Replaced std::vector with HostDeviceVector in MetaInfo and SparsePage. (#3446)

* Replaced std::vector with HostDeviceVector in MetaInfo and SparsePage.

- added distributions to HostDeviceVector
- using HostDeviceVector for labels, weights and base margings in MetaInfo
- using HostDeviceVector for offset and data in SparsePage
- other necessary refactoring

* Added const version of HostDeviceVector API calls.

- const versions added to calls that can trigger data transfers, e.g. DevicePointer()
- updated the code that uses HostDeviceVector
- objective functions now accept const HostDeviceVector<bst_float>& for predictions

* Updated src/linear/updater_gpu_coordinate.cu.

* Added read-only state for HostDeviceVector sync.

- this means no copies are performed if both host and devices access
  the HostDeviceVector read-only

* Fixed linter and test errors.

- updated the lz4 plugin
- added ConstDeviceSpan to HostDeviceVector
- using device % dh::NVisibleDevices() for the physical device number,
  e.g. in calls to cudaSetDevice()

* Fixed explicit template instantiation errors for HostDeviceVector.

- replaced HostDeviceVector<unsigned int> with HostDeviceVector<int>

* Fixed HostDeviceVector tests that require multiple GPUs.

- added a mock set device handler; when set, it is called instead of cudaSetDevice()

src/gbm/gblinear.cc

@@ -143,7 +143,7 @@ class GBLinear : public GradientBooster {
     model_.LazyInitModel();
     CHECK_EQ(ntree_limit, 0U)
         << "GBLinear::PredictContribution: ntrees is only valid for gbtree predictor";
-    const std::vector<bst_float>& base_margin = p_fmat->Info().base_margin_;
+    const auto& base_margin = p_fmat->Info().base_margin_.ConstHostVector();
     const int ngroup = model_.param.num_output_group;
     const size_t ncolumns = model_.param.num_feature + 1;
     // allocate space for (#features + bias) times #groups times #rows
@@ -201,7 +201,7 @@ class GBLinear : public GradientBooster {
     monitor_.Start("PredictBatchInternal");
       model_.LazyInitModel();
     std::vector<bst_float> &preds = *out_preds;
-    const std::vector<bst_float>& base_margin = p_fmat->Info().base_margin_;
+    const auto& base_margin = p_fmat->Info().base_margin_.ConstHostVector();
     // start collecting the prediction
      auto iter = p_fmat->RowIterator();
     const int ngroup = model_.param.num_output_group;

src/gbm/gbtree.cc

@@ -195,8 +195,8 @@ class GBTree : public GradientBooster {
           << "must have exactly ngroup*nrow gpairs";
       // TODO(canonizer): perform this on GPU if HostDeviceVector has device set.
       HostDeviceVector<GradientPair> tmp(in_gpair->Size() / ngroup,
-                                      GradientPair(), in_gpair->Devices());
-      std::vector<GradientPair>& gpair_h = in_gpair->HostVector();
+                                      GradientPair(), in_gpair->Distribution());
+      const auto& gpair_h = in_gpair->ConstHostVector();
       auto nsize = static_cast<bst_omp_uint>(tmp.Size());
       for (int gid = 0; gid < ngroup; ++gid) {
         std::vector<GradientPair>& tmp_h = tmp.HostVector();
@@ -402,7 +402,8 @@ class Dart : public GBTree {
 
     if (init_out_preds) {
       size_t n = num_group * p_fmat->Info().num_row_;
-      const std::vector<bst_float>& base_margin = p_fmat->Info().base_margin_;
+      const auto& base_margin =
+        p_fmat->Info().base_margin_.ConstHostVector();
       out_preds->resize(n);
       if (base_margin.size() != 0) {
         CHECK_EQ(out_preds->size(), n);