Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Separate compilation not working on Windows

See original GitHub issue

The following code runs fine on Linux but fails on Windows:

import cupy
kernel = cupy.RawKernel('extern "C" __global__ void test() {}', 'test', backend='nvcc', enable_cooperative_groups=True)
kernel.compile()

Error raised in Windows:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "cupy\core\raw.pyx", line 282, in cupy.core.raw.RawKernel.compile
  File "cupy\core\raw.pyx", line 110, in cupy.core.raw.RawKernel._kernel
  File "cupy\cuda\function.pyx", line 234, in cupy.cuda.function.Module.get_function
  File "cupy\cuda\function.pyx", line 175, in cupy.cuda.function.Function.__init__
  File "cupy_backends\cuda\api\driver.pyx", line 262, in cupy_backends.cuda.api.driver.moduleGetFunction
  File "cupy_backends\cuda\api\driver.pyx", line 124, in cupy_backends.cuda.api.driver.check_status
cupy_backends.cuda.api.driver.CUDADriverError: CUDA_ERROR_NOT_FOUND: named symbol not found

Internally, the above code does the following nvcc separate compilation:

nvcc -gencode=arch=compute_70,code=sm_70 -ftz=true -rdc=true -Xcompiler -fPIC -shared -o test_rdc.o test.cu
nvcc -gencode=arch=compute_70,code=sm_70 --cubin -ftz=true -rdc=true -Xcompiler -fPIC -shared test_rdc.o --device-link -o test_rdc.cubin

It generates a correct cubin on Linux:

% cuobjdump -symbols test_rdc.cubin

symbols:
STT_FUNC         STB_GLOBAL STO_ENTRY      test

But on Windows it generates an empty (?) cubin, which is causing named symbol not found error:

C:\Users\chainer\Documents>cuobjdump -all -symbols test_rdc.cubin

symbols:

It seems we need some other commands to do separate compilation on Windows.

cc/ @leofang @anaruse

Issue Analytics

State:
Created 3 years ago
Reactions:2
Comments:9 (9 by maintainers)

Top GitHub Comments

2reactions

kmaehashicommented, Dec 21, 2020

--device-c magically solved the issue! (#4470) Thank you very much 👍🏼

1reaction

anarusecommented, Dec 10, 2020

I don’t know well about Windows, but the Windows compiler probably dosen’t have a -fPIC option, does it?

nvcc -gencode=arch=compute_70,code=sm_70 -ftz=true -rdc=true -Xcompiler -fPIC -shared -o test_rdc.o test.cu
nvcc -gencode=arch=compute_70,code=sm_70 --cubin -ftz=true -rdc=true -Xcompiler -fPIC -shared test_rdc.o --device-link -o test_rdc.cubin