fix sanity

src/operator/nn/dnnl/dnnl_where-inl.h

@@ -25,8 +25,9 @@
 #define MXNET_OPERATOR_NN_DNNL_DNNL_WHERE_INL_H_
 
 #if MXNET_USE_ONEDNN == 1
+#include <memory>
+#include <unordered_map>
 #include <vector>
-
 #include "./dnnl_base-inl.h"
 #include "./dnnl_ops-inl.h"
 
@@ -45,7 +46,7 @@ class DNNLWhereFwd {
 
   static DNNLWhereFwd GetCached(const Tensors& tensors);
 
-  DNNLWhereFwd(const Tensors& tensors);
+  explicit DNNLWhereFwd(const Tensors& tensors);
 
   void Execute(const Tensors& tensors,
                const std::vector<OpReqType>& req,
@@ -69,4 +70,4 @@ bool SupportDNNLWhere(const std::vector<NDArray>& inputs);
 }  // namespace op
 }  // namespace mxnet
 #endif
-#endif  // MXNET_OPERATOR_NN_DNNL_DNNL_WHERE_INL_H_
+#endif  // MXNET_OPERATOR_NN_DNNL_DNNL_WHERE_INL_H_

src/operator/nn/dnnl/dnnl_where.cc

@@ -23,8 +23,11 @@
 
 #if MXNET_USE_ONEDNN == 1
 
+#include <algorithm>
+#include <set>
+#include <unordered_set>
 #include "dnnl_where-inl.h"
-#include "src/operator/operator_common.h"
+#include "operator/operator_common.h"
 
 namespace mxnet {
 namespace op {
@@ -168,7 +171,7 @@ void DNNLWhereFwd::Execute(const Tensors& tensors,
 
   // allocate temporary memory for 4 additional tensors
   mshadow::Tensor<cpu, 1> tmp_workspace = ctx.requested[0].get_space<cpu>(
-      mshadow::Shape1(tensors.output.shape().Size() * 4 * dtype_size, cpu_stream);
+      mshadow::Shape1(tensors.output.shape().Size() * 4 * dtype_size), cpu_stream);
   char* workspace_ptr   = reinterpret_cast<char*>(tmp_workspace.dptr_);
   const int offset_size = tensors.output.shape().Size() * dtype_size;
 
@@ -207,4 +210,4 @@ void DNNLWhereFwd::Execute(const Tensors& tensors,
 
 }  // namespace op
 }  // namespace mxnet
-#endif
+#endif