Initial support for external memory in gradient index. (#7183)

* Add hessian to batch param in preparation of new approx impl.
* Extract a push method for gradient index matrix.
* Use span instead of vector ref for hessian in sketching.
* Create a binary format for gradient index.

tests/cpp/data/test_gradient_index.cc

@@ -0,0 +1,26 @@
+/*!
+ * Copyright 2021 XGBoost contributors
+ */
+#include <gtest/gtest.h>
+#include <xgboost/data.h>
+
+#include "../helpers.h"
+#include "../../../src/data/gradient_index.h"
+
+namespace xgboost {
+namespace data {
+TEST(GradientIndex, ExternalMemory) {
+  std::unique_ptr<DMatrix> dmat = CreateSparsePageDMatrix(10000);
+  std::vector<size_t> base_rowids;
+  std::vector<float> hessian(dmat->Info().num_row_, 1);
+  for (auto const& page : dmat->GetBatches<GHistIndexMatrix>({0, 64, hessian})) {
+    base_rowids.push_back(page.base_rowid);
+  }
+  size_t i = 0;
+  for (auto const& page : dmat->GetBatches<SparsePage>()) {
+    ASSERT_EQ(base_rowids[i], page.base_rowid);
+    ++i;
+  }
+}
+}  // namespace data
+}  // namespace xgboost

tests/cpp/data/test_gradient_index_page_raw_format.cc

@@ -0,0 +1,48 @@
+/*!
+ * Copyright 2021 XGBoost contributors
+ */
+#include <gtest/gtest.h>
+
+#include "../../../src/data/gradient_index.h"
+#include "../../../src/data/sparse_page_source.h"
+#include "../helpers.h"
+
+namespace xgboost {
+namespace data {
+TEST(GHistIndexPageRawFormat, IO) {
+  std::unique_ptr<SparsePageFormat<GHistIndexMatrix>> format{
+      CreatePageFormat<GHistIndexMatrix>("raw")};
+  auto m = RandomDataGenerator{100, 14, 0.5}.GenerateDMatrix();
+  dmlc::TemporaryDirectory tmpdir;
+  std::string path = tmpdir.path + "/ghistindex.page";
+
+  {
+    std::unique_ptr<dmlc::Stream> fo{dmlc::Stream::Create(path.c_str(), "w")};
+    for (auto const &index :
+         m->GetBatches<GHistIndexMatrix>({GenericParameter::kCpuId, 256})) {
+      format->Write(index, fo.get());
+    }
+  }
+
+  GHistIndexMatrix page;
+  std::unique_ptr<dmlc::SeekStream> fi{
+      dmlc::SeekStream::CreateForRead(path.c_str())};
+  format->Read(&page, fi.get());
+
+  for (auto const &gidx :
+       m->GetBatches<GHistIndexMatrix>({GenericParameter::kCpuId, 256})) {
+    auto const &loaded = gidx;
+    ASSERT_EQ(loaded.cut.Ptrs(), page.cut.Ptrs());
+    ASSERT_EQ(loaded.cut.MinValues(), page.cut.MinValues());
+    ASSERT_EQ(loaded.cut.Values(), page.cut.Values());
+    ASSERT_EQ(loaded.base_rowid, page.base_rowid);
+    ASSERT_EQ(loaded.IsDense(), page.IsDense());
+    ASSERT_TRUE(std::equal(loaded.index.begin(), loaded.index.end(),
+                           page.index.begin()));
+    ASSERT_TRUE(std::equal(loaded.index.Offset(),
+                           loaded.index.Offset() + loaded.index.OffsetSize(),
+                           page.index.Offset()));
+  }
+}
+} // namespace data
+} // namespace xgboost