Fix input_channels divisibility check in concat_split_op (pytorch#31448)

Summary: Pull Request resolved: pytorch#31448 Replace `(!x%y)` with `(x%y != 0)` Test Plan: CI Reviewed By: orionr Differential Revision: D19165492 fbshipit-source-id: 246635fb8ddd5823196bcef9d0e6cdf1c349015e

Author: malfet

caffe2/operators/concat_split_op.cc

@@ -55,7 +55,7 @@ vector<TensorShape> TensorInferenceForSplit(
  // We cannot infer output shape until we see the value of split input
  return ret_invalid_shape();
  } else if (split.empty()) {
- if (!input_channels % output_size) {
+ if (input_channels % output_size != 0) {
  LOG(WARNING) << "Input channels (" << input_channels
  << ") should be divisible by number of outputs ("
  << output_size << ")";