error when make

[TianheLu]

Hello, I tried to make the nccl, but I met error as follows: transport/p2p.cc: In function ‘ncclResult_t ncclP2pFreeShareableBuffer(ncclIpcDesc*)’: transport/p2p.cc:220:5: error: ‘CUmemAllocationHandleType’ was not declared in this scope 220 | CUmemAllocationHandleType type = NCCL_P2P_HANDLE_TYPE; | ^~~~~~~~~ transport/p2p.cc:222:9: error: ‘type’ was not declared in this scope 222 | if (type == CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR) { | ^~~~ make[2]: Entering directory '/root/nccl/nccl/src/collectives/device' transport/p2p.cc:222:17: error: ‘CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR’ was not declared in this scope 222 | if (type == CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR) { | ^~~~~~~~~~~~ make[1]: *** [Makefile:119: /root/nccl/nccl/build/obj/transport/p2p.o] Error 1

I don't know why it happens, and I believe there will be many errors like this. Thanks so much.

[KaimingOuyang]

Perhaps you at least need to upgrade your cuda toolkit to 11.0

[TianheLu]

Thanks for helping. I use "nvcc -V" and the result is: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2022 NVIDIA Corporation Built on Mon_Oct_24_19:12:58_PDT_2022 Cuda compilation tools, release 12.0, V12.0.76 Build cuda_12.0.r12.0/compiler.31968024_0

And I use "nvidia-smi", the result is: NVIDIA-SMI 525.60.13 Driver Version: 525.60.13 CUDA Version: 12.0

It seems that my CUDA version is 12.0. I still don't know why this error happens.

Thank you in advance.

[KaimingOuyang]

what is your compilation command?

Can you show me echo $CUDA_HOME?

[TianheLu]

what is your compilation command?

Can you show me echo $CUDA_HOME?

The result of echo $CUDA_HOME is: :/usr/local/cuda I use 'make -j src.build' to make.

[KaimingOuyang]

Just want to confirm your CUDA_HOME is /usr/local/cuda instead of :/usr/local/cuda?

And then, can you try make -j src.build CUDA_HOME=$CUDA_HOME?

[TianheLu]

I tried to change my CUDA_HOME from :/usr/local/cuda to /usr/local/cuda.

And also tried make -j src.build CUDA_HOME=$CUDA_HOME.

But I met the same error.

Just want to confirm your CUDA_HOME is /usr/local/cuda instead of :/usr/local/cuda?

And then, can you try make -j src.build CUDA_HOME=$CUDA_HOME?

[KaimingOuyang]

Looks weird. Could you please share your env variable PATH and output of $CUDA_HOME/bin/nvcc -V? and do you have any env variable that sets up the include path such as C_INCLUDE_PATH or CXX_INCLUDE_PATH

[TianheLu]

The PATH is: /opt/anaconda3/bin:/opt/anaconda3/condabin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/usr/local/cuda/bin

The result of $CUDA_HOME/bin/nvcc -V is: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2022 NVIDIA Corporation Built on Mon_Oct_24_19:12:58_PDT_2022 Cuda compilation tools, release 12.0, V12.0.76 Build cuda_12.0.r12.0/compiler.31968024_0

And I have no include path such as C_INCLUDE_PATH or CXX_INCLUDE_PATH.

I also think it's weird. Thanks.

Looks weird. Could you please share your env variable PATH and output of $CUDA_HOME/bin/nvcc -V? and do you have any env variable that sets up the include path such as C_INCLUDE_PATH or CXX_INCLUDE_PATH

[KaimingOuyang]

If so, could you please upgrade your cuda to 12.1? I am afraid your previous cuda installation has a problem.

[TianheLu]

If so, could you please upgrade your cuda to 12.1? I am afraid your previous cuda installation has a problem.

Yeah, I will have a try. Thank you very much.