Cupyを実行したときに「NVRTCError: NVRTC_ERROR

前提・実現したいこと

「Quadro K620」にnvidia社のCUDA toolkit 11.3をインストールし、python3.9.5にてCupyを実行したときエラーメッセージが表示されました。
自己解決できず困り果てています。

発生している問題

tarminal
1Traceback (most recent call last):
2  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 623, in compile
3    nvrtc.compileProgram(self.ptr, options)
4  File "cupy_backends\cuda\libs\nvrtc.pyx", line 133, in cupy_backends.cuda.libs.nvrtc.compileProgram
5  File "cupy_backends\cuda\libs\nvrtc.pyx", line 145, in cupy_backends.cuda.libs.nvrtc.compileProgram
6  File "cupy_backends\cuda\libs\nvrtc.pyx", line 64, in cupy_backends.cuda.libs.nvrtc.check_status
7cupy_backends.cuda.libs.nvrtc.NVRTCError: NVRTC_ERROR_COMPILATION (6)
8
9During handling of the above exception, another exception occurred:
10
11Traceback (most recent call last):
12  File "<stdin>", line 1, in <module>
13  File "cupy\_core\core.pyx", line 1070, in cupy._core.core.ndarray.__add__
14  File "cupy\_core\_kernel.pyx", line 1104, in cupy._core._kernel.ufunc.__call__
15  File "cupy\_core\_kernel.pyx", line 1127, in cupy._core._kernel.ufunc._get_ufunc_kernel
16  File "cupy\_core\_kernel.pyx", line 901, in cupy._core._kernel._get_ufunc_kernel
17  File "cupy\_core\_kernel.pyx", line 61, in cupy._core._kernel._get_simple_elementwise_kernel
18  File "cupy\_core\core.pyx", line 1995, in cupy._core.core.compile_with_cache
19  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 432, in compile_with_cache
20    return _compile_with_cache_cuda(
21  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 509, in _compile_with_cache_cuda
22    ptx, mapping = compile_using_nvrtc(
23  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 271, in compile_using_nvrtc
24    return _compile(source, options, cu_path,
25  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 255, in _compile
26    ptx, mapping = prog.compile(options, log_stream)
27  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 635, in compile
28    raise CompileException(log, self.src, self.name, options,
29cupy.cuda.compiler.CompileException: Catastrophic error: cannot open source file "C:\Users\User\AppData\Local\Temp\tmpyi95wkqt\df1479848d067b9d322bed29495eb57e_2.cubin.cu"
30
311 catastrophic error detected in the compilation of "C:\Users\User\AppData\Local\Temp\tmpyi95wkqt\df1479848d067b9d322bed29495eb57e_2.cubin.cu".
32Compilation terminated.

該当のソースコード

Python
1>>> Python 3.9.5 (tags/v3.9.5:0a7dcbd, May  3 2021, 17:27:52) [MSC v.1928 64 bit (AMD64)] on win32
2>>> Type "help", "copyright", "credits" or "license" for more information.
3
4import cupy,cupyx
5a=cupy.array([[1,2],[2,1]])
6b=cupy.array([[1,2],[2,1]])
7a+b

試したこと

「a+b」と同様に、「a*b」を実行したときも同じエラーが発生します。

Python
1import cupy,cupyx
2a=cupy.array([[1,2],[2,1]])
3b=cupy.array([[1,2],[2,1]])
4a+b

↓変更

Python
1import cupy,cupyx
2a=cupy.array([[1,2],[2,1]])
3b=cupy.array([[1,2],[2,1]])
4a*b

tarminal
1Traceback (most recent call last):
2  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 623, in compile
3    nvrtc.compileProgram(self.ptr, options)
4  File "cupy_backends\cuda\libs\nvrtc.pyx", line 133, in cupy_backends.cuda.libs.nvrtc.compileProgram
5  File "cupy_backends\cuda\libs\nvrtc.pyx", line 145, in cupy_backends.cuda.libs.nvrtc.compileProgram
6  File "cupy_backends\cuda\libs\nvrtc.pyx", line 64, in cupy_backends.cuda.libs.nvrtc.check_status
7cupy_backends.cuda.libs.nvrtc.NVRTCError: NVRTC_ERROR_COMPILATION (6)
8
9During handling of the above exception, another exception occurred:
10
11Traceback (most recent call last):
12  File "<stdin>", line 1, in <module>
13  File "cupy\_core\core.pyx", line 1086, in cupy._core.core.ndarray.__mul__
14  File "cupy\_core\_kernel.pyx", line 1104, in cupy._core._kernel.ufunc.__call__
15  File "cupy\_core\_kernel.pyx", line 1127, in cupy._core._kernel.ufunc._get_ufunc_kernel
16  File "cupy\_core\_kernel.pyx", line 901, in cupy._core._kernel._get_ufunc_kernel
17  File "cupy\_core\_kernel.pyx", line 61, in cupy._core._kernel._get_simple_elementwise_kernel
18  File "cupy\_core\core.pyx", line 1995, in cupy._core.core.compile_with_cache
19  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 432, in compile_with_cache
20    return _compile_with_cache_cuda(
21  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 509, in _compile_with_cache_cuda
22    ptx, mapping = compile_using_nvrtc(
23  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 271, in compile_using_nvrtc
24    return _compile(source, options, cu_path,
25  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 255, in _compile
26    ptx, mapping = prog.compile(options, log_stream)
27  File "C:\Users\User\AppData\Local\Programs\Python\Python39\lib\site-packages\cupy\cuda\compiler.py", line 635, in compile
28    raise CompileException(log, self.src, self.name, options,
29cupy.cuda.compiler.CompileException: Catastrophic error: cannot open source file "C:\Users\User\AppData\Local\Temp\tmprosp_cev\587d834fef5a82250d88756cc66d608b_2.cubin.cu"
30
311 catastrophic error detected in the compilation of "C:\Users\User\AppData\Local\Temp\tmprosp_cev\587d834fef5a82250d88756cc66d608b_2.cubin.cu".
32Compilation terminated.

補足情報（FW/ツールのバージョンなど）

windows 10
Python 3.9.5
cupy-cuda112 9.0.0

Python
1>>> import cupyx
2>>> print(cupyx.get_runtime_info())
3OS                           : Windows-10-10.0.19041-SP0
4CuPy Version                 : 9.0.0
5NumPy Version                : 1.20.1
6SciPy Version                : 1.6.1
7Cython Build Version         : 0.29.23
8Cython Runtime Version       : 0.29.23
9CUDA Root                    : C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3
10CUDA Build Version           : 11020
11CUDA Driver Version          : 11030
12CUDA Runtime Version         : 11030
13cuBLAS Version               : 11402
14cuFFT Version                : 10402
15cuRAND Version               : 10204
16cuSOLVER Version             : (11, 1, 1)
17cuSPARSE Version             : 11500
18NVRTC Version                : (11, 3)
19Thrust Version               : 101000
20CUB Build Version            : 101000
21Jitify Build Version         : 60e9e72
22cuDNN Build Version          : 8101
23cuDNN Version                : 8101
24NCCL Build Version           : None
25NCCL Runtime Version         : None
26cuTENSOR Version             : 10202
27cuSPARSELt Build Version     : None
28Device 0 Name                : NVIDIA Quadro K620
29Device 0 Compute Capability  : 50
30Device 0 PCI Bus ID          : 0000:01:00.0

jbpb0

2021/05/15 15:02

> cannot open source file "C:\Users\User\AppData\Local\Temp\tmpyi95wkqt\df1479848d067b9d322bed29495eb57e_2.cubin.cu" ユーザー名(上記の「User」のところの実際)に、日本語とかスペースとか記号とかが入ってませんでしょうか？ (そういうものが入ってたらダメかは分かりませんが)

jbpb0

2021/05/15 15:11

> CUDA toolkit 11.3をインストールし、python3.9.5にてCupyを実行下記を見ても、precompiled binaryはCUDA 11.2用までで、11.3用は用意されてないみたいですが、どのようにしてインストールしたのでしょうか？ https://pypi.org/project/cupy/ また、下記を見ると、正式にサポートされてるCUDAは11.2までのようです https://docs.cupy.dev/en/stable/install.html

yutatayu

2021/05/15 15:12 編集

jbpb0様へ確かにUserには日本語名である漢字が入っています。 User（日本語名）を変えて試してみたいと思います。

yutatayu

2021/05/15 15:17

jbpb0様へ CUDA toolkit 11.3がwindows10システムにインストールされています。ご記載のURLからCUDA toolkitをバージョン11.3から11.2へダウングレードを行い、動作確認したいと思います。

jbpb0

2021/05/15 15:23

CUDAのバージョン違いが、この質問のエラーの原因ではないとしても、他に何か支障が生じるかもしれないので、可能であればバージョンは正式にサポートされてるものに合わせることをお勧めします

yutatayu

2021/05/16 23:13 編集

jbpb0様へご返答、誠にありがとうございました。 CUDA toolkitをバージョン11.3から11.2へダウングレードを行いました。また、ユーザー名を日本語名（漢字）から英文字に切り替えるために、新たにユーザーアカウントを作成しました。結果として、正常動作いたしました。 ```Python Python 3.9.5 (tags/v3.9.5:0a7dcbd, May 3 2021, 17:27:52) [MSC v.1928 64 bit (AMD64)] on win32 Type "help", "copyright", "credits" or "license" for more information. >>> import cupy,cupyx >>> a=cupy.array([[1,2],[2,1]]) >>> b=cupy.array([[1.,2.],[2.,1.]]) >>> a+b array([[2., 4.], [4., 2.]]) >>> a*b array([[1., 4.], [4., 1.]]) >>> print(cupyx.get_runtime_info()) OS : Windows-10-10.0.19041-SP0 CuPy Version : 9.0.0 NumPy Version : 1.20.3 SciPy Version : 1.6.3 Cython Build Version : 0.29.23 Cython Runtime Version : 0.29.23 CUDA Root : C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2 CUDA Build Version : 11020 CUDA Driver Version : 11030 CUDA Runtime Version : 11020 cuBLAS Version : 11301 cuFFT Version : 10400 cuRAND Version : 10203 cuSOLVER Version : (11, 0, 2) cuSPARSE Version : 11301 NVRTC Version : (11, 2) Thrust Version : 101000 CUB Build Version : 101000 Jitify Build Version : 60e9e72 cuDNN Build Version : 8101 cuDNN Version : 8101 NCCL Build Version : None NCCL Runtime Version : None cuTENSOR Version : 10202 cuSPARSELt Build Version : None Device 0 Name : NVIDIA Quadro K620 Device 0 Compute Capability : 50 Device 0 PCI Bus ID : 0000:01:00.0 ```