Fix the potential crash on L2NormOp

Dragon/python/setup.py

@@ -36,7 +36,7 @@ find_packages('dragon')
 find_modules()
 find_modules()
 
 
 setup(name = 'dragon',
 setup(name = 'dragon',
-      version='0.2.1.4',
+      version='0.2.1.5',
       description = 'Dragon: A Computation Graph Virtual Machine Based Deep Learning Framework',
       description = 'Dragon: A Computation Graph Virtual Machine Based Deep Learning Framework',
       url='https://github.com/neopenx/Dragon',
       url='https://github.com/neopenx/Dragon',
       author='Ting Pan',
       author='Ting Pan',

Dragon/src/contrib/rcnn/proposal_op.cc

@@ -104,7 +104,7 @@ void ProposalOp<Context>::RunWithType() {
     //  distribute rois into K bins
     //  distribute rois into K bins
     if (OutputSize() > 1) {
     if (OutputSize() > 1) {
         CHECK_EQ(max_level - min_level + 1, (int)OutputSize())
         CHECK_EQ(max_level - min_level + 1, (int)OutputSize())
-            << "Excepted " << OutputSize() << " outputs for levels between "
+            << "\nExcepted " << OutputSize() << " outputs for levels between "
             << "[" << min_level << ", " << max_level << "].";
             << "[" << min_level << ", " << max_level << "].";
         vector< vector<TIndex> > roi_bins(OutputSize(), vector<TIndex>());
         vector< vector<TIndex> > roi_bins(OutputSize(), vector<TIndex>());
         vector<T*> outputs;
         vector<T*> outputs;
@@ -130,7 +130,7 @@ template <class Context>
 void ProposalOp<Context>::RunOnDevice() {
 void ProposalOp<Context>::RunOnDevice() {
     num_images = input(0).dim(0);
     num_images = input(0).dim(0);
     CHECK_EQ(input(-1).count(), num_images * 3)
     CHECK_EQ(input(-1).count(), num_images * 3)
-        << "Excepted " << num_images * 3 << " groups image info, "
+        << "\nExcepted " << num_images * 3 << " groups image info, "
         << "but got " << input(-1).count() / 3 << ".";
         << "but got " << input(-1).count() / 3 << ".";
     roi_indices_.Reshape(vector<TIndex>(1, post_nms_top_n));
     roi_indices_.Reshape(vector<TIndex>(1, post_nms_top_n));
     output(0)->Reshape(vector<TIndex>({ num_images * post_nms_top_n, 5 }));
     output(0)->Reshape(vector<TIndex>({ num_images * post_nms_top_n, 5 }));

Dragon/src/contrib/rcnn/proposal_op.h

@@ -42,4 +42,4 @@ class ProposalOp final : public Operator<Context> {
 
 
 }    // namespace dragon
 }    // namespace dragon
 
 
-#endif    // DRAGON_OPERATORS_CONTRIB_RCNN_PROPOSAL_OP_H_
+#endif    // DRAGON_CONTRIB_RCNN_PROPOSAL_OP_H_
\ No newline at end of file

Dragon/src/operators/norm/l2_norm_op.cc

@@ -25,6 +25,7 @@ void L2NormOp<Context>::RunWithType() {
     auto* Ydata = output(0)->template mutable_data<T, Context>();
     auto* Ydata = output(0)->template mutable_data<T, Context>();
     auto* Bdata = buffer->template mutable_data<T, Context>();
     auto* Bdata = buffer->template mutable_data<T, Context>();
     auto* Ndata = norm->template mutable_data<T, Context>();
     auto* Ndata = norm->template mutable_data<T, Context>();
+    math::Set<T, Context>(norm->count(), dragon_cast<T, float>(eps), Ndata);
 
 
     for (int n = 0; n < outer_dim; n++) {
     for (int n = 0; n < outer_dim; n++) {
         if (across_inner) {
         if (across_inner) {
@@ -34,7 +35,6 @@ void L2NormOp<Context>::RunWithType() {
             Ndata_[n] = pow(sum_of_sqr + eps, 0.5);
             Ndata_[n] = pow(sum_of_sqr + eps, 0.5);
             math::Scale<T, Context>(buffer->count(), 1.0 / Ndata_[n], Xdata, Ydata);
             math::Scale<T, Context>(buffer->count(), 1.0 / Ndata_[n], Xdata, Ydata);
         } else {
         } else {
-            math::Set<T, Context>(norm->count(), dragon_cast<T, float>(eps), Ndata);
             math::Square<T, Context>(buffer->count(), Xdata, Bdata);
             math::Square<T, Context>(buffer->count(), Xdata, Bdata);
             //  compute T1 = \sum_{i} x_{i,j}^{2}
             //  compute T1 = \sum_{i} x_{i,j}^{2}
             math::Gemv<T, Context>(CblasTrans, dim, inner_dim,
             math::Gemv<T, Context>(CblasTrans, dim, inner_dim,