Conv double backwards with groups != 1 fails with size mismatch #1972
Description
Here's a commit that adds conv double backwards checks with groups != 1: gchanan@5b1269d
Applying this and running:
python test/test_nn.py TestNN.test_conv_double_backward
gives me:
======================================================================
ERROR: test_conv_double_backward (__main__.TestNN)
----------------------------------------------------------------------
Traceback (most recent call last):
File "test/test_nn.py", line 2439, in test_conv_double_backward
no_weight, groups = chan_in)
File "test/test_nn.py", line 2425, in run_conv_double_back_test
dummy_out = func(*inputs)
File "test/test_nn.py", line 2414, in func
out = F.conv2d(lx, lweight, lbias, stride, padding, dilation, groups)
File "/.../torch/nn/functional.py", line 50, in conv2d
return f(input, weight, bias)
RuntimeError: Need input of dimension 4 and input.size[1] == 2 but got input to be of shape: [2 x 1 x 6 x 6] at /.../torch/lib/THNN/generic/SpatialConvolutionMM.c:47
is this expected @albanD?