[Misc] clang-format auto fix. (#4810)

* [Misc] clang-format auto fix.

* manual

* manual
Co-authored-by: Steve <ubuntu@ip-172-31-34-29.ap-northeast-1.compute.internal>

src/graph/serialize/zerocopy_serializer.cc

@@ -13,8 +13,8 @@ namespace dgl {
 
 using dgl::runtime::NDArray;
 
-NDArray CreateNDArrayFromRawData(std::vector<int64_t> shape, DGLDataType dtype,
-                                 DGLContext ctx, void* raw) {
+NDArray CreateNDArrayFromRawData(
+    std::vector<int64_t> shape, DGLDataType dtype, DGLContext ctx, void* raw) {
   return NDArray::CreateFromRaw(shape, dtype, ctx, raw, true);
 }
 
@@ -40,7 +40,8 @@ void StreamWithBuffer::PushNDArray(const NDArray& tensor) {
     // If the stream is for remote communication or the data is not stored in
     // shared memory, serialize the data content as a buffer.
     this->Write<bool>(false);
-    // If this is a null ndarray, we will not push it into the underlying buffer_list
+    // If this is a null ndarray, we will not push it into the underlying
+    // buffer_list
     if (data_byte_size != 0) {
       buffer_list_.emplace_back(tensor, tensor->data, data_byte_size);
     }
@@ -90,8 +91,8 @@ NDArray StreamWithBuffer::PopNDArray() {
       // Mean this is a null ndarray
       ret = CreateNDArrayFromRawData(shape, dtype, cpu_ctx, nullptr);
     } else {
-      ret = CreateNDArrayFromRawData(shape, dtype, cpu_ctx,
-                                     buffer_list_.front().data);
+      ret = CreateNDArrayFromRawData(
+          shape, dtype, cpu_ctx, buffer_list_.front().data);
       buffer_list_.pop_front();
     }
     return ret;

src/graph/shared_mem_manager.cc

@@ -31,8 +31,8 @@ using namespace dgl::aten;
 namespace dgl {
 
 template <>
-NDArray SharedMemManager::CopyToSharedMem<NDArray>(const NDArray &data,
-                                                   std::string name) {
+NDArray SharedMemManager::CopyToSharedMem<NDArray>(
+    const NDArray &data, std::string name) {
   DGLContext ctx = {kDGLCPU, 0};
   std::vector<int64_t> shape(data->shape, data->shape + data->ndim);
   strm_->Write(data->ndim);
@@ -53,21 +53,22 @@ NDArray SharedMemManager::CopyToSharedMem<NDArray>(const NDArray &data,
 }
 
 template <>
-CSRMatrix SharedMemManager::CopyToSharedMem<CSRMatrix>(const CSRMatrix &csr,
-                                                       std::string name) {
+CSRMatrix SharedMemManager::CopyToSharedMem<CSRMatrix>(
+    const CSRMatrix &csr, std::string name) {
   auto indptr_shared_mem = CopyToSharedMem(csr.indptr, name + "_indptr");
   auto indices_shared_mem = CopyToSharedMem(csr.indices, name + "_indices");
   auto data_shared_mem = CopyToSharedMem(csr.data, name + "_data");
   strm_->Write(csr.num_rows);
   strm_->Write(csr.num_cols);
   strm_->Write(csr.sorted);
-  return CSRMatrix(csr.num_rows, csr.num_cols, indptr_shared_mem,
-                   indices_shared_mem, data_shared_mem, csr.sorted);
+  return CSRMatrix(
+      csr.num_rows, csr.num_cols, indptr_shared_mem, indices_shared_mem,
+      data_shared_mem, csr.sorted);
 }
 
 template <>
-COOMatrix SharedMemManager::CopyToSharedMem<COOMatrix>(const COOMatrix &coo,
-                                                       std::string name) {
+COOMatrix SharedMemManager::CopyToSharedMem<COOMatrix>(
+    const COOMatrix &coo, std::string name) {
   auto row_shared_mem = CopyToSharedMem(coo.row, name + "_row");
   auto col_shared_mem = CopyToSharedMem(coo.col, name + "_col");
   auto data_shared_mem = CopyToSharedMem(coo.data, name + "_data");
@@ -75,13 +76,14 @@ COOMatrix SharedMemManager::CopyToSharedMem<COOMatrix>(const COOMatrix &coo,
   strm_->Write(coo.num_cols);
   strm_->Write(coo.row_sorted);
   strm_->Write(coo.col_sorted);
-  return COOMatrix(coo.num_rows, coo.num_cols, row_shared_mem, col_shared_mem,
+  return COOMatrix(
+      coo.num_rows, coo.num_cols, row_shared_mem, col_shared_mem,
       data_shared_mem, coo.row_sorted, coo.col_sorted);
 }
 
 template <>
-bool SharedMemManager::CreateFromSharedMem<NDArray>(NDArray *nd,
-                                                    std::string name) {
+bool SharedMemManager::CreateFromSharedMem<NDArray>(
+    NDArray *nd, std::string name) {
   int ndim;
   DGLContext ctx = {kDGLCPU, 0};
   DGLDataType dtype;
@@ -98,15 +100,14 @@ bool SharedMemManager::CreateFromSharedMem<NDArray>(NDArray *nd,
   if (is_null) {
     *nd = NDArray::Empty(shape, dtype, ctx);
   } else {
-    *nd =
-      NDArray::EmptyShared(graph_name_ + name, shape, dtype, ctx, false);
+    *nd = NDArray::EmptyShared(graph_name_ + name, shape, dtype, ctx, false);
   }
   return true;
 }
 
 template <>
-bool SharedMemManager::CreateFromSharedMem<COOMatrix>(COOMatrix *coo,
-                                                      std::string name) {
+bool SharedMemManager::CreateFromSharedMem<COOMatrix>(
+    COOMatrix *coo, std::string name) {
   CreateFromSharedMem(&coo->row, name + "_row");
   CreateFromSharedMem(&coo->col, name + "_col");
   CreateFromSharedMem(&coo->data, name + "_data");
@@ -118,8 +119,8 @@ bool SharedMemManager::CreateFromSharedMem<COOMatrix>(COOMatrix *coo,
 }
 
 template <>
-bool SharedMemManager::CreateFromSharedMem<CSRMatrix>(CSRMatrix *csr,
-                                                      std::string name) {
+bool SharedMemManager::CreateFromSharedMem<CSRMatrix>(
+    CSRMatrix *csr, std::string name) {
   CreateFromSharedMem(&csr->indptr, name + "_indptr");
   CreateFromSharedMem(&csr->indices, name + "_indices");
   CreateFromSharedMem(&csr->data, name + "_data");

src/graph/shared_mem_manager.h

@@ -29,8 +29,7 @@ const size_t SHARED_MEM_METAINFO_SIZE_MAX = 1024 * 32;
 class SharedMemManager : public dmlc::Stream {
  public:
   explicit SharedMemManager(std::string graph_name, dmlc::Stream* strm)
-      : graph_name_(graph_name),
-        strm_(strm) {}
+      : graph_name_(graph_name), strm_(strm) {}
 
   template <typename T>
   T CopyToSharedMem(const T& data, std::string name);

src/graph/subgraph.cc

@@ -11,7 +11,8 @@ namespace dgl {
 HeteroSubgraph InEdgeGraphRelabelNodes(
     const HeteroGraphPtr graph, const std::vector<IdArray>& vids) {
   CHECK_EQ(vids.size(), graph->NumVertexTypes())
-    << "Invalid input: the input list size must be the same as the number of vertex types.";
+      << "Invalid input: the input list size must be the same as the number of "
+         "vertex types.";
   std::vector<IdArray> eids(graph->NumEdgeTypes());
   DGLContext ctx = aten::GetContextOf(vids);
   for (dgl_type_t etype = 0; etype < graph->NumEdgeTypes(); ++etype) {
@@ -29,9 +30,11 @@ HeteroSubgraph InEdgeGraphRelabelNodes(
 
 HeteroSubgraph InEdgeGraphNoRelabelNodes(
     const HeteroGraphPtr graph, const std::vector<IdArray>& vids) {
-  // TODO(mufei): This should also use EdgeSubgraph once it is supported for CSR graphs
+  // TODO(mufei): This should also use EdgeSubgraph once it is supported for CSR
+  // graphs
   CHECK_EQ(vids.size(), graph->NumVertexTypes())
-    << "Invalid input: the input list size must be the same as the number of vertex types.";
+      << "Invalid input: the input list size must be the same as the number of "
+         "vertex types.";
   std::vector<HeteroGraphPtr> subrels(graph->NumEdgeTypes());
   std::vector<IdArray> induced_edges(graph->NumEdgeTypes());
   DGLContext ctx = aten::GetContextOf(vids);
@@ -43,30 +46,28 @@ HeteroSubgraph InEdgeGraphNoRelabelNodes(
     if (aten::IsNullArray(vids[dst_vtype])) {
       // create a placeholder graph
       subrels[etype] = UnitGraph::Empty(
-        relgraph->NumVertexTypes(),
-        graph->NumVertices(src_vtype),
-        graph->NumVertices(dst_vtype),
-        graph->DataType(), ctx);
-      induced_edges[etype] = IdArray::Empty({0}, graph->DataType(), graph->Context());
+          relgraph->NumVertexTypes(), graph->NumVertices(src_vtype),
+          graph->NumVertices(dst_vtype), graph->DataType(), ctx);
+      induced_edges[etype] =
+          IdArray::Empty({0}, graph->DataType(), graph->Context());
     } else {
       const auto& earr = graph->InEdges(etype, {vids[dst_vtype]});
       subrels[etype] = UnitGraph::CreateFromCOO(
-        relgraph->NumVertexTypes(),
-        graph->NumVertices(src_vtype),
-        graph->NumVertices(dst_vtype),
-        earr.src,
-        earr.dst);
+          relgraph->NumVertexTypes(), graph->NumVertices(src_vtype),
+          graph->NumVertices(dst_vtype), earr.src, earr.dst);
       induced_edges[etype] = earr.id;
     }
   }
   HeteroSubgraph ret;
-  ret.graph = CreateHeteroGraph(graph->meta_graph(), subrels, graph->NumVerticesPerType());
+  ret.graph = CreateHeteroGraph(
+      graph->meta_graph(), subrels, graph->NumVerticesPerType());
   ret.induced_edges = std::move(induced_edges);
   return ret;
 }
 
 HeteroSubgraph InEdgeGraph(
-    const HeteroGraphPtr graph, const std::vector<IdArray>& vids, bool relabel_nodes) {
+    const HeteroGraphPtr graph, const std::vector<IdArray>& vids,
+    bool relabel_nodes) {
   if (relabel_nodes) {
     return InEdgeGraphRelabelNodes(graph, vids);
   } else {
@@ -77,7 +78,8 @@ HeteroSubgraph InEdgeGraph(
 HeteroSubgraph OutEdgeGraphRelabelNodes(
     const HeteroGraphPtr graph, const std::vector<IdArray>& vids) {
   CHECK_EQ(vids.size(), graph->NumVertexTypes())
-    << "Invalid input: the input list size must be the same as the number of vertex types.";
+      << "Invalid input: the input list size must be the same as the number of "
+         "vertex types.";
   std::vector<IdArray> eids(graph->NumEdgeTypes());
   DGLContext ctx = aten::GetContextOf(vids);
   for (dgl_type_t etype = 0; etype < graph->NumEdgeTypes(); ++etype) {
@@ -95,9 +97,11 @@ HeteroSubgraph OutEdgeGraphRelabelNodes(
 
 HeteroSubgraph OutEdgeGraphNoRelabelNodes(
     const HeteroGraphPtr graph, const std::vector<IdArray>& vids) {
-  // TODO(mufei): This should also use EdgeSubgraph once it is supported for CSR graphs
+  // TODO(mufei): This should also use EdgeSubgraph once it is supported for CSR
+  // graphs
   CHECK_EQ(vids.size(), graph->NumVertexTypes())
-    << "Invalid input: the input list size must be the same as the number of vertex types.";
+      << "Invalid input: the input list size must be the same as the number of "
+         "vertex types.";
   std::vector<HeteroGraphPtr> subrels(graph->NumEdgeTypes());
   std::vector<IdArray> induced_edges(graph->NumEdgeTypes());
   DGLContext ctx = aten::GetContextOf(vids);
@@ -109,30 +113,28 @@ HeteroSubgraph OutEdgeGraphNoRelabelNodes(
     if (aten::IsNullArray(vids[src_vtype])) {
       // create a placeholder graph
       subrels[etype] = UnitGraph::Empty(
-        relgraph->NumVertexTypes(),
-        graph->NumVertices(src_vtype),
-        graph->NumVertices(dst_vtype),
-        graph->DataType(), ctx);
-      induced_edges[etype] = IdArray::Empty({0}, graph->DataType(), graph->Context());
+          relgraph->NumVertexTypes(), graph->NumVertices(src_vtype),
+          graph->NumVertices(dst_vtype), graph->DataType(), ctx);
+      induced_edges[etype] =
+          IdArray::Empty({0}, graph->DataType(), graph->Context());
     } else {
       const auto& earr = graph->OutEdges(etype, {vids[src_vtype]});
       subrels[etype] = UnitGraph::CreateFromCOO(
-          relgraph->NumVertexTypes(),
-          graph->NumVertices(src_vtype),
-          graph->NumVertices(dst_vtype),
-          earr.src,
-          earr.dst);
+          relgraph->NumVertexTypes(), graph->NumVertices(src_vtype),
+          graph->NumVertices(dst_vtype), earr.src, earr.dst);
       induced_edges[etype] = earr.id;
     }
   }
   HeteroSubgraph ret;
-  ret.graph = CreateHeteroGraph(graph->meta_graph(), subrels, graph->NumVerticesPerType());
+  ret.graph = CreateHeteroGraph(
+      graph->meta_graph(), subrels, graph->NumVerticesPerType());
   ret.induced_edges = std::move(induced_edges);
   return ret;
 }
 
 HeteroSubgraph OutEdgeGraph(
-    const HeteroGraphPtr graph, const std::vector<IdArray>& vids, bool relabel_nodes) {
+    const HeteroGraphPtr graph, const std::vector<IdArray>& vids,
+    bool relabel_nodes) {
   if (relabel_nodes) {
     return OutEdgeGraphRelabelNodes(graph, vids);
   } else {

src/graph/transform/compact.cc

@@ -19,18 +19,20 @@
 
 #include "compact.h"
 
-#include <dgl/base_heterograph.h>
-#include <dgl/transform.h>
 #include <dgl/array.h>
+#include <dgl/base_heterograph.h>
 #include <dgl/packed_func_ext.h>
-#include <dgl/runtime/registry.h>
 #include <dgl/runtime/container.h>
-#include <vector>
+#include <dgl/runtime/registry.h>
+#include <dgl/transform.h>
+
 #include <utility>
+#include <vector>
+
 #include "../../c_api_common.h"
 #include "../unit_graph.h"
-// TODO(BarclayII): currently CompactGraphs depend on IdHashMap implementation which
-// only works on CPU.  Should fix later to make it device agnostic.
+// TODO(BarclayII): currently CompactGraphs depend on IdHashMap implementation
+// which only works on CPU.  Should fix later to make it device agnostic.
 #include "../../array/cpu/array_utils.h"
 
 namespace dgl {
@@ -42,16 +44,16 @@ namespace transform {
 
 namespace {
 
-template<typename IdType>
-std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
-CompactGraphsCPU(
+template <typename IdType>
+std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>> CompactGraphsCPU(
     const std::vector<HeteroGraphPtr> &graphs,
     const std::vector<IdArray> &always_preserve) {
-  // TODO(BarclayII): check whether the node space and metagraph of each graph is the same.
-  // Step 1: Collect the nodes that has connections for each type.
+  // TODO(BarclayII): check whether the node space and metagraph of each graph
+  // is the same. Step 1: Collect the nodes that has connections for each type.
   const int64_t num_ntypes = graphs[0]->NumVertexTypes();
   std::vector<aten::IdHashMap<IdType>> hashmaps(num_ntypes);
-  std::vector<std::vector<EdgeArray>> all_edges(graphs.size());   // all_edges[i][etype]
+  std::vector<std::vector<EdgeArray>> all_edges(
+      graphs.size());  // all_edges[i][etype]
 
   std::vector<int64_t> max_vertex_cnt(num_ntypes, 0);
   for (size_t i = 0; i < graphs.size(); ++i) {
@@ -98,7 +100,8 @@ CompactGraphsCPU(
     }
   }
 
-  // Step 2: Relabel the nodes for each type to a smaller ID space and save the mapping.
+  // Step 2: Relabel the nodes for each type to a smaller ID space and save the
+  // mapping.
   std::vector<IdArray> induced_nodes(num_ntypes);
   std::vector<int64_t> num_induced_nodes(num_ntypes);
   for (int64_t i = 0; i < num_ntypes; ++i) {
@@ -123,14 +126,12 @@ CompactGraphsCPU(
       const IdArray mapped_cols = hashmaps[dsttype].Map(edges.dst, -1);
 
       rel_graphs.push_back(UnitGraph::CreateFromCOO(
-          srctype == dsttype ? 1 : 2,
-          induced_nodes[srctype]->shape[0],
-          induced_nodes[dsttype]->shape[0],
-          mapped_rows,
-          mapped_cols));
+          srctype == dsttype ? 1 : 2, induced_nodes[srctype]->shape[0],
+          induced_nodes[dsttype]->shape[0], mapped_rows, mapped_cols));
     }
 
-    new_graphs.push_back(CreateHeteroGraph(meta_graph, rel_graphs, num_induced_nodes));
+    new_graphs.push_back(
+        CreateHeteroGraph(meta_graph, rel_graphs, num_induced_nodes));
   }
 
   return std::make_pair(new_graphs, induced_nodes);
@@ -138,7 +139,7 @@ CompactGraphsCPU(
 
 };  // namespace
 
-template<>
+template <>
 std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
 CompactGraphs<kDGLCPU, int32_t>(
     const std::vector<HeteroGraphPtr> &graphs,
@@ -146,7 +147,7 @@ CompactGraphs<kDGLCPU, int32_t>(
   return CompactGraphsCPU<int32_t>(graphs, always_preserve);
 }
 
-template<>
+template <>
 std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
 CompactGraphs<kDGLCPU, int64_t>(
     const std::vector<HeteroGraphPtr> &graphs,
@@ -155,26 +156,26 @@ CompactGraphs<kDGLCPU, int64_t>(
 }
 
 DGL_REGISTER_GLOBAL("transform._CAPI_DGLCompactGraphs")
-.set_body([] (DGLArgs args, DGLRetValue* rv) {
+    .set_body([](DGLArgs args, DGLRetValue *rv) {
       List<HeteroGraphRef> graph_refs = args[0];
       List<Value> always_preserve_refs = args[1];
 
       std::vector<HeteroGraphPtr> graphs;
       std::vector<IdArray> always_preserve;
-    for (HeteroGraphRef gref : graph_refs)
-      graphs.push_back(gref.sptr());
+      for (HeteroGraphRef gref : graph_refs) graphs.push_back(gref.sptr());
       for (Value array : always_preserve_refs)
         always_preserve.push_back(array->data);
 
       // TODO(BarclayII): check for all IdArrays
-    CHECK(graphs[0]->DataType() == always_preserve[0]->dtype) << "data type mismatch.";
+      CHECK(graphs[0]->DataType() == always_preserve[0]->dtype)
+          << "data type mismatch.";
 
       std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>> result_pair;
 
-    ATEN_XPU_SWITCH_CUDA(graphs[0]->Context().device_type, XPU, "CompactGraphs", {
+      ATEN_XPU_SWITCH_CUDA(
+          graphs[0]->Context().device_type, XPU, "CompactGraphs", {
             ATEN_ID_TYPE_SWITCH(graphs[0]->DataType(), IdType, {
-        result_pair = CompactGraphs<XPU, IdType>(
-          graphs, always_preserve);
+              result_pair = CompactGraphs<XPU, IdType>(graphs, always_preserve);
             });
           });
 

src/graph/transform/compact.h

@@ -24,8 +24,8 @@
 #include <dgl/array.h>
 #include <dgl/base_heterograph.h>
 
-#include <vector>
 #include <utility>
+#include <vector>
 
 namespace dgl {
 namespace transform {
@@ -41,9 +41,8 @@ namespace transform {
  *
  * @return The vector of compacted graphs and the vector of induced nodes.
  */
-template<DGLDeviceType XPU, typename IdType>
-std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
-CompactGraphs(
+template <DGLDeviceType XPU, typename IdType>
+std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>> CompactGraphs(
     const std::vector<HeteroGraphPtr> &graphs,
     const std::vector<IdArray> &always_preserve);
 

src/graph/transform/cpu/kdtree_ndarray_adapter.h

@@ -9,7 +9,9 @@
 
 #include <dgl/array.h>
 #include <dmlc/logging.h>
+
 #include <nanoflann.hpp>
+
 #include "../../../c_api_common.h"
 
 namespace dgl {
@@ -17,29 +19,34 @@ namespace transform {
 namespace knn_utils {
 
 /*!
- * \brief A simple 2D NDArray adapter for nanoflann, without duplicating the storage.
+ * \brief A simple 2D NDArray adapter for nanoflann, without duplicating the
+ *        storage.
  *
- * \tparam FloatType: The type of the point coordinates (typically, double or float).
- * \tparam IdType: The type for indices in the KD-tree index (typically, size_t of int)
- * \tparam FeatureDim: If set to > 0, it specifies a compile-time fixed dimensionality
- *         for the points in the data set, allowing more compiler optimizations.
- * \tparam Dist: The distance metric to use: nanoflann::metric_L1, nanoflann::metric_L2,
- *         nanoflann::metric_L2_Simple, etc.
- * \note The spelling of dgl's adapter ("adapter") is different from naneflann ("adaptor")
+ * \tparam FloatType: The type of the point coordinates (typically, double or
+ *         float).
+ * \tparam IdType: The type for indices in the KD-tree index (typically,
+ *         size_t of int)
+ * \tparam FeatureDim: If set to > 0, it specifies a compile-time fixed
+ *         dimensionality for the points in the data set, allowing more compiler
+ *         optimizations.
+ * \tparam Dist: The distance metric to use: nanoflann::metric_L1,
+           nanoflann::metric_L2, nanoflann::metric_L2_Simple, etc.
+ * \note The spelling of dgl's adapter ("adapter") is different from naneflann
+ *       ("adaptor")
  */
-template <typename FloatType,
-          typename IdType,
-          int FeatureDim = -1,
+template <
+    typename FloatType, typename IdType, int FeatureDim = -1,
     typename Dist = nanoflann::metric_L2>
 class KDTreeNDArrayAdapter {
  public:
   using self_type = KDTreeNDArrayAdapter<FloatType, IdType, FeatureDim, Dist>;
-  using metric_type = typename Dist::template traits<FloatType, self_type>::distance_t;
+  using metric_type =
+      typename Dist::template traits<FloatType, self_type>::distance_t;
   using index_type = nanoflann::KDTreeSingleIndexAdaptor<
       metric_type, self_type, FeatureDim, IdType>;
 
-  KDTreeNDArrayAdapter(const size_t /* dims */,
-                       const NDArray data_points,
+  KDTreeNDArrayAdapter(
+      const size_t /* dims */, const NDArray data_points,
       const int leaf_max_size = 10)
       : data_(data_points) {
     CHECK(data_points->shape[0] != 0 && data_points->shape[1] != 0)
@@ -49,46 +56,38 @@ class KDTreeNDArrayAdapter {
         << "Data set feature dimension does not match the 'FeatureDim' "
         << "template argument.";
     index_ = new index_type(
-      static_cast<int>(dims), *this, nanoflann::KDTreeSingleIndexAdaptorParams(leaf_max_size));
+        static_cast<int>(dims), *this,
+        nanoflann::KDTreeSingleIndexAdaptorParams(leaf_max_size));
     index_->buildIndex();
   }
 
-  ~KDTreeNDArrayAdapter() {
-    delete index_;
-  }
+  ~KDTreeNDArrayAdapter() { delete index_; }
 
-  index_type* GetIndex() {
-    return index_;
-  }
+  index_type* GetIndex() { return index_; }
 
   /*!
    * \brief Query for the \a num_closest points to a given point
    *  Note that this is a short-cut method for GetIndex()->findNeighbors().
    */
-  void query(const FloatType* query_pt, const size_t num_closest,
-             IdType* out_idxs, FloatType* out_dists) const {
+  void query(
+      const FloatType* query_pt, const size_t num_closest, IdType* out_idxs,
+      FloatType* out_dists) const {
     nanoflann::KNNResultSet<FloatType, IdType> resultSet(num_closest);
     resultSet.init(out_idxs, out_dists);
     index_->findNeighbors(resultSet, query_pt, nanoflann::SearchParams());
   }
 
   /*! \brief Interface expected by KDTreeSingleIndexAdaptor */
-  const self_type& derived() const {
-    return *this;
-  }
+  const self_type& derived() const { return *this; }
 
   /*! \brief Interface expected by KDTreeSingleIndexAdaptor */
-  self_type& derived() {
-    return *this;
-  }
+  self_type& derived() { return *this; }
 
   /*!
    * \brief Interface expected by KDTreeSingleIndexAdaptor,
    *  return the number of data points
    */
-  size_t kdtree_get_point_count() const {
-    return data_->shape[0];
-  }
+  size_t kdtree_get_point_count() const { return data_->shape[0]; }
 
   /*!
    * \brief Interface expected by KDTreeSingleIndexAdaptor,

src/graph/transform/cuda/cuda_compact_graph.cu

@@ -18,13 +18,13 @@
  * all given graphs with the same set of nodes.
  */
 
-
-#include <dgl/runtime/device_api.h>
-#include <dgl/immutable_graph.h>
 #include <cuda_runtime.h>
-#include <utility>
+#include <dgl/immutable_graph.h>
+#include <dgl/runtime/device_api.h>
+
 #include <algorithm>
 #include <memory>
+#include <utility>
 
 #include "../../../runtime/cuda/cuda_common.h"
 #include "../../heterograph.h"
@@ -41,54 +41,45 @@ namespace transform {
 namespace {
 
 /**
-* \brief This function builds node maps for each node type, preserving the
-* order of the input nodes. Here it is assumed the nodes are not unique,
-* and thus a unique list is generated.
-*
-* \param input_nodes The set of input nodes.
-* \param node_maps The node maps to be constructed.
-* \param count_unique_device The number of unique nodes (on the GPU).
-* \param unique_nodes_device The unique nodes (on the GPU).
-* \param stream The stream to operate on.
-*/
-template<typename IdType>
+ * \brief This function builds node maps for each node type, preserving the
+ * order of the input nodes. Here it is assumed the nodes are not unique,
+ * and thus a unique list is generated.
+ *
+ * \param input_nodes The set of input nodes.
+ * \param node_maps The node maps to be constructed.
+ * \param count_unique_device The number of unique nodes (on the GPU).
+ * \param unique_nodes_device The unique nodes (on the GPU).
+ * \param stream The stream to operate on.
+ */
+template <typename IdType>
 void BuildNodeMaps(
-    const std::vector<IdArray>& input_nodes,
-    DeviceNodeMap<IdType> * const node_maps,
-    int64_t * const count_unique_device,
-    std::vector<IdArray>* const unique_nodes_device,
-    cudaStream_t stream) {
+    const std::vector<IdArray> &input_nodes,
+    DeviceNodeMap<IdType> *const node_maps, int64_t *const count_unique_device,
+    std::vector<IdArray> *const unique_nodes_device, cudaStream_t stream) {
   const int64_t num_ntypes = static_cast<int64_t>(input_nodes.size());
 
   CUDA_CALL(cudaMemsetAsync(
-    count_unique_device,
-    0,
-    num_ntypes*sizeof(*count_unique_device),
+      count_unique_device, 0, num_ntypes * sizeof(*count_unique_device),
       stream));
 
   // possibly duplicated nodes
   for (int64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-    const IdArray& nodes = input_nodes[ntype];
+    const IdArray &nodes = input_nodes[ntype];
     if (nodes->shape[0] > 0) {
       CHECK_EQ(nodes->ctx.device_type, kDGLCUDA);
       node_maps->LhsHashTable(ntype).FillWithDuplicates(
-          nodes.Ptr<IdType>(),
-          nodes->shape[0],
+          nodes.Ptr<IdType>(), nodes->shape[0],
           (*unique_nodes_device)[ntype].Ptr<IdType>(),
-          count_unique_device+ntype,
-          stream);
+          count_unique_device + ntype, stream);
     }
   }
 }
 
-
-template<typename IdType>
-std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
-CompactGraphsGPU(
+template <typename IdType>
+std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>> CompactGraphsGPU(
     const std::vector<HeteroGraphPtr> &graphs,
     const std::vector<IdArray> &always_preserve) {
-
-  const auto& ctx = graphs[0]->Context();
+  const auto &ctx = graphs[0]->Context();
   auto device = runtime::DeviceAPI::Get(ctx);
   cudaStream_t stream = runtime::getCurrentCUDAStream();
 
@@ -96,7 +87,8 @@ CompactGraphsGPU(
 
   // Step 1: Collect the nodes that has connections for each type.
   const uint64_t num_ntypes = graphs[0]->NumVertexTypes();
-  std::vector<std::vector<EdgeArray>> all_edges(graphs.size());   // all_edges[i][etype]
+  std::vector<std::vector<EdgeArray>> all_edges(
+      graphs.size());  // all_edges[i][etype]
 
   // count the number of nodes per type
   std::vector<int64_t> max_vertex_cnt(num_ntypes, 0);
@@ -123,19 +115,18 @@ CompactGraphsGPU(
   std::vector<int64_t> node_offsets(num_ntypes, 0);
 
   for (uint64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-    all_nodes[ntype] = NewIdArray(max_vertex_cnt[ntype], ctx,
-      sizeof(IdType)*8);
+    all_nodes[ntype] =
+        NewIdArray(max_vertex_cnt[ntype], ctx, sizeof(IdType) * 8);
     // copy the nodes in always_preserve
-    if (ntype < always_preserve.size() && always_preserve[ntype]->shape[0] > 0) {
+    if (ntype < always_preserve.size() &&
+        always_preserve[ntype]->shape[0] > 0) {
       device->CopyDataFromTo(
           always_preserve[ntype].Ptr<IdType>(), 0,
-          all_nodes[ntype].Ptr<IdType>(),
-          node_offsets[ntype],
-          sizeof(IdType)*always_preserve[ntype]->shape[0],
-          always_preserve[ntype]->ctx,
-          all_nodes[ntype]->ctx,
+          all_nodes[ntype].Ptr<IdType>(), node_offsets[ntype],
+          sizeof(IdType) * always_preserve[ntype]->shape[0],
+          always_preserve[ntype]->ctx, all_nodes[ntype]->ctx,
           always_preserve[ntype]->dtype);
-      node_offsets[ntype] += sizeof(IdType)*always_preserve[ntype]->shape[0];
+      node_offsets[ntype] += sizeof(IdType) * always_preserve[ntype]->shape[0];
     }
   }
 
@@ -152,25 +143,17 @@ CompactGraphsGPU(
 
       if (edges.src.defined()) {
         device->CopyDataFromTo(
-            edges.src.Ptr<IdType>(), 0,
-            all_nodes[srctype].Ptr<IdType>(),
-            node_offsets[srctype],
-            sizeof(IdType)*edges.src->shape[0],
-            edges.src->ctx,
-            all_nodes[srctype]->ctx,
-            edges.src->dtype);
-        node_offsets[srctype] += sizeof(IdType)*edges.src->shape[0];
+            edges.src.Ptr<IdType>(), 0, all_nodes[srctype].Ptr<IdType>(),
+            node_offsets[srctype], sizeof(IdType) * edges.src->shape[0],
+            edges.src->ctx, all_nodes[srctype]->ctx, edges.src->dtype);
+        node_offsets[srctype] += sizeof(IdType) * edges.src->shape[0];
       }
       if (edges.dst.defined()) {
         device->CopyDataFromTo(
-            edges.dst.Ptr<IdType>(), 0,
-            all_nodes[dsttype].Ptr<IdType>(),
-            node_offsets[dsttype],
-            sizeof(IdType)*edges.dst->shape[0],
-            edges.dst->ctx,
-            all_nodes[dsttype]->ctx,
-            edges.dst->dtype);
-        node_offsets[dsttype] += sizeof(IdType)*edges.dst->shape[0];
+            edges.dst.Ptr<IdType>(), 0, all_nodes[dsttype].Ptr<IdType>(),
+            node_offsets[dsttype], sizeof(IdType) * edges.dst->shape[0],
+            edges.dst->ctx, all_nodes[dsttype]->ctx, edges.dst->dtype);
+        node_offsets[dsttype] += sizeof(IdType) * edges.dst->shape[0];
       }
       all_edges[i].push_back(edges);
     }
@@ -185,29 +168,22 @@ CompactGraphsGPU(
   // number of unique nodes per type on CPU
   std::vector<int64_t> num_induced_nodes(num_ntypes);
   // number of unique nodes per type on GPU
-  int64_t * count_unique_device = static_cast<int64_t*>(
-      device->AllocWorkspace(ctx, sizeof(int64_t)*num_ntypes));
+  int64_t *count_unique_device = static_cast<int64_t *>(
+      device->AllocWorkspace(ctx, sizeof(int64_t) * num_ntypes));
   // the set of unique nodes per type
   std::vector<IdArray> induced_nodes(num_ntypes);
   for (uint64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-    induced_nodes[ntype] = NewIdArray(max_vertex_cnt[ntype], ctx,
-      sizeof(IdType)*8);
+    induced_nodes[ntype] =
+        NewIdArray(max_vertex_cnt[ntype], ctx, sizeof(IdType) * 8);
   }
 
   BuildNodeMaps(
-    all_nodes,
-    &node_maps,
-    count_unique_device,
-    &induced_nodes,
-    stream);
+      all_nodes, &node_maps, count_unique_device, &induced_nodes, stream);
 
   device->CopyDataFromTo(
-    count_unique_device, 0,
-    num_induced_nodes.data(), 0,
-    sizeof(*num_induced_nodes.data())*num_ntypes,
-    ctx,
-    DGLContext{kDGLCPU, 0},
-    DGLDataType{kDGLInt, 64, 1});
+      count_unique_device, 0, num_induced_nodes.data(), 0,
+      sizeof(*num_induced_nodes.data()) * num_ntypes, ctx,
+      DGLContext{kDGLCPU, 0}, DGLDataType{kDGLInt, 64, 1});
   device->StreamSync(ctx, stream);
 
   // wait for the node counts to finish transferring
@@ -230,22 +206,20 @@ CompactGraphsGPU(
 
     std::vector<IdArray> new_src;
     std::vector<IdArray> new_dst;
-    std::tie(new_src, new_dst) = MapEdges(
-      curr_graph, all_edges[i], node_maps, stream);
+    std::tie(new_src, new_dst) =
+        MapEdges(curr_graph, all_edges[i], node_maps, stream);
 
     for (IdType etype = 0; etype < num_etypes; ++etype) {
       IdType srctype, dsttype;
       std::tie(srctype, dsttype) = curr_graph->GetEndpointTypes(etype);
 
       rel_graphs.push_back(UnitGraph::CreateFromCOO(
-          srctype == dsttype ? 1 : 2,
-          induced_nodes[srctype]->shape[0],
-          induced_nodes[dsttype]->shape[0],
-          new_src[etype],
-          new_dst[etype]));
+          srctype == dsttype ? 1 : 2, induced_nodes[srctype]->shape[0],
+          induced_nodes[dsttype]->shape[0], new_src[etype], new_dst[etype]));
     }
 
-    new_graphs.push_back(CreateHeteroGraph(meta_graph, rel_graphs, num_induced_nodes));
+    new_graphs.push_back(
+        CreateHeteroGraph(meta_graph, rel_graphs, num_induced_nodes));
   }
 
   return std::make_pair(new_graphs, induced_nodes);
@@ -253,7 +227,7 @@ CompactGraphsGPU(
 
 }  // namespace
 
-template<>
+template <>
 std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
 CompactGraphs<kDGLCUDA, int32_t>(
     const std::vector<HeteroGraphPtr> &graphs,
@@ -261,7 +235,7 @@ CompactGraphs<kDGLCUDA, int32_t>(
   return CompactGraphsGPU<int32_t>(graphs, always_preserve);
 }
 
-template<>
+template <>
 std::pair<std::vector<HeteroGraphPtr>, std::vector<IdArray>>
 CompactGraphs<kDGLCUDA, int64_t>(
     const std::vector<HeteroGraphPtr> &graphs,

src/graph/transform/cuda/cuda_map_edges.cuh

@@ -20,13 +20,14 @@
 #ifndef DGL_GRAPH_TRANSFORM_CUDA_CUDA_MAP_EDGES_CUH_
 #define DGL_GRAPH_TRANSFORM_CUDA_CUDA_MAP_EDGES_CUH_
 
-#include <dgl/runtime/c_runtime_api.h>
 #include <cuda_runtime.h>
+#include <dgl/runtime/c_runtime_api.h>
+
 #include <algorithm>
 #include <memory>
 #include <tuple>
-#include <vector>
 #include <utility>
+#include <vector>
 
 #include "../../../runtime/cuda/cuda_common.h"
 #include "../../../runtime/cuda/cuda_hashtable.cuh"
@@ -39,48 +40,46 @@ namespace transform {
 
 namespace cuda {
 
-template<typename IdType, int BLOCK_SIZE, IdType TILE_SIZE>
+template <typename IdType, int BLOCK_SIZE, IdType TILE_SIZE>
 __device__ void map_vertex_ids(
-    const IdType * const global,
-    IdType * const new_global,
-    const IdType num_vertices,
-    const DeviceOrderedHashTable<IdType>& table) {
+    const IdType* const global, IdType* const new_global,
+    const IdType num_vertices, const DeviceOrderedHashTable<IdType>& table) {
   assert(BLOCK_SIZE == blockDim.x);
 
   using Mapping = typename OrderedHashTable<IdType>::Mapping;
 
-  const IdType tile_start = TILE_SIZE*blockIdx.x;
-  const IdType tile_end = min(TILE_SIZE*(blockIdx.x+1), num_vertices);
+  const IdType tile_start = TILE_SIZE * blockIdx.x;
+  const IdType tile_end = min(TILE_SIZE * (blockIdx.x + 1), num_vertices);
 
-  for (IdType idx = threadIdx.x+tile_start; idx < tile_end; idx+=BLOCK_SIZE) {
+  for (IdType idx = threadIdx.x + tile_start; idx < tile_end;
+       idx += BLOCK_SIZE) {
     const Mapping& mapping = *table.Search(global[idx]);
     new_global[idx] = mapping.local;
   }
 }
 
 /**
-* \brief Generate mapped edge endpoint ids.
-*
-* \tparam IdType The type of id.
-* \tparam BLOCK_SIZE The size of each thread block.
-* \tparam TILE_SIZE The number of edges to process per thread block.
-* \param global_srcs_device The source ids to map.
-* \param new_global_srcs_device The mapped source ids (output).
-* \param global_dsts_device The destination ids to map.
-* \param new_global_dsts_device The mapped destination ids (output).
-* \param num_edges The number of edges to map.
-* \param src_mapping The mapping of sources ids.
-* \param src_hash_size The the size of source id hash table/mapping.
-* \param dst_mapping The mapping of destination ids.
-* \param dst_hash_size The the size of destination id hash table/mapping.
-*/
-template<typename IdType, int BLOCK_SIZE, IdType TILE_SIZE>
+ * \brief Generate mapped edge endpoint ids.
+ *
+ * \tparam IdType The type of id.
+ * \tparam BLOCK_SIZE The size of each thread block.
+ * \tparam TILE_SIZE The number of edges to process per thread block.
+ * \param global_srcs_device The source ids to map.
+ * \param new_global_srcs_device The mapped source ids (output).
+ * \param global_dsts_device The destination ids to map.
+ * \param new_global_dsts_device The mapped destination ids (output).
+ * \param num_edges The number of edges to map.
+ * \param src_mapping The mapping of sources ids.
+ * \param src_hash_size The the size of source id hash table/mapping.
+ * \param dst_mapping The mapping of destination ids.
+ * \param dst_hash_size The the size of destination id hash table/mapping.
+ */
+template <typename IdType, int BLOCK_SIZE, IdType TILE_SIZE>
 __global__ void map_edge_ids(
-    const IdType * const global_srcs_device,
-    IdType * const new_global_srcs_device,
-    const IdType * const global_dsts_device,
-    IdType * const new_global_dsts_device,
-    const IdType num_edges,
+    const IdType* const global_srcs_device,
+    IdType* const new_global_srcs_device,
+    const IdType* const global_dsts_device,
+    IdType* const new_global_dsts_device, const IdType num_edges,
     DeviceOrderedHashTable<IdType> src_mapping,
     DeviceOrderedHashTable<IdType> dst_mapping) {
   assert(BLOCK_SIZE == blockDim.x);
@@ -88,39 +87,32 @@ __global__ void map_edge_ids(
 
   if (blockIdx.y == 0) {
     map_vertex_ids<IdType, BLOCK_SIZE, TILE_SIZE>(
-        global_srcs_device,
-        new_global_srcs_device,
-        num_edges,
-        src_mapping);
+        global_srcs_device, new_global_srcs_device, num_edges, src_mapping);
   } else {
     map_vertex_ids<IdType, BLOCK_SIZE, TILE_SIZE>(
-        global_dsts_device,
-        new_global_dsts_device,
-        num_edges,
-        dst_mapping);
+        global_dsts_device, new_global_dsts_device, num_edges, dst_mapping);
   }
 }
 
 /**
-* \brief Device level node maps for each node type.
-*
-* \param num_nodes Number of nodes per type.
-* \param offset When offset is set to 0, LhsHashTable is identical to RhsHashTable. 
-* Or set to num_nodes.size()/2 to use seperated LhsHashTable and RhsHashTable.
-* \param ctx The DGL context.
-* \param stream The stream to operate on.
-*/
-template<typename IdType>
+ * \brief Device level node maps for each node type.
+ *
+ * \param num_nodes Number of nodes per type.
+ * \param offset When offset is set to 0, LhsHashTable is identical to
+ *        RhsHashTable. Or set to num_nodes.size()/2 to use seperated
+ *        LhsHashTable and RhsHashTable.
+ * \param ctx The DGL context.
+ * \param stream The stream to operate on.
+ */
+template <typename IdType>
 class DeviceNodeMap {
  public:
   using Mapping = typename OrderedHashTable<IdType>::Mapping;
 
   DeviceNodeMap(
-      const std::vector<int64_t>& num_nodes,
-      const int64_t offset,
-      DGLContext ctx,
-      cudaStream_t stream) :
-    num_types_(num_nodes.size()),
+      const std::vector<int64_t>& num_nodes, const int64_t offset,
+      DGLContext ctx, cudaStream_t stream)
+      : num_types_(num_nodes.size()),
         rhs_offset_(offset),
         hash_tables_(),
         ctx_(ctx) {
@@ -129,46 +121,33 @@ class DeviceNodeMap {
     hash_tables_.reserve(num_types_);
     for (int64_t i = 0; i < num_types_; ++i) {
       hash_tables_.emplace_back(
-          new OrderedHashTable<IdType>(
-            num_nodes[i],
-            ctx_,
-            stream));
+          new OrderedHashTable<IdType>(num_nodes[i], ctx_, stream));
     }
   }
 
-  OrderedHashTable<IdType>& LhsHashTable(
-      const size_t index) {
+  OrderedHashTable<IdType>& LhsHashTable(const size_t index) {
     return HashData(index);
   }
 
-  OrderedHashTable<IdType>& RhsHashTable(
-      const size_t index) {
-    return HashData(index+rhs_offset_);
+  OrderedHashTable<IdType>& RhsHashTable(const size_t index) {
+    return HashData(index + rhs_offset_);
   }
 
-  const OrderedHashTable<IdType>& LhsHashTable(
-      const size_t index) const {
+  const OrderedHashTable<IdType>& LhsHashTable(const size_t index) const {
     return HashData(index);
   }
 
-  const OrderedHashTable<IdType>& RhsHashTable(
-      const size_t index) const {
-    return HashData(index+rhs_offset_);
+  const OrderedHashTable<IdType>& RhsHashTable(const size_t index) const {
+    return HashData(index + rhs_offset_);
   }
 
-  IdType LhsHashSize(
-      const size_t index) const {
-    return HashSize(index);
-  }
+  IdType LhsHashSize(const size_t index) const { return HashSize(index); }
 
-  IdType RhsHashSize(
-      const size_t index) const {
-    return HashSize(rhs_offset_+index);
+  IdType RhsHashSize(const size_t index) const {
+    return HashSize(rhs_offset_ + index);
   }
 
-  size_t Size() const {
-    return hash_tables_.size();
-  }
+  size_t Size() const { return hash_tables_.size(); }
 
  private:
   int64_t num_types_;
@@ -176,45 +155,35 @@ class DeviceNodeMap {
   std::vector<std::unique_ptr<OrderedHashTable<IdType>>> hash_tables_;
   DGLContext ctx_;
 
-  inline OrderedHashTable<IdType>& HashData(
-      const size_t index) {
+  inline OrderedHashTable<IdType>& HashData(const size_t index) {
     CHECK_LT(index, hash_tables_.size());
     return *hash_tables_[index];
   }
 
-  inline const OrderedHashTable<IdType>& HashData(
-      const size_t index) const {
+  inline const OrderedHashTable<IdType>& HashData(const size_t index) const {
     CHECK_LT(index, hash_tables_.size());
     return *hash_tables_[index];
   }
 
-  inline IdType HashSize(
-      const size_t index) const {
+  inline IdType HashSize(const size_t index) const {
     return HashData(index).size();
   }
 };
 
-template<typename IdType>
-inline size_t RoundUpDiv(
-    const IdType num,
-    const size_t divisor) {
-  return static_cast<IdType>(num/divisor) + (num % divisor == 0 ? 0 : 1);
+template <typename IdType>
+inline size_t RoundUpDiv(const IdType num, const size_t divisor) {
+  return static_cast<IdType>(num / divisor) + (num % divisor == 0 ? 0 : 1);
 }
 
-template<typename IdType>
-inline IdType RoundUp(
-    const IdType num,
-    const size_t unit) {
-  return RoundUpDiv(num, unit)*unit;
+template <typename IdType>
+inline IdType RoundUp(const IdType num, const size_t unit) {
+  return RoundUpDiv(num, unit) * unit;
 }
 
-template<typename IdType>
-std::tuple<std::vector<IdArray>, std::vector<IdArray>>
-MapEdges(
-    HeteroGraphPtr graph,
-    const std::vector<EdgeArray>& edge_sets,
-    const DeviceNodeMap<IdType>& node_map,
-    cudaStream_t stream) {
+template <typename IdType>
+std::tuple<std::vector<IdArray>, std::vector<IdArray>> MapEdges(
+    HeteroGraphPtr graph, const std::vector<EdgeArray>& edge_sets,
+    const DeviceNodeMap<IdType>& node_map, cudaStream_t stream) {
   constexpr const int BLOCK_SIZE = 128;
   constexpr const size_t TILE_SIZE = 1024;
 
@@ -233,8 +202,8 @@ MapEdges(
     if (edges.id.defined() && edges.src->shape[0] > 0) {
       const int64_t num_edges = edges.src->shape[0];
 
-      new_lhs.emplace_back(NewIdArray(num_edges, ctx, sizeof(IdType)*8));
-      new_rhs.emplace_back(NewIdArray(num_edges, ctx, sizeof(IdType)*8));
+      new_lhs.emplace_back(NewIdArray(num_edges, ctx, sizeof(IdType) * 8));
+      new_rhs.emplace_back(NewIdArray(num_edges, ctx, sizeof(IdType) * 8));
 
       const auto src_dst_types = graph->GetEndpointTypes(etype);
       const int src_type = src_dst_types.first;
@@ -244,20 +213,17 @@ MapEdges(
       const dim3 block(BLOCK_SIZE);
 
       // map the srcs
-      CUDA_KERNEL_CALL((map_edge_ids<IdType, BLOCK_SIZE, TILE_SIZE>),
-        grid, block, 0, stream,
-        edges.src.Ptr<IdType>(),
-        new_lhs.back().Ptr<IdType>(),
-        edges.dst.Ptr<IdType>(),
-        new_rhs.back().Ptr<IdType>(),
-        num_edges,
+      CUDA_KERNEL_CALL(
+          (map_edge_ids<IdType, BLOCK_SIZE, TILE_SIZE>), grid, block, 0, stream,
+          edges.src.Ptr<IdType>(), new_lhs.back().Ptr<IdType>(),
+          edges.dst.Ptr<IdType>(), new_rhs.back().Ptr<IdType>(), num_edges,
           node_map.LhsHashTable(src_type).DeviceHandle(),
           node_map.RhsHashTable(dst_type).DeviceHandle());
     } else {
       new_lhs.emplace_back(
-          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType)*8, 1}, ctx));
+          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType) * 8, 1}, ctx));
       new_rhs.emplace_back(
-          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType)*8, 1}, ctx));
+          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType) * 8, 1}, ctx));
     }
   }
 
@@ -265,7 +231,6 @@ MapEdges(
       std::move(new_lhs), std::move(new_rhs));
 }
 
-
 }  // namespace cuda
 }  // namespace transform
 }  // namespace dgl

src/graph/transform/cuda/cuda_to_block.cu

@@ -18,13 +18,13 @@
  * ids.
  */
 
-
-#include <dgl/runtime/device_api.h>
-#include <dgl/immutable_graph.h>
 #include <cuda_runtime.h>
-#include <utility>
+#include <dgl/immutable_graph.h>
+#include <dgl/runtime/device_api.h>
+
 #include <algorithm>
 #include <memory>
+#include <utility>
 
 #include "../../../runtime/cuda/cuda_common.h"
 #include "../../heterograph.h"
@@ -40,14 +40,13 @@ namespace transform {
 
 namespace {
 
-template<typename IdType>
+template <typename IdType>
 class DeviceNodeMapMaker {
  public:
-  explicit DeviceNodeMapMaker(
-      const std::vector<int64_t>& maxNodesPerType) :
-      max_num_nodes_(0) {
-    max_num_nodes_ = *std::max_element(maxNodesPerType.begin(),
-        maxNodesPerType.end());
+  explicit DeviceNodeMapMaker(const std::vector<int64_t>& maxNodesPerType)
+      : max_num_nodes_(0) {
+    max_num_nodes_ =
+        *std::max_element(maxNodesPerType.begin(), maxNodesPerType.end());
   }
 
   /**
@@ -65,17 +64,12 @@ class DeviceNodeMapMaker {
   void Make(
       const std::vector<IdArray>& lhs_nodes,
       const std::vector<IdArray>& rhs_nodes,
-      DeviceNodeMap<IdType> * const node_maps,
-      int64_t * const count_lhs_device,
-      std::vector<IdArray>* const lhs_device,
-      cudaStream_t stream) {
+      DeviceNodeMap<IdType>* const node_maps, int64_t* const count_lhs_device,
+      std::vector<IdArray>* const lhs_device, cudaStream_t stream) {
     const int64_t num_ntypes = lhs_nodes.size() + rhs_nodes.size();
 
     CUDA_CALL(cudaMemsetAsync(
-      count_lhs_device,
-      0,
-      num_ntypes*sizeof(*count_lhs_device),
-      stream));
+        count_lhs_device, 0, num_ntypes * sizeof(*count_lhs_device), stream));
 
     // possibly dublicate lhs nodes
     const int64_t lhs_num_ntypes = static_cast<int64_t>(lhs_nodes.size());
@@ -84,10 +78,8 @@ class DeviceNodeMapMaker {
       if (nodes->shape[0] > 0) {
         CHECK_EQ(nodes->ctx.device_type, kDGLCUDA);
         node_maps->LhsHashTable(ntype).FillWithDuplicates(
-            nodes.Ptr<IdType>(),
-            nodes->shape[0],
-            (*lhs_device)[ntype].Ptr<IdType>(),
-            count_lhs_device+ntype,
+            nodes.Ptr<IdType>(), nodes->shape[0],
+            (*lhs_device)[ntype].Ptr<IdType>(), count_lhs_device + ntype,
             stream);
       }
     }
@@ -98,9 +90,7 @@ class DeviceNodeMapMaker {
       const IdArray& nodes = rhs_nodes[ntype];
       if (nodes->shape[0] > 0) {
         node_maps->RhsHashTable(ntype).FillWithUnique(
-            nodes.Ptr<IdType>(),
-            nodes->shape[0],
-            stream);
+            nodes.Ptr<IdType>(), nodes->shape[0], stream);
       }
     }
   }
@@ -118,8 +108,7 @@ class DeviceNodeMapMaker {
   void Make(
       const std::vector<IdArray>& lhs_nodes,
       const std::vector<IdArray>& rhs_nodes,
-      DeviceNodeMap<IdType> * const node_maps,
-      cudaStream_t stream) {
+      DeviceNodeMap<IdType>* const node_maps, cudaStream_t stream) {
     const int64_t num_ntypes = lhs_nodes.size() + rhs_nodes.size();
 
     // unique lhs nodes
@@ -129,9 +118,7 @@ class DeviceNodeMapMaker {
       if (nodes->shape[0] > 0) {
         CHECK_EQ(nodes->ctx.device_type, kDGLCUDA);
         node_maps->LhsHashTable(ntype).FillWithUnique(
-            nodes.Ptr<IdType>(),
-            nodes->shape[0],
-            stream);
+            nodes.Ptr<IdType>(), nodes->shape[0], stream);
       }
     }
 
@@ -141,9 +128,7 @@ class DeviceNodeMapMaker {
       const IdArray& nodes = rhs_nodes[ntype];
       if (nodes->shape[0] > 0) {
         node_maps->RhsHashTable(ntype).FillWithUnique(
-            nodes.Ptr<IdType>(),
-            nodes->shape[0],
-            stream);
+            nodes.Ptr<IdType>(), nodes->shape[0], stream);
       }
     }
   }
@@ -152,20 +137,15 @@ class DeviceNodeMapMaker {
   IdType max_num_nodes_;
 };
 
-
 // Since partial specialization is not allowed for functions, use this as an
 // intermediate for ToBlock where XPU = kDGLCUDA.
-template<typename IdType>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlockGPU(
-    HeteroGraphPtr graph,
-    const std::vector<IdArray> &rhs_nodes,
-    const bool include_rhs_in_lhs,
-    std::vector<IdArray>* const lhs_nodes_ptr) {
+template <typename IdType>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlockGPU(
+    HeteroGraphPtr graph, const std::vector<IdArray>& rhs_nodes,
+    const bool include_rhs_in_lhs, std::vector<IdArray>* const lhs_nodes_ptr) {
   std::vector<IdArray>& lhs_nodes = *lhs_nodes_ptr;
   const bool generate_lhs_nodes = lhs_nodes.empty();
 
-
   const auto& ctx = graph->Context();
   auto device = runtime::DeviceAPI::Get(ctx);
   cudaStream_t stream = runtime::getCurrentCUDAStream();
@@ -176,10 +156,11 @@ ToBlockGPU(
   }
 
   // Since DST nodes are included in SRC nodes, a common requirement is to fetch
-  // the DST node features from the SRC nodes features. To avoid expensive sparse lookup,
-  // the function assures that the DST nodes in both SRC and DST sets have the same ids.
-  // As a result, given the node feature tensor ``X`` of type ``utype``,
-  // the following code finds the corresponding DST node features of type ``vtype``:
+  // the DST node features from the SRC nodes features. To avoid expensive
+  // sparse lookup, the function assures that the DST nodes in both SRC and DST
+  // sets have the same ids. As a result, given the node feature tensor ``X`` of
+  // type ``utype``, the following code finds the corresponding DST node
+  // features of type ``vtype``:
 
   const int64_t num_etypes = graph->NumEdgeTypes();
   const int64_t num_ntypes = graph->NumVertexTypes();
@@ -197,9 +178,9 @@ ToBlockGPU(
   }
 
   // count lhs and rhs nodes
-  std::vector<int64_t> maxNodesPerType(num_ntypes*2, 0);
+  std::vector<int64_t> maxNodesPerType(num_ntypes * 2, 0);
   for (int64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-    maxNodesPerType[ntype+num_ntypes] += rhs_nodes[ntype]->shape[0];
+    maxNodesPerType[ntype + num_ntypes] += rhs_nodes[ntype]->shape[0];
 
     if (generate_lhs_nodes) {
       if (include_rhs_in_lhs) {
@@ -226,16 +207,16 @@ ToBlockGPU(
   if (generate_lhs_nodes) {
     std::vector<int64_t> src_node_offsets(num_ntypes, 0);
     for (int64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-      src_nodes[ntype] = NewIdArray(maxNodesPerType[ntype], ctx,
-          sizeof(IdType)*8);
+      src_nodes[ntype] =
+          NewIdArray(maxNodesPerType[ntype], ctx, sizeof(IdType) * 8);
       if (include_rhs_in_lhs) {
         // place rhs nodes first
-        device->CopyDataFromTo(rhs_nodes[ntype].Ptr<IdType>(), 0,
-            src_nodes[ntype].Ptr<IdType>(), src_node_offsets[ntype],
-            sizeof(IdType)*rhs_nodes[ntype]->shape[0],
-            rhs_nodes[ntype]->ctx, src_nodes[ntype]->ctx,
-            rhs_nodes[ntype]->dtype);
-        src_node_offsets[ntype] += sizeof(IdType)*rhs_nodes[ntype]->shape[0];
+        device->CopyDataFromTo(
+            rhs_nodes[ntype].Ptr<IdType>(), 0, src_nodes[ntype].Ptr<IdType>(),
+            src_node_offsets[ntype],
+            sizeof(IdType) * rhs_nodes[ntype]->shape[0], rhs_nodes[ntype]->ctx,
+            src_nodes[ntype]->ctx, rhs_nodes[ntype]->dtype);
+        src_node_offsets[ntype] += sizeof(IdType) * rhs_nodes[ntype]->shape[0];
       }
     }
     for (int64_t etype = 0; etype < num_etypes; ++etype) {
@@ -244,14 +225,13 @@ ToBlockGPU(
       if (edge_arrays[etype].src.defined()) {
         device->CopyDataFromTo(
             edge_arrays[etype].src.Ptr<IdType>(), 0,
-            src_nodes[srctype].Ptr<IdType>(),
-            src_node_offsets[srctype],
-            sizeof(IdType)*edge_arrays[etype].src->shape[0],
-            rhs_nodes[srctype]->ctx,
-            src_nodes[srctype]->ctx,
+            src_nodes[srctype].Ptr<IdType>(), src_node_offsets[srctype],
+            sizeof(IdType) * edge_arrays[etype].src->shape[0],
+            rhs_nodes[srctype]->ctx, src_nodes[srctype]->ctx,
             rhs_nodes[srctype]->dtype);
 
-        src_node_offsets[srctype] += sizeof(IdType)*edge_arrays[etype].src->shape[0];
+        src_node_offsets[srctype] +=
+            sizeof(IdType) * edge_arrays[etype].src->shape[0];
       }
     }
   } else {
@@ -267,47 +247,35 @@ ToBlockGPU(
   if (generate_lhs_nodes) {
     lhs_nodes.reserve(num_ntypes);
     for (int64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-      lhs_nodes.emplace_back(NewIdArray(
-          maxNodesPerType[ntype], ctx, sizeof(IdType)*8));
+      lhs_nodes.emplace_back(
+          NewIdArray(maxNodesPerType[ntype], ctx, sizeof(IdType) * 8));
     }
   }
 
-  std::vector<int64_t> num_nodes_per_type(num_ntypes*2);
+  std::vector<int64_t> num_nodes_per_type(num_ntypes * 2);
   // populate RHS nodes from what we already know
   for (int64_t ntype = 0; ntype < num_ntypes; ++ntype) {
-    num_nodes_per_type[num_ntypes+ntype] = rhs_nodes[ntype]->shape[0];
+    num_nodes_per_type[num_ntypes + ntype] = rhs_nodes[ntype]->shape[0];
   }
 
   // populate the mappings
   if (generate_lhs_nodes) {
-    int64_t * count_lhs_device = static_cast<int64_t*>(
-        device->AllocWorkspace(ctx, sizeof(int64_t)*num_ntypes*2));
+    int64_t* count_lhs_device = static_cast<int64_t*>(
+        device->AllocWorkspace(ctx, sizeof(int64_t) * num_ntypes * 2));
 
     maker.Make(
-        src_nodes,
-        rhs_nodes,
-        &node_maps,
-        count_lhs_device,
-        &lhs_nodes,
-        stream);
+        src_nodes, rhs_nodes, &node_maps, count_lhs_device, &lhs_nodes, stream);
 
     device->CopyDataFromTo(
-        count_lhs_device, 0,
-        num_nodes_per_type.data(), 0,
-        sizeof(*num_nodes_per_type.data())*num_ntypes,
-        ctx,
-        DGLContext{kDGLCPU, 0},
-        DGLDataType{kDGLInt, 64, 1});
+        count_lhs_device, 0, num_nodes_per_type.data(), 0,
+        sizeof(*num_nodes_per_type.data()) * num_ntypes, ctx,
+        DGLContext{kDGLCPU, 0}, DGLDataType{kDGLInt, 64, 1});
     device->StreamSync(ctx, stream);
 
     // wait for the node counts to finish transferring
     device->FreeWorkspace(ctx, count_lhs_device);
   } else {
-    maker.Make(
-        lhs_nodes,
-        rhs_nodes,
-        &node_maps,
-        stream);
+    maker.Make(lhs_nodes, rhs_nodes, &node_maps, stream);
 
     for (int64_t ntype = 0; ntype < num_ntypes; ++ntype) {
       num_nodes_per_type[ntype] = lhs_nodes[ntype]->shape[0];
@@ -321,7 +289,7 @@ ToBlockGPU(
       induced_edges.push_back(edge_arrays[etype].id);
     } else {
       induced_edges.push_back(
-            aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType)*8, 1}, ctx));
+          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType) * 8, 1}, ctx));
     }
   }
 
@@ -329,8 +297,8 @@ ToBlockGPU(
   const auto meta_graph = graph->meta_graph();
   const EdgeArray etypes = meta_graph->Edges("eid");
   const IdArray new_dst = Add(etypes.dst, num_ntypes);
-  const auto new_meta_graph = ImmutableGraph::CreateFromCOO(
-      num_ntypes * 2, etypes.src, new_dst);
+  const auto new_meta_graph =
+      ImmutableGraph::CreateFromCOO(num_ntypes * 2, etypes.src, new_dst);
 
   // allocate vector for graph relations while GPU is busy
   std::vector<HeteroGraphPtr> rel_graphs;
@@ -358,20 +326,17 @@ ToBlockGPU(
       // No rhs nodes are given for this edge type. Create an empty graph.
       rel_graphs.push_back(CreateFromCOO(
           2, lhs_nodes[srctype]->shape[0], rhs_nodes[dsttype]->shape[0],
-          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType)*8, 1}, ctx),
-          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType)*8, 1}, ctx)));
+          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType) * 8, 1}, ctx),
+          aten::NullArray(DGLDataType{kDGLInt, sizeof(IdType) * 8, 1}, ctx)));
     } else {
       rel_graphs.push_back(CreateFromCOO(
-          2,
-          lhs_nodes[srctype]->shape[0],
-          rhs_nodes[dsttype]->shape[0],
-          new_lhs[etype],
-          new_rhs[etype]));
+          2, lhs_nodes[srctype]->shape[0], rhs_nodes[dsttype]->shape[0],
+          new_lhs[etype], new_rhs[etype]));
     }
   }
 
-  HeteroGraphPtr new_graph = CreateHeteroGraph(
-      new_meta_graph, rel_graphs, num_nodes_per_type);
+  HeteroGraphPtr new_graph =
+      CreateHeteroGraph(new_meta_graph, rel_graphs, num_nodes_per_type);
 
   // return the new graph, the new src nodes, and new edges
   return std::make_tuple(new_graph, induced_edges);
@@ -379,26 +344,22 @@ ToBlockGPU(
 
 }  // namespace
 
-// Use explicit names to get around MSVC's broken mangling that thinks the following two
-// functions are the same.
-// Using template<> fails to export the symbols.
+// Use explicit names to get around MSVC's broken mangling that thinks the
+// following two functions are the same. Using template<> fails to export the
+// symbols.
 std::tuple<HeteroGraphPtr, std::vector<IdArray>>
 // ToBlock<kDGLCUDA, int32_t>
 ToBlockGPU32(
-    HeteroGraphPtr graph,
-    const std::vector<IdArray> &rhs_nodes,
-    bool include_rhs_in_lhs,
-    std::vector<IdArray>* const lhs_nodes) {
+    HeteroGraphPtr graph, const std::vector<IdArray>& rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray>* const lhs_nodes) {
   return ToBlockGPU<int32_t>(graph, rhs_nodes, include_rhs_in_lhs, lhs_nodes);
 }
 
 std::tuple<HeteroGraphPtr, std::vector<IdArray>>
 // ToBlock<kDGLCUDA, int64_t>
 ToBlockGPU64(
-    HeteroGraphPtr graph,
-    const std::vector<IdArray> &rhs_nodes,
-    bool include_rhs_in_lhs,
-    std::vector<IdArray>* const lhs_nodes) {
+    HeteroGraphPtr graph, const std::vector<IdArray>& rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray>* const lhs_nodes) {
   return ToBlockGPU<int64_t>(graph, rhs_nodes, include_rhs_in_lhs, lhs_nodes);
 }
 

src/graph/transform/knn.cc

@@ -4,9 +4,11 @@
  * \brief k-nearest-neighbor (KNN) interface
  */
 
-#include <dgl/runtime/registry.h>
-#include <dgl/runtime/packed_func.h>
 #include "knn.h"
+
+#include <dgl/runtime/packed_func.h>
+#include <dgl/runtime/registry.h>
+
 #include "../../array/check.h"
 
 using namespace dgl::runtime;
@@ -14,7 +16,7 @@ namespace dgl {
 namespace transform {
 
 DGL_REGISTER_GLOBAL("transform._CAPI_DGLKNN")
-.set_body([] (DGLArgs args, DGLRetValue* rv) {
+    .set_body([](DGLArgs args, DGLRetValue* rv) {
       const NDArray data_points = args[0];
       const IdArray data_offsets = args[1];
       const NDArray query_points = args[2];
@@ -25,7 +27,8 @@ DGL_REGISTER_GLOBAL("transform._CAPI_DGLKNN")
 
       aten::CheckContiguous(
           {data_points, data_offsets, query_points, query_offsets, result},
-      {"data_points", "data_offsets", "query_points", "query_offsets", "result"});
+          {"data_points", "data_offsets", "query_points", "query_offsets",
+           "result"});
       aten::CheckCtx(
           data_points->ctx, {data_offsets, query_points, query_offsets, result},
           {"data_offsets", "query_points", "query_offsets", "result"});
@@ -34,15 +37,15 @@ DGL_REGISTER_GLOBAL("transform._CAPI_DGLKNN")
         ATEN_FLOAT_TYPE_SWITCH(data_points->dtype, FloatType, "data_points", {
           ATEN_ID_TYPE_SWITCH(result->dtype, IdType, {
             KNN<XPU, FloatType, IdType>(
-            data_points, data_offsets, query_points,
-            query_offsets, k, result, algorithm);
+                data_points, data_offsets, query_points, query_offsets, k,
+                result, algorithm);
           });
         });
       });
     });
 
 DGL_REGISTER_GLOBAL("transform._CAPI_DGLNNDescent")
-.set_body([] (DGLArgs args, DGLRetValue* rv) {
+    .set_body([](DGLArgs args, DGLRetValue* rv) {
       const NDArray points = args[0];
       const IdArray offsets = args[1];
       const IdArray result = args[2];
@@ -54,7 +57,8 @@ DGL_REGISTER_GLOBAL("transform._CAPI_DGLNNDescent")
       aten::CheckContiguous(
           {points, offsets, result}, {"points", "offsets", "result"});
       aten::CheckCtx(
-      points->ctx, {points, offsets, result}, {"points", "offsets", "result"});
+          points->ctx, {points, offsets, result},
+          {"points", "offsets", "result"});
 
       ATEN_XPU_SWITCH_CUDA(points->ctx.device_type, XPU, "NNDescent", {
         ATEN_FLOAT_TYPE_SWITCH(points->dtype, FloatType, "points", {

src/graph/transform/knn.h

@@ -8,6 +8,7 @@
 #define DGL_GRAPH_TRANSFORM_KNN_H_
 
 #include <dgl/array.h>
+
 #include <string>
 
 namespace dgl {
@@ -15,22 +16,24 @@ namespace transform {
 
 /*!
  * \brief For each point in each segment in \a query_points, find \a k nearest
- *  points in the same segment in \a data_points. \a data_offsets and \a query_offsets
- *  determine the start index of each segment in \a data_points and \a query_points.
+ *        points in the same segment in \a data_points. \a data_offsets and \a
+ *        query_offsets determine the start index of each segment in \a
+ *        data_points and \a query_points.
  *
  * \param data_points dataset points.
  * \param data_offsets offsets of point index in \a data_points.
  * \param query_points query points.
  * \param query_offsets offsets of point index in \a query_points.
  * \param k the number of nearest points.
- * \param result output array. A 2D tensor indicating the index
- *  relation between \a query_points and \a data_points.
+ * \param result output array. A 2D tensor indicating the index  relation
+ *        between \a query_points and \a data_points.
  * \param algorithm algorithm used to compute the k-nearest neighbors.
  */
 template <DGLDeviceType XPU, typename FloatType, typename IdType>
-void KNN(const NDArray& data_points, const IdArray& data_offsets,
-         const NDArray& query_points, const IdArray& query_offsets,
-         const int k, IdArray result, const std::string& algorithm);
+void KNN(
+    const NDArray& data_points, const IdArray& data_offsets,
+    const NDArray& query_points, const IdArray& query_offsets, const int k,
+    IdArray result, const std::string& algorithm);
 
 /*!
  * \brief For each input point, find \a k approximate nearest points in the same
@@ -38,19 +41,20 @@ void KNN(const NDArray& data_points, const IdArray& data_offsets,
  *
  * \param points input points.
  * \param offsets offsets of point index.
- * \param result output array. A 2D tensor indicating the index relation between points.
+ * \param result output array. A 2D tensor indicating the index relation between
+ *        points.
  * \param k the number of nearest points.
  * \param num_iters The maximum number of NN-descent iterations to perform.
- * \param num_candidates The maximum number of candidates to be considered during one iteration.
+ * \param num_candidates The maximum number of candidates to be considered
+ *        during one iteration.
  * \param delta A value controls the early abort.
  */
 template <DGLDeviceType XPU, typename FloatType, typename IdType>
-void NNDescent(const NDArray& points, const IdArray& offsets,
-               IdArray result, const int k, const int num_iters,
-               const int num_candidates, const double delta);
+void NNDescent(
+    const NDArray& points, const IdArray& offsets, IdArray result, const int k,
+    const int num_iters, const int num_candidates, const double delta);
 
 }  // namespace transform
 }  // namespace dgl
 
-
 #endif  // DGL_GRAPH_TRANSFORM_KNN_H_

src/graph/transform/line_graph.cc

@@ -4,12 +4,14 @@
  * \brief Line graph implementation
  */
 
-#include <dgl/base_heterograph.h>
-#include <dgl/transform.h>
 #include <dgl/array.h>
+#include <dgl/base_heterograph.h>
 #include <dgl/packed_func_ext.h>
-#include <vector>
+#include <dgl/transform.h>
+
 #include <utility>
+#include <vector>
+
 #include "../../c_api_common.h"
 #include "../heterograph.h"
 
@@ -21,26 +23,25 @@ using namespace dgl::aten;
 namespace transform {
 
 /*!
- * \brief Create Line Graph
- * \param hg Graph
- * \param backtracking whether the pair of (v, u) (u, v) edges are treated as linked
- * \return The Line Graph
+ * \brief Create Line Graph.
+ * \param hg Graph.
+ * \param backtracking whether the pair of (v, u) (u, v) edges are treated as
+ *        linked.
+ * \return The Line Graph.
  */
-HeteroGraphPtr CreateLineGraph(
-    HeteroGraphPtr hg,
-    bool backtracking) {
+HeteroGraphPtr CreateLineGraph(HeteroGraphPtr hg, bool backtracking) {
   const auto hgp = std::dynamic_pointer_cast<HeteroGraph>(hg);
   return hgp->LineGraph(backtracking);
 }
 
 DGL_REGISTER_GLOBAL("transform._CAPI_DGLHeteroLineGraph")
-.set_body([] (DGLArgs args, DGLRetValue* rv) {
+    .set_body([](DGLArgs args, DGLRetValue* rv) {
       HeteroGraphRef hg = args[0];
       bool backtracking = args[1];
 
       auto hgptr = CreateLineGraph(hg.sptr(), backtracking);
       *rv = HeteroGraphRef(hgptr);
-});
+    });
 
 };  // namespace transform
 };  // namespace dgl

src/graph/transform/metis_partition_hetero.cc

@@ -19,8 +19,9 @@ namespace transform {
 
 #if !defined(_WIN32)
 
-IdArray MetisPartition(UnitGraphPtr g, int k, NDArray vwgt_arr,
-                       const std::string &mode, bool obj_cut) {
+IdArray MetisPartition(
+    UnitGraphPtr g, int k, NDArray vwgt_arr, const std::string &mode,
+    bool obj_cut) {
   // Mode can only be "k-way" or "recursive"
   CHECK(mode == "k-way" || mode == "recursive")
       << "mode can only be \"k-way\" or \"recursive\"";
@@ -51,7 +52,8 @@ IdArray MetisPartition(UnitGraphPtr g, int k, NDArray vwgt_arr,
     vwgt = static_cast<idx_t *>(vwgt_arr->data);
   }
 
-  auto partition_func = (mode == "k-way") ? METIS_PartGraphKway : METIS_PartGraphRecursive;
+  auto partition_func =
+      (mode == "k-way") ? METIS_PartGraphKway : METIS_PartGraphRecursive;
 
   idx_t options[METIS_NOPTIONS];
   METIS_SetDefaultOptions(options);

src/graph/transform/partition_hetero.cc

@@ -37,20 +37,23 @@ HeteroGraphPtr ReorderUnitGraph(UnitGraphPtr ug, IdArray new_order) {
   if (format & CSC_CODE) {
     auto cscmat = ug->GetCSCMatrix(0);
     auto new_cscmat = aten::CSRReorder(cscmat, new_order, new_order);
-    return UnitGraph::CreateFromCSC(ug->NumVertexTypes(), new_cscmat, ug->GetAllowedFormats());
+    return UnitGraph::CreateFromCSC(
+        ug->NumVertexTypes(), new_cscmat, ug->GetAllowedFormats());
   } else if (format & CSR_CODE) {
     auto csrmat = ug->GetCSRMatrix(0);
     auto new_csrmat = aten::CSRReorder(csrmat, new_order, new_order);
-    return UnitGraph::CreateFromCSR(ug->NumVertexTypes(), new_csrmat, ug->GetAllowedFormats());
+    return UnitGraph::CreateFromCSR(
+        ug->NumVertexTypes(), new_csrmat, ug->GetAllowedFormats());
   } else {
     auto coomat = ug->GetCOOMatrix(0);
     auto new_coomat = aten::COOReorder(coomat, new_order, new_order);
-    return UnitGraph::CreateFromCOO(ug->NumVertexTypes(), new_coomat, ug->GetAllowedFormats());
+    return UnitGraph::CreateFromCOO(
+        ug->NumVertexTypes(), new_coomat, ug->GetAllowedFormats());
   }
 }
 
-HaloHeteroSubgraph GetSubgraphWithHalo(std::shared_ptr<HeteroGraph> hg,
-                                       IdArray nodes, int num_hops) {
+HaloHeteroSubgraph GetSubgraphWithHalo(
+    std::shared_ptr<HeteroGraph> hg, IdArray nodes, int num_hops) {
   CHECK_EQ(hg->NumBits(), 64) << "halo subgraph only supports 64bits graph";
   CHECK_EQ(hg->relation_graphs().size(), 1)
       << "halo subgraph only supports homogeneous graph";
@@ -113,8 +116,8 @@ HaloHeteroSubgraph GetSubgraphWithHalo(std::shared_ptr<HeteroGraph> hg,
     const dgl_id_t *eid_data = static_cast<dgl_id_t *>(eid->data);
     for (int64_t i = 0; i < num_edges; i++) {
       auto it1 = orig_nodes.find(src_data[i]);
-      // If the source node is in the partition, we have got this edge when we iterate over
-      // the out-edges above.
+      // If the source node is in the partition, we have got this edge when we
+      // iterate over the out-edges above.
       if (it1 == orig_nodes.end()) {
         edge_src.push_back(src_data[i]);
         edge_dst.push_back(dst_data[i]);
@@ -164,10 +167,10 @@ HaloHeteroSubgraph GetSubgraphWithHalo(std::shared_ptr<HeteroGraph> hg,
   }
 
   num_edges = edge_src.size();
-  IdArray new_src = IdArray::Empty({num_edges}, DGLDataType{kDGLInt, 64, 1},
-                                   DGLContext{kDGLCPU, 0});
-  IdArray new_dst = IdArray::Empty({num_edges}, DGLDataType{kDGLInt, 64, 1},
-                                   DGLContext{kDGLCPU, 0});
+  IdArray new_src = IdArray::Empty(
+      {num_edges}, DGLDataType{kDGLInt, 64, 1}, DGLContext{kDGLCPU, 0});
+  IdArray new_dst = IdArray::Empty(
+      {num_edges}, DGLDataType{kDGLInt, 64, 1}, DGLContext{kDGLCPU, 0});
   dgl_id_t *new_src_data = static_cast<dgl_id_t *>(new_src->data);
   dgl_id_t *new_dst_data = static_cast<dgl_id_t *>(new_dst->data);
   for (size_t i = 0; i < edge_src.size(); i++) {
@@ -180,8 +183,8 @@ HaloHeteroSubgraph GetSubgraphWithHalo(std::shared_ptr<HeteroGraph> hg,
     dgl_id_t old_nid = old_node_ids[i];
     inner_nodes[i] = all_nodes[old_nid];
   }
-  aten::COOMatrix coo(old_node_ids.size(), old_node_ids.size(), new_src,
-                      new_dst);
+  aten::COOMatrix coo(
+      old_node_ids.size(), old_node_ids.size(), new_src, new_dst);
   HeteroGraphPtr ugptr = UnitGraph::CreateFromCOO(1, coo);
   HeteroGraphPtr subg = CreateHeteroGraph(hg->meta_graph(), {ugptr});
   HaloHeteroSubgraph halo_subg;
@@ -245,10 +248,10 @@ DGL_REGISTER_GLOBAL("partition._CAPI_DGLPartitionWithHalo_Hetero")
         part_ids.push_back(it->first);
         part_nodes.push_back(it->second);
       }
-    // When we construct subgraphs, we need to access both in-edges and out-edges.
-    // We need to make sure the in-CSR and out-CSR exist. Otherwise, we'll
-    // try to construct in-CSR and out-CSR in openmp for loop, which will lead
-    // to some unexpected results.
+      // When we construct subgraphs, we need to access both in-edges and
+      // out-edges. We need to make sure the in-CSR and out-CSR exist.
+      // Otherwise, we'll try to construct in-CSR and out-CSR in openmp for
+      // loop, which will lead to some unexpected results.
       ugptr->GetInCSR();
       ugptr->GetOutCSR();
       std::vector<std::shared_ptr<HaloHeteroSubgraph>> subgs(max_part_id + 1);
@@ -270,7 +273,7 @@ DGL_REGISTER_GLOBAL("partition._CAPI_DGLPartitionWithHalo_Hetero")
       *rv = ret_list;
     });
 
-template<class IdType>
+template <class IdType>
 struct EdgeProperty {
   IdType eid;
   int64_t idx;
@@ -315,15 +318,18 @@ DGL_REGISTER_GLOBAL("partition._CAPI_DGLReassignEdges_Hetero")
             indexed_eids[j].part_id = part_id_data[i];
           }
         }
-      auto comp = [etype_data](const EdgeProperty<IdType> &a, const EdgeProperty<IdType> &b) {
+        auto comp = [etype_data](
+                        const EdgeProperty<IdType> &a,
+                        const EdgeProperty<IdType> &b) {
           if (a.part_id == b.part_id) {
             return etype_data[a.eid] < etype_data[b.eid];
           } else {
             return a.part_id < b.part_id;
           }
         };
-      // We only need to sort the edges if the input graph has multiple relations.
-      // If it's a homogeneous grap, we'll just assign edge Ids based on its previous order.
+        // We only need to sort the edges if the input graph has multiple
+        // relations. If it's a homogeneous grap, we'll just assign edge Ids
+        // based on its previous order.
         if (etype->shape[0] > 0) {
           std::sort(indexed_eids.begin(), indexed_eids.end(), comp);
         }
@@ -345,7 +351,6 @@ DGL_REGISTER_GLOBAL("partition._CAPI_GetHaloSubgraphInnerNodes_Hetero")
       *rv = gptr->inner_nodes[0];
     });
 
-
 DGL_REGISTER_GLOBAL("partition._CAPI_DGLMakeSymmetric_Hetero")
     .set_body([](DGLArgs args, DGLRetValue *rv) {
       HeteroGraphRef g = args[0];
@@ -363,13 +368,14 @@ DGL_REGISTER_GLOBAL("partition._CAPI_DGLMakeSymmetric_Hetero")
       gk_csr_Free(&gk_csr);
       gk_csr_Free(&sym_gk_csr);
 
-    auto new_ugptr = UnitGraph::CreateFromCSC(ugptr->NumVertexTypes(), mat,
-                                              ugptr->GetAllowedFormats());
+      auto new_ugptr = UnitGraph::CreateFromCSC(
+          ugptr->NumVertexTypes(), mat, ugptr->GetAllowedFormats());
       std::vector<HeteroGraphPtr> rel_graphs = {new_ugptr};
       *rv = HeteroGraphRef(std::make_shared<HeteroGraph>(
           hgptr->meta_graph(), rel_graphs, hgptr->NumVerticesPerType()));
 #else
-    LOG(FATAL) << "The fast version of making symmetric graph is not supported in Windows.";
+      LOG(FATAL) << "The fast version of making symmetric graph is not "
+                    "supported in Windows.";
 #endif  // !defined(_WIN32)
     });
 

src/graph/transform/remove_edges.cc

@@ -4,15 +4,16 @@
  * \brief Remove edges.
  */
 
-#include <dgl/base_heterograph.h>
-#include <dgl/transform.h>
 #include <dgl/array.h>
+#include <dgl/base_heterograph.h>
 #include <dgl/packed_func_ext.h>
-#include <dgl/runtime/registry.h>
 #include <dgl/runtime/container.h>
-#include <vector>
-#include <utility>
+#include <dgl/runtime/registry.h>
+#include <dgl/transform.h>
+
 #include <tuple>
+#include <utility>
+#include <vector>
 
 namespace dgl {
 
@@ -21,8 +22,8 @@ using namespace dgl::aten;
 
 namespace transform {
 
-std::pair<HeteroGraphPtr, std::vector<IdArray>>
-RemoveEdges(const HeteroGraphPtr graph, const std::vector<IdArray> &eids) {
+std::pair<HeteroGraphPtr, std::vector<IdArray>> RemoveEdges(
+    const HeteroGraphPtr graph, const std::vector<IdArray> &eids) {
   std::vector<IdArray> induced_eids;
   std::vector<HeteroGraphPtr> rel_graphs;
   const int64_t num_etypes = graph->NumEdgeTypes();
@@ -40,23 +41,30 @@ RemoveEdges(const HeteroGraphPtr graph, const std::vector<IdArray> &eids) {
       const COOMatrix &coo = graph->GetCOOMatrix(etype);
       const COOMatrix &result = COORemove(coo, eids[etype]);
       new_rel_graph = CreateFromCOO(
-          num_ntypes_rel, result.num_rows, result.num_cols, result.row, result.col);
+          num_ntypes_rel, result.num_rows, result.num_cols, result.row,
+          result.col);
       induced_eids_rel = result.data;
     } else if (fmt == SparseFormat::kCSR) {
       const CSRMatrix &csr = graph->GetCSRMatrix(etype);
       const CSRMatrix &result = CSRRemove(csr, eids[etype]);
       new_rel_graph = CreateFromCSR(
-          num_ntypes_rel, result.num_rows, result.num_cols, result.indptr, result.indices,
+          num_ntypes_rel, result.num_rows, result.num_cols, result.indptr,
+          result.indices,
           // TODO(BarclayII): make CSR support null eid array
-          Range(0, result.indices->shape[0], result.indices->dtype.bits, result.indices->ctx));
+          Range(
+              0, result.indices->shape[0], result.indices->dtype.bits,
+              result.indices->ctx));
       induced_eids_rel = result.data;
     } else if (fmt == SparseFormat::kCSC) {
       const CSRMatrix &csc = graph->GetCSCMatrix(etype);
       const CSRMatrix &result = CSRRemove(csc, eids[etype]);
       new_rel_graph = CreateFromCSC(
-          num_ntypes_rel, result.num_rows, result.num_cols, result.indptr, result.indices,
+          num_ntypes_rel, result.num_rows, result.num_cols, result.indptr,
+          result.indices,
           // TODO(BarclayII): make CSR support null eid array
-          Range(0, result.indices->shape[0], result.indices->dtype.bits, result.indices->ctx));
+          Range(
+              0, result.indices->shape[0], result.indices->dtype.bits,
+              result.indices->ctx));
       induced_eids_rel = result.data;
     }
 
@@ -70,7 +78,7 @@ RemoveEdges(const HeteroGraphPtr graph, const std::vector<IdArray> &eids) {
 }
 
 DGL_REGISTER_GLOBAL("transform._CAPI_DGLRemoveEdges")
-.set_body([] (DGLArgs args, DGLRetValue *rv) {
+    .set_body([](DGLArgs args, DGLRetValue *rv) {
       const HeteroGraphRef graph_ref = args[0];
       const std::vector<IdArray> &eids = ListValueToVector<IdArray>(args[1]);
 

src/graph/transform/to_bipartite.cc

@@ -19,16 +19,18 @@
 
 #include "to_bipartite.h"
 
-#include <dgl/base_heterograph.h>
-#include <dgl/transform.h>
 #include <dgl/array.h>
-#include <dgl/packed_func_ext.h>
+#include <dgl/base_heterograph.h>
 #include <dgl/immutable_graph.h>
-#include <dgl/runtime/registry.h>
+#include <dgl/packed_func_ext.h>
 #include <dgl/runtime/container.h>
-#include <vector>
+#include <dgl/runtime/registry.h>
+#include <dgl/transform.h>
+
 #include <tuple>
 #include <utility>
+#include <vector>
+
 #include "../../array/cpu/array_utils.h"
 
 namespace dgl {
@@ -42,11 +44,11 @@ namespace {
 
 // Since partial specialization is not allowed for functions, use this as an
 // intermediate for ToBlock where XPU = kDGLCPU.
-template<typename IdType>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
-    bool include_rhs_in_lhs, std::vector<IdArray>* const lhs_nodes_ptr) {
-  std::vector<IdArray>& lhs_nodes = *lhs_nodes_ptr;
+template <typename IdType>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlockCPU(
+    HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray> *const lhs_nodes_ptr) {
+  std::vector<IdArray> &lhs_nodes = *lhs_nodes_ptr;
   const bool generate_lhs_nodes = lhs_nodes.empty();
 
   const int64_t num_etypes = graph->NumEdgeTypes();
@@ -56,7 +58,8 @@ ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
   CHECK(rhs_nodes.size() == static_cast<size_t>(num_ntypes))
       << "rhs_nodes not given for every node type";
 
-  const std::vector<IdHashMap<IdType>> rhs_node_mappings(rhs_nodes.begin(), rhs_nodes.end());
+  const std::vector<IdHashMap<IdType>> rhs_node_mappings(
+      rhs_nodes.begin(), rhs_nodes.end());
   std::vector<IdHashMap<IdType>> lhs_node_mappings;
 
   if (generate_lhs_nodes) {
@@ -66,16 +69,16 @@ ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
     else
       lhs_node_mappings.resize(num_ntypes);
   } else {
-    lhs_node_mappings = std::vector<IdHashMap<IdType>>(lhs_nodes.begin(), lhs_nodes.end());
+    lhs_node_mappings =
+        std::vector<IdHashMap<IdType>>(lhs_nodes.begin(), lhs_nodes.end());
   }
 
-
   for (int64_t etype = 0; etype < num_etypes; ++etype) {
     const auto src_dst_types = graph->GetEndpointTypes(etype);
     const dgl_type_t srctype = src_dst_types.first;
     const dgl_type_t dsttype = src_dst_types.second;
     if (!aten::IsNullArray(rhs_nodes[dsttype])) {
-      const EdgeArray& edges = graph->Edges(etype);
+      const EdgeArray &edges = graph->Edges(etype);
       if (generate_lhs_nodes) {
         lhs_node_mappings[srctype].Update(edges.src);
       }
@@ -89,8 +92,8 @@ ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
   const auto meta_graph = graph->meta_graph();
   const EdgeArray etypes = meta_graph->Edges("eid");
   const IdArray new_dst = Add(etypes.dst, num_ntypes);
-  const auto new_meta_graph = ImmutableGraph::CreateFromCOO(
-      num_ntypes * 2, etypes.src, new_dst);
+  const auto new_meta_graph =
+      ImmutableGraph::CreateFromCOO(num_ntypes * 2, etypes.src, new_dst);
 
   for (int64_t ntype = 0; ntype < num_ntypes; ++ntype)
     num_nodes_per_type.push_back(lhs_node_mappings[ntype].Size());
@@ -108,8 +111,8 @@ ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
     if (rhs_map.Size() == 0) {
       // No rhs nodes are given for this edge type. Create an empty graph.
       rel_graphs.push_back(CreateFromCOO(
-          2, lhs_map.Size(), rhs_map.Size(),
-          aten::NullArray(), aten::NullArray()));
+          2, lhs_map.Size(), rhs_map.Size(), aten::NullArray(),
+          aten::NullArray()));
       induced_edges.push_back(aten::NullArray());
     } else {
       IdArray new_src = lhs_map.Map(edge_arrays[etype].src, -1);
@@ -117,18 +120,18 @@ ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
       // Check whether there are unmapped IDs and raise error.
       for (int64_t i = 0; i < new_dst->shape[0]; ++i)
         CHECK_NE(new_dst.Ptr<IdType>()[i], -1)
-          << "Node " << edge_arrays[etype].dst.Ptr<IdType>()[i] << " does not exist"
+            << "Node " << edge_arrays[etype].dst.Ptr<IdType>()[i]
+            << " does not exist"
             << " in `rhs_nodes`. Argument `rhs_nodes` must contain all the edge"
             << " destination nodes.";
-      rel_graphs.push_back(CreateFromCOO(
-          2, lhs_map.Size(), rhs_map.Size(),
-          new_src, new_dst));
+      rel_graphs.push_back(
+          CreateFromCOO(2, lhs_map.Size(), rhs_map.Size(), new_src, new_dst));
       induced_edges.push_back(edge_arrays[etype].id);
     }
   }
 
-  const HeteroGraphPtr new_graph = CreateHeteroGraph(
-      new_meta_graph, rel_graphs, num_nodes_per_type);
+  const HeteroGraphPtr new_graph =
+      CreateHeteroGraph(new_meta_graph, rel_graphs, num_nodes_per_type);
 
   if (generate_lhs_nodes) {
     CHECK_EQ(lhs_nodes.size(), 0) << "InteralError: lhs_nodes should be empty "
@@ -141,59 +144,56 @@ ToBlockCPU(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
 
 }  // namespace
 
-template<>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlock<kDGLCPU, int32_t>(HeteroGraphPtr graph,
-                         const std::vector<IdArray> &rhs_nodes,
-                         bool include_rhs_in_lhs,
-                         std::vector<IdArray>* const lhs_nodes) {
+template <>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlock<kDGLCPU, int32_t>(
+    HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray> *const lhs_nodes) {
   return ToBlockCPU<int32_t>(graph, rhs_nodes, include_rhs_in_lhs, lhs_nodes);
 }
 
-template<>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlock<kDGLCPU, int64_t>(HeteroGraphPtr graph,
-                         const std::vector<IdArray> &rhs_nodes,
-                         bool include_rhs_in_lhs,
-                         std::vector<IdArray>* const lhs_nodes) {
+template <>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlock<kDGLCPU, int64_t>(
+    HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray> *const lhs_nodes) {
   return ToBlockCPU<int64_t>(graph, rhs_nodes, include_rhs_in_lhs, lhs_nodes);
 }
 
 #ifdef DGL_USE_CUDA
 
-// Forward declaration of GPU ToBlock implementations - actual implementation is in
+// Forward declaration of GPU ToBlock implementations - actual implementation is
+// in
 // ./cuda/cuda_to_block.cu
-// This is to get around the broken name mangling in VS2019 CL 16.5.5 + CUDA 11.3
-// which complains that the two template specializations have the same signature.
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlockGPU32(HeteroGraphPtr, const std::vector<IdArray>&, bool, std::vector<IdArray>* const);
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlockGPU64(HeteroGraphPtr, const std::vector<IdArray>&, bool, std::vector<IdArray>* const);
-
-template<>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlock<kDGLCUDA, int32_t>(HeteroGraphPtr graph,
-                         const std::vector<IdArray> &rhs_nodes,
-                         bool include_rhs_in_lhs,
-                         std::vector<IdArray>* const lhs_nodes) {
+// This is to get around the broken name mangling in VS2019 CL 16.5.5 +
+// CUDA 11.3 which complains that the two template specializations have the same
+// signature.
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlockGPU32(
+    HeteroGraphPtr, const std::vector<IdArray> &, bool,
+    std::vector<IdArray> *const);
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlockGPU64(
+    HeteroGraphPtr, const std::vector<IdArray> &, bool,
+    std::vector<IdArray> *const);
+
+template <>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlock<kDGLCUDA, int32_t>(
+    HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray> *const lhs_nodes) {
   return ToBlockGPU32(graph, rhs_nodes, include_rhs_in_lhs, lhs_nodes);
 }
 
-template<>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlock<kDGLCUDA, int64_t>(HeteroGraphPtr graph,
-                         const std::vector<IdArray> &rhs_nodes,
-                         bool include_rhs_in_lhs,
-                         std::vector<IdArray>* const lhs_nodes) {
+template <>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlock<kDGLCUDA, int64_t>(
+    HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
+    bool include_rhs_in_lhs, std::vector<IdArray> *const lhs_nodes) {
   return ToBlockGPU64(graph, rhs_nodes, include_rhs_in_lhs, lhs_nodes);
 }
 
 #endif  // DGL_USE_CUDA
 
 DGL_REGISTER_GLOBAL("transform._CAPI_DGLToBlock")
-.set_body([] (DGLArgs args, DGLRetValue *rv) {
+    .set_body([](DGLArgs args, DGLRetValue *rv) {
       const HeteroGraphRef graph_ref = args[0];
-    const std::vector<IdArray> &rhs_nodes = ListValueToVector<IdArray>(args[1]);
+      const std::vector<IdArray> &rhs_nodes =
+          ListValueToVector<IdArray>(args[1]);
       const bool include_rhs_in_lhs = args[2];
       std::vector<IdArray> lhs_nodes = ListValueToVector<IdArray>(args[3]);
 
@@ -203,8 +203,7 @@ DGL_REGISTER_GLOBAL("transform._CAPI_DGLToBlock")
       ATEN_XPU_SWITCH_CUDA(graph_ref->Context().device_type, XPU, "ToBlock", {
         ATEN_ID_TYPE_SWITCH(graph_ref->DataType(), IdType, {
           std::tie(new_graph, induced_edges) = ToBlock<XPU, IdType>(
-          graph_ref.sptr(), rhs_nodes, include_rhs_in_lhs,
-          &lhs_nodes);
+              graph_ref.sptr(), rhs_nodes, include_rhs_in_lhs, &lhs_nodes);
         });
       });
 

src/graph/transform/to_bipartite.h

@@ -44,9 +44,9 @@ namespace transform {
  *
  * @return The block and the induced edges.
  */
-template<DGLDeviceType XPU, typename IdType>
-std::tuple<HeteroGraphPtr, std::vector<IdArray>>
-ToBlock(HeteroGraphPtr graph, const std::vector<IdArray> &rhs_nodes,
+template <DGLDeviceType XPU, typename IdType>
+std::tuple<HeteroGraphPtr, std::vector<IdArray>> ToBlock(
+    HeteroGraphPtr graph, const std::vector<IdArray>& rhs_nodes,
     bool include_rhs_in_lhs, std::vector<IdArray>* lhs_nodes);
 
 }  // namespace transform