Remove omp_get_max_threads in tree updaters. (#7590)

src/data/data.cc

@@ -1035,6 +1035,16 @@ SparsePage SparsePage::GetTranspose(int num_columns, int32_t n_threads) const {
   return transpose;
 }
 
+void SparsePage::SortRows(int32_t n_threads) {
+  auto& h_offset = this->offset.HostVector();
+  auto& h_data = this->data.HostVector();
+  common::ParallelFor(this->Size(), n_threads, [&](auto i) {
+    if (h_offset[i] < h_offset[i + 1]) {
+      std::sort(h_data.begin() + h_offset[i], h_data.begin() + h_offset[i + 1], Entry::CmpValue);
+    }
+  });
+}
+
 void SparsePage::Push(const SparsePage &batch) {
   auto& data_vec = data.HostVector();
   auto& offset_vec = offset.HostVector();

src/data/simple_dmatrix.cc

@@ -67,7 +67,7 @@ BatchSet<SortedCSCPage> SimpleDMatrix::GetSortedColumnBatches() {
   if (!sorted_column_page_) {
     sorted_column_page_.reset(
         new SortedCSCPage(sparse_page_->GetTranspose(info_.num_col_, ctx_.Threads())));
-    sorted_column_page_->SortRows();
+    sorted_column_page_->SortRows(ctx_.Threads());
   }
   auto begin_iter = BatchIterator<SortedCSCPage>(
       new SimpleBatchIteratorImpl<SortedCSCPage>(sorted_column_page_));

src/data/sparse_page_source.h

@@ -339,7 +339,7 @@ class SortedCSCPageSource : public PageSourceIncMixIn<SortedCSCPage> {
       this->page_->PushCSC(csr->GetTranspose(n_features_, nthreads_));
       CHECK_EQ(this->page_->Size(), n_features_);
       CHECK_EQ(this->page_->data.Size(), csr->data.Size());
-      this->page_->SortRows();
+      this->page_->SortRows(this->nthreads_);
       page_->SetBaseRowId(csr->base_rowid);
       this->WriteCache();
     }