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
This commit is contained in:
Duncan P. N. Exon Smith
2014-07-28 21:19:41 +00:00
parent 0aed6e729d
commit bd24fe8c7e
9 changed files with 287 additions and 103 deletions

View File

@ -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)