aidge_core.Tensor not working on Windows

After installing aidge, I tried to follow the tutorial and load the onnx but it crashed. Further investigation shows that it comes from the aigde_core.Tensor() call. To get the same crash I can just launch the code:

import aidge_backend_cpu
import aidge_core

print(aidge_core.Tensor([5, 5]))

PyCharm returns

Process finished with exit code -1073741819 (0xC0000005)

but otherwise I have no error message.

I don't know if this is related or if this is another issue but when I run aidge_core.Tensor() (empty arg) it returns

Assertion failed: mImpl && mImpl->hostPtr() != nullptr in C:\Users\AL253370\Documents\Projets\DeepGreen\aidge\aidge\aidge_core\include\aidge/data/Tensor.hpp:394Traceback (most recent c
all last):
  File "C:\Users\AL253370\Documents\Projets\DeepGreen\tuto_aidge\bug.py", line 4, in <module>
    print(aidge_core.Tensor())
RuntimeError: tensor should have a valid host pointer

Hello @alemesle,

If you look at available backends do you see aidge_backend_cpu ?

Edit: I have tested this on my Window machine and it works. I tought the issue was importing backend cpu before core but it works either way on my side.

Doing aidge_core.Tensor.get_available_backends() I get {'cpu'} and exchanging the import changes nothing

So looking at this post about you error code it looks like this error code is raised when Python fails to deduce the function to call due to ambiguous arg/return type.

The return type should be obvious (Tensor). So I feel like the issue would be on the argument type !

At the moment PyBind have always pointed to the right function to call with a default to int32 data when unspecified... (see).

But maybe this rely on an undefined behavior ? Which I don't understand why it would work on my windows computer ?

Maybe you can try:

>>> import aidge_core
>>> import aidge_backend_cpu
>>> import numpy as np
>>> a = aidge_core.Tensor(np.arange(5)).astype(np.int32)
>>> print(a)
{ 0, 1, 2, 3, 4}

And then try to change the dtype to check if we manage well some common type ?

Some dtype you can try:

• .astype(np.int64)
• .astype(np.float32)
• .astype(np.float64)

Cheers, Cyril

Ok well I've updated my version of Aidge and something fixed it

However, I still get the error while trying to print an empty Tensor print(aidge_core.Tensor()).

aidge_core!71 (merged) Should fix the issue !

closed