Add categorical data support to GPU Hist. (#6164)

src/common/device_helpers.cuh

@@ -536,6 +536,21 @@ void CopyDeviceSpanToVector(std::vector<T> *dst, xgboost::common::Span<const T>
                                 cudaMemcpyDeviceToHost));
 }
 
+template <class HContainer, class DContainer>
+void CopyToD(HContainer const &h, DContainer *d) {
+  if (h.empty()) {
+    d->clear();
+    return;
+  }
+  d->resize(h.size());
+  using HVT = std::remove_cv_t<typename HContainer::value_type>;
+  using DVT = std::remove_cv_t<typename DContainer::value_type>;
+  static_assert(std::is_same<HVT, DVT>::value,
+                "Host and device containers must have same value type.");
+  dh::safe_cuda(cudaMemcpyAsync(d->data().get(), h.data(), h.size() * sizeof(HVT),
+                                cudaMemcpyHostToDevice));
+}
+
 // Keep track of pinned memory allocation
 struct PinnedMemory {
   void *temp_storage{nullptr};

src/common/hist_util.cu

@@ -178,7 +178,7 @@ void ProcessBatch(int device, MetaInfo const &info, const SparsePage &page,
   dh::XGBCachingDeviceAllocator<char> alloc;
   const auto& host_data = page.data.ConstHostVector();
   dh::device_vector<Entry> sorted_entries(host_data.begin() + begin,
-                                                  host_data.begin() + end);
+                                          host_data.begin() + end);
   thrust::sort(thrust::cuda::par(alloc), sorted_entries.begin(),
                sorted_entries.end(), detail::EntryCompareOp());