build error due to legacy shuffle API

[jeng1220]

it threw:

nvcc -O3 -std=c++11 --compiler-options=-fPIC -use_fast_math --ptxas-options="-v" -gencode arch=compute_70,code=sm_70 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_30,code=sm_30 -DOMPI_SKIP_MPICXX --keep-dir /test/workspace/ctr/amazon-dsstne/build/tmp/engine/cuda -I/usr/local/include -isystem /usr/local/cuda/include -isystem /usr/lib/openmpi/include -isystem /usr/include/jsoncpp -IB40C -IB40C/KernelCommon -I/test/workspace/ctr/amazon-dsstne/build/include -I../utils  -c kernels.cu -o /test/workspace/ctr/amazon-dsstne/build/tmp/engine/cuda/kernels.o
ptxas /tmp/tmpxft_00000a17_00000000-8_kernels.compute_70.ptx, line 61962; error   : Instruction 'shfl' without '.sync' is not supported on .target sm_70 and higher from PTX ISA version 6.4

the old CUDA shuffle API have been deprecated in CUDA 9.0, and not available after CUDA 10 (link)

• CUDA: 10.1
• GPU: NVIDIA Tesla V100

[jeng1220]

after investigation, this problem was caused by old version of CUB, which referred by setup instructions (https://github.com/amzn/amazon-dsstne/blob/master/docs/getting_started/setup.md#cub-setup)

version 1.5.2 is too old for Tesla V100. could anyone update setup instructions?

[scottlegrand]

So first problem is that I do not believe anyone is left at Amazon to accept pull requests. But since you seem to work at NVIDIA and I do too, we could at least maintain my fork of it. I am surprised it's not working under CUDA 10 because I have had no issues building for Turing. But I thought we already handled this correctly because it was an issue as of CUDA 9? Also where is the CUB version in the source? I am betting my local version is newer than 1.5.2 and that might have obscured this for some time.

[jeng1220]

this line also uses old shuffle API, which needs to be fixed https://github.com/amzn/amazon-dsstne/blob/master/src/amazon/dsstne/knn/topk.cu#L141