CI fix

src/common/hist_util.cc

@@ -659,6 +659,13 @@ void GHistIndexBlockMatrix::Init(const GHistIndexMatrix& gmat,
   }
 }
 
+/*!
+ * \brief fill a histogram by zeroes
+ */
+void InitilizeHistByZeroes(GHistRow hist) {
+  memset(hist.data(), '\0', hist.size()*sizeof(tree::GradStats));
+}
+
 void GHistBuilder::BuildHist(const std::vector<GradientPair>& gpair,
                              const RowSetCollection::Elem row_indices,
                              const GHistIndexMatrix& gmat,

src/common/hist_util.h

@@ -347,6 +347,11 @@ class GHistIndexBlockMatrix {
  */
 using GHistRow = Span<tree::GradStats>;
 
+/*!
+ * \brief fill a histogram by zeroes
+ */
+void InitilizeHistByZeroes(GHistRow hist);
+
 /*!
  * \brief histogram of gradient statistics for multiple nodes
  */
@@ -369,9 +374,12 @@ class HistCollection {
 
   // initialize histogram collection
   void Init(uint32_t nbins) {
-    nbins_ = nbins;
+    if (nbins_ != nbins) {
+      nbins_ = nbins;
+      // quite expensive operation, so let's do this only once
+      data_.clear();
+    }
     row_ptr_.clear();
-    data_.clear();
   }
 
   // create an empty histogram for i-th node
@@ -382,13 +390,16 @@ class HistCollection {
     }
     CHECK_EQ(row_ptr_[nid], kMax);
 
-    row_ptr_[nid] = data_.size();
-    data_.resize(data_.size() + nbins_);
+    row_ptr_[nid] = nbins_ * nid;
+
+    if (data_.size() <= nbins_ * (nid + 1)) {
+      data_.resize(nbins_ * (nid + 1));
+    }
   }
 
  private:
   /*! \brief number of all bins over all features */
-  uint32_t nbins_;
+  uint32_t nbins_ = 0;
 
   std::vector<tree::GradStats> data_;
 
@@ -433,7 +444,7 @@ class HistBuffer {
     GHistRow hist = hist_[tid * nodes_ + nid];
 
     if (!hist_was_used_[tid * nodes_ + nid]) {
-      InitilizeHist(hist);
+      InitilizeHistByZeroes(hist);
       hist_was_used_[tid * nodes_ + nid] = true;
     }
 
@@ -457,13 +468,6 @@ class HistBuffer {
   }
 
  protected:
-  // fill hist by zeroes
-  void InitilizeHist(GHistRow hist) {
-    for (auto& val : hist) {
-      val = tree::GradStats(0.0f, 0.0f);
-    }
-  }
-
   /*! \brief number of bins in each histogram */
   size_t nbins_ = 0;
   /*! \brief number of threads for parallel computation */

src/tree/updater_quantile_hist.cc

@@ -100,19 +100,22 @@ void QuantileHistMaker::Builder::SyncHistograms(
   builder_monitor_.Start("SyncHistograms");
 
   const bool isDistributed = rabit::IsDistributed();
+  builder_monitor_.Start("AddHistRow");
   for (auto const& node : nodes_for_subtraction_trick_) {
     hist_.AddHistRow(node.nid);
   }
+  builder_monitor_.Stop("AddHistRow");
 
   const size_t nbins = gmat.cut.Ptrs().back();
   common::BlockedSpace2d space(nodes_for_explicit_hist_build_.size(), [&](size_t node) {
     return nbins;
-  }, 512);
+  }, 1024);
 
   common::ParallelFor2d(space, [&](size_t node, common::Range1d r) {
     const auto entry = nodes_for_explicit_hist_build_[node];
     auto this_hist = hist_[entry.nid];
     // Merging histograms from each thread into once
+    common::InitilizeHistByZeroes({this_hist.data() + r.begin(), this_hist.data() + r.end()});
     hist_buffer_.ReduceHist(this_hist, node, r.begin(), r.end());
 
     if (!(*p_tree)[entry.nid].IsRoot() && entry.sibling_nid > -1 && !isDistributed) {
@@ -156,9 +159,8 @@ void QuantileHistMaker::Builder::BuildHistogramsLossGuide(
   if (entry.sibling_nid > -1) {
     nodes_for_subtraction_trick_.emplace_back(entry.sibling_nid, entry.nid,
         p_tree->GetDepth(entry.sibling_nid), 0.0f, 0);
-
-    SyncHistograms(starting_index, sync_count, gmat, p_tree);
   }
+  SyncHistograms(starting_index, sync_count, gmat, p_tree);
 }
 
 void QuantileHistMaker::Builder::BuildLocalHistograms(
@@ -172,6 +174,7 @@ void QuantileHistMaker::Builder::BuildLocalHistograms(
 
   hist_buffer_.Reset(this->nthread_, nodes_for_explicit_hist_build_.size());
 
+  builder_monitor_.Start("AddHistRow");
   size_t node_count = 0;
   for (auto const& entry : nodes_for_explicit_hist_build_) {
     int nid = entry.nid;
@@ -180,6 +183,7 @@ void QuantileHistMaker::Builder::BuildLocalHistograms(
     node_count++;
   }
   (*sync_count) = node_count;
+  builder_monitor_.Stop("AddHistRow");
 
   // create space of size (# rows in each node)
   common::BlockedSpace2d space(nodes_for_explicit_hist_build_.size(), [&](size_t node) {