Forgetting retain_variables=True causes segfault in conv2d #1288
Description
Usually calling backwards twice gives this message:
RuntimeError: Trying to backward through the graph second time, but the buffers have already been freed. Please specify retain_variables=True when calling backward for the first time.
But conv2d just segfaults.
import torch
import torch.nn as nn
from torch.autograd import Variable
l1 = nn.Conv2d(3, 10, 4, 2, 1, bias=False).cuda()
input = Variable(torch.cuda.FloatTensor(5, 3, 8, 8))
o1 = l1(input)
print("A")
o1.sum().backward()
print("B")
o1.sum().backward()
print("C")$ pt test.py
A
B
Segmentation fault (core dumped)