cupy.RawModule error: "AttributeError: 'NoneType' object has no attribute 'split'"

• Conditions (you can just paste the output of python -c 'import cupy; cupy.show_config()')

OS                           : Linux-5.9.10-1-MANJARO-x86_64-with-glibc2.10
CuPy Version                 : 8.1.0
NumPy Version                : 1.19.4
SciPy Version                : None
CUDA Root                    : /opt/cuda-10.1
CUDA Build Version           : 10010
CUDA Driver Version          : 11010
CUDA Runtime Version         : 10010
cuBLAS Version               : 10201
cuFFT Version                : 10101
cuRAND Version               : 10101
cuSOLVER Version             : (10, 2, 0)
cuSPARSE Version             : 10300
NVRTC Version                : (10, 1)
Thrust Version               : 100906
CUB Build Version            : 100800
cuDNN Build Version          : None
cuDNN Version                : None
NCCL Build Version           : 2708
NCCL Runtime Version         : 2708
cuTENSOR Version             : None
Device 0 Name                : GeForce GTX 1080 Ti
Device 0 Compute Capability  : 61

• Code to reproduce Testing class cupy.RawModule, I was trying to reproduce the example from https://github.com/cupy/cupy/issues/3728#issuecomment-670018685.

import cupy
N = 100
code = """
#include <thrust/count.h>

extern "C" __global__
void xyzw_frequency_thrust_device(int *count, char *text, int n)
{
  const char letters[] { 'x','y','z','w' };

  *count = thrust::count_if(thrust::device, text, text+n, [=](char c) {
    for (const auto x : letters) 
      if (c == x) return true;
    return false;
  });
}"""
kernel=cupy.RawModule(code=code,backend='nvcc')
code = kernel.get_function("xyzw_frequency_thrust_device")

• Error messages, stack traces, or logs

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-32-9903428f9696> in <module>
     16 }"""
     17 kernel=cupy.RawModule(code=code,backend='nvcc')
---> 18 code = kernel.get_function("xyzw_frequency_thrust_device")

cupy/core/raw.pyx in cupy.core.raw.RawModule.get_function()

cupy/core/raw.pyx in cupy.core.raw.RawKernel.kernel.__get__()

cupy/core/raw.pyx in cupy.core.raw.RawKernel._kernel()

cupy/_util.pyx in cupy._util.memoize.decorator.ret()

cupy/core/raw.pyx in cupy.core.raw._get_raw_module()

cupy/core/core.pyx in cupy.core.core.compile_with_cache()

cupy/core/core.pyx in cupy.core.core.compile_with_cache()

/opt/miniconda3/lib/python3.8/site-packages/cupy/cuda/compiler.py in compile_with_cache(source, options, arch, cache_dir, extra_source, backend, enable_cooperative_groups, name_expressions, log_stream)
    333             name_expressions, log_stream, cache_in_memory)
    334     else:
--> 335         return _compile_with_cache_cuda(
    336             source, options, arch, cache_dir, extra_source, backend,
    337             enable_cooperative_groups, name_expressions, log_stream,

/opt/miniconda3/lib/python3.8/site-packages/cupy/cuda/compiler.py in _compile_with_cache_cuda(source, options, arch, cache_dir, extra_source, backend, enable_cooperative_groups, name_expressions, log_stream, cache_in_memory)
    365     if base is None:
    366         # This is checking of NVRTC compiler internal version
--> 367         base = _preprocess('', options, arch, backend)
    368         _empty_file_preprocess_cache[env] = base
    369 

/opt/miniconda3/lib/python3.8/site-packages/cupy/cuda/compiler.py in _preprocess(source, options, arch, backend)
    280     elif backend == 'nvcc':
    281         try:
--> 282             result = compile_using_nvcc(source, options, arch, 'preprocess.cu',
    283                                         code_type='ptx')
    284         except CompileException as e:

/opt/miniconda3/lib/python3.8/site-packages/cupy/cuda/compiler.py in compile_using_nvcc(source, options, arch, filename, code_type, separate_compilation, log_stream)
    194     _nvcc = get_nvcc_path()
    195     # split() is needed because _nvcc could come from the env var NVCC
--> 196     cmd = _nvcc.split()
    197     cmd.append(arch_str)
    198 

AttributeError: 'NoneType' object has no attribute 'split'

[lucasrodes]

Hi @leofang, indeed the code from the first example is working, thanks a lot.

Your 2nd example is illegal code. Only kernels and device functions can be compiled and loaded by RawModule, not host functions.

Now I see what you mean. I am new to CUDA programming, so all these issues are new to me.

Once again, thanks for your time & help.

[leofang]

Glad it worked! Welcome to the GPU wonderland!