CPUAllocator running out of memory while assigning complex tensors

1

Hello all,
I just got started with complex tensors and tried assigning a complex tensor of shape (2,2) and it runs out of memory. My code is -

import torch
import torch.nn as nn

a = torch.rand(size=(2,2), dtype=torch.complex32)
print(a)

and the output error is -

Traceback (most recent call last):
  File "/home/atharva/PycharmProjects/DeepComplexNetwork/main.py", line 4, in <module>
    a = torch.rand(size=(2,2), dtype=torch.complex32)
RuntimeError: [enforce fail at CPUAllocator.cpp:64] . DefaultCPUAllocator: can't allocate memory: you tried to allocate 34359738368 bytes. Error code 12 (Cannot allocate memory)

I was just wondering if complex tensor of type complex32 even of size (2,2) 34359738368 bytes big ?
TIA…

2

I just realized that it works with complex128 and comples64 but not with complex32… complex32 tensors are bigger?

3

Hi,

Thanks for the report! This definitely looks like a bug: https://github.com/pytorch/pytorch/issues/43143
Note that complex32 is most likely not ready for general use.

Also could you share on the issue your pytorch version and how you installed it please?

4

Hi
Sorry for the late response… pytorch version is 1.6.0 and it was installed simply through pip3 - pip3 install torch torchvision

5

Hi,

Yes we were able to reproduce it and it should be fixed now if you use the nightly build :slight_smile:

6

Thanks a lot @albanD