by adding a space before that option in this file and also in dependencies/tiny-cuda-nn/CMakeLists.txt.

Now I am stuck at nvcc fatal : Unsupported gpu architecture 'compute_75+PTX'. I have an RTX 3060, so it should be supported according to the README, but maybe the nvcc version which NVIDIA ships by default on Ubuntu 20.04 is too old. I am reluctant to upgrade since I tried that a few times in the past and it never went well.

EDIT: I upgraded nvcc to 11.6 and for the first time in history, it did not fry my entire system.

New issues so far:

The installation instructions on https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_local are missing reboot
nvcc version 11.6 is now shadowed by the default nvcc, which is still 10.1. Fortunately, the fix is partially covered in the README (I replaced 11.4 with 11.6)

export PATH="/usr/local/cuda-11.6/bin:$PATH"
export LD_LIBRARY_PATH="/usr/local/cuda-11.6/lib64:$LD_LIBRARY_PATH"

Next, I got:

CMake Error in instant-ngp/build/CMakeFiles/CMakeTmp/CMakeLists.txt:
  CUDA_ARCHITECTURES is empty for target "cmTC_97a7d".

This could be fixed by defining CMAKE_CUDA_COMPILER as descriped in #25

cmake . -B build -DCMAKE_CUDA_COMPILER=/usr/local/cuda-11.6/bin/nvcc

After compiling for maybe 15 minutes, I get the next error:

/usr/bin/ld: libngp.a(testbed_image.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::Trainer<float, __half, __half>::training_step(CUstream_st*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, float*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const*)::{lambda()#1}>(CUstream_st*, bool, tcnn::Trainer<float, __half, __half>::training_step(CUstream_st*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const&, float*, tcnn::GPUMatrix<float, (tcnn::MatrixLayout)1> const*)::{lambda()#1})':
tmpxft_00001340_00000000-6_testbed_image.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_7TrainerIf6__halfS3_E13training_stepEP11CUstream_stRKNS_9GPUMatrixIfLNS_12MatrixLayoutE1EEESB_PfPSA_EUlvE_EEvS6_bT_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_7TrainerIf6__halfS3_E13training_stepEP11CUstream_stRKNS_9GPUMatrixIfLNS_12MatrixLayoutE1EEESB_PfPSA_EUlvE_EEvS6_bT_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `tcnn::ShampooOptimizer<float>::step(CUstream_st*, float, float*, float*, float const*)::{lambda()#1}::operator()() const':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv]+0xacf): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv]+0xebb): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerIfE4stepEP11CUstream_stfPfS4_PKfENKUlvE_clEv]+0x12c1): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `tcnn::ShampooOptimizer<__half>::step(CUstream_st*, float, float*, __half*, __half const*)::{lambda()#1}::operator()() const':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv]+0xaec): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv]+0xed8): undefined reference to `cublasSetWorkspace_v2'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o):tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv[_ZZN4tcnn16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS1_PKS1_ENKUlvE_clEv]+0x12de): more undefined references to `cublasSetWorkspace_v2' follow
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<float>::step(CUstream_st*, float, float*, float*, float const*)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<float>::step(CUstream_st*, float, float*, float*, float const*)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE4stepEP11CUstream_stfPfS6_PKfEUlvE_EEvS5_bT_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE4stepEP11CUstream_stfPfS6_PKfEUlvE_EEvS5_bT_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS6_bS9_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS6_bS9_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#2}>(CUstream_st*, bool, tcnn::ShampooOptimizer<float>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#2})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS6_bS9_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerIfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS6_bS9_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<__half>::step(CUstream_st*, float, float*, __half*, __half const*)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<__half>::step(CUstream_st*, float, float*, __half*, __half const*)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS3_PKS3_EUlvE_EEvS6_bT_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE4stepEP11CUstream_stfPfPS3_PKS3_EUlvE_EEvS6_bT_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o): in function `void tcnn::CudaGraph::capture_and_execute<tcnn::ShampooOptimizer<__half>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1}>(CUstream_st*, bool, tcnn::ShampooOptimizer<__half>::inverse_pth_root_batched<float>(CUstream_st*, unsigned int, float*, tcnn::GPUMemory<float>&, unsigned int, unsigned int)::{lambda()#1})':
tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS7_bSA_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE_EEvS7_bSA_]+0x406): undefined reference to `cudaGraphExecUpdate'
/usr/bin/ld: dependencies/tiny-cuda-nn/src/libtiny-cuda-nn.a(optimizer.cu.o):tmpxft_00001087_00000000-6_optimizer.cudafe1.cpp:(.text._ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS7_bSA_[_ZN4tcnn9CudaGraph19capture_and_executeIZNS_16ShampooOptimizerI6__halfE24inverse_pth_root_batchedIfEEvP11CUstream_stjPfRNS_9GPUMemoryIT_EEjjEUlvE0_EEvS7_bSA_]+0x406): more undefined references to `cudaGraphExecUpdate' follow
collect2: error: ld returned 1 exit status

Another issue mentions that a similar error could be fixed by defining CMAKE_CUDA_COMPILER, which I already do, so I am stuck again.

my new issue is: nvcc fatal: A single input file is required for a non-link phase when an outputfile is specified

Thomas Germer · Answer 6 · Fri Jan 21 2022 17:37:40 GMT+0800 (China Standard Time)

I give up. My computer is broken now because of driver update. Of course no offense to the authors. It is not their fault that NVIDIA drivers are trash.

Best luck to all of you.

Carlos Gómez · Answer 7 · Mon Jan 31 2022 02:11:40 GMT+0800 (China Standard Time)

my new issue is: nvcc fatal: A single input file is required for a non-link phase when an outputfile is specified

Hi, has anyone been able to solve this issue?

SaschaHornauer · Answer 8 · Wed Feb 09 2022 00:48:18 GMT+0800 (China Standard Time)

By adding a -v to cmake, it will show you the line it is trying to execute. For me, I followed the advice to add a space but that will actually change the parameters which are generated. Where before, the parameters would be

nvcc ... -fPIC ...

now it was

nvcc ... " -fPIC" ...

and nvcc does not seem to understand that. So after updating my cmake I removed all spaces in both CMakeLists files again and this time it finishes compiling. You can check what your cmake command produces with:

cmake . -B build -DCMAKE_CUDA_COMPILER=/usr/local/cuda-11.6/bin/nvcc -v

Depending on your cuda version, for me it's 11.6. I think the culprit here could be cmake which changes how it adds these command line parameters between versions but I don't know for sure.

NcMasters · Answer 9 · Sat Feb 19 2022 00:40:48 GMT+0800 (China Standard Time)

you cant have file path with spaces. that solved my problem

Thomas Müller · Answer 10 · Sun Feb 20 2022 22:27:31 GMT+0800 (China Standard Time)