Better error messages for blas ops on torch.cuda.LongTensor #4157
Description
Reported at https://discuss.pytorch.org/t/runtimeerror-unimplemented-data-type/11127.
import torch
from torch.autograd import Variable
x = Variable(torch.zeros(3, 3).long().random_(0, 10).cuda())
y = Variable(torch.zeros(3, 3).long().random_(0, 10).cuda())
torch.mm(x, y.t())
>>>
RuntimeError Traceback (most recent call last)
<ipython-input-3-472380edcc50> in <module>()
3 x = Variable(torch.zeros(3, 3).long().random_(0, 10).cuda())
4 y = Variable(torch.zeros(3, 3).long().random_(0, 10).cuda())
----> 5 torch.mm(x, y.t())
RuntimeError: unimplemented data type at /opt/conda/conda-bld/pytorch_1512386481460/work/torch/lib/THC/generic/THCTensorMathBlas.cu:377
unimplemented data type is a little confusing.
Will take care of this later.