Bitcode: Serialize (and recover) use-list order

Predict and serialize use-list order in bitcode.  This makes the option
`-preserve-bc-use-list-order` work *most* of the time, but this is still
experimental.

  - Builds a full value-table up front in the writer, sets up a list of
    use-list orders to write out, and discards the table.  This is a
    simpler first step than determining the order from the various
    overlapping IDs of values on-the-fly.

  - The shuffles stored in the use-list order list have an unnecessarily
    large memory footprint.

  - `blockaddress` expressions cause functions to be materialized
    out-of-order.  For now I've ignored this problem, so use-list orders
    will be wrong for constants used by functions that have block
    addresses taken.  There are a couple of ways to fix this, but I
    don't have a concrete plan yet.

  - When materializing functions lazily, the use-lists for constants
    will not be correct.  This use case is out of scope: what should the
    use-list order be, if it's incomplete?

This is part of PR5680.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@214125 91177308-0d34-0410-b5e6-96231b3b80d8

Author: dexonsmith

include/llvm/Bitcode/LLVMBitCodes.h

@@ -330,7 +330,8 @@ namespace bitc {
   };
 
   enum UseListCodes {
-    USELIST_CODE_ENTRY = 1   // USELIST_CODE_ENTRY: TBD.
+    USELIST_CODE_DEFAULT = 1, // DEFAULT: [index..., value-id]
+    USELIST_CODE_BB      = 2  // BB: [index..., bb-id]
   };
 
   enum AttributeKindCodes {

include/llvm/IR/UseListOrder.h

@@ -16,10 +16,23 @@
 #define LLVM_IR_USELISTORDER_H
 
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/SmallVector.h"
+#include <vector>
 
 namespace llvm {
 
 class Module;
+class Function;
+class Value;
+
+/// \brief Structure to hold a use-list order.
+struct UseListOrder {
+  const Function *F;
+  const Value *V;
+  SmallVector<unsigned, 8> Shuffle;
+};
+
+typedef std::vector<UseListOrder> UseListOrderStack;
 
 /// \brief Whether to preserve use-list ordering.
 bool shouldPreserveBitcodeUseListOrder();

lib/Bitcode/Reader/BitcodeReader.cpp

@@ -1620,9 +1620,8 @@ std::error_code BitcodeReader::ParseUseLists() {
   if (Stream.EnterSubBlock(bitc::USELIST_BLOCK_ID))
     return Error(InvalidRecord);
 
-  SmallVector<uint64_t, 64> Record;
-
   // Read all the records.
+  SmallVector<uint64_t, 64> Record;
   while (1) {
     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
 
@@ -1639,14 +1638,42 @@ std::error_code BitcodeReader::ParseUseLists() {
 
     // Read a use list record.
     Record.clear();
+    bool IsBB = false;
     switch (Stream.readRecord(Entry.ID, Record)) {
     default:  // Default behavior: unknown type.
       break;
-    case bitc::USELIST_CODE_ENTRY: { // USELIST_CODE_ENTRY: TBD.
+    case bitc::USELIST_CODE_BB:
+      IsBB = true;
+      // fallthrough
+    case bitc::USELIST_CODE_DEFAULT: {
       unsigned RecordLength = Record.size();
-      if (RecordLength < 1)
-        return Error(InvalidRecord);
-      UseListRecords.push_back(Record);
+      if (RecordLength < 3)
+        // Records should have at least an ID and two indexes.
+        return Error(InvalidRecord);
+      unsigned ID = Record.back();
+      Record.pop_back();
+
+      Value *V;
+      if (IsBB) {
+        assert(ID < FunctionBBs.size() && "Basic block not found");
+        V = FunctionBBs[ID];
+      } else
+        V = ValueList[ID];
+      unsigned NumUses = 0;
+      SmallDenseMap<const Use *, unsigned, 16> Order;
+      for (const Use &U : V->uses()) {
+        if (NumUses > Record.size())
+          break;
+        Order[&U] = Record[NumUses++];
+      }
+      if (Order.size() != Record.size() || NumUses > Record.size())
+        // Mismatches can happen if the functions are being materialized lazily
+        // (out-of-order), or a value has been upgraded.
+        break;
+
+      V->sortUseList([&](const Use &L, const Use &R) {
+        return Order.lookup(&L) < Order.lookup(&R);
+      });
       break;
     }
     }
@@ -2298,6 +2325,10 @@ std::error_code BitcodeReader::ParseFunctionBody(Function *F) {
         if (std::error_code EC = ParseMetadata())
           return EC;
         break;
+      case bitc::USELIST_BLOCK_ID:
+        if (std::error_code EC = ParseUseLists())
+          return EC;
+        break;
       }
       continue;
 

lib/Bitcode/Reader/BitcodeReader.h

@@ -138,7 +138,6 @@ class BitcodeReader : public GVMaterializer {
   BitcodeReaderMDValueList MDValueList;
   std::vector<Comdat *> ComdatList;
   SmallVector<Instruction *, 64> InstructionList;
-  SmallVector<SmallVector<uint64_t, 64>, 64> UseListRecords;
 
   std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInits;
   std::vector<std::pair<GlobalAlias*, unsigned> > AliasInits;

lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -1602,6 +1602,39 @@ static void WriteValueSymbolTable(const ValueSymbolTable &VST,
   Stream.ExitBlock();
 }
 
+static void WriteUseList(ValueEnumerator &VE, UseListOrder &&Order,
+                         BitstreamWriter &Stream) {
+  assert(Order.Shuffle.size() >= 2 && "Shuffle too small");
+  unsigned Code;
+  if (isa<BasicBlock>(Order.V))
+    Code = bitc::USELIST_CODE_BB;
+  else
+    Code = bitc::USELIST_CODE_DEFAULT;
+
+  SmallVector<uint64_t, 64> Record;
+  for (unsigned I : Order.Shuffle)
+    Record.push_back(I);
+  Record.push_back(VE.getValueID(Order.V));
+  Stream.EmitRecord(Code, Record);
+}
+
+static void WriteUseListBlock(const Function *F, ValueEnumerator &VE,
+                              BitstreamWriter &Stream) {
+  auto hasMore = [&]() {
+    return !VE.UseListOrders.empty() && VE.UseListOrders.back().F == F;
+  };
+  if (!hasMore())
+    // Nothing to do.
+    return;
+
+  Stream.EnterSubblock(bitc::USELIST_BLOCK_ID, 3);
+  while (hasMore()) {
+    WriteUseList(VE, std::move(VE.UseListOrders.back()), Stream);
+    VE.UseListOrders.pop_back();
+  }
+  Stream.ExitBlock();
+}
+
 /// WriteFunction - Emit a function body to the module stream.
 static void WriteFunction(const Function &F, ValueEnumerator &VE,
                           BitstreamWriter &Stream) {
@@ -1670,6 +1703,8 @@ static void WriteFunction(const Function &F, ValueEnumerator &VE,
 
   if (NeedsMetadataAttachment)
     WriteMetadataAttachment(F, VE, Stream);
+  if (shouldPreserveBitcodeUseListOrder())
+    WriteUseListBlock(&F, VE, Stream);
   VE.purgeFunction();
   Stream.ExitBlock();
 }
@@ -1835,98 +1870,6 @@ static void WriteBlockInfo(const ValueEnumerator &VE, BitstreamWriter &Stream) {
   Stream.ExitBlock();
 }
 
-// Sort the Users based on the order in which the reader parses the bitcode
-// file.
-static bool bitcodereader_order(const User *lhs, const User *rhs) {
-  // TODO: Implement.
-  return true;
-}
-
-static void WriteUseList(const Value *V, const ValueEnumerator &VE,
-                         BitstreamWriter &Stream) {
-
-  // One or zero uses can't get out of order.
-  if (V->use_empty() || V->hasNUses(1))
-    return;
-
-  // Make a copy of the in-memory use-list for sorting.
-  SmallVector<const User*, 8> UserList(V->user_begin(), V->user_end());
-
-  // Sort the copy based on the order read by the BitcodeReader.
-  std::sort(UserList.begin(), UserList.end(), bitcodereader_order);
-
-  // TODO: Generate a diff between the BitcodeWriter in-memory use-list and the
-  // sorted list (i.e., the expected BitcodeReader in-memory use-list).
-
-  // TODO: Emit the USELIST_CODE_ENTRYs.
-}
-
-static void WriteFunctionUseList(const Function *F, ValueEnumerator &VE,
-                                 BitstreamWriter &Stream) {
-  VE.incorporateFunction(*F);
-
-  for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
-       AI != AE; ++AI)
-    WriteUseList(AI, VE, Stream);
-  for (Function::const_iterator BB = F->begin(), FE = F->end(); BB != FE;
-       ++BB) {
-    WriteUseList(BB, VE, Stream);
-    for (BasicBlock::const_iterator II = BB->begin(), IE = BB->end(); II != IE;
-         ++II) {
-      WriteUseList(II, VE, Stream);
-      for (User::const_op_iterator OI = II->op_begin(), E = II->op_end();
-           OI != E; ++OI) {
-        if ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
-            isa<InlineAsm>(*OI))
-          WriteUseList(*OI, VE, Stream);
-      }
-    }
-  }
-  VE.purgeFunction();
-}
-
-// Emit use-lists.
-static void WriteModuleUseLists(const Module *M, ValueEnumerator &VE,
-                                BitstreamWriter &Stream) {
-  Stream.EnterSubblock(bitc::USELIST_BLOCK_ID, 3);
-
-  // XXX: this modifies the module, but in a way that should never change the
-  // behavior of any pass or codegen in LLVM. The problem is that GVs may
-  // contain entries in the use_list that do not exist in the Module and are
-  // not stored in the .bc file.
-  for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
-       I != E; ++I)
-    I->removeDeadConstantUsers();
-
-  // Write the global variables.
-  for (Module::const_global_iterator GI = M->global_begin(),
-         GE = M->global_end(); GI != GE; ++GI) {
-    WriteUseList(GI, VE, Stream);
-
-    // Write the global variable initializers.
-    if (GI->hasInitializer())
-      WriteUseList(GI->getInitializer(), VE, Stream);
-  }
-
-  // Write the functions.
-  for (Module::const_iterator FI = M->begin(), FE = M->end(); FI != FE; ++FI) {
-    WriteUseList(FI, VE, Stream);
-    if (!FI->isDeclaration())
-      WriteFunctionUseList(FI, VE, Stream);
-    if (FI->hasPrefixData())
-      WriteUseList(FI->getPrefixData(), VE, Stream);
-  }
-
-  // Write the aliases.
-  for (Module::const_alias_iterator AI = M->alias_begin(), AE = M->alias_end();
-       AI != AE; ++AI) {
-    WriteUseList(AI, VE, Stream);
-    WriteUseList(AI->getAliasee(), VE, Stream);
-  }
-
-  Stream.ExitBlock();
-}
-
 /// WriteModule - Emit the specified module to the bitstream.
 static void WriteModule(const Module *M, BitstreamWriter &Stream) {
   Stream.EnterSubblock(bitc::MODULE_BLOCK_ID, 3);
@@ -1969,9 +1912,9 @@ static void WriteModule(const Module *M, BitstreamWriter &Stream) {
   // Emit names for globals/functions etc.
   WriteValueSymbolTable(M->getValueSymbolTable(), VE, Stream);
 
-  // Emit use-lists.
+  // Emit module-level use-lists.
   if (shouldPreserveBitcodeUseListOrder())
-    WriteModuleUseLists(M, VE, Stream);
+    WriteUseListBlock(nullptr, VE, Stream);
 
   // Emit function bodies.
   for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F)