Share "scan16" between Node16 and Node48

ConflictSet.cpp

@@ -236,11 +236,6 @@ constexpr int kNodeCopyBegin = offsetof(Node, entry);
 constexpr int kNodeCopySize =
     offsetof(Node, parentsIndex) + sizeof(Node::parentsIndex) - kNodeCopyBegin;
 
-struct Child {
-  int64_t childMaxVersion;
-  Node *child;
-};
-
 // copyChildrenAndKeyFrom is responsible for copying all
 // public members of Node, copying the partial key, logically copying the
 // children (converting representation if necessary), and updating all the
@@ -262,7 +257,8 @@ struct Node3 : Node {
   constexpr static auto kType = Type_Node3;
   // Sorted
   uint8_t index[kMaxNodes];
-  Child children[kMaxNodes];
+  Node *children[kMaxNodes];
+  int64_t childMaxVersion[kMaxNodes];
   uint8_t *partialKey() { return (uint8_t *)(this + 1); }
 
   void copyChildrenAndKeyFrom(const Node0 &other);
@@ -277,7 +273,8 @@ struct Node16 : Node {
   constexpr static auto kMaxNodes = 16;
   // Sorted
   uint8_t index[kMaxNodes];
-  Child children[kMaxNodes];
+  Node *children[kMaxNodes];
+  int64_t childMaxVersion[kMaxNodes];
   uint8_t *partialKey() { return (uint8_t *)(this + 1); }
 
   void copyChildrenAndKeyFrom(const Node3 &other);
@@ -293,9 +290,10 @@ struct Node48 : Node {
   BitSet bitSet;
   int8_t nextFree;
   int8_t index[256];
-  Child children[kMaxNodes];
+  Node *children[kMaxNodes];
+  int64_t childMaxVersion[kMaxNodes];
   uint8_t reverseIndex[kMaxNodes];
-  constexpr static int kMaxOfMaxPageSize = 8;
+  constexpr static int kMaxOfMaxPageSize = 16;
   constexpr static int kMaxOfMaxShift =
       std::countr_zero(uint32_t(kMaxOfMaxPageSize));
   constexpr static int kMaxOfMaxTotalPages = kMaxNodes / kMaxOfMaxPageSize;
@@ -313,7 +311,8 @@ struct Node48 : Node {
 struct Node256 : Node {
   constexpr static auto kType = Type_Node256;
   BitSet bitSet;
-  Child children[256];
+  Node *children[256];
+  int64_t childMaxVersion[256];
   constexpr static int kMaxOfMaxPageSize = 8;
   constexpr static int kMaxOfMaxShift =
       std::countr_zero(uint32_t(kMaxOfMaxPageSize));
@@ -350,8 +349,8 @@ inline void Node3::copyChildrenAndKeyFrom(const Node3 &other) {
   memcpy(index, other.index, sizeof(*this) - sizeof(Node));
   memcpy(partialKey(), &other + 1, partialKeyLen);
   for (int i = 0; i < numChildren; ++i) {
-    assert(children[i].child->parent == &other);
-    children[i].child->parent = this;
+    assert(children[i]->parent == &other);
+    children[i]->parent = this;
   }
 }
 
@@ -359,11 +358,13 @@ inline void Node3::copyChildrenAndKeyFrom(const Node16 &other) {
   memcpy((char *)this + kNodeCopyBegin, (char *)&other + kNodeCopyBegin,
          kNodeCopySize);
   memcpy(index, other.index, kMaxNodes);
-  memcpy(children, other.children, kMaxNodes * sizeof(Child));
+  memcpy(children, other.children, kMaxNodes * sizeof(children[0])); // NOLINT
+  memcpy(childMaxVersion, other.childMaxVersion,
+         kMaxNodes * sizeof(childMaxVersion[0]));
   memcpy(partialKey(), &other + 1, partialKeyLen);
   for (int i = 0; i < numChildren; ++i) {
-    assert(children[i].child->parent == &other);
-    children[i].child->parent = this;
+    assert(children[i]->parent == &other);
+    children[i]->parent = this;
   }
 }
 
@@ -371,12 +372,15 @@ inline void Node16::copyChildrenAndKeyFrom(const Node3 &other) {
   memcpy((char *)this + kNodeCopyBegin, (char *)&other + kNodeCopyBegin,
          kNodeCopySize);
   memcpy(index, other.index, Node3::kMaxNodes);
-  memcpy(children, other.children, Node3::kMaxNodes * sizeof(Child));
+  memcpy(children, other.children,
+         Node3::kMaxNodes * sizeof(children[0])); // NOLINT
+  memcpy(childMaxVersion, other.childMaxVersion,
+         Node3::kMaxNodes * sizeof(childMaxVersion[0]));
   memcpy(partialKey(), &other + 1, partialKeyLen);
   assert(numChildren == Node3::kMaxNodes);
   for (int i = 0; i < Node3::kMaxNodes; ++i) {
-    assert(children[i].child->parent == &other);
-    children[i].child->parent = this;
+    assert(children[i]->parent == &other);
+    children[i]->parent = this;
   }
 }
 
@@ -386,8 +390,9 @@ inline void Node16::copyChildrenAndKeyFrom(const Node16 &other) {
   memcpy(index, other.index, sizeof(index));
   for (int i = 0; i < numChildren; ++i) {
     children[i] = other.children[i];
-    assert(children[i].child->parent == &other);
-    children[i].child->parent = this;
+    childMaxVersion[i] = other.childMaxVersion[i];
+    assert(children[i]->parent == &other);
+    children[i]->parent = this;
   }
   memcpy(partialKey(), &other + 1, partialKeyLen);
 }
@@ -403,8 +408,9 @@ inline void Node16::copyChildrenAndKeyFrom(const Node48 &other) {
         assume(i < Node16::kMaxNodes);
         index[i] = c;
         children[i] = other.children[other.index[c]];
-        assert(children[i].child->parent == &other);
-        children[i].child->parent = this;
+        childMaxVersion[i] = other.childMaxVersion[other.index[c]];
+        assert(children[i]->parent == &other);
+        children[i]->parent = this;
         ++i;
       },
       0, 256);
@@ -417,6 +423,7 @@ inline void Node48::copyChildrenAndKeyFrom(const Node16 &other) {
   assert(numChildren == Node16::kMaxNodes);
   memset(index, -1, sizeof(index));
   memset(children, 0, sizeof(children));
+  memset(childMaxVersion, 0, sizeof(childMaxVersion));
   memcpy(partialKey(), &other + 1, partialKeyLen);
   bitSet.init();
   nextFree = Node16::kMaxNodes;
@@ -425,11 +432,12 @@ inline void Node48::copyChildrenAndKeyFrom(const Node16 &other) {
     bitSet.set(x);
     index[x] = i;
     children[i] = other.children[i];
-    assert(children[i].child->parent == &other);
-    children[i].child->parent = this;
+    childMaxVersion[i] = other.childMaxVersion[i];
+    assert(children[i]->parent == &other);
+    children[i]->parent = this;
     reverseIndex[i] = x;
-    maxOfMax[i >> Node48::kMaxOfMaxShift] = std::max(
-        maxOfMax[i >> Node48::kMaxOfMaxShift], children[i].childMaxVersion);
+    maxOfMax[i >> Node48::kMaxOfMaxShift] =
+        std::max(maxOfMax[i >> Node48::kMaxOfMaxShift], childMaxVersion[i]);
     ++i;
   }
 }
@@ -441,10 +449,12 @@ inline void Node48::copyChildrenAndKeyFrom(const Node48 &other) {
   nextFree = other.nextFree;
   memcpy(index, other.index, sizeof(index));
   memset(children, 0, sizeof(children));
+  memset(childMaxVersion, 0, sizeof(childMaxVersion));
   for (int i = 0; i < numChildren; ++i) {
     children[i] = other.children[i];
-    assert(children[i].child->parent == &other);
-    children[i].child->parent = this;
+    childMaxVersion[i] = other.childMaxVersion[i];
+    assert(children[i]->parent == &other);
+    children[i]->parent = this;
   }
   memcpy(reverseIndex, other.reverseIndex, sizeof(reverseIndex));
   memcpy(maxOfMax, other.maxOfMax, sizeof(maxOfMax));
@@ -456,6 +466,7 @@ inline void Node48::copyChildrenAndKeyFrom(const Node256 &other) {
          kNodeCopySize);
   memset(index, -1, sizeof(index));
   memset(children, 0, sizeof(children));
+  memset(childMaxVersion, 0, sizeof(childMaxVersion));
   nextFree = other.numChildren;
   bitSet = other.bitSet;
   int i = 0;
@@ -466,11 +477,12 @@ inline void Node48::copyChildrenAndKeyFrom(const Node256 &other) {
         assume(i < Node48::kMaxNodes);
         index[c] = i;
         children[i] = other.children[c];
-        assert(children[i].child->parent == &other);
-        children[i].child->parent = this;
+        childMaxVersion[i] = other.childMaxVersion[c];
+        assert(children[i]->parent == &other);
+        children[i]->parent = this;
         reverseIndex[i] = c;
-        maxOfMax[i >> Node48::kMaxOfMaxShift] = std::max(
-            maxOfMax[i >> Node48::kMaxOfMaxShift], children[i].childMaxVersion);
+        maxOfMax[i >> Node48::kMaxOfMaxShift] =
+            std::max(maxOfMax[i >> Node48::kMaxOfMaxShift], childMaxVersion[i]);
         ++i;
       },
       0, 256);
@@ -482,15 +494,16 @@ inline void Node256::copyChildrenAndKeyFrom(const Node48 &other) {
          kNodeCopySize);
   bitSet = other.bitSet;
   memset(children, 0, sizeof(children));
+  memset(childMaxVersion, 0, sizeof(childMaxVersion));
   memset(maxOfMax, 0, sizeof(maxOfMax));
   bitSet.forEachInRange(
       [&](int c) {
         children[c] = other.children[other.index[c]];
-        assert(children[c].child->parent == &other);
-        children[c].child->parent = this;
-        maxOfMax[c >> Node256::kMaxOfMaxShift] =
-            std::max(maxOfMax[c >> Node256::kMaxOfMaxShift],
-                     children[c].childMaxVersion);
+        childMaxVersion[c] = other.childMaxVersion[other.index[c]];
+        assert(children[c]->parent == &other);
+        children[c]->parent = this;
+        maxOfMax[c >> Node256::kMaxOfMaxShift] = std::max(
+            maxOfMax[c >> Node256::kMaxOfMaxShift], childMaxVersion[c]);
       },
       0, 256);
   memcpy(partialKey(), &other + 1, partialKeyLen);
@@ -500,12 +513,14 @@ inline void Node256::copyChildrenAndKeyFrom(const Node256 &other) {
   memcpy((char *)this + kNodeCopyBegin, (char *)&other + kNodeCopyBegin,
          kNodeCopySize);
   memset(children, 0, sizeof(children));
+  memset(childMaxVersion, 0, sizeof(childMaxVersion));
   bitSet = other.bitSet;
   bitSet.forEachInRange(
       [&](int c) {
         children[c] = other.children[c];
-        assert(children[c].child->parent == &other);
-        children[c].child->parent = this;
+        childMaxVersion[c] = other.childMaxVersion[c];
+        assert(children[c]->parent == &other);
+        children[c]->parent = this;
       },
       0, 256);
   memcpy(maxOfMax, other.maxOfMax, sizeof(maxOfMax));
@@ -592,6 +607,7 @@ template <class T> struct BoundedFreeListAllocator {
     T *result = allocate_helper(partialKeyCapacity);
     if constexpr (!std::is_same_v<T, Node0>) {
       memset(result->children, 0, sizeof(result->children));
+      memset(result->childMaxVersion, 0, sizeof(result->childMaxVersion));
     }
     if constexpr (std::is_same_v<T, Node48> || std::is_same_v<T, Node256>) {
       memset(result->maxOfMax, 0, sizeof(result->maxOfMax));
@@ -750,21 +766,21 @@ Node *&getChildExists(Node *self, uint8_t index) {
     __builtin_unreachable(); // GCOVR_EXCL_LINE
   case Type_Node3: {
     auto *self3 = static_cast<Node3 *>(self);
-    return self3->children[getNodeIndex(self3, index)].child;
+    return self3->children[getNodeIndex(self3, index)];
   }
   case Type_Node16: {
     auto *self16 = static_cast<Node16 *>(self);
-    return self16->children[getNodeIndex(self16, index)].child;
+    return self16->children[getNodeIndex(self16, index)];
   }
   case Type_Node48: {
     auto *self48 = static_cast<Node48 *>(self);
     assert(self48->bitSet.test(index));
-    return self48->children[self48->index[index]].child;
+    return self48->children[self48->index[index]];
   }
   case Type_Node256: {
     auto *self256 = static_cast<Node256 *>(self);
     assert(self256->bitSet.test(index));
-    return self256->children[index].child;
+    return self256->children[index];
   }
   default:                   // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
@@ -784,21 +800,21 @@ Node *getChild(Node *self, uint8_t index) {
   case Type_Node3: {
     auto *self3 = static_cast<Node3 *>(self);
     int i = getNodeIndex(self3, index);
-    return i < 0 ? nullptr : self3->children[i].child;
+    return i < 0 ? nullptr : self3->children[i];
   }
   case Type_Node16: {
     auto *self16 = static_cast<Node16 *>(self);
     int i = getNodeIndex(self16, index);
-    return i < 0 ? nullptr : self16->children[i].child;
+    return i < 0 ? nullptr : self16->children[i];
   }
   case Type_Node48: {
     auto *self48 = static_cast<Node48 *>(self);
     int i = self48->index[index];
-    return i < 0 ? nullptr : self48->children[i].child;
+    return i < 0 ? nullptr : self48->children[i];
   }
   case Type_Node256: {
     auto *self256 = static_cast<Node256 *>(self);
-    return self256->children[index].child;
+    return self256->children[index];
   }
   default:                   // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
@@ -914,26 +930,26 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     auto *self3 = static_cast<Node3 *>(self);
     int i = getNodeIndex(self3, index);
     if (i >= 0) {
-      return self3->children[i].child;
+      return self3->children[i];
     }
   } break;
   case Type_Node16: {
     auto *self16 = static_cast<Node16 *>(self);
     int i = getNodeIndex(self16, index);
     if (i >= 0) {
-      return self16->children[i].child;
+      return self16->children[i];
     }
   } break;
   case Type_Node48: {
     auto *self48 = static_cast<Node48 *>(self);
     int secondIndex = self48->index[index];
     if (secondIndex >= 0) {
-      return self48->children[secondIndex].child;
+      return self48->children[secondIndex];
     }
   } break;
   case Type_Node256: {
     auto *self256 = static_cast<Node256 *>(self);
-    if (auto &result = self256->children[index].child; result != nullptr) {
+    if (auto &result = self256->children[index]; result != nullptr) {
       return result;
     }
   } break;
@@ -971,12 +987,16 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
         memmove(self3->index + i + 1, self3->index + i,
                 self->numChildren - (i + 1));
         memmove(self3->children + i + 1, self3->children + i,
-                (self->numChildren - (i + 1)) * sizeof(Child));
+                (self->numChildren - (i + 1)) *
+                    sizeof(self3->children[0])); // NOLINT
+        memmove(self3->childMaxVersion + i + 1, self3->childMaxVersion + i,
+                (self->numChildren - (i + 1)) *
+                    sizeof(self3->childMaxVersion[0]));
         break;
       }
     }
     self3->index[i] = index;
-    auto &result = self3->children[i].child;
+    auto &result = self3->children[i];
     result = nullptr;
     return result;
   }
@@ -1008,7 +1028,10 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
       memmove(self16->index + i + 1, self16->index + i,
               self->numChildren - (i + 1));
       memmove(self16->children + i + 1, self16->children + i,
-              (self->numChildren - (i + 1)) * sizeof(Child));
+              (self->numChildren - (i + 1)) * sizeof(self16->children[0]));
+      memmove(self16->childMaxVersion + i + 1, self16->childMaxVersion + i,
+              (self->numChildren - (i + 1)) *
+                  sizeof(self16->childMaxVersion[0]));
     }
 #elif defined(HAS_ARM_NEON)
     uint8x16_t indices;
@@ -1028,7 +1051,11 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
       memmove(self16->index + i + 1, self16->index + i,
               self->numChildren - (i + 1));
       memmove(self16->children + i + 1, self16->children + i,
-              (self->numChildren - (i + 1)) * sizeof(Child));
+              (self->numChildren - (i + 1)) *
+                  sizeof(self16->children[0])); // NOLINT
+      memmove(self16->childMaxVersion + i + 1, self16->childMaxVersion + i,
+              (self->numChildren - (i + 1)) *
+                  sizeof(self16->childMaxVersion[0]));
     }
 #else
     int i = 0;
@@ -1037,13 +1064,16 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
         memmove(self16->index + i + 1, self16->index + i,
                 self->numChildren - (i + 1));
         memmove(self16->children + i + 1, self16->children + i,
-                (self->numChildren - (i + 1)) * sizeof(Child));
+                (self->numChildren - (i + 1)) * sizeof(self16->children[0]));
+        memmove(self16->childMaxVersion + i + 1, self16->childMaxVersion + i,
+                (self->numChildren - (i + 1)) *
+                    sizeof(self16->childMaxVersion[0]));
         break;
       }
     }
 #endif
     self16->index[i] = index;
-    auto &result = self16->children[i].child;
+    auto &result = self16->children[i];
     result = nullptr;
     return result;
   }
@@ -1066,7 +1096,7 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     int nextFree = self48->nextFree++;
     self48->index[index] = nextFree;
     self48->reverseIndex[nextFree] = index;
-    auto &result = self48->children[nextFree].child;
+    auto &result = self48->children[nextFree];
     result = nullptr;
     return result;
   }
@@ -1076,7 +1106,7 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     auto *self256 = static_cast<Node256 *>(self);
     ++self->numChildren;
     self256->bitSet.set(index);
-    return self256->children[index].child;
+    return self256->children[index];
   }
   default:                   // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
@@ -1219,7 +1249,7 @@ void maybeDownsize(Node *self, NodeAllocators *allocators,
       getInTree(self, impl) = newSelf;
       allocators->node3.release(self3);
     } else if (self->numChildren == 1 && !self->entryPresent) {
-      auto *child = self3->children[0].child;
+      auto *child = self3->children[0];
       int minCapacity = self3->partialKeyLen + 1 + child->partialKeyLen;
 
       if (minCapacity > child->getCapacity()) {
@@ -1332,7 +1362,11 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
             sizeof(parent3->index[0]) *
                 (parent->numChildren - (nodeIndex + 1)));
     memmove(parent3->children + nodeIndex, parent3->children + nodeIndex + 1,
-            sizeof(parent3->children[0]) *
+            sizeof(parent3->children[0]) * // NOLINT
+                (parent->numChildren - (nodeIndex + 1)));
+    memmove(parent3->childMaxVersion + nodeIndex,
+            parent3->childMaxVersion + nodeIndex + 1,
+            sizeof(parent3->childMaxVersion[0]) *
                 (parent->numChildren - (nodeIndex + 1)));
 
     --parent->numChildren;
@@ -1346,7 +1380,11 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
             sizeof(parent16->index[0]) *
                 (parent->numChildren - (nodeIndex + 1)));
     memmove(parent16->children + nodeIndex, parent16->children + nodeIndex + 1,
-            sizeof(parent16->children[0]) *
+            sizeof(parent16->children[0]) * // NOLINT
+                (parent->numChildren - (nodeIndex + 1)));
+    memmove(parent16->childMaxVersion + nodeIndex,
+            parent16->childMaxVersion + nodeIndex + 1,
+            sizeof(parent16->childMaxVersion[0]) *
                 (parent->numChildren - (nodeIndex + 1)));
 
     --parent->numChildren;
@@ -1365,16 +1403,18 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
     if (toRemoveChildrenIndex != lastChildrenIndex) {
       parent48->children[toRemoveChildrenIndex] =
           parent48->children[lastChildrenIndex];
+      parent48->childMaxVersion[toRemoveChildrenIndex] =
+          parent48->childMaxVersion[lastChildrenIndex];
       parent48->maxOfMax[toRemoveChildrenIndex >> Node48::kMaxOfMaxShift] =
           std::max(parent48->maxOfMax[toRemoveChildrenIndex >>
                                       Node48::kMaxOfMaxShift],
-                   parent48->children[toRemoveChildrenIndex].childMaxVersion);
+                   parent48->childMaxVersion[toRemoveChildrenIndex]);
       auto parentIndex =
-          parent48->children[toRemoveChildrenIndex].child->parentsIndex;
+          parent48->children[toRemoveChildrenIndex]->parentsIndex;
       parent48->index[parentIndex] = toRemoveChildrenIndex;
       parent48->reverseIndex[toRemoveChildrenIndex] = parentIndex;
     }
-    parent48->children[lastChildrenIndex].childMaxVersion = 0;
+    parent48->childMaxVersion[lastChildrenIndex] = 0;
 
     --parent->numChildren;
 
@@ -1384,7 +1424,7 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
   case Type_Node256: {
     auto *parent256 = static_cast<Node256 *>(parent);
     parent256->bitSet.reset(parentsIndex);
-    parent256->children[parentsIndex].child = nullptr;
+    parent256->children[parentsIndex] = nullptr;
 
     --parent->numChildren;
 
@@ -1669,6 +1709,47 @@ downLeftSpine:
   }
 }
 
+// Returns true if all in-bounds vs are <= readVersion
+bool scan16(const int64_t *vs, const uint8_t *is, int begin, int end,
+            int64_t readVersion) {
+
+#ifdef HAS_ARM_NEON
+
+  assert(end - begin < 256);
+  uint8x16_t indices;
+  memcpy(&indices, is, 16);
+  // 0xff for each in bounds
+  auto results =
+      vcltq_u8(vsubq_u8(indices, vdupq_n_u8(begin)), vdupq_n_u8(end - begin));
+  // 0xf for each 0xff
+  uint64_t mask = vget_lane_u64(
+      vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(results), 4)), 0);
+
+  uint64_t compared = 0;
+  for (int i = 0; i < 16; ++i) {
+    compared |= (uint64_t(vs[i] > readVersion) << (i << 2));
+  }
+  return !(compared & mask);
+
+#else
+
+  const unsigned shiftUpperBound = end - begin;
+  const unsigned shiftAmount = begin;
+  auto inBounds = [&](unsigned c) { return c - shiftAmount < shiftUpperBound; };
+
+  uint32_t compared = 0;
+  for (int i = 0; i < 16; ++i) {
+    compared |= ((vs[i] > readVersion) << i);
+  }
+  uint32_t mask = 0;
+  for (int i = 0; i < 16; ++i) {
+    mask |= inBounds(is[i]) << i;
+  }
+  return !(compared & mask);
+
+#endif
+}
+
 // Return whether or not the max version among all keys starting with the search
 // path of n + [child], where child in (begin, end) is <= readVersion. Does not
 // account for the range version of firstGt(searchpath(n) + [end - 1])
@@ -1679,6 +1760,9 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
   assume(-1 <= end);
   assume(end <= 256);
   assume(begin < end);
+
+  assert(!(begin == -1 && end == 256));
+
   {
     int c = getChildGeq(n, begin + 1);
     if (c >= 0 && c < end) {
@@ -1708,34 +1792,31 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
     auto *self = static_cast<Node3 *>(n);
     bool result = true;
     for (int i = 0; i < 3; ++i) {
-      result &= !((self->children[i].childMaxVersion > readVersion) &
+      result &= !((self->childMaxVersion[i] > readVersion) &
                   inBounds(self->index[i]));
     }
     return result;
   } break;
   case Type_Node16: {
     auto *self = static_cast<Node16 *>(n);
-    bool result = true;
-    for (int i = 0; i < 16; ++i) {
-      result &= !((self->children[i].childMaxVersion > readVersion) &
-                  inBounds(self->index[i]));
-    }
-    return result;
+
+    return scan16(self->childMaxVersion, self->index, begin, end, readVersion);
+
   } break;
   case Type_Node48: {
     auto *self = static_cast<Node48 *>(n);
     // Check all pages
-    bool result = true;
+    static_assert(Node48::kMaxOfMaxPageSize == 16);
     for (int i = 0; i < Node48::kMaxOfMaxTotalPages; ++i) {
       if (self->maxOfMax[i] > readVersion) {
-        for (int j = 0; j < Node48::kMaxOfMaxPageSize; ++j) {
-          int k = (i << Node48::kMaxOfMaxShift) + j;
-          result &= !((self->children[k].childMaxVersion > readVersion) &
-                      inBounds(self->reverseIndex[k]));
+        if (!scan16(self->childMaxVersion + (i << Node48::kMaxOfMaxShift),
+                    self->reverseIndex + (i << Node48::kMaxOfMaxShift), begin,
+                    end, readVersion)) {
+          return false;
         }
       }
     }
-    return result;
+    return true;
   }
   case Type_Node256: {
     auto *self = static_cast<Node256 *>(n);
@@ -1744,8 +1825,7 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
       bool result = true;
       for (int i = 0; i < Node256::kMaxOfMaxPageSize; ++i) {
         int j = (begin & ~(Node256::kMaxOfMaxPageSize - 1)) + i;
-        result &=
-            !((self->children[j].childMaxVersion > readVersion) & inBounds(j));
+        result &= !((self->childMaxVersion[j] > readVersion) & inBounds(j));
       }
       if (!result) {
         return result;
@@ -1757,8 +1837,7 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
       bool result = true;
       for (int i = 0; i < Node256::kMaxOfMaxPageSize; ++i) {
         int j = ((end - 1) & ~(Node256::kMaxOfMaxPageSize - 1)) + i;
-        result &=
-            !((self->children[j].childMaxVersion > readVersion) & inBounds(j));
+        result &= !((self->childMaxVersion[j] > readVersion) & inBounds(j));
       }
       if (!result) {
         return result;
@@ -2649,22 +2728,22 @@ int64_t maxVersion(Node *n, ConflictSet::Impl *impl) {
   case Type_Node3: {
     auto *n3 = static_cast<Node3 *>(n);
     int i = getNodeIndex(n3, index);
-    return n3->children[i].childMaxVersion;
+    return n3->childMaxVersion[i];
   }
   case Type_Node16: {
     auto *n16 = static_cast<Node16 *>(n);
     int i = getNodeIndex(n16, index);
-    return n16->children[i].childMaxVersion;
+    return n16->childMaxVersion[i];
   }
   case Type_Node48: {
     auto *n48 = static_cast<Node48 *>(n);
     assert(n48->bitSet.test(index));
-    return n48->children[n48->index[index]].childMaxVersion;
+    return n48->childMaxVersion[n48->index[index]];
   }
   case Type_Node256: {
     auto *n256 = static_cast<Node256 *>(n);
     assert(n256->bitSet.test(index));
-    return n256->children[index].childMaxVersion;
+    return n256->childMaxVersion[index];
   }
   default:                   // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
@@ -2684,20 +2763,20 @@ void setMaxVersion(Node *n, ConflictSet::Impl *impl, int64_t newMax) {
   case Type_Node3: {
     auto *n3 = static_cast<Node3 *>(n);
     int i = getNodeIndex(n3, index);
-    n3->children[i].childMaxVersion = newMax;
+    n3->childMaxVersion[i] = newMax;
     return;
   }
   case Type_Node16: {
     auto *n16 = static_cast<Node16 *>(n);
     int i = getNodeIndex(n16, index);
-    n16->children[i].childMaxVersion = newMax;
+    n16->childMaxVersion[i] = newMax;
     return;
   }
   case Type_Node48: {
     auto *n48 = static_cast<Node48 *>(n);
     assert(n48->bitSet.test(index));
     int i = n48->index[index];
-    n48->children[i].childMaxVersion = newMax;
+    n48->childMaxVersion[i] = newMax;
     n48->maxOfMax[i >> Node48::kMaxOfMaxShift] =
         std::max(n48->maxOfMax[i >> Node48::kMaxOfMaxShift], newMax);
     return;
@@ -2705,7 +2784,7 @@ void setMaxVersion(Node *n, ConflictSet::Impl *impl, int64_t newMax) {
   case Type_Node256: {
     auto *n256 = static_cast<Node256 *>(n);
     assert(n256->bitSet.test(index));
-    n256->children[index].childMaxVersion = newMax;
+    n256->childMaxVersion[index] = newMax;
     n256->maxOfMax[index >> Node256::kMaxOfMaxShift] =
         std::max(n256->maxOfMax[index >> Node256::kMaxOfMaxShift], newMax);
     return;