CUDNN_STATUS_BAD_PARAM in BatchNorm backward #1866
Description
import torch
from torch import nn
import torch.nn.functional as F
from torch.autograd import Variable
bn = nn.BatchNorm1d(10)
bn = bn.cuda()
input = Variable(torch.Tensor(4, 10).cuda())
o = bn(input)
o.sum().backward()This throws a CUDNN_STATUS_BAD_PARAM on the backward call
Its fine in v0.12 and is fine in 5c74534 but broken in c573d53
Still trying to track it down but compiling takes ages