Use std::uint64_t for row index.

include/xgboost/base.h

@@ -1,20 +1,18 @@
 /**
- * Copyright 2015-2023 by XGBoost Contributors
+ * Copyright 2015-2024, XGBoost Contributors
  * \file base.h
  * \brief Defines configuration macros and basic types for xgboost.
  */
 #ifndef XGBOOST_BASE_H_
 #define XGBOOST_BASE_H_
 
-#include <dmlc/base.h>
-#include <dmlc/omp.h>
+#include <dmlc/omp.h>  // for omp_uint, omp_ulong
 
-#include <cmath>
-#include <cstdint>
-#include <iostream>
-#include <string>
-#include <utility>
-#include <vector>
+#include <cstdint>  // for int32_t, uint64_t, int16_t
+#include <ostream>  // for ostream
+#include <string>   // for string
+#include <utility>  // for pair
+#include <vector>   // for vector
 
 /*!
  * \brief string flag for R library, to leave hooks when needed.
@@ -86,34 +84,31 @@
 
 #endif  // !defined(XGBOOST_MM_PREFETCH_PRESENT) && !defined()
 
-/*! \brief namespace of xgboost*/
 namespace xgboost {
-
 /*! \brief unsigned integer type used for feature index. */
-using bst_uint = uint32_t;  // NOLINT
+using bst_uint = std::uint32_t;  // NOLINT
 /*! \brief unsigned long integers */
-using bst_ulong = uint64_t;  // NOLINT
+using bst_ulong = std::uint64_t;  // NOLINT
 /*! \brief float type, used for storing statistics */
 using bst_float = float;  // NOLINT
 /*! \brief Categorical value type. */
-using bst_cat_t = int32_t;  // NOLINT
+using bst_cat_t = std::int32_t;  // NOLINT
 /*! \brief Type for data column (feature) index. */
-using bst_feature_t = uint32_t;  // NOLINT
-/*! \brief Type for histogram bin index. */
-using bst_bin_t = int32_t;  // NOLINT
-/*! \brief Type for data row index.
- *
- * Be careful `std::size_t' is implementation-defined.  Meaning that the binary
- * representation of DMatrix might not be portable across platform.  Booster model should
- * be portable as parameters are floating points.
+using bst_feature_t = std::uint32_t;  // NOLINT
+/**
+ * @brief Type for histogram bin index.  We sometimes use -1 to indicate invalid bin.
  */
-using bst_row_t = std::size_t;   // NOLINT
+using bst_bin_t = std::int32_t;  // NOLINT
+/**
+ * @brief Type for data row index (sample).
+ */
+using bst_idx_t = std::uint64_t;  // NOLINT
 /*! \brief Type for tree node index. */
 using bst_node_t = std::int32_t;      // NOLINT
 /*! \brief Type for ranking group index. */
 using bst_group_t = std::uint32_t;  // NOLINT
 /**
- * \brief Type for indexing into output targets.
+ * @brief Type for indexing into output targets.
  */
 using bst_target_t = std::uint32_t;  // NOLINT
 /**
@@ -306,8 +301,7 @@ class GradientPairInt64 {
   XGBOOST_DEVICE bool operator==(const GradientPairInt64 &rhs) const {
     return grad_ == rhs.grad_ && hess_ == rhs.hess_;
   }
-  friend std::ostream &operator<<(std::ostream &os,
-                                  const GradientPairInt64 &g) {
+  friend std::ostream &operator<<(std::ostream &os, const GradientPairInt64 &g) {
     os << g.GetQuantisedGrad() << "/" << g.GetQuantisedHess();
     return os;
   }
@@ -323,7 +317,7 @@ using omp_ulong = dmlc::omp_ulong;  // NOLINT
 /*! \brief define unsigned int for openmp loop */
 using bst_omp_uint = dmlc::omp_uint;  // NOLINT
 /*! \brief Type used for representing version number in binary form.*/
-using XGBoostVersionT = int32_t;
+using XGBoostVersionT = std::int32_t;
 }  // namespace xgboost
 
 #endif  // XGBOOST_BASE_H_

include/xgboost/data.h

@@ -315,7 +315,7 @@ struct BatchParam {
 struct HostSparsePageView {
   using Inst = common::Span<Entry const>;
 
-  common::Span<bst_row_t const> offset;
+  common::Span<bst_idx_t const> offset;
   common::Span<Entry const> data;
 
   Inst operator[](size_t i) const {
@@ -333,7 +333,7 @@ struct HostSparsePageView {
 class SparsePage {
  public:
   // Offset for each row.
-  HostDeviceVector<bst_row_t> offset;
+  HostDeviceVector<bst_idx_t> offset;
   /*! \brief the data of the segments */
   HostDeviceVector<Entry> data;
 

include/xgboost/json.h

@@ -60,9 +60,7 @@ class Value {
   virtual Json& operator[](int ind);
 
   virtual bool operator==(Value const& rhs) const = 0;
-#if !defined(__APPLE__)
   virtual Value& operator=(Value const& rhs) = delete;
-#endif  // !defined(__APPLE__)
 
   std::string TypeStr() const;
 
@@ -105,6 +103,7 @@ class JsonString : public Value {
   std::string&       GetString()       & { return str_; }
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   static bool IsClassOf(Value const* value) {
     return value->Type() == ValueKind::kString;
@@ -134,6 +133,7 @@ class JsonArray : public Value {
   std::vector<Json>&       GetArray()       & { return vec_; }
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   static bool IsClassOf(Value const* value) {
     return value->Type() == ValueKind::kArray;
@@ -158,6 +158,7 @@ class JsonTypedArray : public Value {
   JsonTypedArray(JsonTypedArray&& that) noexcept : Value{kind}, vec_{std::move(that.vec_)} {}
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   void Set(size_t i, T v) { vec_[i] = v; }
   size_t Size() const { return vec_.size(); }
@@ -216,6 +217,7 @@ class JsonObject : public Value {
   Map& GetObject() & { return object_; }
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   static bool IsClassOf(Value const* value) { return value->Type() == ValueKind::kObject; }
   ~JsonObject() override = default;
@@ -249,6 +251,7 @@ class JsonNumber : public Value {
   Float&       GetNumber()       & { return number_; }
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   static bool IsClassOf(Value const* value) {
     return value->Type() == ValueKind::kNumber;
@@ -287,6 +290,7 @@ class JsonInteger : public Value {
       : Value{ValueKind::kInteger}, integer_{that.integer_} {}
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   Int const& GetInteger() &&      { return integer_; }
   Int const& GetInteger() const & { return integer_; }
@@ -307,6 +311,7 @@ class JsonNull : public Value {
   void Save(JsonWriter* writer) const override;
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   static bool IsClassOf(Value const* value) {
     return value->Type() == ValueKind::kNull;
@@ -336,6 +341,7 @@ class JsonBoolean : public Value {
   bool&       GetBoolean()       & { return boolean_; }
 
   bool operator==(Value const& rhs) const override;
+  Value& operator=(Value const& rhs) override = delete;
 
   static bool IsClassOf(Value const* value) {
     return value->Type() == ValueKind::kBoolean;

plugin/sycl/data/gradient_index.cc

@@ -57,7 +57,7 @@ void GHistIndexMatrix::SetIndexData(::sycl::queue qu,
                                     uint32_t* offsets) {
   if (nbins == 0) return;
   const xgboost::Entry *data_ptr = dmat.data.DataConst();
-  const bst_row_t *offset_vec = dmat.row_ptr.DataConst();
+  const bst_idx_t *offset_vec = dmat.row_ptr.DataConst();
   const size_t num_rows = dmat.row_ptr.Size() - 1;
   const bst_float* cut_values = cut_device.Values().DataConst();
   const uint32_t* cut_ptrs = cut_device.Ptrs().DataConst();

src/collective/device_communicator_adapter.cuh

@@ -3,6 +3,8 @@
  */
 #pragma once
 
+#include <numeric>  // for accumulate
+
 #include "communicator.h"
 #include "device_communicator.cuh"
 

src/common/column_matrix.h

@@ -72,7 +72,7 @@ class SparseColumnIter : public Column<BinIdxT> {
 
  public:
   SparseColumnIter(common::Span<const BinIdxT> index, bst_bin_t least_bin_idx,
-                   common::Span<const size_t> row_ind, bst_row_t first_row_idx)
+                   common::Span<const size_t> row_ind, bst_idx_t first_row_idx)
       : Base{index, least_bin_idx}, row_ind_(row_ind) {
     // first_row_id is the first row in the leaf partition
     const size_t* row_data = RowIndices();
@@ -301,7 +301,7 @@ class ColumnMatrix {
   }
 
   template <typename BinIdxType>
-  auto SparseColumn(bst_feature_t fidx, bst_row_t first_row_idx) const {
+  auto SparseColumn(bst_feature_t fidx, bst_idx_t first_row_idx) const {
     const size_t feature_offset = feature_offsets_[fidx];  // to get right place for certain feature
     const size_t column_size = feature_offsets_[fidx + 1] - feature_offset;
     common::Span<const BinIdxType> bin_index = {
@@ -325,7 +325,7 @@ class ColumnMatrix {
   // all columns are dense column and has no missing value
   // FIXME(jiamingy): We don't need a column matrix if there's no missing value.
   template <typename RowBinIdxT>
-  void SetIndexNoMissing(bst_row_t base_rowid, RowBinIdxT const* row_index, const size_t n_samples,
+  void SetIndexNoMissing(bst_idx_t base_rowid, RowBinIdxT const* row_index, const size_t n_samples,
                          const size_t n_features, int32_t n_threads) {
     missing_.GrowTo(feature_offsets_[n_features], false);
 

src/common/device_helpers.cuh

@@ -19,19 +19,16 @@
 #include <thrust/unique.h>
 
 #include <algorithm>
-#include <chrono>
 #include <cstddef>  // for size_t
 #include <cub/cub.cuh>
 #include <cub/util_allocator.cuh>
-#include <numeric>
 #include <sstream>
 #include <string>
 #include <tuple>
 #include <vector>
 
 #include "../collective/communicator-inl.h"
 #include "common.h"
-#include "xgboost/global_config.h"
 #include "xgboost/host_device_vector.h"
 #include "xgboost/logging.h"
 #include "xgboost/span.h"

src/common/hist_util.cc

@@ -34,7 +34,7 @@ HistogramCuts SketchOnDMatrix(Context const *ctx, DMatrix *m, bst_bin_t max_bins
   HistogramCuts out;
   auto const &info = m->Info();
   auto n_threads = ctx->Threads();
-  std::vector<bst_row_t> reduced(info.num_col_, 0);
+  std::vector<bst_idx_t> reduced(info.num_col_, 0);
   for (auto const &page : m->GetBatches<SparsePage>()) {
     auto const &entries_per_column =
         CalcColumnSize(data::SparsePageAdapterBatch{page.GetView()}, info.num_col_, n_threads,
@@ -209,10 +209,10 @@ void RowsWiseBuildHistKernel(Span<GradientPair const> gpair,
     CHECK(offsets);
   }
 
-  auto get_row_ptr = [&](bst_row_t ridx) {
+  auto get_row_ptr = [&](bst_idx_t ridx) {
     return kFirstPage ? row_ptr[ridx] : row_ptr[ridx - base_rowid];
   };
-  auto get_rid = [&](bst_row_t ridx) { return kFirstPage ? ridx : (ridx - base_rowid); };
+  auto get_rid = [&](bst_idx_t ridx) { return kFirstPage ? ridx : (ridx - base_rowid); };
 
   const size_t n_features =
       get_row_ptr(row_indices.begin[0] + 1) - get_row_ptr(row_indices.begin[0]);
@@ -275,10 +275,10 @@ void ColsWiseBuildHistKernel(Span<GradientPair const> gpair,
   auto const &row_ptr = gmat.row_ptr.data();
   auto base_rowid = gmat.base_rowid;
   const uint32_t *offsets = gmat.index.Offset();
-  auto get_row_ptr = [&](bst_row_t ridx) {
+  auto get_row_ptr = [&](bst_idx_t ridx) {
     return kFirstPage ? row_ptr[ridx] : row_ptr[ridx - base_rowid];
   };
-  auto get_rid = [&](bst_row_t ridx) { return kFirstPage ? ridx : (ridx - base_rowid); };
+  auto get_rid = [&](bst_idx_t ridx) { return kFirstPage ? ridx : (ridx - base_rowid); };
 
   const size_t n_features = gmat.cut.Ptrs().size() - 1;
   const size_t n_columns = n_features;

src/common/hist_util.cu

@@ -13,8 +13,6 @@
 #include <xgboost/logging.h>
 
 #include <cstddef>  // for size_t
-#include <memory>
-#include <mutex>
 #include <utility>
 #include <vector>
 
@@ -39,15 +37,15 @@ size_t RequiredSampleCutsPerColumn(int max_bins, size_t num_rows) {
   return std::min(num_cuts, num_rows);
 }
 
-size_t RequiredSampleCuts(bst_row_t num_rows, bst_feature_t num_columns,
+size_t RequiredSampleCuts(bst_idx_t num_rows, bst_feature_t num_columns,
                           size_t max_bins, size_t nnz) {
   auto per_column = RequiredSampleCutsPerColumn(max_bins, num_rows);
   auto if_dense = num_columns * per_column;
   auto result = std::min(nnz, if_dense);
   return result;
 }
 
-size_t RequiredMemory(bst_row_t num_rows, bst_feature_t num_columns, size_t nnz,
+size_t RequiredMemory(bst_idx_t num_rows, bst_feature_t num_columns, size_t nnz,
                       size_t num_bins, bool with_weights) {
   size_t peak = 0;
   // 0. Allocate cut pointer in quantile container by increasing: n_columns + 1
@@ -85,7 +83,7 @@ size_t RequiredMemory(bst_row_t num_rows, bst_feature_t num_columns, size_t nnz,
   return peak;
 }
 
-size_t SketchBatchNumElements(size_t sketch_batch_num_elements, bst_row_t num_rows,
+size_t SketchBatchNumElements(size_t sketch_batch_num_elements, bst_idx_t num_rows,
                               bst_feature_t columns, size_t nnz, int device, size_t num_cuts,
                               bool has_weight) {
   auto constexpr kIntMax = static_cast<std::size_t>(std::numeric_limits<std::int32_t>::max());
@@ -123,7 +121,7 @@ void SortByWeight(dh::device_vector<float>* weights, dh::device_vector<Entry>* s
       [=] __device__(const Entry& a, const Entry& b) { return a.index == b.index; });
 }
 
-void RemoveDuplicatedCategories(DeviceOrd device, MetaInfo const& info, Span<bst_row_t> d_cuts_ptr,
+void RemoveDuplicatedCategories(DeviceOrd device, MetaInfo const& info, Span<bst_idx_t> d_cuts_ptr,
                                 dh::device_vector<Entry>* p_sorted_entries,
                                 dh::device_vector<float>* p_sorted_weights,
                                 dh::caching_device_vector<size_t>* p_column_sizes_scan) {
@@ -210,7 +208,7 @@ void ProcessWeightedBatch(Context const* ctx, const SparsePage& page, MetaInfo c
     sorted_entries = dh::device_vector<Entry>(h_data.begin() + begin, h_data.begin() + end);
   }
 
-  bst_row_t base_rowid = page.base_rowid;
+  bst_idx_t base_rowid = page.base_rowid;
 
   dh::device_vector<float> entry_weight;
   auto cuctx = ctx->CUDACtx();

src/common/hist_util.cuh

@@ -186,7 +186,7 @@ inline size_t constexpr BytesPerElement(bool has_weight) {
  *        directly if it's not 0.
  */
 size_t SketchBatchNumElements(size_t sketch_batch_num_elements,
-                              bst_row_t num_rows, bst_feature_t columns,
+                              bst_idx_t num_rows, bst_feature_t columns,
                               size_t nnz, int device,
                               size_t num_cuts, bool has_weight);
 
@@ -209,7 +209,7 @@ size_t RequiredSampleCutsPerColumn(int max_bins, size_t num_rows);
  *
  * \return The estimated bytes
  */
-size_t RequiredMemory(bst_row_t num_rows, bst_feature_t num_columns, size_t nnz,
+size_t RequiredMemory(bst_idx_t num_rows, bst_feature_t num_columns, size_t nnz,
                       size_t num_bins, bool with_weights);
 
 // Count the valid entries in each column and copy them out.
@@ -240,7 +240,7 @@ void MakeEntriesFromAdapter(AdapterBatch const& batch, BatchIter batch_iter, Ran
 void SortByWeight(dh::device_vector<float>* weights,
                   dh::device_vector<Entry>* sorted_entries);
 
-void RemoveDuplicatedCategories(DeviceOrd device, MetaInfo const& info, Span<bst_row_t> d_cuts_ptr,
+void RemoveDuplicatedCategories(DeviceOrd device, MetaInfo const& info, Span<bst_idx_t> d_cuts_ptr,
                                 dh::device_vector<Entry>* p_sorted_entries,
                                 dh::device_vector<float>* p_sorted_weights,
                                 dh::caching_device_vector<size_t>* p_column_sizes_scan);

src/common/host_device_vector.cc

@@ -178,7 +178,7 @@ template class HostDeviceVector<uint8_t>;
 template class HostDeviceVector<int8_t>;
 template class HostDeviceVector<FeatureType>;
 template class HostDeviceVector<Entry>;
-template class HostDeviceVector<uint64_t>;  // bst_row_t
+template class HostDeviceVector<bst_idx_t>;
 template class HostDeviceVector<uint32_t>;  // bst_feature_t
 
 #if defined(__APPLE__) || defined(__EMSCRIPTEN__)

src/common/host_device_vector.cu

@@ -412,7 +412,7 @@ template class HostDeviceVector<uint8_t>;
 template class HostDeviceVector<int8_t>;
 template class HostDeviceVector<FeatureType>;
 template class HostDeviceVector<Entry>;
-template class HostDeviceVector<uint64_t>;  // bst_row_t
+template class HostDeviceVector<bst_idx_t>;
 template class HostDeviceVector<uint32_t>;  // bst_feature_t
 template class HostDeviceVector<RegTree::Node>;
 template class HostDeviceVector<RegTree::CategoricalSplitMatrix::Segment>;