[turbofan] Rework Node guts to save space.

This reduces the storage per-Node storage from 7 words to 6 and per-edge
storage from 6 words to 4.
On average this is about 10%-15% space savings over the whole graph.

Remove the use of std::deque as the out-of-line storage for inputs.
Reduce size of Use links and use pointer arithmetic to find Node
from Use.

R=mstarzinger@chromium.org,jarin@chromium.org
BUG=

Review URL: https://codereview.chromium.org/1150923003

Cr-Commit-Position: refs/heads/master@{#28583}

src/compiler/node.cc

@@ -4,37 +4,105 @@
 
 #include "src/compiler/node.h"
 
-#include <algorithm>
-
 namespace v8 {
 namespace internal {
 namespace compiler {
 
+Node::OutOfLineInputs* Node::OutOfLineInputs::New(Zone* zone, int capacity) {
+  size_t size =
+      sizeof(OutOfLineInputs) + capacity * (sizeof(Node*) + sizeof(Use));
+  intptr_t raw_buffer = reinterpret_cast<intptr_t>(zone->New(size));
+  Node::OutOfLineInputs* outline =
+      reinterpret_cast<OutOfLineInputs*>(raw_buffer + capacity * sizeof(Use));
+  outline->capacity_ = capacity;
+  outline->count_ = 0;
+  return outline;
+}
+
+
+void Node::OutOfLineInputs::ExtractFrom(Use* old_use_ptr, Node** old_input_ptr,
+                                        int count) {
+  // Extract the inputs from the old use and input pointers and copy them
+  // to this out-of-line-storage.
+  Use* new_use_ptr = reinterpret_cast<Use*>(this) - 1;
+  Node** new_input_ptr = inputs_;
+  for (int current = 0; current < count; current++) {
+    new_use_ptr->bit_field_ =
+        Use::InputIndexField::encode(current) | Use::InlineField::encode(false);
+    DCHECK_EQ(old_input_ptr, old_use_ptr->input_ptr());
+    DCHECK_EQ(new_input_ptr, new_use_ptr->input_ptr());
+    Node* old_to = *old_input_ptr;
+    if (old_to) {
+      *old_input_ptr = nullptr;
+      old_to->RemoveUse(old_use_ptr);
+      *new_input_ptr = old_to;
+      old_to->AppendUse(new_use_ptr);
+    } else {
+      *new_input_ptr = nullptr;
+    }
+    old_input_ptr++;
+    new_input_ptr++;
+    old_use_ptr--;
+    new_use_ptr--;
+  }
+  this->count_ = count;
+}
+
+
 Node* Node::New(Zone* zone, NodeId id, const Operator* op, int input_count,
                 Node** inputs, bool has_extensible_inputs) {
-  size_t node_size = sizeof(Node) - sizeof(Input);
-  int reserve_input_count = has_extensible_inputs ? kDefaultReservedInputs : 0;
-  size_t inputs_size = std::max<size_t>(
-      (input_count + reserve_input_count) * sizeof(Input), sizeof(InputDeque*));
-  size_t uses_size = input_count * sizeof(Use);
-  int size = static_cast<int>(node_size + inputs_size + uses_size);
-  void* buffer = zone->New(size);
-  Node* result = new (buffer) Node(id, op, input_count, reserve_input_count);
-  Input* input = result->inputs_.static_;
-  Use* use =
-      reinterpret_cast<Use*>(reinterpret_cast<char*>(input) + inputs_size);
+  Node** input_ptr;
+  Use* use_ptr;
+  Node* node;
+  bool is_inline;
+
+  if (input_count > kMaxInlineCapacity) {
+    // Allocate out-of-line inputs.
+    int capacity =
+        has_extensible_inputs ? input_count + kMaxInlineCapacity : input_count;
+    OutOfLineInputs* outline = OutOfLineInputs::New(zone, capacity);
+
+    // Allocate node.
+    void* node_buffer = zone->New(sizeof(Node));
+    node = new (node_buffer) Node(id, op, kOutlineMarker, 0);
+    node->inputs_.outline_ = outline;
+
+    outline->node_ = node;
+    outline->count_ = input_count;
+
+    input_ptr = outline->inputs_;
+    use_ptr = reinterpret_cast<Use*>(outline);
+    is_inline = false;
+  } else {
+    // Allocate node with inline inputs.
+    int capacity = input_count;
+    if (has_extensible_inputs) {
+      const int max = kMaxInlineCapacity;
+      capacity = std::min(input_count + 3, max);
+    }
+
+    size_t size = sizeof(Node) + capacity * (sizeof(Node*) + sizeof(Use));
+    intptr_t raw_buffer = reinterpret_cast<intptr_t>(zone->New(size));
+    void* node_buffer =
+        reinterpret_cast<void*>(raw_buffer + capacity * sizeof(Use));
+
+    node = new (node_buffer) Node(id, op, input_count, capacity);
+    input_ptr = node->inputs_.inline_;
+    use_ptr = reinterpret_cast<Use*>(node);
+    is_inline = true;
+  }
 
+  // Initialize the input pointers and the uses.
   for (int current = 0; current < input_count; ++current) {
     Node* to = *inputs++;
-    input->to = to;
-    input->use = use;
-    use->input_index = current;
-    use->from = result;
+    input_ptr[current] = to;
+    Use* use = use_ptr - 1 - current;
+    use->bit_field_ = Use::InputIndexField::encode(current) |
+                      Use::InlineField::encode(is_inline);
     to->AppendUse(use);
-    ++use;
-    ++input;
   }
-  return result;
+  node->Verify();
+  return node;
 }
 
 
@@ -48,22 +116,47 @@ void Node::Kill() {
 void Node::AppendInput(Zone* zone, Node* new_to) {
   DCHECK_NOT_NULL(zone);
   DCHECK_NOT_NULL(new_to);
-  Use* new_use = new (zone) Use;
-  Input new_input;
-  new_input.to = new_to;
-  new_input.use = new_use;
-  if (reserved_input_count() > 0) {
-    DCHECK(!has_appendable_inputs());
-    set_reserved_input_count(reserved_input_count() - 1);
-    inputs_.static_[input_count()] = new_input;
+
+  int inline_count = InlineCountField::decode(bit_field_);
+  int inline_capacity = InlineCapacityField::decode(bit_field_);
+  if (inline_count < inline_capacity) {
+    // Append inline input.
+    bit_field_ = InlineCountField::update(bit_field_, inline_count + 1);
+    *GetInputPtr(inline_count) = new_to;
+    Use* use = GetUsePtr(inline_count);
+    use->bit_field_ = Use::InputIndexField::encode(inline_count) |
+                      Use::InlineField::encode(true);
+    new_to->AppendUse(use);
   } else {
-    EnsureAppendableInputs(zone);
-    inputs_.appendable_->push_back(new_input);
+    // Append out-of-line input.
+    int input_count = InputCount();
+    OutOfLineInputs* outline = nullptr;
+    if (inline_count != kOutlineMarker) {
+      // switch to out of line inputs.
+      outline = OutOfLineInputs::New(zone, input_count * 2 + 3);
+      outline->node_ = this;
+      outline->ExtractFrom(GetUsePtr(0), GetInputPtr(0), input_count);
+      bit_field_ = InlineCountField::update(bit_field_, kOutlineMarker);
+      inputs_.outline_ = outline;
+    } else {
+      // use current out of line inputs.
+      outline = inputs_.outline_;
+      if (input_count >= outline->capacity_) {
+        // out of space in out-of-line inputs.
+        outline = OutOfLineInputs::New(zone, input_count * 2 + 3);
+        outline->node_ = this;
+        outline->ExtractFrom(GetUsePtr(0), GetInputPtr(0), input_count);
+        inputs_.outline_ = outline;
+      }
+    }
+    outline->count_++;
+    *GetInputPtr(input_count) = new_to;
+    Use* use = GetUsePtr(input_count);
+    use->bit_field_ = Use::InputIndexField::encode(input_count) |
+                      Use::InlineField::encode(false);
+    new_to->AppendUse(use);
   }
-  new_use->input_index = input_count();
-  new_use->from = this;
-  new_to->AppendUse(new_use);
-  set_input_count(input_count() + 1);
+  Verify();
 }
 
 
@@ -76,6 +169,7 @@ void Node::InsertInput(Zone* zone, int index, Node* new_to) {
     ReplaceInput(i, InputAt(i - 1));
   }
   ReplaceInput(index, new_to);
+  Verify();
 }
 
 
@@ -86,28 +180,38 @@ void Node::RemoveInput(int index) {
     ReplaceInput(index, InputAt(index + 1));
   }
   TrimInputCount(InputCount() - 1);
+  Verify();
 }
 
 
-void Node::NullAllInputs() {
-  for (Edge edge : input_edges()) edge.UpdateTo(nullptr);
+void Node::ClearInputs(int start, int count) {
+  Node** input_ptr = GetInputPtr(start);
+  Use* use_ptr = GetUsePtr(start);
+  while (count-- > 0) {
+    DCHECK_EQ(input_ptr, use_ptr->input_ptr());
+    Node* input = *input_ptr;
+    *input_ptr = nullptr;
+    if (input) input->RemoveUse(use_ptr);
+    input_ptr++;
+    use_ptr--;
+  }
+  Verify();
 }
 
 
+void Node::NullAllInputs() { ClearInputs(0, InputCount()); }
+
+
 void Node::TrimInputCount(int new_input_count) {
-  DCHECK_LE(new_input_count, input_count());
-  if (new_input_count == input_count()) return;  // Nothing to do.
-  for (int index = new_input_count; index < input_count(); ++index) {
-    ReplaceInput(index, nullptr);
-  }
-  if (has_appendable_inputs()) {
-    inputs_.appendable_->resize(new_input_count);
+  int current_count = InputCount();
+  DCHECK_LE(new_input_count, current_count);
+  if (new_input_count == current_count) return;  // Nothing to do.
+  ClearInputs(new_input_count, current_count - new_input_count);
+  if (has_inline_inputs()) {
+    bit_field_ = InlineCountField::update(bit_field_, new_input_count);
   } else {
-    set_reserved_input_count(std::min<int>(
-        ReservedInputCountField::kMax,
-        reserved_input_count() + (input_count() - new_input_count)));
+    inputs_.outline_->count_ = new_input_count;
   }
-  set_input_count(new_input_count);
 }
 
 
@@ -127,7 +231,7 @@ void Node::ReplaceUses(Node* that) {
   // Update the pointers to {this} to point to {that}.
   Use* last_use = nullptr;
   for (Use* use = this->first_use_; use; use = use->next) {
-    use->from->GetInputRecordPtr(use->input_index)->to = that;
+    *use->input_ptr() = that;
     last_use = use;
   }
   if (last_use) {
@@ -143,9 +247,10 @@ void Node::ReplaceUses(Node* that) {
 bool Node::OwnedBy(Node const* owner1, Node const* owner2) const {
   unsigned mask = 0;
   for (Use* use = first_use_; use; use = use->next) {
-    if (use->from == owner1) {
+    Node* from = use->from();
+    if (from == owner1) {
       mask |= 1;
-    } else if (use->from == owner2) {
+    } else if (from == owner2) {
       mask |= 2;
     } else {
       return false;
@@ -155,50 +260,21 @@ bool Node::OwnedBy(Node const* owner1, Node const* owner2) const {
 }
 
 
-void Node::Input::Update(Node* new_to) {
-  Node* old_to = this->to;
-  if (new_to == old_to) return;  // Nothing to do.
-  // Snip out the use from where it used to be
-  if (old_to) {
-    old_to->RemoveUse(use);
-  }
-  to = new_to;
-  // And put it into the new node's use list.
-  if (new_to) {
-    new_to->AppendUse(use);
-  } else {
-    use->next = nullptr;
-    use->prev = nullptr;
-  }
-}
-
-
-Node::Node(NodeId id, const Operator* op, int input_count,
-           int reserved_input_count)
+Node::Node(NodeId id, const Operator* op, int inline_count, int inline_capacity)
     : op_(op),
       mark_(0),
-      id_(id),
-      bit_field_(InputCountField::encode(input_count) |
-                 ReservedInputCountField::encode(reserved_input_count) |
-                 HasAppendableInputsField::encode(false)),
-      first_use_(nullptr) {}
-
-
-void Node::EnsureAppendableInputs(Zone* zone) {
-  if (!has_appendable_inputs()) {
-    void* deque_buffer = zone->New(sizeof(InputDeque));
-    InputDeque* deque = new (deque_buffer) InputDeque(zone);
-    for (int i = 0; i < input_count(); ++i) {
-      deque->push_back(inputs_.static_[i]);
-    }
-    inputs_.appendable_ = deque;
-    set_has_appendable_inputs(true);
-  }
+      bit_field_(IdField::encode(id) | InlineCountField::encode(inline_count) |
+                 InlineCapacityField::encode(inline_capacity)),
+      first_use_(nullptr) {
+  // Inputs must either be out of line or within the inline capacity.
+  DCHECK(inline_capacity <= kMaxInlineCapacity);
+  DCHECK(inline_count == kOutlineMarker || inline_count <= inline_capacity);
 }
 
 
-void Node::AppendUse(Use* const use) {
+void Node::AppendUse(Use* use) {
   DCHECK(first_use_ == nullptr || first_use_->prev == nullptr);
+  DCHECK_EQ(this, *use->input_ptr());
   use->next = first_use_;
   use->prev = nullptr;
   if (first_use_) first_use_->prev = use;
@@ -206,7 +282,7 @@ void Node::AppendUse(Use* const use) {
 }
 
 
-void Node::RemoveUse(Use* const use) {
+void Node::RemoveUse(Use* use) {
   DCHECK(first_use_ == nullptr || first_use_->prev == nullptr);
   if (use->prev) {
     DCHECK_NE(first_use_, use);
@@ -221,6 +297,44 @@ void Node::RemoveUse(Use* const use) {
 }
 
 
+#if DEBUG
+void Node::Verify() {
+  // Check basic sanity of input data structures.
+  fflush(stdout);
+  int count = this->InputCount();
+  // Avoid quadratic explosion for mega nodes; only verify if the input
+  // count is less than 200 or is a round number of 100s.
+  if (count > 200 && count % 100) return;
+
+  for (int i = 0; i < count; i++) {
+    CHECK_EQ(i, this->GetUsePtr(i)->input_index());
+    CHECK_EQ(this->GetInputPtr(i), this->GetUsePtr(i)->input_ptr());
+    CHECK_EQ(count, this->InputCount());
+  }
+  {  // Direct input iteration.
+    int index = 0;
+    for (Node* input : this->inputs()) {
+      CHECK_EQ(this->InputAt(index), input);
+      index++;
+    }
+    CHECK_EQ(count, index);
+    CHECK_EQ(this->InputCount(), index);
+  }
+  {  // Input edge iteration.
+    int index = 0;
+    for (Edge edge : this->input_edges()) {
+      CHECK_EQ(edge.from(), this);
+      CHECK_EQ(index, edge.index());
+      CHECK_EQ(this->InputAt(index), edge.to());
+      index++;
+    }
+    CHECK_EQ(count, index);
+    CHECK_EQ(this->InputCount(), index);
+  }
+}
+#endif
+
+
 std::ostream& operator<<(std::ostream& os, const Node& n) {
   os << n.id() << ": " << *n.op();
   if (n.InputCount() > 0) {

src/compiler/node.h

@@ -55,19 +55,49 @@ class Node final {
     return static_cast<IrOpcode::Value>(op_->opcode());
   }
 
-  NodeId id() const { return id_; }
+  NodeId id() const { return IdField::decode(bit_field_); }
+
+  int InputCount() const {
+    return has_inline_inputs() ? InlineCountField::decode(bit_field_)
+                               : inputs_.outline_->count_;
+  }
 
-  int InputCount() const { return input_count(); }
-  Node* InputAt(int index) const {
 #if DEBUG
-    if (index < 0 || index >= InputCount()) {
-      V8_Fatal(__FILE__, __LINE__, "Node #%d:%s->InputAt(%d) out of bounds",
-               id(), op()->mnemonic(), index);
-    }
+  void Verify();
+#define BOUNDS_CHECK(index)                                                  \
+  do {                                                                       \
+    if (index < 0 || index >= InputCount()) {                                \
+      V8_Fatal(__FILE__, __LINE__, "Node #%d:%s->InputAt(%d) out of bounds", \
+               id(), op()->mnemonic(), index);                               \
+    }                                                                        \
+  } while (false)
+#else
+  // No bounds checks or verification in release mode.
+  inline void Verify() {}
+#define BOUNDS_CHECK(index) \
+  do {                      \
+  } while (false)
 #endif
-    return GetInputRecordPtr(index)->to;
+
+  Node* InputAt(int index) const {
+    BOUNDS_CHECK(index);
+    return *GetInputPtrConst(index);
   }
-  inline void ReplaceInput(int index, Node* new_to);
+
+  void ReplaceInput(int index, Node* new_to) {
+    BOUNDS_CHECK(index);
+    Node** input_ptr = GetInputPtr(index);
+    Node* old_to = *input_ptr;
+    if (old_to != new_to) {
+      Use* use = GetUsePtr(index);
+      if (old_to) old_to->RemoveUse(use);
+      *input_ptr = new_to;
+      if (new_to) new_to->AppendUse(use);
+    }
+  }
+
+#undef BOUNDS_CHECK
+
   void AppendInput(Zone* zone, Node* new_to);
   void InsertInput(Zone* zone, int index, Node* new_to);
   void RemoveInput(int index);
@@ -151,49 +181,108 @@ class Node final {
 
   // Returns true if {owner} is the user of {this} node.
   bool OwnedBy(Node* owner) const {
-    return first_use_ && first_use_->from == owner && !first_use_->next;
+    return first_use_ && first_use_->from() == owner && !first_use_->next;
   }
 
   // Returns true if {owner1} and {owner2} are the only users of {this} node.
   bool OwnedBy(Node const* owner1, Node const* owner2) const;
 
  private:
-  struct Use final : public ZoneObject {
-    Node* from;
+  struct Use;
+  // Out of line storage for inputs when the number of inputs overflowed the
+  // capacity of the inline-allocated space.
+  struct OutOfLineInputs {
+    Node* node_;
+    int count_;
+    int capacity_;
+    Node* inputs_[1];
+
+    static OutOfLineInputs* New(Zone* zone, int capacity);
+    void ExtractFrom(Use* use_ptr, Node** input_ptr, int count);
+  };
+
+  // A link in the use chain for a node. Every input {i} to a node {n} has an
+  // associated {Use} which is linked into the use chain of the {i} node.
+  struct Use {
     Use* next;
     Use* prev;
-    int input_index;
-  };
+    uint32_t bit_field_;
+
+    int input_index() const { return InputIndexField::decode(bit_field_); }
+    int output_index() const { return OutputIndexField::decode(bit_field_); }
+    bool is_inline_use() const { return InlineField::decode(bit_field_); }
+    Node** input_ptr() {
+      int index = input_index();
+      Use* start = this + 1 + index;
+      Node** inputs = is_inline_use()
+                          ? reinterpret_cast<Node*>(start)->inputs_.inline_
+                          : reinterpret_cast<OutOfLineInputs*>(start)->inputs_;
+      return &inputs[index];
+    }
 
-  class Input final {
-   public:
-    Node* to;
-    Use* use;
+    Node* from() {
+      Use* start = this + 1 + input_index();
+      return is_inline_use() ? reinterpret_cast<Node*>(start)
+                             : reinterpret_cast<OutOfLineInputs*>(start)->node_;
+    }
 
-    void Update(Node* new_to);
+    typedef BitField<bool, 0, 1> InlineField;
+    typedef BitField<unsigned, 1, 17> InputIndexField;
+    typedef BitField<unsigned, 17, 14> OutputIndexField;
   };
 
-  inline Node(NodeId id, const Operator* op, int input_count,
-              int reserve_input_count);
-
-  inline void EnsureAppendableInputs(Zone* zone);
-
-  Input* GetInputRecordPtr(int index) {
-    return has_appendable_inputs() ? &((*inputs_.appendable_)[index])
-                                   : &inputs_.static_[index];
+  //============================================================================
+  //== Memory layout ===========================================================
+  //============================================================================
+  // Saving space for big graphs is important. We use a memory layout trick to
+  // be able to map {Node} objects to {Use} objects and vice-versa in a
+  // space-efficient manner.
+  //
+  // {Use} links are laid out in memory directly before a {Node}, followed by
+  // direct pointers to input {Nodes}.
+  //
+  // inline case:
+  // |Use #N  |Use #N-1|...|Use #1  |Use #0  |Node xxxx |I#0|I#1|...|I#N-1|I#N|
+  //          ^                              ^                  ^
+  //          + Use                          + Node             + Input
+  //
+  // Since every {Use} instance records its {input_index}, pointer arithmetic
+  // can compute the {Node}.
+  //
+  // out-of-line case:
+  //     |Node xxxx |
+  //     ^       + outline ------------------+
+  //     +----------------------------------------+
+  //                                         |    |
+  //                                         v    | node
+  // |Use #N  |Use #N-1|...|Use #1  |Use #0  |OOL xxxxx |I#0|I#1|...|I#N-1|I#N|
+  //          ^                                                 ^
+  //          + Use                                             + Input
+  //
+  // Out-of-line storage of input lists is needed if appending an input to
+  // a node exceeds the maximum inline capacity.
+
+  Node(NodeId id, const Operator* op, int inline_count, int inline_capacity);
+
+  Node* const* GetInputPtrConst(int input_index) const {
+    return has_inline_inputs() ? &(inputs_.inline_[input_index])
+                               : &inputs_.outline_->inputs_[input_index];
   }
-  const Input* GetInputRecordPtr(int index) const {
-    return has_appendable_inputs() ? &((*inputs_.appendable_)[index])
-                                   : &inputs_.static_[index];
+  Node** GetInputPtr(int input_index) {
+    return has_inline_inputs() ? &(inputs_.inline_[input_index])
+                               : &inputs_.outline_->inputs_[input_index];
+  }
+  Use* GetUsePtr(int input_index) {
+    Use* ptr = has_inline_inputs() ? reinterpret_cast<Use*>(this)
+                                   : reinterpret_cast<Use*>(inputs_.outline_);
+    return &ptr[-1 - input_index];
   }
 
-  inline void AppendUse(Use* const use);
-  inline void RemoveUse(Use* const use);
+  void AppendUse(Use* use);
+  void RemoveUse(Use* use);
 
   void* operator new(size_t, void* location) { return location; }
 
-  typedef ZoneDeque<Input> InputDeque;
-
   // Only NodeProperties should manipulate the bounds.
   Bounds bounds() { return bounds_; }
   void set_bounds(Bounds b) { bounds_ = b; }
@@ -202,47 +291,28 @@ class Node final {
   Mark mark() { return mark_; }
   void set_mark(Mark mark) { mark_ = mark; }
 
-  int input_count() const { return InputCountField::decode(bit_field_); }
-  void set_input_count(int input_count) {
-    DCHECK_LE(0, input_count);
-    bit_field_ = InputCountField::update(bit_field_, input_count);
-  }
-
-  int reserved_input_count() const {
-    return ReservedInputCountField::decode(bit_field_);
-  }
-  void set_reserved_input_count(int reserved_input_count) {
-    DCHECK_LE(0, reserved_input_count);
-    bit_field_ =
-        ReservedInputCountField::update(bit_field_, reserved_input_count);
+  inline bool has_inline_inputs() const {
+    return InlineCountField::decode(bit_field_) != kOutlineMarker;
   }
 
-  bool has_appendable_inputs() const {
-    return HasAppendableInputsField::decode(bit_field_);
-  }
-  void set_has_appendable_inputs(bool has_appendable_inputs) {
-    bit_field_ =
-        HasAppendableInputsField::update(bit_field_, has_appendable_inputs);
-  }
+  void ClearInputs(int start, int count);
 
-  typedef BitField<unsigned, 0, 29> InputCountField;
-  typedef BitField<unsigned, 29, 2> ReservedInputCountField;
-  typedef BitField<unsigned, 31, 1> HasAppendableInputsField;
-  static const int kDefaultReservedInputs = ReservedInputCountField::kMax;
+  typedef BitField<NodeId, 0, 24> IdField;
+  typedef BitField<unsigned, 24, 4> InlineCountField;
+  typedef BitField<unsigned, 28, 4> InlineCapacityField;
+  static const int kOutlineMarker = InlineCountField::kMax;
+  static const int kMaxInlineCount = InlineCountField::kMax - 1;
+  static const int kMaxInlineCapacity = InlineCapacityField::kMax - 1;
 
   const Operator* op_;
   Bounds bounds_;
   Mark mark_;
-  NodeId const id_;
-  unsigned bit_field_;
+  uint32_t bit_field_;
   Use* first_use_;
   union {
-    // When a node is initially allocated, it uses a static buffer to hold its
-    // inputs under the assumption that the number of outputs will not increase.
-    // When the first input is appended, the static buffer is converted into a
-    // deque to allow for space-efficient growing.
-    Input static_[1];
-    InputDeque* appendable_;
+    // Inline storage for inputs or out-of-line storage.
+    Node* inline_[1];
+    OutOfLineInputs* outline_;
   } inputs_;
 
   friend class Edge;
@@ -275,26 +345,38 @@ static inline const T& OpParameter(const Node* node) {
 // the node having the input.
 class Edge final {
  public:
-  Node* from() const { return input_->use->from; }
-  Node* to() const { return input_->to; }
+  Node* from() const { return use_->from(); }
+  Node* to() const { return *input_ptr_; }
   int index() const {
-    int const index = input_->use->input_index;
-    DCHECK_LT(index, input_->use->from->input_count());
+    int const index = use_->input_index();
+    DCHECK_LT(index, use_->from()->InputCount());
     return index;
   }
 
-  bool operator==(const Edge& other) { return input_ == other.input_; }
+  bool operator==(const Edge& other) { return input_ptr_ == other.input_ptr_; }
   bool operator!=(const Edge& other) { return !(*this == other); }
 
-  void UpdateTo(Node* new_to) { input_->Update(new_to); }
+  void UpdateTo(Node* new_to) {
+    Node* old_to = *input_ptr_;
+    if (old_to != new_to) {
+      if (old_to) old_to->RemoveUse(use_);
+      *input_ptr_ = new_to;
+      if (new_to) new_to->AppendUse(use_);
+    }
+  }
 
  private:
   friend class Node::UseEdges::iterator;
   friend class Node::InputEdges::iterator;
 
-  explicit Edge(Node::Input* input) : input_(input) { DCHECK_NOT_NULL(input); }
+  Edge(Node::Use* use, Node** input_ptr) : use_(use), input_ptr_(input_ptr) {
+    DCHECK_NOT_NULL(use);
+    DCHECK_NOT_NULL(input_ptr);
+    DCHECK_EQ(input_ptr, use->input_ptr());
+  }
 
-  Node::Input* input_;
+  Node::Use* use_;
+  Node** input_ptr_;
 };
 
 
@@ -307,16 +389,18 @@ class Node::InputEdges::iterator final {
   typedef Edge* pointer;
   typedef Edge& reference;
 
-  iterator() : input_(nullptr) {}
-  iterator(const iterator& other) : input_(other.input_) {}
+  iterator() : use_(nullptr), input_ptr_(nullptr) {}
+  iterator(const iterator& other)
+      : use_(other.use_), input_ptr_(other.input_ptr_) {}
 
-  Edge operator*() const { return Edge(input_); }
+  Edge operator*() const { return Edge(use_, input_ptr_); }
   bool operator==(const iterator& other) const {
-    return input_ == other.input_;
+    return input_ptr_ == other.input_ptr_;
   }
   bool operator!=(const iterator& other) const { return !(*this == other); }
   iterator& operator++() {
-    SetInput(Edge(input_).from(), input_->use->input_index + 1);
+    input_ptr_++;
+    use_--;
     return *this;
   }
   iterator operator++(int);
@@ -324,20 +408,11 @@ class Node::InputEdges::iterator final {
  private:
   friend class Node;
 
-  explicit iterator(Node* from, int index = 0) : input_(nullptr) {
-    SetInput(from, index);
-  }
-
-  void SetInput(Node* from, int index) {
-    DCHECK(index >= 0 && index <= from->InputCount());
-    if (index < from->InputCount()) {
-      input_ = from->GetInputRecordPtr(index);
-    } else {
-      input_ = nullptr;
-    }
-  }
+  explicit iterator(Node* from, int index = 0)
+      : use_(from->GetUsePtr(index)), input_ptr_(from->GetInputPtr(index)) {}
 
-  Input* input_;
+  Use* use_;
+  Node** input_ptr_;
 };
 
 
@@ -400,10 +475,7 @@ class Node::UseEdges::iterator final {
   iterator(const iterator& other)
       : current_(other.current_), next_(other.next_) {}
 
-  Edge operator*() const {
-    return Edge(current_->from->GetInputRecordPtr(current_->input_index));
-  }
-
+  Edge operator*() const { return Edge(current_, current_->input_ptr()); }
   bool operator==(const iterator& other) const {
     return current_ == other.current_;
   }
@@ -450,7 +522,7 @@ class Node::Uses::const_iterator final {
 
   const_iterator(const const_iterator& other) : current_(other.current_) {}
 
-  Node* operator*() const { return current_->from; }
+  Node* operator*() const { return current_->from(); }
   bool operator==(const const_iterator& other) const {
     return other.current_ == current_;
   }
@@ -481,11 +553,6 @@ Node::Uses::const_iterator Node::Uses::begin() const {
 
 Node::Uses::const_iterator Node::Uses::end() const { return const_iterator(); }
 
-
-void Node::ReplaceInput(int index, Node* new_to) {
-  GetInputRecordPtr(index)->Update(new_to);
-}
-
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8

test/cctest/compiler/test-control-reducer.cc

@@ -864,8 +864,8 @@ TEST(CMergeReduce_dead_rm1b) {
   Node* t = R.graph.NewNode(R.common.IfTrue(), R.start);
   Node* f = R.graph.NewNode(R.common.IfTrue(), R.start);
   for (int i = 0; i < 2; i++) {
-    Node* merge = R.graph.NewNode(R.common.Merge(3), R.dead, R.dead, R.dead);
     for (int j = i + 1; j < 3; j++) {
+      Node* merge = R.graph.NewNode(R.common.Merge(3), R.dead, R.dead, R.dead);
       merge->ReplaceInput(i, t);
       merge->ReplaceInput(j, f);
       R.ReduceMerge(merge, merge);

test/unittests/compiler/node-unittest.cc

@@ -233,6 +233,29 @@ TEST_F(NodeTest, TrimThenAppend) {
   EXPECT_THAT(node->inputs(), ElementsAre(n1, n3, n4, n5, n8, n9));
 }
 
+
+TEST_F(NodeTest, BigNodes) {
+  static const int kMaxSize = 512;
+  Node* inputs[kMaxSize];
+
+  Node* n0 = Node::New(zone(), 0, &kOp0, 0, nullptr, false);
+  Node* n1 = Node::New(zone(), 1, &kOp1, 1, &n0, false);
+
+  for (int i = 0; i < kMaxSize; i++) {
+    inputs[i] = i & 1 ? n0 : n1;
+  }
+
+  for (int size = 13; size <= kMaxSize; size += 9) {
+    Node* node = Node::New(zone(), 12345, &kOp0, size, inputs, false);
+    EXPECT_EQ(size, node->InputCount());
+
+    for (int i = 0; i < size; i++) {
+      EXPECT_EQ(inputs[i], node->InputAt(i));
+    }
+  }
+}
+
+
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8