[sycl] add split applications and tests (#10636)

Co-authored-by: Dmitry Razdoburdin <>

plugin/sycl/tree/hist_updater.cc

@@ -9,6 +9,8 @@
 
 #include <functional>
 
+#include "../../src/tree/common_row_partitioner.h"
+
 #include "../common/hist_util.h"
 #include "../../src/collective/allreduce.h"
 
@@ -250,6 +252,55 @@ void HistUpdater<GradientSumT>::InitData(
   builder_monitor_.Stop("InitData");
 }
 
+template <typename GradientSumT>
+void HistUpdater<GradientSumT>::AddSplitsToRowSet(
+                                                const std::vector<ExpandEntry>& nodes,
+                                                RegTree* p_tree) {
+  const size_t n_nodes = nodes.size();
+  for (size_t i = 0; i < n_nodes; ++i) {
+    const int32_t nid = nodes[i].nid;
+    const size_t n_left = partition_builder_.GetNLeftElems(i);
+    const size_t n_right = partition_builder_.GetNRightElems(i);
+
+    row_set_collection_.AddSplit(nid, (*p_tree)[nid].LeftChild(),
+        (*p_tree)[nid].RightChild(), n_left, n_right);
+  }
+}
+
+template <typename GradientSumT>
+void HistUpdater<GradientSumT>::ApplySplit(
+                      const std::vector<ExpandEntry> nodes,
+                      const common::GHistIndexMatrix& gmat,
+                      RegTree* p_tree) {
+  using CommonRowPartitioner = xgboost::tree::CommonRowPartitioner;
+  builder_monitor_.Start("ApplySplit");
+
+  const size_t n_nodes = nodes.size();
+  std::vector<int32_t> split_conditions(n_nodes);
+  CommonRowPartitioner::FindSplitConditions(nodes, *p_tree, gmat, &split_conditions);
+
+  partition_builder_.Init(&qu_, n_nodes, [&](size_t node_in_set) {
+    const int32_t nid = nodes[node_in_set].nid;
+    return row_set_collection_[nid].Size();
+  });
+
+  ::sycl::event event;
+  partition_builder_.Partition(gmat, nodes, row_set_collection_,
+                               split_conditions, p_tree, &event);
+  qu_.wait_and_throw();
+
+  for (size_t node_in_set = 0; node_in_set < n_nodes; node_in_set++) {
+    const int32_t nid = nodes[node_in_set].nid;
+    size_t* data_result = const_cast<size_t*>(row_set_collection_[nid].begin);
+    partition_builder_.MergeToArray(node_in_set, data_result, &event);
+  }
+  qu_.wait_and_throw();
+
+  AddSplitsToRowSet(nodes, p_tree);
+
+  builder_monitor_.Stop("ApplySplit");
+}
+
 template <typename GradientSumT>
 void HistUpdater<GradientSumT>::InitNewNode(int nid,
                                             const common::GHistIndexMatrix& gmat,

plugin/sycl/tree/hist_updater.h

@@ -101,6 +101,12 @@ class HistUpdater {
       typename TreeEvaluator<GradientSumT>::SplitEvaluator const &evaluator,
       float min_child_weight);
 
+  void ApplySplit(std::vector<ExpandEntry> nodes,
+                      const common::GHistIndexMatrix& gmat,
+                      RegTree* p_tree);
+
+  void AddSplitsToRowSet(const std::vector<ExpandEntry>& nodes, RegTree* p_tree);
+
   void InitData(const common::GHistIndexMatrix& gmat,
                 const USMVector<GradientPair, MemoryType::on_device> &gpair,
                 const DMatrix& fmat,
@@ -179,6 +185,8 @@ class HistUpdater {
 
   uint64_t seed_ = 0;
 
+  common::PartitionBuilder partition_builder_;
+
   // key is the node id which should be calculated by Subtraction Trick, value is the node which
   // provides the evidence for substracts
   std::vector<ExpandEntry> nodes_for_subtraction_trick_;

src/tree/common_row_partitioner.h

@@ -144,9 +144,10 @@ class CommonRowPartitioner {
     }
   }
 
-  template <typename ExpandEntry>
-  void FindSplitConditions(const std::vector<ExpandEntry>& nodes, const RegTree& tree,
-                           const GHistIndexMatrix& gmat, std::vector<bst_bin_t>* split_conditions) {
+  /* Making GHistIndexMatrix_t a templete parameter allows reuse this function for sycl-plugin */
+  template <typename ExpandEntry, typename GHistIndexMatrix_t>
+  static void FindSplitConditions(const std::vector<ExpandEntry>& nodes, const RegTree& tree,
+                        const GHistIndexMatrix_t& gmat, std::vector<int32_t>* split_conditions) {
     auto const& ptrs = gmat.cut.Ptrs();
     auto const& vals = gmat.cut.Values();
 

tests/cpp/plugin/test_sycl_hist_updater.cc

@@ -8,6 +8,8 @@
 #include "../../../plugin/sycl/tree/hist_updater.h"
 #include "../../../plugin/sycl/device_manager.h"
 
+#include "../../../src/tree/common_row_partitioner.h"
+
 #include "../helpers.h"
 
 namespace xgboost::sycl::tree {
@@ -61,6 +63,12 @@ class TestHistUpdater : public HistUpdater<GradientSumT> {
     HistUpdater<GradientSumT>::EvaluateSplits(nodes_set, gmat, tree);
     return HistUpdater<GradientSumT>::snode_host_;
   }
+
+  void TestApplySplit(const std::vector<ExpandEntry> nodes,
+                      const common::GHistIndexMatrix& gmat,
+                      RegTree* p_tree) {
+    HistUpdater<GradientSumT>::ApplySplit(nodes, gmat, p_tree);
+  }
 };
 
 void GenerateRandomGPairs(::sycl::queue* qu, GradientPair* gpair_ptr, size_t num_rows, bool has_neg_hess) {
@@ -131,7 +139,6 @@ void TestHistUpdaterSampling(const xgboost::tree::TrainParam& param) {
 
     ASSERT_NE(num_diffs, 0);
   }
-
 }
 
 template <typename GradientSumT>
@@ -392,6 +399,95 @@ void TestHistUpdaterEvaluateSplits(const xgboost::tree::TrainParam& param) {
   ASSERT_NEAR(best_loss_chg_des[0], best_loss_chg, 1e-6);
 }
 
+template <typename GradientSumT>
+void TestHistUpdaterApplySplit(const xgboost::tree::TrainParam& param, float sparsity, int max_bins) {
+  const size_t num_rows = 1024;
+  const size_t num_columns = 2;
+
+  Context ctx;
+  ctx.UpdateAllowUnknown(Args{{"device", "sycl"}});
+
+  DeviceManager device_manager;
+  auto qu = device_manager.GetQueue(ctx.Device());
+
+  ObjInfo task{ObjInfo::kRegression}; 
+
+  auto p_fmat = RandomDataGenerator{num_rows, num_columns, sparsity}.GenerateDMatrix();
+  sycl::DeviceMatrix dmat;
+  dmat.Init(qu, p_fmat.get());
+
+  common::GHistIndexMatrix gmat;
+  gmat.Init(qu, &ctx, dmat, max_bins);
+
+  RegTree tree;
+  tree.ExpandNode(0, 0, 0, false, 0, 0, 0, 0, 0, 0, 0);
+
+  std::vector<tree::ExpandEntry> nodes;
+  nodes.emplace_back(tree::ExpandEntry(0, tree.GetDepth(0)));
+
+  FeatureInteractionConstraintHost int_constraints;
+  std::unique_ptr<TreeUpdater> pruner{TreeUpdater::Create("prune", &ctx, &task)};
+  TestHistUpdater<GradientSumT> updater(&ctx, qu, param, std::move(pruner), int_constraints, p_fmat.get());
+  USMVector<GradientPair, MemoryType::on_device> gpair(&qu, num_rows);
+  GenerateRandomGPairs(&qu, gpair.Data(), num_rows, false);
+
+  auto* row_set_collection = updater.TestInitData(gmat, gpair, *p_fmat, tree);
+  updater.TestApplySplit(nodes, gmat, &tree);
+
+  // Copy indexes to host
+  std::vector<size_t> row_indices_host(num_rows);
+  qu.memcpy(row_indices_host.data(), row_set_collection->Data().Data(), sizeof(size_t)*num_rows).wait();
+
+  // Reference Implementation
+  std::vector<size_t> row_indices_desired_host(num_rows);
+  size_t n_left, n_right;
+  {
+    std::unique_ptr<TreeUpdater> pruner4verification{TreeUpdater::Create("prune", &ctx, &task)};
+    TestHistUpdater<GradientSumT> updater4verification(&ctx, qu, param, std::move(pruner4verification), int_constraints, p_fmat.get());
+    auto* row_set_collection4verification = updater4verification.TestInitData(gmat, gpair, *p_fmat, tree);
+
+    size_t n_nodes = nodes.size();
+    std::vector<int32_t> split_conditions(n_nodes);
+    xgboost::tree::CommonRowPartitioner::FindSplitConditions(nodes, tree, gmat, &split_conditions);
+
+    common::PartitionBuilder partition_builder;
+    partition_builder.Init(&qu, n_nodes, [&](size_t node_in_set) {
+      const int32_t nid = nodes[node_in_set].nid;
+      return (*row_set_collection4verification)[nid].Size();
+    });
+
+    ::sycl::event event;
+    partition_builder.Partition(gmat, nodes, (*row_set_collection4verification),
+                                split_conditions, &tree, &event);
+    qu.wait_and_throw();
+
+    for (size_t node_in_set = 0; node_in_set < n_nodes; node_in_set++) {
+      const int32_t nid = nodes[node_in_set].nid;
+      size_t* data_result = const_cast<size_t*>((*row_set_collection4verification)[nid].begin);
+      partition_builder.MergeToArray(node_in_set, data_result, &event);
+    }
+    qu.wait_and_throw();
+
+    const int32_t nid = nodes[0].nid;
+    n_left = partition_builder.GetNLeftElems(0);
+    n_right = partition_builder.GetNRightElems(0);
+
+    row_set_collection4verification->AddSplit(nid, tree[nid].LeftChild(),
+        tree[nid].RightChild(), n_left, n_right);
+
+    qu.memcpy(row_indices_desired_host.data(), row_set_collection4verification->Data().Data(), sizeof(size_t)*num_rows).wait();
+  }
+
+  std::sort(row_indices_desired_host.begin(), row_indices_desired_host.begin() + n_left);
+  std::sort(row_indices_host.begin(), row_indices_host.begin() + n_left);
+  std::sort(row_indices_desired_host.begin() + n_left, row_indices_desired_host.end());
+  std::sort(row_indices_host.begin() + n_left, row_indices_host.end());
+
+  for (size_t row = 0; row < num_rows; ++row) {
+    ASSERT_EQ(row_indices_desired_host[row], row_indices_host[row]);
+  }
+}
+
 TEST(SyclHistUpdater, Sampling) {
   xgboost::tree::TrainParam param;
   param.UpdateAllowUnknown(Args{{"subsample", "0.7"}});
@@ -439,4 +535,24 @@ TEST(SyclHistUpdater, EvaluateSplits) {
   TestHistUpdaterEvaluateSplits<double>(param);
 }
 
+TEST(SyclHistUpdater, ApplySplitSparce) {
+  xgboost::tree::TrainParam param;
+  param.UpdateAllowUnknown(Args{{"max_depth", "3"}});
+
+  TestHistUpdaterApplySplit<float>(param, 0.3, 256);
+  TestHistUpdaterApplySplit<double>(param, 0.3, 256);
+}
+
+TEST(SyclHistUpdater, ApplySplitDence) {
+  xgboost::tree::TrainParam param;
+  param.UpdateAllowUnknown(Args{{"max_depth", "3"}});
+
+  TestHistUpdaterApplySplit<float>(param, 0.0, 256);
+  TestHistUpdaterApplySplit<float>(param, 0.0, 256+1);
+  TestHistUpdaterApplySplit<float>(param, 0.0, (1u << 16) + 1);
+  TestHistUpdaterApplySplit<double>(param, 0.0, 256);
+  TestHistUpdaterApplySplit<double>(param, 0.0, 256+1);
+  TestHistUpdaterApplySplit<double>(param, 0.0, (1u << 16) + 1);
+}
+
 }  // namespace xgboost::sycl::tree