common.o fails build with cudaStreamCaptureModeThreadLocal undefined - CUDA 10 #90
Description
In some of our ML pipelines we still have dependencies on CUDA 10 while others use CUDA 11. This is an automated build pipeline and as of a couple of days ago the CUDA 10 docker image with the nccl-tests fails to compile common.o in the nccl-tests in the sm_70 gencode
Step 15/16 : RUN git clone https://github.com/NVIDIA/nccl-tests.git $HOME/nccl-tests && cd $HOME/nccl-tests && git checkout ${NCCL_TESTS_VERSION} && make MPI=1 MPI_HOME=/opt/amazon/openmpi/ CUDA_HOME=/usr/local/cuda NCCL_HOME=/opt/nccl/build NVCC_GENCODE="-gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_70,code=sm_70"
---> Running in 01fdf75d8d36
Cloning into '/tmp/nccl-tests'...
Already on 'master'
Your branch is up to date with 'origin/master'.
make -C src build
make[1]: Entering directory '/tmp/nccl-tests/src'
Compiling all_reduce.cu > ../build/all_reduce.o
Compiling common.cu > ../build/common.o
common.cu(542): error: identifier "cudaStreamCaptureModeThreadLocal" is undefined
common.cu(542): error: identifier "cudaStreamCaptureModeGlobal" is undefined
common.cu(542): error: too many arguments in function call
common.cu(606): error: identifier "cudaStreamCaptureModeThreadLocal" is undefined
common.cu(606): error: too many arguments in function call
5 errors detected in the compilation of "/tmp/tmpxft_0000004b_00000000-7_common.compute_70.cpp1.ii".
make[1]: *** [../build/common.o] Error 1
Makefile:82: recipe for target '../build/common.o' failed
make[1]: Leaving directory '/tmp/nccl-tests/src'
make: *** [src.build] Error 2