[clang] Introduce paged vector (PR #66430)

Giulio Eulisse via cfe-commits cfe-commits at lists.llvm.org
Thu Sep 21 04:55:20 PDT 2023


https://github.com/ktf updated https://github.com/llvm/llvm-project/pull/66430

>From 44b01fb22b4a44a2de35299e7649626cea30852a Mon Sep 17 00:00:00 2001
From: Giulio Eulisse <10544+ktf at users.noreply.github.com>
Date: Thu, 14 Sep 2023 21:58:21 +0200
Subject: [PATCH] Introduce PagedVector class

The goal of the class is to be an (almost) drop in replacement for
SmallVector and std::vector when those are presized and filled later,
as it happens in SourceManager and ASTReader.

By splitting the actual vector in pages of the same size and allocating
the pages only when they are needed, using this containers reduces the
memory usage by a factor 4 for the cases relevant to the ALICE
experiment ROOT / cling usage.
---
 clang/include/clang/Basic/SourceManager.h     |   3 +-
 clang/include/clang/Serialization/ASTReader.h |   5 +-
 clang/lib/Basic/SourceManager.cpp             |  10 +-
 clang/lib/Serialization/ASTReader.cpp         |   5 +-
 llvm/docs/ProgrammersManual.rst               |  33 ++
 llvm/include/llvm/ADT/PagedVector.h           | 304 ++++++++++++++++
 llvm/unittests/ADT/CMakeLists.txt             |   1 +
 llvm/unittests/ADT/PagedVectorTest.cpp        | 329 ++++++++++++++++++
 8 files changed, 680 insertions(+), 10 deletions(-)
 create mode 100644 llvm/include/llvm/ADT/PagedVector.h
 create mode 100644 llvm/unittests/ADT/PagedVectorTest.cpp

diff --git a/clang/include/clang/Basic/SourceManager.h b/clang/include/clang/Basic/SourceManager.h
index 2f846502d6f3327..e37caa2252532f9 100644
--- a/clang/include/clang/Basic/SourceManager.h
+++ b/clang/include/clang/Basic/SourceManager.h
@@ -43,6 +43,7 @@
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/IntrusiveRefCntPtr.h"
+#include "llvm/ADT/PagedVector.h"
 #include "llvm/ADT/PointerIntPair.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
@@ -699,7 +700,7 @@ class SourceManager : public RefCountedBase<SourceManager> {
   ///
   /// Negative FileIDs are indexes into this table. To get from ID to an index,
   /// use (-ID - 2).
-  SmallVector<SrcMgr::SLocEntry, 0> LoadedSLocEntryTable;
+  llvm::PagedVector<SrcMgr::SLocEntry> LoadedSLocEntryTable;
 
   /// The starting offset of the next local SLocEntry.
   ///
diff --git a/clang/include/clang/Serialization/ASTReader.h b/clang/include/clang/Serialization/ASTReader.h
index dc1eb21c27801fe..65e19c6e44cf571 100644
--- a/clang/include/clang/Serialization/ASTReader.h
+++ b/clang/include/clang/Serialization/ASTReader.h
@@ -38,6 +38,7 @@
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/IntrusiveRefCntPtr.h"
 #include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/PagedVector.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallPtrSet.h"
@@ -487,7 +488,7 @@ class ASTReader
   ///
   /// When the pointer at index I is non-NULL, the type with
   /// ID = (I + 1) << FastQual::Width has already been loaded
-  std::vector<QualType> TypesLoaded;
+  llvm::PagedVector<QualType> TypesLoaded;
 
   using GlobalTypeMapType =
       ContinuousRangeMap<serialization::TypeID, ModuleFile *, 4>;
@@ -501,7 +502,7 @@ class ASTReader
   ///
   /// When the pointer at index I is non-NULL, the declaration with ID
   /// = I + 1 has already been loaded.
-  std::vector<Decl *> DeclsLoaded;
+  llvm::PagedVector<Decl *> DeclsLoaded;
 
   using GlobalDeclMapType =
       ContinuousRangeMap<serialization::DeclID, ModuleFile *, 4>;
diff --git a/clang/lib/Basic/SourceManager.cpp b/clang/lib/Basic/SourceManager.cpp
index 0521ac7b30339ab..7fa8b8096ac4931 100644
--- a/clang/lib/Basic/SourceManager.cpp
+++ b/clang/lib/Basic/SourceManager.cpp
@@ -2344,11 +2344,11 @@ SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const {
 }
 
 size_t SourceManager::getDataStructureSizes() const {
-  size_t size = llvm::capacity_in_bytes(MemBufferInfos)
-    + llvm::capacity_in_bytes(LocalSLocEntryTable)
-    + llvm::capacity_in_bytes(LoadedSLocEntryTable)
-    + llvm::capacity_in_bytes(SLocEntryLoaded)
-    + llvm::capacity_in_bytes(FileInfos);
+  size_t size = llvm::capacity_in_bytes(MemBufferInfos) +
+                llvm::capacity_in_bytes(LocalSLocEntryTable) +
+                llvm::capacity_in_bytes(LoadedSLocEntryTable) +
+                llvm::capacity_in_bytes(SLocEntryLoaded) +
+                llvm::capacity_in_bytes(FileInfos);
 
   if (OverriddenFilesInfo)
     size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles);
diff --git a/clang/lib/Serialization/ASTReader.cpp b/clang/lib/Serialization/ASTReader.cpp
index 0952244d037a77c..badd54987af18dd 100644
--- a/clang/lib/Serialization/ASTReader.cpp
+++ b/clang/lib/Serialization/ASTReader.cpp
@@ -7944,9 +7944,10 @@ void ASTReader::PrintStats() {
   std::fprintf(stderr, "*** AST File Statistics:\n");
 
   unsigned NumTypesLoaded =
-      TypesLoaded.size() - llvm::count(TypesLoaded, QualType());
+      TypesLoaded.size() - llvm::count(TypesLoaded.materialised(), QualType());
   unsigned NumDeclsLoaded =
-      DeclsLoaded.size() - llvm::count(DeclsLoaded, (Decl *)nullptr);
+      DeclsLoaded.size() -
+      llvm::count(DeclsLoaded.materialised(), (Decl *)nullptr);
   unsigned NumIdentifiersLoaded =
       IdentifiersLoaded.size() -
       llvm::count(IdentifiersLoaded, (IdentifierInfo *)nullptr);
diff --git a/llvm/docs/ProgrammersManual.rst b/llvm/docs/ProgrammersManual.rst
index 43dd985d9779ed2..1096701288c033c 100644
--- a/llvm/docs/ProgrammersManual.rst
+++ b/llvm/docs/ProgrammersManual.rst
@@ -1625,6 +1625,39 @@ SmallVector has grown a few other minor advantages over std::vector, causing
    and is no longer "private to the implementation". A name like
    ``SmallVectorHeader`` might be more appropriate.
 
+.. _dss_pagedvector:
+
+llvm/ADT/PagedVector.h
+^^^^^^^^^^^^^^^^^^^^^^
+
+``PagedVector<Type, PageSize>`` is a random access container that allocates
+(PageSize) elements of type Type when the first element of a page is accessed
+via the ``operator[]``.  This is useful for the case in which the number of
+elements is known in advance and their actual initialization is expensive and
+sparse so that it's only done lazily when the element is accessed. When the
+number of used pages is small significant memory savings can be achieved.
+
+The main advantage is that a ``PagedVector`` allows to delay the actual
+allocation of the page until it's needed, at the extra cost of one integer per
+page and one extra indirection when accessing elements with their positional
+index. 
+
+In order to maximise the memory footprint of this container, it's important to
+balance the PageSize so that it's not too small (otherwise the overhead of the
+integer per page might become too high) and not too big (otherwise the memory is
+wasted if the page is not fully used).
+
+Moreover, while retaining the oder of the elements based on their insertion
+index, like a vector, iterating over the elements via ``begin()`` and ``end()``
+is not provided in the API, due to the fact accessing the elements in order
+would allocate all the iterated pages, defeating memory savings and the purpose
+of the ``PagedVector``.
+
+Finally a ``materialised_begin()`` and ``materialised_end`` iterators are
+provided to access the elements associated to the accessed pages, which could
+speed up operations that need to iterate over initialized elements in a
+non-ordered manner.
+
 .. _dss_vector:
 
 <vector>
diff --git a/llvm/include/llvm/ADT/PagedVector.h b/llvm/include/llvm/ADT/PagedVector.h
new file mode 100644
index 000000000000000..c63e2083d4ef786
--- /dev/null
+++ b/llvm/include/llvm/ADT/PagedVector.h
@@ -0,0 +1,304 @@
+//===- llvm/ADT/PagedVector.h - 'Lazyly allocated' vectors --------*- C++
+//-*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the PagedVector class.
+//
+//===----------------------------------------------------------------------===//
+#ifndef LLVM_ADT_PAGEDVECTOR_H
+#define LLVM_ADT_PAGEDVECTOR_H
+
+#include "llvm/ADT/PointerIntPair.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/iterator_range.h"
+#include "llvm/Support/Allocator.h"
+#include <cassert>
+#include <vector>
+
+namespace llvm {
+// A vector that allocates memory in pages.
+//
+// Order is kept, but memory is allocated only when one element of the page is
+// accessed. This introduces a level of indirection, but it is useful when you
+// have a sparsely initialised vector where the full size is allocated upfront.
+//
+// As a side effect the elements are initialised later than in a normal vector.
+// On the first access to one of the elements of a given page all, the elements
+// of the page are initialised. This also means that the elements of the page
+// are initialised beyond the size of the vector.
+//
+// Similarly on destruction the elements are destroyed only when the page is
+// not needed anymore, delaying invoking the destructor of the elements.
+//
+// Notice that this does not have iterators, because if you have iterators it
+// probably means you are going to touch all the memory in any case, so better
+// use a std::vector in the first place.
+template <typename T, size_t PageSize = 1024 / sizeof(T)> class PagedVector {
+  static_assert(PageSize > 0, "PageSize must be greater than 0. Most likely "
+                              "you want it to be greater than 16.");
+  // The actual number of element in the vector which can be accessed.
+  size_t Size = 0;
+
+  // The position of the initial element of the page in the Data vector.
+  // Pages are allocated contiguously in the Data vector.
+  mutable SmallVector<uintptr_t, 0> PageToDataIdx;
+  // Actual page data. All the page elements are added to this vector on the
+  // first access of any of the elements of the page. Elements default
+  // constructed and elements of the page are stored contiguously. The order of
+  // the elements however depends on the order of access of the pages.
+  PointerIntPair<BumpPtrAllocator *, 1, bool> Allocator;
+
+  constexpr static uintptr_t InvalidPage = SIZE_MAX;
+
+public:
+  using value_type = T;
+
+  // Default constructor. We build our own allocator and mark it as such with
+  // `true` in the second pair element.
+  PagedVector() : Allocator(new BumpPtrAllocator, true) {}
+  PagedVector(BumpPtrAllocator *A) : Allocator(A, false) {
+    assert(A != nullptr && "Allocator cannot be null");
+  }
+
+  ~PagedVector() {
+    clear();
+    // If we own the allocator, delete it.
+    if (Allocator.getInt())
+      delete Allocator.getPointer();
+  }
+
+  // Look up an element at position `Index`.
+  // If the associated page is not filled, it will be filled with default
+  // constructed elements. If the associated page is filled, return the element.
+  T &operator[](size_t Index) const {
+    assert(Index < Size);
+    assert(Index / PageSize < PageToDataIdx.size());
+    uintptr_t &PagePtr = PageToDataIdx[Index / PageSize];
+    // If the page was not yet allocated, allocate it.
+    if (PagePtr == InvalidPage) {
+      T *NewPagePtr = Allocator.getPointer()->template Allocate<T>(PageSize);
+      // We need to invoke the default constructor on all the elements of the
+      // page.
+      for (size_t I = 0; I < PageSize; ++I)
+        new (NewPagePtr + I) T();
+
+      PagePtr = reinterpret_cast<uintptr_t>(NewPagePtr);
+    }
+    // Dereference the element in the page.
+    return *((Index % PageSize) + reinterpret_cast<T *>(PagePtr));
+  }
+
+  // Return the capacity of the vector. I.e. the maximum size it can be expanded
+  // to with the resize method without allocating more pages.
+  [[nodiscard]] size_t capacity() const {
+    return PageToDataIdx.size() * PageSize;
+  }
+
+  // Return the size of the vector. I.e. the maximum index that can be
+  // accessed, i.e. the maximum value which was used as argument of the
+  // resize method.
+  [[nodiscard]] size_t size() const { return Size; }
+
+  // Resize the vector. Notice that the constructor of the elements will not
+  // be invoked until an element of a given page is accessed, at which point
+  // all the elements of the page will be constructed.
+  // If the new size is smaller than the current size, the elements of the
+  // pages that are not needed anymore will be destroyed, however, elements of 
+  // the last page will not be destroyed.
+  // For these reason the usage of this vector is discouraged if you rely
+  // on the construction / destructor of the elements to be invoked.
+  void resize(size_t NewSize) {
+    if (NewSize == 0) {
+      clear();
+      return;
+    }
+    // Handle shrink case: destroy the elements in the pages that are not
+    // needed anymore and deallocate the pages.
+    //
+    // On the other hand, we do not destroy the extra elements in the last page,
+    // because we might need them later and the logic is simpler if we do not
+    // destroy them. This means that elements are only destroyed only when the
+    // page they belong to is destroyed. This is similar to what happens on
+    // access of the elements of a page, where all the elements of the page are
+    // constructed not only the one effectively neeeded.
+    if (NewSize < Size) {
+      size_t NewLastPage = (NewSize - 1) / PageSize;
+      for (size_t I = NewLastPage + 1, E = PageToDataIdx.size(); I < E; ++I) {
+        uintptr_t PagePtr = PageToDataIdx[I];
+        if (PagePtr == InvalidPage)
+          continue;
+        T *Page = reinterpret_cast<T *>(PagePtr);
+        // We need to invoke the destructor on all the elements of the page.
+        for (size_t J = 0; J < PageSize; ++J)
+          Page[J].~T();
+        Allocator.getPointer()->Deallocate(Page);
+        // We mark the page invalid, to avoid double deletion.
+        PageToDataIdx[I] = InvalidPage;
+      }
+      PageToDataIdx.resize(NewLastPage + 1);
+    }
+    Size = NewSize;
+    // If the capacity is enough, just update the size and continue
+    // with the currently allocated pages.
+    if (Size <= capacity())
+      return;
+    // The number of pages to allocate. The Remainder is calculated
+    // for the case in which the NewSize is not a multiple of PageSize.
+    // In that case we need one more page.
+    size_t Pages = Size / PageSize;
+    size_t Remainder = Size % PageSize;
+    if (Remainder != 0)
+      Pages += 1;
+    assert(Pages > PageToDataIdx.size());
+    // We use InvalidPage to indicate that a page has not been allocated yet.
+    // This cannot be 0, because 0 is a valid page id.
+    // We use InvalidPage instead of a separate bool to avoid wasting space.
+    PageToDataIdx.resize(Pages, InvalidPage);
+  }
+
+  // Return true if the vector is empty
+  [[nodiscard]] bool empty() const { return Size == 0; }
+
+  /// Clear the vector, i.e. clear the allocated pages, the whole page
+  /// lookup index and reset the size.
+  void clear() {
+    Size = 0;
+    for (uintptr_t Page : PageToDataIdx) {
+      for (size_t I = 0; I < PageSize; ++I)
+        reinterpret_cast<T *>(Page)[I].~T();
+      // If we do not own the allocator, deallocate the pages one by one.
+      if (!Allocator.getInt()) {
+        Allocator.getPointer()->Deallocate(reinterpret_cast<T *>(Page));
+      }
+    }
+    // If we own the allocator, simply reset it.
+    if (Allocator.getInt() == true) {
+      Allocator.getPointer()->Reset();
+    }
+    PageToDataIdx.clear();
+  }
+
+  // Iterator on all the elements of the vector
+  // which have actually being constructed.
+  class MaterialisedIterator {
+    const PagedVector *PV;
+    size_t ElementIdx;
+
+  public:
+    using iterator_category = std::forward_iterator_tag;
+    using value_type = T;
+    using difference_type = std::ptrdiff_t;
+    using pointer = T *;
+    using reference = T &;
+
+    MaterialisedIterator(PagedVector const *PV, size_t ElementIdx)
+        : PV(PV), ElementIdx(ElementIdx) {}
+
+    // When incrementing the iterator, we skip the elements which have not
+    // been materialised yet.
+    MaterialisedIterator &operator++() {
+      ++ElementIdx;
+      if (ElementIdx % PageSize == 0) {
+        while (ElementIdx < PV->Size &&
+               PV->PageToDataIdx[ElementIdx / PageSize] == InvalidPage)
+          ElementIdx += PageSize;
+        if (ElementIdx > PV->Size)
+          ElementIdx = PV->Size;
+      }
+
+      return *this;
+    }
+    // Post increment operator.
+    MaterialisedIterator operator++(int) {
+      MaterialisedIterator Copy = *this;
+      ++*this;
+      return Copy;
+    }
+
+    // When dereferencing the iterator, we materialise the page if needed.
+    T const &operator*() const {
+      assert(ElementIdx < PV->Size);
+      assert(PV->PageToDataIdx[ElementIdx / PageSize] != InvalidPage);
+      T *PagePtr =
+          reinterpret_cast<T *>(PV->PageToDataIdx[ElementIdx / PageSize]);
+      return *((ElementIdx % PageSize) + PagePtr);
+    }
+
+    // Equality operator.
+    bool operator==(MaterialisedIterator const &Other) const {
+      // Iterators of two different vectors are never equal.
+      if (PV != Other.PV)
+        return false;
+      // Any iterator for an empty vector is equal to any other iterator.
+      if (PV->empty())
+        return true;
+      // Get the pages of the two iterators. If between the two pages there
+      // are no valid pages, we can condider the iterators equal.
+      size_t PageMin = std::min(ElementIdx, Other.ElementIdx) / PageSize;
+      size_t PageMax = std::max(ElementIdx, Other.ElementIdx) / PageSize;
+      // If the two pages are past the end, the iterators are equal.
+      if (PageMin >= PV->PageToDataIdx.size())
+        return true;
+      // If only the last page is past the end, the iterators are equal if
+      // all the pages up to the end are invalid.
+      if (PageMax >= PV->PageToDataIdx.size()) {
+        for (size_t PageIdx = PageMin; PageIdx < PV->PageToDataIdx.size();
+             ++PageIdx)
+          if (PV->PageToDataIdx[PageIdx] != InvalidPage)
+            return false;
+        return true;
+      }
+
+      uintptr_t Page1 = PV->PageToDataIdx[PageMin];
+      uintptr_t Page2 = PV->PageToDataIdx[PageMax];
+      if (Page1 == InvalidPage && Page2 == InvalidPage)
+        return true;
+
+      // If the two pages are the same, the iterators are equal if they point
+      // to the same element.
+      if (PageMin == PageMax)
+        return ElementIdx == Other.ElementIdx;
+
+      // If the two pages are different, the iterators are equal if all the
+      // pages between them are invalid.
+      return std::all_of(PV->PageToDataIdx.begin() + PageMin,
+                         PV->PageToDataIdx.begin() + PageMax,
+                         [](uintptr_t Page) { return Page == InvalidPage; });
+    }
+
+    bool operator!=(MaterialisedIterator const &Other) const {
+      return !(*this == Other);
+    }
+
+    [[nodiscard]] size_t getIndex() const { return ElementIdx; }
+  };
+
+  // Iterators over the materialised elements of the vector.
+  // This includes all the elements belonging to allocated pages,
+  // even if they have not been accessed yet. It's enough to access
+  // one element of a page to materialise all the elements of the page.
+  MaterialisedIterator materialised_begin() const {
+    // Look for the first valid page
+    for (size_t ElementIdx = 0; ElementIdx < Size; ElementIdx += PageSize)
+      if (PageToDataIdx[ElementIdx / PageSize] != InvalidPage)
+        return MaterialisedIterator(this, ElementIdx);
+
+    return MaterialisedIterator(this, Size);
+  }
+
+  MaterialisedIterator materialised_end() const {
+    return MaterialisedIterator(this, Size);
+  }
+
+  [[nodiscard]] llvm::iterator_range<MaterialisedIterator>
+  materialised() const {
+    return {materialised_begin(), materialised_end()};
+  }
+};
+} // namespace llvm
+#endif // LLVM_ADT_PAGEDVECTOR_H
diff --git a/llvm/unittests/ADT/CMakeLists.txt b/llvm/unittests/ADT/CMakeLists.txt
index c5190255ba7738e..9aa7120f30696da 100644
--- a/llvm/unittests/ADT/CMakeLists.txt
+++ b/llvm/unittests/ADT/CMakeLists.txt
@@ -51,6 +51,7 @@ add_llvm_unittest(ADTTests
   MapVectorTest.cpp
   MoveOnly.cpp
   PackedVectorTest.cpp
+  PagedVectorTest.cpp
   PointerEmbeddedIntTest.cpp
   PointerIntPairTest.cpp
   PointerSumTypeTest.cpp
diff --git a/llvm/unittests/ADT/PagedVectorTest.cpp b/llvm/unittests/ADT/PagedVectorTest.cpp
new file mode 100644
index 000000000000000..c79c80b98802ee4
--- /dev/null
+++ b/llvm/unittests/ADT/PagedVectorTest.cpp
@@ -0,0 +1,329 @@
+//===- llvm/unittest/ADT/PagedVectorTest.cpp ------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// PagedVector unit tests.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/PagedVector.h"
+#include "gtest/gtest.h"
+#include <iterator>
+
+namespace llvm {
+TEST(PagedVectorTest, EmptyTest) {
+  PagedVector<int, 10> V;
+  EXPECT_EQ(V.empty(), true);
+  EXPECT_EQ(V.size(), 0ULL);
+  EXPECT_EQ(V.capacity(), 0ULL);
+  EXPECT_EQ(V.materialised_begin().getIndex(), 0ULL);
+  EXPECT_EQ(V.materialised_end().getIndex(), 0ULL);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 0LL);
+}
+
+TEST(PagedVectorTest, ExpandTest) {
+  PagedVector<int, 10> V;
+  V.resize(2);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 2ULL);
+  EXPECT_EQ(V.capacity(), 10ULL);
+  EXPECT_EQ(V.materialised_begin().getIndex(), 2ULL);
+  EXPECT_EQ(V.materialised_end().getIndex(), 2ULL);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 0LL);
+}
+
+TEST(PagedVectorTest, FullPageFillingTest) {
+  PagedVector<int, 10> V;
+  V.resize(10);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 10ULL);
+  EXPECT_EQ(V.capacity(), 10ULL);
+  for (int I = 0; I < 10; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 10ULL);
+  EXPECT_EQ(V.capacity(), 10ULL);
+  EXPECT_EQ(V.materialised_begin().getIndex(), 0ULL);
+  EXPECT_EQ(V.materialised_end().getIndex(), 10ULL);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 10LL);
+  for (int I = 0; I < 10; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+}
+
+TEST(PagedVectorTest, HalfPageFillingTest) {
+  PagedVector<int, 10> V;
+  V.resize(5);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 5ULL);
+  EXPECT_EQ(V.capacity(), 10ULL);
+  for (int I = 0; I < 5; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 5LL);
+  for (int I = 0; I < 5; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  for (int I = 5; I < 10; ++I) {
+    EXPECT_DEATH(V[I], "Index < Size");
+  }
+}
+
+TEST(PagedVectorTest, FillFullMultiPageTest) {
+  PagedVector<int, 10> V;
+  V.resize(20);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 20ULL);
+  EXPECT_EQ(V.capacity(), 20ULL);
+  for (int I = 0; I < 20; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 20LL);
+  for (auto MI = V.materialised_begin(), ME = V.materialised_end(); MI != ME;
+       ++MI) {
+    EXPECT_EQ(*MI, std::distance(V.materialised_begin(), MI));
+  }
+}
+
+TEST(PagedVectorTest, FillHalfMultiPageTest) {
+  PagedVector<int, 10> V;
+  V.resize(20);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 20ULL);
+  EXPECT_EQ(V.capacity(), 20ULL);
+  for (int I = 0; I < 5; ++I) {
+    V[I] = I;
+  }
+  for (int I = 10; I < 15; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 20LL);
+  for (int I = 0; I < 5; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  for (int I = 10; I < 15; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+}
+
+TEST(PagedVectorTest, FillLastMultiPageTest) {
+  PagedVector<int, 10> V;
+  V.resize(20);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 20ULL);
+  EXPECT_EQ(V.capacity(), 20ULL);
+  for (int I = 10; I < 15; ++I) {
+    V[I] = I;
+  }
+  for (int I = 10; I < 15; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+
+  // Since we fill the last page only, the materialised vector
+  // should contain only the last page.
+  int J = 10;
+  for (auto MI = V.materialised_begin(), ME = V.materialised_end(); MI != ME;
+       ++MI) {
+    if (J < 15) {
+      EXPECT_EQ(*MI, J);
+    } else {
+      EXPECT_EQ(*MI, 0);
+    }
+    ++J;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 10LL);
+}
+
+// Filling the first element of all the pages
+// will allocate all of them
+TEST(PagedVectorTest, FillSparseMultiPageTest) {
+  PagedVector<int, 10> V;
+  V.resize(100);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 100ULL);
+  EXPECT_EQ(V.capacity(), 100ULL);
+  for (int I = 0; I < 10; ++I) {
+    V[I * 10] = I;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 100LL);
+  for (int I = 0; I < 100; ++I) {
+    if (I % 10 == 0) {
+      EXPECT_EQ(V[I], I / 10);
+    } else {
+      EXPECT_EQ(V[I], 0);
+    }
+  }
+}
+
+struct TestHelper {
+  int A = -1;
+};
+
+// Use this to count how many times the constructor / destructor are called
+struct TestHelper2 {
+  int A = -1;
+  static int constructed;
+  static int destroyed;
+
+  TestHelper2() { constructed++; }
+  ~TestHelper2() { destroyed++; }
+};
+
+int TestHelper2::constructed = 0;
+int TestHelper2::destroyed = 0;
+
+TEST(PagedVectorTest, FillNonTrivialConstructor) {
+  PagedVector<TestHelper, 10> V;
+  V.resize(10);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 10ULL);
+  EXPECT_EQ(V.capacity(), 10ULL);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 0LL);
+  for (int I = 0; I < 10; ++I) {
+    EXPECT_EQ(V[I].A, -1);
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 10LL);
+}
+
+// Elements are constructed, destructed in pages, so we expect
+// the number of constructed / destructed elements to be a multiple of the
+// page size and the constructor is invoked when the page is actually accessed
+// the first time.
+TEST(PagedVectorTest, FillNonTrivialConstructorDestructor) {
+  PagedVector<TestHelper2, 10> V;
+  V.resize(19);
+  EXPECT_EQ(TestHelper2::constructed, 0);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 19ULL);
+  EXPECT_EQ(V.capacity(), 20ULL);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 0LL);
+  EXPECT_EQ(V[0].A, -1);
+  EXPECT_EQ(TestHelper2::constructed, 10);
+
+  for (int I = 0; I < 10; ++I) {
+    EXPECT_EQ(V[I].A, -1);
+    EXPECT_EQ(TestHelper2::constructed, 10);
+  }
+  for (int I = 10; I < 11; ++I) {
+    EXPECT_EQ(V[I].A, -1);
+    EXPECT_EQ(TestHelper2::constructed, 20);
+  }
+  for (int I = 0; I < 19; ++I) {
+    EXPECT_EQ(V[I].A, -1);
+    EXPECT_EQ(TestHelper2::constructed, 20);
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 19LL);
+  // We initialise the whole page, not just the materialised part
+  // EXPECT_EQ(TestHelper2::constructed, 20);
+  V.resize(18);
+  EXPECT_EQ(TestHelper2::destroyed, 0);
+  V.resize(1);
+  EXPECT_EQ(TestHelper2::destroyed, 10);
+  V.resize(0);
+  EXPECT_EQ(TestHelper2::destroyed, 20);
+}
+
+TEST(PagedVectorTest, ShrinkTest) {
+  PagedVector<int, 10> V;
+  V.resize(20);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 20ULL);
+  EXPECT_EQ(V.capacity(), 20ULL);
+  for (int I = 0; I < 20; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 20LL);
+  V.resize(9);
+  EXPECT_EQ(V.empty(), false);
+  EXPECT_EQ(V.size(), 9ULL);
+  EXPECT_EQ(V.capacity(), 10ULL);
+  for (int I = 0; I < 9; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 9LL);
+  V.resize(0);
+  EXPECT_EQ(V.empty(), true);
+  EXPECT_EQ(V.size(), 0ULL);
+  EXPECT_EQ(V.capacity(), 0ULL);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 0LL);
+  EXPECT_DEATH(V[0], "Index < Size");
+}
+
+TEST(PagedVectorTest, FunctionalityTest) {
+  PagedVector<int, 10> V;
+  EXPECT_EQ(V.empty(), true);
+
+  // Next ten numbers are 10..19
+  V.resize(2);
+  EXPECT_EQ(V.empty(), false);
+  V.resize(10);
+  V.resize(20);
+  V.resize(30);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 0LL);
+
+  EXPECT_EQ(V.size(), 30ULL);
+  for (int I = 0; I < 10; ++I) {
+    V[I] = I;
+  }
+  for (int I = 0; I < 10; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 10LL);
+  for (int I = 20; I < 30; ++I) {
+    V[I] = I;
+  }
+  for (int I = 20; I < 30; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 20LL);
+
+  for (int I = 10; I < 20; ++I) {
+    V[I] = I;
+  }
+  for (int I = 10; I < 20; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 30LL);
+  V.resize(35);
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 30LL);
+  for (int I = 30; I < 35; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(std::distance(V.materialised_begin(), V.materialised_end()), 35LL);
+  EXPECT_EQ(V.size(), 35ULL);
+  EXPECT_EQ(V.capacity(), 40ULL);
+  V.resize(37);
+  for (int I = 30; I < 37; ++I) {
+    V[I] = I;
+  }
+  EXPECT_EQ(V.size(), 37ULL);
+  EXPECT_EQ(V.capacity(), 40ULL);
+  for (int I = 0; I < 37; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+
+  V.resize(41);
+  V[40] = 40;
+  EXPECT_EQ(V.size(), 41ULL);
+  EXPECT_EQ(V.capacity(), 50ULL);
+  for (int I = 0; I < 36; ++I) {
+    EXPECT_EQ(V[I], I);
+  }
+  for (int I = 37; I < 40; ++I) {
+    EXPECT_EQ(V[I], 0);
+  }
+  V.resize(50);
+  EXPECT_EQ(V.capacity(), 50ULL);
+  EXPECT_EQ(V.size(), 50ULL);
+  EXPECT_EQ(V[40], 40);
+  V.resize(50ULL);
+  V.clear();
+  EXPECT_EQ(V.size(), 0ULL);
+  EXPECT_EQ(V.capacity(), 0ULL);
+}
+} // namespace llvm



More information about the cfe-commits mailing list