nvcc linker issue?

[jamesyoungdigital]

nvcc -dlink -o cuda_libs.o *.cu.o -lcudadevrt -lcudart
nvlink warning : SM Arch ('sm_30') not found in 'CudaAtomicList.cu.o'
nvlink warning : SM Arch ('sm_30') not found in 'CudaDeviceKeys.cu.o'
nvlink warning : SM Arch ('sm_30') not found in 'CudaHashLookup.cu.o'
nvlink warning : SM Arch ('sm_30') not found in 'CudaKeySearchDevice.cu.o'
nvlink warning : SM Arch ('sm_30') not found in 'cudabridge.cu.o'

FYI I built with COMPUTE_CAP=70, so maybe it's not supposed to find sm_30 in the object files.

[prodnet]

What is this COMPUTE_CAP=70 ?

[jamesyoungdigital]

It's the sm_70 or compute capability guide for certain CUDA capable cards. In the case of a Tesla V100, COMPUTE_CAP = 70. This guides the nvidia CUDA compiler.

…

On Thu, 29 Nov. 2018, 8:52 am prodnet ***@***.*** wrote: What is this COMPUTE_CAP=70 ? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#90 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Ao5ZJ9wB6mPBKyej_7FHwedM8ZN27CjBks5uzwWtgaJpZM4Y4Xvs> .

[brichard19]

Nvcc by default assumes sm_30 architecture.

The warning goes away if you pass ${NVCCFLAGS}, which contains the arch flags.