`bench_cg.py` fails with NVRTC Error

首先，使用 issue 中回報的 Python 和 ArrayFire 版本執行 examples/bench_cg.py，然後檢查第 144 行失敗的操作以及來自 src/backend/cuda/jit.cpp:298 的 NVRTC 錯誤。確定 benchmark 為什麼無法在回報的 CUDA 環境中編譯此操作；完成的標準是 benchmark 在沒有 NVRTC 編譯錯誤的情況下完成其測試。

Running bench_cg. py from the examples, gives me this error:

python bench_cg.py
ArrayFire v3.6.2 (CUDA, 64-bit Linux, build dc38ef1)
Platform: CUDA Toolkit 10, Driver: 410.79
[0] Tesla K80, 11442 MB, CUDA Compute 3.7
-1- Tesla K80, 11442 MB, CUDA Compute 3.7
-2- Tesla K80, 11442 MB, CUDA Compute 3.7
-3- Tesla K80, 11442 MB, CUDA Compute 3.7

Testing benchmark functions...
Traceback (most recent call last):
  File "bench_cg.py", line 197, in <module>
    test()
  File "bench_cg.py", line 144, in test
    if af.sum(af.abs(x1 - x2)/x2 > 1e-6):
  File "/home/vanderaa/miniconda3/envs/arrayfire/lib/python3.7/site-packages/arrayfire/array.py", line 894, in __truediv__
    return _binary_func(self, other, backend.get().af_div)
  File "/home/vanderaa/miniconda3/envs/arrayfire/lib/python3.7/site-packages/arrayfire/array.py", line 175, in _binary_func
    safe_call(c_func(c_pointer(out.arr), lhs.arr, other.arr, _bcast_var.get()))
  File "/home/vanderaa/miniconda3/envs/arrayfire/lib/python3.7/site-packages/arrayfire/util.py", line 79, in safe_call
    raise RuntimeError(to_str(err_str))
RuntimeError: In function std::vector<char> cuda::compileToPTX(const char*, std::string)
In file src/backend/cuda/jit.cpp:298
NVRTC Error(6): NVRTC_ERROR_COMPILATION

System: Linux K80 4.15.0-1035-azure #36~16.04.1-Ubuntu SMP Fri Nov 30 15:25:49 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
python --version:
Python 3.7.1

$pip  show arrayfire
Name: arrayfire
Version: 3.6.20181017
Summary: Python bindings for ArrayFire
Home-page: http://arrayfire.com
Author: Pavan Yalamanchili
Author-email: contact@pavanky.com
License: BSD
Location: /mnt/vanderaa/miniconda3/envs/arrayfire/lib/python3.7/site-packages
Requires:
Required-by: