[MC] Replace fragment ilist with singly-linked lists

Fragments are allocated with `operator new` and stored in an ilist with
Prev/Next/Parent pointers. A more efficient representation would be an
array of fragments without the overhead of Prev/Next pointers.

As the first step, replace ilist with singly-linked lists.

* `getPrevNode` uses have been eliminated by previous changes.
* The last use of the `Prev` pointer remains: for each subsection, there is an insertion point and
  the current insertion point is stored at `CurInsertionPoint`.
* `HexagonAsmBackend::finishLayout` needs a backward iterator. Save all
  fragments within `Frags`. Hexagon programs are usually small, and the
  performance does not matter that much.

To eliminate `Prev`, change the subsection representation to
singly-linked lists for subsections and a pointer to the active
singly-linked list. The fragments from all subsections will be chained
together at layout time.

Since fragment lists are disconnected before layout time, we can remove
`MCFragment::SubsectionNumber` (https://reviews.llvm.org/D69411). The
current implementation of `AttemptToFoldSymbolOffsetDifference` requires
future improvement for robustness.

Pull Request: #95077

Author: MaskRay

llvm/include/llvm/MC/MCFragment.h

@@ -23,12 +23,15 @@
 
 namespace llvm {
 
+class MCAssembler;
 class MCSection;
 class MCSubtargetInfo;
 class MCSymbol;
 
-class MCFragment : public ilist_node_with_parent<MCFragment, MCSection> {
+class MCFragment {
   friend class MCAsmLayout;
+  friend class MCAssembler;
+  friend class MCSection;
 
 public:
   enum FragmentType : uint8_t {
@@ -51,6 +54,9 @@ class MCFragment : public ilist_node_with_parent<MCFragment, MCSection> {
   };
 
 private:
+  // The next fragment within the section.
+  MCFragment *Next = nullptr;
+
   /// The data for the section this fragment is in.
   MCSection *Parent;
 
@@ -64,10 +70,6 @@ class MCFragment : public ilist_node_with_parent<MCFragment, MCSection> {
   /// The layout order of this fragment.
   unsigned LayoutOrder;
 
-  /// The subsection this fragment belongs to. This is 0 if the fragment is not
-  // in any subsection.
-  unsigned SubsectionNumber = 0;
-
   FragmentType Kind;
 
 protected:
@@ -88,6 +90,8 @@ class MCFragment : public ilist_node_with_parent<MCFragment, MCSection> {
   /// This method will dispatch to the appropriate subclass.
   void destroy();
 
+  MCFragment *getNext() const { return Next; }
+
   FragmentType getKind() const { return Kind; }
 
   MCSection *getParent() const { return Parent; }
@@ -104,9 +108,6 @@ class MCFragment : public ilist_node_with_parent<MCFragment, MCSection> {
   bool hasInstructions() const { return HasInstructions; }
 
   void dump() const;
-
-  void setSubsectionNumber(unsigned Value) { SubsectionNumber = Value; }
-  unsigned getSubsectionNumber() const { return SubsectionNumber; }
 };
 
 class MCDummyFragment : public MCFragment {

llvm/include/llvm/MC/MCObjectStreamer.h

@@ -41,7 +41,6 @@ class raw_pwrite_stream;
 /// implementation.
 class MCObjectStreamer : public MCStreamer {
   std::unique_ptr<MCAssembler> Assembler;
-  MCSection::iterator CurInsertionPoint;
   bool EmitEHFrame;
   bool EmitDebugFrame;
   SmallVector<MCSymbol *, 2> PendingLabels;
@@ -94,7 +93,7 @@ class MCObjectStreamer : public MCStreamer {
   void insert(MCFragment *F) {
     flushPendingLabels(F);
     MCSection *CurSection = getCurrentSectionOnly();
-    CurSection->getFragmentList().insert(CurInsertionPoint, F);
+    CurSection->addFragment(*F);
     F->setParent(CurSection);
   }
 

llvm/include/llvm/MC/MCSection.h

@@ -14,7 +14,6 @@
 #define LLVM_MC_MCSECTION_H
 
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/ilist.h"
 #include "llvm/MC/MCFragment.h"
 #include "llvm/MC/SectionKind.h"
 #include "llvm/Support/Alignment.h"
@@ -24,20 +23,18 @@
 namespace llvm {
 
 class MCAsmInfo;
+class MCAssembler;
 class MCContext;
 class MCExpr;
 class MCSymbol;
 class raw_ostream;
 class Triple;
 
-template <> struct ilist_alloc_traits<MCFragment> {
-  static void deleteNode(MCFragment *V);
-};
-
 /// Instances of this class represent a uniqued identifier for a section in the
 /// current translation unit.  The MCContext class uniques and creates these.
 class MCSection {
 public:
+  friend MCAssembler;
   static constexpr unsigned NonUniqueID = ~0U;
 
   enum SectionVariant {
@@ -58,12 +55,29 @@ class MCSection {
     BundleLockedAlignToEnd
   };
 
-  using FragmentListType = iplist<MCFragment>;
+  struct iterator {
+    MCFragment *F = nullptr;
+    iterator() = default;
+    explicit iterator(MCFragment *F) : F(F) {}
+    MCFragment &operator*() const { return *F; }
+    bool operator==(const iterator &O) const { return F == O.F; }
+    bool operator!=(const iterator &O) const { return F != O.F; }
+    iterator &operator++() {
+      F = F->Next;
+      return *this;
+    }
+    iterator operator++(int) { return iterator(F->Next); }
+  };
 
-  using const_iterator = FragmentListType::const_iterator;
-  using iterator = FragmentListType::iterator;
+  struct FragList {
+    MCFragment *Head = nullptr;
+    MCFragment *Tail = nullptr;
+  };
 
 private:
+  // At parse time, this holds the fragment list of the current subsection. At
+  // layout time, this holds the concatenated fragment lists of all subsections.
+  FragList *CurFragList;
   MCSymbol *Begin;
   MCSymbol *End = nullptr;
   /// The alignment requirement of this section.
@@ -92,11 +106,10 @@ class MCSection {
 
   MCDummyFragment DummyFragment;
 
-  FragmentListType Fragments;
-
-  /// Mapping from subsection number to insertion point for subsection numbers
-  /// below that number.
-  SmallVector<std::pair<unsigned, MCFragment *>, 1> SubsectionFragmentMap;
+  // Mapping from subsection number to fragment list. At layout time, the
+  // subsection 0 list is replaced with concatenated fragments from all
+  // subsections.
+  SmallVector<std::pair<unsigned, FragList>, 1> Subsections;
 
   /// State for tracking labels that don't yet have Fragments
   struct PendingLabel {
@@ -171,29 +184,27 @@ class MCSection {
   bool isRegistered() const { return IsRegistered; }
   void setIsRegistered(bool Value) { IsRegistered = Value; }
 
-  MCSection::FragmentListType &getFragmentList() { return Fragments; }
-  const MCSection::FragmentListType &getFragmentList() const {
-    return const_cast<MCSection *>(this)->getFragmentList();
-  }
-
-  /// Support for MCFragment::getNextNode().
-  static FragmentListType MCSection::*getSublistAccess(MCFragment *) {
-    return &MCSection::Fragments;
-  }
-
   const MCDummyFragment &getDummyFragment() const { return DummyFragment; }
   MCDummyFragment &getDummyFragment() { return DummyFragment; }
 
-  iterator begin() { return Fragments.begin(); }
-  const_iterator begin() const { return Fragments.begin(); }
-
-  iterator end() { return Fragments.end(); }
-  const_iterator end() const { return Fragments.end(); }
-  bool empty() const { return Fragments.empty(); }
-
-  void addFragment(MCFragment &F) { Fragments.push_back(&F); }
+  FragList *curFragList() const { return CurFragList; }
+  iterator begin() const { return iterator(CurFragList->Head); }
+  iterator end() const { return {}; }
+  bool empty() const { return !CurFragList->Head; }
+
+  void addFragment(MCFragment &F) {
+    // The formal layout order will be finalized in MCAssembler::layout.
+    if (CurFragList->Tail) {
+      CurFragList->Tail->Next = &F;
+      F.setLayoutOrder(CurFragList->Tail->getLayoutOrder() + 1);
+    } else {
+      CurFragList->Head = &F;
+      assert(F.getLayoutOrder() == 0);
+    }
+    CurFragList->Tail = &F;
+  }
 
-  MCSection::iterator getSubsectionInsertionPoint(unsigned Subsection);
+  void switchSubsection(unsigned Subsection);
 
   void dump() const;
 

llvm/lib/MC/MCAssembler.cpp

@@ -831,6 +831,19 @@ void MCAssembler::layout(MCAsmLayout &Layout) {
     MCSection *Sec = Layout.getSectionOrder()[i];
     Sec->setLayoutOrder(i);
 
+    // Chain together fragments from all subsections.
+    MCDummyFragment Dummy(Sec);
+    MCFragment *Tail = &Dummy;
+    for (auto &[_, List] : Sec->Subsections) {
+      if (!List.Head)
+        continue;
+      Tail->Next = List.Head;
+      Tail = List.Tail;
+    }
+    Sec->Subsections.clear();
+    Sec->Subsections.push_back({0u, {Dummy.getNext(), Tail}});
+    Sec->CurFragList = &Sec->Subsections[0].second;
+
     unsigned FragmentIndex = 0;
     for (MCFragment &Frag : *Sec)
       Frag.setLayoutOrder(FragmentIndex++);
@@ -1094,7 +1107,7 @@ bool MCAssembler::relaxBoundaryAlign(MCAsmLayout &Layout,
 
   uint64_t AlignedOffset = Layout.getFragmentOffset(&BF);
   uint64_t AlignedSize = 0;
-  for (const MCFragment *F = BF.getNextNode();; F = F->getNextNode()) {
+  for (const MCFragment *F = BF.getNext();; F = F->getNext()) {
     AlignedSize += computeFragmentSize(Layout, *F);
     if (F == BF.getLastFragment())
       break;

llvm/lib/MC/MCExpr.cpp

@@ -661,25 +661,16 @@ static void AttemptToFoldSymbolOffsetDifference(
     // this is important when the Subtarget is changed and a new MCDataFragment
     // is created in the case of foo: instr; .arch_extension ext; instr .if . -
     // foo.
-    if (SA.isVariable() || SB.isVariable() ||
-        FA->getSubsectionNumber() != FB->getSubsectionNumber())
+    if (SA.isVariable() || SB.isVariable())
       return;
 
     // Try to find a constant displacement from FA to FB, add the displacement
     // between the offset in FA of SA and the offset in FB of SB.
     bool Reverse = false;
-    if (FA == FB) {
+    if (FA == FB)
       Reverse = SA.getOffset() < SB.getOffset();
-    } else if (!isa<MCDummyFragment>(FA)) {
-      // Testing FA < FB is slow. Use setLayoutOrder to speed up computation.
-      // The formal layout order will be finalized in MCAssembler::layout.
-      if (FA->getLayoutOrder() == 0 || FB->getLayoutOrder()== 0) {
-        unsigned LayoutOrder = 0;
-        for (MCFragment &F : *FA->getParent())
-          F.setLayoutOrder(++LayoutOrder);
-      }
+    else if (!isa<MCDummyFragment>(FA))
       Reverse = FA->getLayoutOrder() < FB->getLayoutOrder();
-    }
 
     uint64_t SAOffset = SA.getOffset(), SBOffset = SB.getOffset();
     int64_t Displacement = SA.getOffset() - SB.getOffset();
@@ -695,7 +686,7 @@ static void AttemptToFoldSymbolOffsetDifference(
     // instruction, the difference cannot be resolved as it may be changed by
     // the linker.
     bool BBeforeRelax = false, AAfterRelax = false;
-    for (auto FI = FB->getIterator(), FE = SecA.end(); FI != FE; ++FI) {
+    for (auto FI = FB; FI; FI = FI->getNext()) {
       auto DF = dyn_cast<MCDataFragment>(FI);
       if (DF && DF->isLinkerRelaxable()) {
         if (&*FI != FB || SBOffset != DF->getContents().size())
@@ -726,12 +717,14 @@ static void AttemptToFoldSymbolOffsetDifference(
         return;
       }
     }
-    // If the previous loop does not find FA, FA must be a dummy fragment not in
-    // the fragment list (which means SA is a pending label (see
-    // flushPendingLabels)). In either case, we can resolve the difference.
-    assert(Found || isa<MCDummyFragment>(FA));
-    Addend += Reverse ? -Displacement : Displacement;
-    FinalizeFolding();
+    // If FA and FB belong to the same subsection, either the previous loop
+    // found FA, or FA is a dummy fragment not in the fragment list (which means
+    // SA is a pending label (see flushPendingLabels)) or FA and FB belong to
+    // different subsections. In either case, we can resolve the difference.
+    if (Found || isa<MCDummyFragment>(FA)) {
+      Addend += Reverse ? -Displacement : Displacement;
+      FinalizeFolding();
+    }
   }
 }
 

llvm/lib/MC/MCFragment.cpp

@@ -141,7 +141,7 @@ const MCSymbol *MCAsmLayout::getBaseSymbol(const MCSymbol &Symbol) const {
 
 uint64_t MCAsmLayout::getSectionAddressSize(const MCSection *Sec) const {
   // The size is the last fragment's end offset.
-  const MCFragment &F = Sec->getFragmentList().back();
+  const MCFragment &F = *Sec->curFragList()->Tail;
   return getFragmentOffset(&F) + getAssembler().computeFragmentSize(*this, F);
 }
 
@@ -197,8 +197,6 @@ uint64_t llvm::computeBundlePadding(const MCAssembler &Assembler,
 
 /* *** */
 
-void ilist_alloc_traits<MCFragment>::deleteNode(MCFragment *V) { V->destroy(); }
-
 MCFragment::MCFragment(FragmentType Kind, bool HasInstructions,
                        MCSection *Parent)
     : Parent(Parent), Atom(nullptr), Offset(~UINT64_C(0)), LayoutOrder(0),

llvm/lib/MC/MCObjectStreamer.cpp

@@ -180,7 +180,6 @@ void MCObjectStreamer::reset() {
     if (getContext().getTargetOptions())
       Assembler->setRelaxAll(getContext().getTargetOptions()->MCRelaxAll);
   }
-  CurInsertionPoint = MCSection::iterator();
   EmitEHFrame = true;
   EmitDebugFrame = false;
   PendingLabels.clear();
@@ -200,12 +199,7 @@ void MCObjectStreamer::emitFrames(MCAsmBackend *MAB) {
 }
 
 MCFragment *MCObjectStreamer::getCurrentFragment() const {
-  assert(getCurrentSectionOnly() && "No current section!");
-
-  if (CurInsertionPoint != getCurrentSectionOnly()->begin())
-    return &*std::prev(CurInsertionPoint);
-
-  return nullptr;
+  return getCurrentSectionOnly()->curFragList()->Tail;
 }
 
 static bool canReuseDataFragment(const MCDataFragment &F,
@@ -391,8 +385,7 @@ bool MCObjectStreamer::changeSectionImpl(MCSection *Section,
   }
 
   CurSubsectionIdx = unsigned(IntSubsection);
-  CurInsertionPoint =
-      Section->getSubsectionInsertionPoint(CurSubsectionIdx);
+  Section->switchSubsection(CurSubsectionIdx);
   return Created;
 }