Fix cpplint. (#4157)

* Add comment after #endif.
* Add missing headers.

src/common/common.cc

@@ -27,6 +27,6 @@ GlobalRandomEngine& GlobalRandom() {
 int AllVisibleImpl::AllVisible() {
   return 0;
 }
-#endif
+#endif  // !defined(XGBOOST_USE_CUDA)
 
 }  // namespace xgboost

src/common/common.h

@@ -26,7 +26,7 @@
 
 #define WITH_CUDA() false
 
-#endif
+#endif  // defined(__CUDACC__)
 
 namespace dh {
 #if defined(__CUDACC__)
@@ -44,7 +44,7 @@ inline cudaError_t ThrowOnCudaError(cudaError_t code, const char *file,
   }
   return code;
 }
-#endif
+#endif  // defined(__CUDACC__)
 }  // namespace dh
 
 namespace xgboost {

src/common/compressed_iterator.h

@@ -10,7 +10,7 @@
 
 #ifdef __CUDACC__
 #include "device_helpers.cuh"
-#endif
+#endif  // __CUDACC__
 
 namespace xgboost {
 namespace common {
@@ -115,7 +115,7 @@ class CompressedBufferWriter {
       symbol >>= 8;
     }
   }
-#endif
+#endif  // __CUDACC__
 
   template <typename IterT>
   void Write(CompressedByteT *buffer, IterT input_begin, IterT input_end) {

src/common/hist_util.cc

@@ -19,7 +19,7 @@
   #define PREFETCH_READ_T0(addr) __builtin_prefetch(reinterpret_cast<const char*>(addr), 0, 3)
 #else  // no SW pre-fetching available; PREFETCH_READ_T0 is no-op
   #define PREFETCH_READ_T0(addr) do {} while (0)
-#endif
+#endif  // defined(XGBOOST_MM_PREFETCH_PRESENT)
 
 namespace xgboost {
 namespace common {
@@ -544,7 +544,7 @@ void GHistBuilder::BuildBlockHist(const std::vector<GradientPair>& gpair,
 
 #if defined(_OPENMP)
   const auto nthread = static_cast<bst_omp_uint>(this->nthread_);
-#endif
+#endif  // defined(_OPENMP)
   tree::GradStats* p_hist = hist.data();
 
 #pragma omp parallel for num_threads(nthread) schedule(guided)
@@ -590,7 +590,7 @@ void GHistBuilder::SubtractionTrick(GHistRow self, GHistRow sibling, GHistRow pa
 
 #if defined(_OPENMP)
   const auto nthread = static_cast<bst_omp_uint>(this->nthread_);
-#endif
+#endif  // defined(_OPENMP)
   tree::GradStats* p_self = self.data();
   tree::GradStats* p_sibling = sibling.data();
   tree::GradStats* p_parent = parent.data();

src/common/host_device_vector.cc

@@ -159,4 +159,4 @@ template class HostDeviceVector<size_t>;
 
 }  // namespace xgboost
 
-#endif
+#endif  // XGBOOST_USE_CUDA

src/common/host_device_vector.h

@@ -57,6 +57,7 @@
 #include <algorithm>
 #include <cstdlib>
 #include <initializer_list>
+#include <utility>
 #include <vector>
 
 #include "common.h"
@@ -66,7 +67,7 @@
 // is included from a .cu file
 #ifdef __CUDACC__
 #include <thrust/device_ptr.h>
-#endif
+#endif  // __CUDACC__
 
 namespace xgboost {
 
@@ -74,7 +75,7 @@ namespace xgboost {
 // Sets a function to call instead of cudaSetDevice();
 // only added for testing
 void SetCudaSetDeviceHandler(void (*handler)(int));
-#endif
+#endif  // __CUDACC__
 
 template <typename T> struct HostDeviceVectorImpl;
 
@@ -233,7 +234,7 @@ class HostDeviceVector {
 
   void ScatterFrom(thrust::device_ptr<const T> begin, thrust::device_ptr<const T> end);
   void GatherTo(thrust::device_ptr<T> begin, thrust::device_ptr<T> end) const;
-#endif
+#endif  // __CUDACC__
 
   void Fill(T v);
   void Copy(const HostDeviceVector<T>& other);

src/common/math.h

@@ -123,7 +123,7 @@ inline bool CheckNAN(T v) {
   return (_isnan(v) != 0);
 #else
   return std::isnan(v);
-#endif
+#endif  // _MSC_VER
 }
 #endif  // XGBOOST_STRICT_R_MODE_
 
@@ -149,7 +149,7 @@ XGBOOST_DEVICE inline T LogGamma(T v) {
 
 #else
   return lgamma(v);
-#endif
+#endif  // _MSC_VER
 }
 
 #endif  // XGBOOST_STRICT_R_MODE && !defined(XGBOOST_USE_CUDA)

src/common/random.h

@@ -66,7 +66,7 @@ typedef CustomGlobalRandomEngine GlobalRandomEngine;
  * \brief global random engine
  */
 using GlobalRandomEngine = RandomEngine;
-#endif
+#endif  // XGBOOST_CUSTOMIZE_GLOBAL_PRNG
 
 /*!
  * \brief global singleton of a random engine.

src/common/span.h

@@ -62,7 +62,7 @@
 
 #define __span_noexcept noexcept
 
-#endif
+#endif  // defined(_MSC_VER) && _MSC_VER < 1910
 
 namespace xgboost {
 namespace common {
@@ -85,7 +85,7 @@ namespace common {
 #define SPAN_CHECK KERNEL_CHECK
 #else
 #define SPAN_CHECK CHECK  // check from dmlc
-#endif
+#endif  // __CUDA_ARCH__
 
 namespace detail {
 /*!
@@ -101,7 +101,7 @@ using ptrdiff_t = int64_t;  // NOLINT
 constexpr const detail::ptrdiff_t dynamic_extent = -1;  // NOLINT
 #else
 constexpr detail::ptrdiff_t dynamic_extent = -1;  // NOLINT
-#endif
+#endif  // defined(_MSC_VER) && _MSC_VER < 1910
 
 enum class byte : unsigned char {};  // NOLINT
 

src/common/timer.h

@@ -81,7 +81,7 @@ struct Monitor {
         cudaSetDevice(device);
         cudaDeviceSynchronize();
       }
-#endif
+#endif  // __CUDACC__
     }
     statistics_map[name].timer.Start();
   }
@@ -96,7 +96,7 @@ struct Monitor {
         cudaSetDevice(device);
         cudaDeviceSynchronize();
       }
-#endif
+#endif  // __CUDACC__
     }
     this->Stop(name);
   }

src/common/transform.h

@@ -6,6 +6,7 @@
 
 #include <dmlc/omp.h>
 #include <xgboost/data.h>
+#include <utility>
 #include <vector>
 #include <type_traits>  // enable_if
 
@@ -15,7 +16,7 @@
 
 #if defined (__CUDACC__)
 #include "device_helpers.cuh"
-#endif
+#endif  // defined (__CUDACC__)
 
 namespace xgboost {
 namespace common {
@@ -32,7 +33,7 @@ __global__ void LaunchCUDAKernel(Functor _func, Range _range,
     _func(i, _spans...);
   }
 }
-#endif
+#endif  // defined(__CUDACC__)
 
 }  // namespace detail
 
@@ -155,7 +156,7 @@ class Transform {
     void LaunchCUDA(Functor _func, HDV*... _vectors) const {
       LOG(FATAL) << "Not part of device code. WITH_CUDA: " << WITH_CUDA();
     }
-#endif
+#endif  // defined(__CUDACC__)
 
     template <typename... HDV>
     void LaunchCPU(Functor func, HDV*... vectors) const {

src/data/data.cc

@@ -15,7 +15,7 @@
 #if DMLC_ENABLE_STD_THREAD
 #include "./sparse_page_source.h"
 #include "./sparse_page_dmatrix.h"
-#endif
+#endif  // DMLC_ENABLE_STD_THREAD
 
 namespace dmlc {
 DMLC_REGISTRY_ENABLE(::xgboost::data::SparsePageFormatReg);
@@ -264,7 +264,7 @@ DMatrix* DMatrix::Create(dmlc::Parser<uint32_t>* parser,
 #else
     LOG(FATAL) << "External memory is not enabled in mingw";
     return nullptr;
-#endif
+#endif  // DMLC_ENABLE_STD_THREAD
   }
 }
 
@@ -285,7 +285,7 @@ DMatrix* DMatrix::Create(std::unique_ptr<DataSource>&& source,
 #else
     LOG(FATAL) << "External memory is not enabled in mingw";
     return nullptr;
-#endif
+#endif  // DMLC_ENABLE_STD_THREAD
   }
 }
 }  // namespace xgboost

src/data/simple_dmatrix.h

@@ -9,9 +9,13 @@
 
 #include <xgboost/base.h>
 #include <xgboost/data.h>
+
 #include <algorithm>
 #include <cstring>
+#include <memory>
+#include <utility>
 #include <vector>
+
 #include "simple_csr_source.h"
 
 namespace xgboost {

src/data/sparse_page_dmatrix.cc

@@ -100,4 +100,4 @@ bool SparsePageDMatrix::SingleColBlock() const {
 }
 }  // namespace data
 }  // namespace xgboost
-#endif
+#endif  // DMLC_ENABLE_STD_THREAD

src/data/sparse_page_dmatrix.h

@@ -9,9 +9,11 @@
 
 #include <xgboost/data.h>
 #include <algorithm>
+#include <memory>
 #include <string>
 #include <utility>
 #include <vector>
+
 #include "sparse_page_source.h"
 
 namespace xgboost {

src/data/sparse_page_source.cc

@@ -29,7 +29,7 @@ GetCacheShards(const std::string& cache_info) {
     cache_shards[0] = cache_info.substr(0, 2) + cache_shards[0];
     return cache_shards;
   }
-#endif
+#endif  // (defined _WIN32) || (defined __CYGWIN__)
   return xgboost::common::Split(cache_info, ':');
 }
 
@@ -291,4 +291,4 @@ void SparsePageSource::CreateColumnPage(DMatrix* src,
 }
 }  // namespace data
 }  // namespace xgboost
-#endif
+#endif  // DMLC_ENABLE_STD_THREAD

src/data/sparse_page_source.h

@@ -10,9 +10,12 @@
 #include <xgboost/base.h>
 #include <xgboost/data.h>
 #include <dmlc/threadediter.h>
-#include <vector>
+
 #include <algorithm>
+#include <memory>
 #include <string>
+#include <vector>
+
 #include "sparse_page_writer.h"
 
 namespace xgboost {

src/data/sparse_page_writer.h

@@ -19,7 +19,7 @@
 #if DMLC_ENABLE_STD_THREAD
 #include <dmlc/concurrency.h>
 #include <thread>
-#endif
+#endif  // DMLC_ENABLE_STD_THREAD
 
 namespace xgboost {
 namespace data {

src/gbm/gbtree_model.h

@@ -5,6 +5,8 @@
 #include <dmlc/parameter.h>
 #include <dmlc/io.h>
 #include <xgboost/tree_model.h>
+
+#include <memory>
 #include <utility>
 #include <string>
 #include <vector>

src/learner.cc

@@ -178,7 +178,7 @@ class LearnerImpl : public Learner {
   static void AssertGPUSupport() {
 #ifndef XGBOOST_USE_CUDA
     LOG(FATAL) << "XGBoost version not compiled with GPU support.";
-#endif
+#endif  // XGBOOST_USE_CUDA
   }
 
 
@@ -387,7 +387,7 @@ class LearnerImpl : public Learner {
             cfg_["predictor"] = "cpu_predictor";
             kv.second = "cpu_predictor";
           }
-#endif
+#endif  // XGBOOST_USE_CUDA
         }
       }
       attributes_ =

src/linear/linear_updater.cc

@@ -30,6 +30,6 @@ DMLC_REGISTRY_LINK_TAG(updater_shotgun);
 DMLC_REGISTRY_LINK_TAG(updater_coordinate);
 #ifdef XGBOOST_USE_CUDA
 DMLC_REGISTRY_LINK_TAG(updater_gpu_coordinate);
-#endif
+#endif  // XGBOOST_USE_CUDA
 }  // namespace linear
 }  // namespace xgboost

src/logging.cc

@@ -35,7 +35,7 @@ TrackerLogger::~TrackerLogger() {
 
 }  // namespace xgboost
 
-#endif
+#endif  // !defined(XGBOOST_STRICT_R_MODE) || XGBOOST_STRICT_R_MODE == 0
 
 namespace xgboost {
 

src/metric/elementwise_metric.cc

@@ -5,4 +5,4 @@
 
 #if !defined(XGBOOST_USE_CUDA)
 #include "elementwise_metric.cu"
-#endif
+#endif  // !defined(XGBOOST_USE_CUDA)

src/metric/elementwise_metric.cu

@@ -144,7 +144,7 @@ class MetricsReduction {
         result.weights_sum_ += res_per_device[i].weights_sum_;
       }
     }
-#endif
+#endif  // defined(XGBOOST_USE_CUDA)
     return result;
   }
 
@@ -152,7 +152,7 @@ class MetricsReduction {
   EvalRow policy_;
 #if defined(XGBOOST_USE_CUDA)
   std::vector<dh::CubMemory> allocators_;
-#endif
+#endif  // defined(XGBOOST_USE_CUDA)
 };
 
 struct EvalRowRMSE {

src/objective/hinge.cc

@@ -15,4 +15,4 @@ DMLC_REGISTRY_FILE_TAG(hinge_obj);
 
 #ifndef XGBOOST_USE_CUDA
 #include "hinge.cu"
-#endif
+#endif  // XGBOOST_USE_CUDA

src/objective/hinge.cu

@@ -16,7 +16,7 @@ namespace obj {
 
 #if defined(XGBOOST_USE_CUDA)
 DMLC_REGISTRY_FILE_TAG(hinge_obj_gpu);
-#endif
+#endif  // defined(XGBOOST_USE_CUDA)
 
 struct HingeObjParam : public dmlc::Parameter<HingeObjParam> {
   int n_gpus;

src/objective/multiclass_obj.cc

@@ -15,4 +15,4 @@ DMLC_REGISTRY_FILE_TAG(multiclass_obj);
 
 #ifndef XGBOOST_USE_CUDA
 #include "multiclass_obj.cu"
-#endif
+#endif  // XGBOOST_USE_CUDA

src/objective/multiclass_obj.cu

@@ -21,7 +21,7 @@ namespace obj {
 
 #if defined(XGBOOST_USE_CUDA)
 DMLC_REGISTRY_FILE_TAG(multiclass_obj_gpu);
-#endif
+#endif  // defined(XGBOOST_USE_CUDA)
 
 struct SoftmaxMultiClassParam : public dmlc::Parameter<SoftmaxMultiClassParam> {
   int num_class;

src/objective/objective.cc

@@ -38,7 +38,7 @@ DMLC_REGISTRY_LINK_TAG(multiclass_obj_gpu);
 DMLC_REGISTRY_LINK_TAG(regression_obj);
 DMLC_REGISTRY_LINK_TAG(hinge_obj);
 DMLC_REGISTRY_LINK_TAG(multiclass_obj);
-#endif
+#endif  // XGBOOST_USE_CUDA
 DMLC_REGISTRY_LINK_TAG(rank_obj);
 }  // namespace obj
 }  // namespace xgboost

src/objective/regression_obj.cc

@@ -15,4 +15,4 @@ DMLC_REGISTRY_FILE_TAG(regression_obj);
 
 #ifndef XGBOOST_USE_CUDA
 #include "regression_obj.cu"
-#endif
+#endif  // XGBOOST_USE_CUDA

src/objective/regression_obj.cu

@@ -24,7 +24,7 @@ namespace obj {
 
 #if defined(XGBOOST_USE_CUDA)
 DMLC_REGISTRY_FILE_TAG(regression_obj_gpu);
-#endif
+#endif  // defined(XGBOOST_USE_CUDA)
 
 struct RegLossParam : public dmlc::Parameter<RegLossParam> {
   float scale_pos_weight;

src/predictor/predictor.cc

@@ -29,7 +29,7 @@ namespace predictor {
 // List of files that will be force linked in static links.
 #ifdef XGBOOST_USE_CUDA
 DMLC_REGISTRY_LINK_TAG(gpu_predictor);
-#endif
+#endif  // XGBOOST_USE_CUDA
 DMLC_REGISTRY_LINK_TAG(cpu_predictor);
 }  // namespace predictor
 }  // namespace xgboost

src/tree/split_evaluator.cc

@@ -10,6 +10,7 @@
 #include <unordered_set>
 #include <vector>
 #include <limits>
+#include <memory>
 #include <string>
 #include <sstream>
 #include <utility>

src/tree/split_evaluator.h

@@ -11,6 +11,7 @@
 #include <dmlc/registry.h>
 #include <xgboost/base.h>
 #include <functional>
+#include <memory>
 #include <string>
 #include <utility>
 #include <vector>

src/tree/tree_updater.cc

@@ -37,6 +37,6 @@ DMLC_REGISTRY_LINK_TAG(updater_sync);
 #ifdef XGBOOST_USE_CUDA
 DMLC_REGISTRY_LINK_TAG(updater_gpu);
 DMLC_REGISTRY_LINK_TAG(updater_gpu_hist);
-#endif
+#endif  // XGBOOST_USE_CUDA
 }  // namespace tree
 }  // namespace xgboost

src/tree/updater_colmaker.cc

@@ -597,7 +597,7 @@ class ColMaker: public TreeUpdater {
       const auto num_features = static_cast<bst_omp_uint>(feat_set.size());
       #if defined(_OPENMP)
       const int batch_size = std::max(static_cast<int>(num_features / this->nthread_ / 32), 1);
-      #endif
+      #endif  // defined(_OPENMP)
       int poption = param_.parallel_option;
       if (poption == 2) {
         poption = static_cast<int>(num_features) * 2 < this->nthread_ ? 1 : 0;

src/tree/updater_quantile_hist.h

@@ -16,6 +16,7 @@
 #include <string>
 #include <queue>
 #include <iomanip>
+#include <unordered_map>
 #include <utility>
 
 #include "./param.h"