Reapply "LTO: add API to set strategy for -internalize"

Reapply r199191, reverted in r199197 because it carelessly broke
Other/link-opts.ll.  The problem was that calling
createInternalizePass("main") would select
createInternalizePass(bool("main")) instead of
createInternalizePass(ArrayRef<const char *>("main")).  This commit
fixes the bug.

The original commit message follows.

Add API to LTOCodeGenerator to specify a strategy for the -internalize
pass.

This is a new attempt at Bill's change in r185882, which he reverted in
r188029 due to problems with the gold linker.  This puts the onus on the
linker to decide whether (and what) to internalize.

In particular, running internalize before outputting an object file may
change a 'weak' symbol into an internal one, even though that symbol
could be needed by an external object file --- e.g., with arclite.

This patch enables three strategies:

- LTO_INTERNALIZE_FULL: the default (and the old behaviour).
- LTO_INTERNALIZE_NONE: skip -internalize.
- LTO_INTERNALIZE_HIDDEN: only -internalize symbols with hidden
  visibility.

LTO_INTERNALIZE_FULL should be used when linking an executable.

Outputting an object file (e.g., via ld -r) is more complicated, and
depends on whether hidden symbols should be internalized.  E.g., for
ld -r, LTO_INTERNALIZE_NONE can be used when -keep_private_externs, and
LTO_INTERNALIZE_HIDDEN can be used otherwise.  However,
LTO_INTERNALIZE_FULL is inappropriate, since the output object file will
eventually need to link with others.

lto_codegen_set_internalize_strategy() sets the strategy for subsequent
calls to lto_codegen_write_merged_modules() and lto_codegen_compile*().

<rdar://problem/14334895>

llvm-svn: 199244

include/llvm-c/lto.h

@@ -40,7 +40,7 @@ typedef bool lto_bool_t;
  * @{
  */
 
-#define LTO_API_VERSION 5
+#define LTO_API_VERSION 6
 
 typedef enum {
     LTO_SYMBOL_ALIGNMENT_MASK              = 0x0000001F, /* log2 of alignment */
@@ -73,6 +73,11 @@ typedef enum {
     LTO_CODEGEN_PIC_MODEL_DYNAMIC_NO_PIC = 2
 } lto_codegen_model;
 
+typedef enum {
+    LTO_INTERNALIZE_FULL   = 0,
+    LTO_INTERNALIZE_NONE   = 1,
+    LTO_INTERNALIZE_HIDDEN = 2
+} lto_internalize_strategy;
 
 /** opaque reference to a loaded object module */
 typedef struct LTOModule*         lto_module_t;
@@ -263,6 +268,14 @@ extern void
 lto_codegen_set_assembler_args(lto_code_gen_t cg, const char **args,
                                int nargs);
 
+/**
+ * Sets the strategy to use during internalize.  Default strategy is
+ * LTO_INTERNALIZE_FULL.
+ */
+extern void
+lto_codegen_set_internalize_strategy(lto_code_gen_t cg,
+                                     lto_internalize_strategy);
+
 /**
  * Tells LTO optimization passes that this symbol must be preserved
  * because it is referenced by native code or a command line option.

include/llvm/LTO/LTOCodeGenerator.h

@@ -70,6 +70,7 @@ struct LTOCodeGenerator {
   void setTargetOptions(llvm::TargetOptions options);
   void setDebugInfo(lto_debug_model);
   void setCodePICModel(lto_codegen_model);
+  void setInternalizeStrategy(lto_internalize_strategy);
 
   void setCpu(const char *mCpu) { MCpu = mCpu; }
 
@@ -114,6 +115,14 @@ struct LTOCodeGenerator {
                       bool disableGVNLoadPRE,
                       std::string &errMsg);
 
+  bool shouldInternalize() const {
+    return InternalizeStrategy != LTO_INTERNALIZE_NONE;
+  }
+
+  bool shouldOnlyInternalizeHidden() const {
+    return InternalizeStrategy == LTO_INTERNALIZE_HIDDEN;
+  }
+
 private:
   void initializeLTOPasses();
 
@@ -138,6 +147,7 @@ private:
   bool EmitDwarfDebugInfo;
   bool ScopeRestrictionsDone;
   lto_codegen_model CodeModel;
+  lto_internalize_strategy InternalizeStrategy;
   StringSet MustPreserveSymbols;
   StringSet AsmUndefinedRefs;
   llvm::MemoryBuffer *NativeObjectFile;

include/llvm/Transforms/IPO.h

@@ -108,14 +108,19 @@ Pass *createPruneEHPass();
 ////
 /// The symbols in \p ExportList are never internalized.
 ///
+/// When OnlyHidden=true, only symbols with hidden visibility are internalized.
+///
 /// The symbol in DSOList are internalized if it is safe to drop them from
 /// the symbol table.
 ///
 /// Note that commandline options that are used with the above function are not
 /// used now!
-ModulePass *createInternalizePass(ArrayRef<const char *> ExportList);
+ModulePass *createInternalizePass(ArrayRef<const char *> ExportList,
+                                  bool OnlyHidden = false);
 /// createInternalizePass - Same as above, but with an empty exportList.
-ModulePass *createInternalizePass();
+ModulePass *createInternalizePass(bool OnlyHidden = false);
+/// createInternalizePass - Resolve ambiguity when passed a const char *.
+ModulePass *createInternalizePass(const char *SingleExport);
 
 //===----------------------------------------------------------------------===//
 /// createDeadArgEliminationPass - This pass removes arguments from functions

lib/LTO/LTOCodeGenerator.cpp

@@ -62,7 +62,8 @@ const char* LTOCodeGenerator::getVersionString() {
 LTOCodeGenerator::LTOCodeGenerator()
     : Context(getGlobalContext()), Linker(new Module("ld-temp.o", Context)),
       TargetMach(NULL), EmitDwarfDebugInfo(false), ScopeRestrictionsDone(false),
-      CodeModel(LTO_CODEGEN_PIC_MODEL_DYNAMIC), NativeObjectFile(NULL) {
+      CodeModel(LTO_CODEGEN_PIC_MODEL_DYNAMIC),
+      InternalizeStrategy(LTO_INTERNALIZE_FULL), NativeObjectFile(NULL) {
   initializeLTOPasses();
 }
 
@@ -164,6 +165,18 @@ void LTOCodeGenerator::setCodePICModel(lto_codegen_model model) {
   llvm_unreachable("Unknown PIC model!");
 }
 
+void
+LTOCodeGenerator::setInternalizeStrategy(lto_internalize_strategy Strategy) {
+  switch (Strategy) {
+  case LTO_INTERNALIZE_FULL:
+  case LTO_INTERNALIZE_NONE:
+  case LTO_INTERNALIZE_HIDDEN:
+    InternalizeStrategy = Strategy;
+    return;
+  }
+  llvm_unreachable("Unknown internalize strategy!");
+}
+
 bool LTOCodeGenerator::writeMergedModules(const char *path,
                                           std::string &errMsg) {
   if (!determineTarget(errMsg))
@@ -377,7 +390,7 @@ static void accumulateAndSortLibcalls(std::vector<StringRef> &Libcalls,
 }
 
 void LTOCodeGenerator::applyScopeRestrictions() {
-  if (ScopeRestrictionsDone)
+  if (ScopeRestrictionsDone || !shouldInternalize())
     return;
   Module *mergedModule = Linker.getModule();
 
@@ -429,7 +442,8 @@ void LTOCodeGenerator::applyScopeRestrictions() {
     LLVMCompilerUsed->setSection("llvm.metadata");
   }
 
-  passes.add(createInternalizePass(MustPreserveList));
+  passes.add(
+      createInternalizePass(MustPreserveList, shouldOnlyInternalizeHidden()));
 
   // apply scope restrictions
   passes.run(*mergedModule);

lib/Transforms/IPO/Internalize.cpp

@@ -54,10 +54,11 @@ APIList("internalize-public-api-list", cl::value_desc("list"),
 namespace {
   class InternalizePass : public ModulePass {
     std::set<std::string> ExternalNames;
+    bool OnlyHidden;
   public:
     static char ID; // Pass identification, replacement for typeid
-    explicit InternalizePass();
-    explicit InternalizePass(ArrayRef<const char *> ExportList);
+    explicit InternalizePass(bool OnlyHidden = false);
+    explicit InternalizePass(ArrayRef<const char *> ExportList, bool OnlyHidden);
     void LoadFile(const char *Filename);
     virtual bool runOnModule(Module &M);
 
@@ -72,16 +73,17 @@ char InternalizePass::ID = 0;
 INITIALIZE_PASS(InternalizePass, "internalize",
                 "Internalize Global Symbols", false, false)
 
-InternalizePass::InternalizePass()
-  : ModulePass(ID) {
+InternalizePass::InternalizePass(bool OnlyHidden)
+  : ModulePass(ID), OnlyHidden(OnlyHidden) {
   initializeInternalizePassPass(*PassRegistry::getPassRegistry());
   if (!APIFile.empty())           // If a filename is specified, use it.
     LoadFile(APIFile.c_str());
   ExternalNames.insert(APIList.begin(), APIList.end());
 }
 
-InternalizePass::InternalizePass(ArrayRef<const char *> ExportList)
-  : ModulePass(ID){
+InternalizePass::InternalizePass(ArrayRef<const char *> ExportList,
+                                 bool OnlyHidden)
+  : ModulePass(ID), OnlyHidden(OnlyHidden) {
   initializeInternalizePassPass(*PassRegistry::getPassRegistry());
   for(ArrayRef<const char *>::const_iterator itr = ExportList.begin();
         itr != ExportList.end(); itr++) {
@@ -106,7 +108,11 @@ void InternalizePass::LoadFile(const char *Filename) {
 }
 
 static bool shouldInternalize(const GlobalValue &GV,
-                              const std::set<std::string> &ExternalNames) {
+                              const std::set<std::string> &ExternalNames,
+                              bool OnlyHidden) {
+  if (OnlyHidden && !GV.hasHiddenVisibility())
+    return false;
+
   // Function must be defined here
   if (GV.isDeclaration())
     return false;
@@ -155,9 +161,8 @@ bool InternalizePass::runOnModule(Module &M) {
   }
 
   // Mark all functions not in the api as internal.
-  // FIXME: maybe use private linkage?
   for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
-    if (!shouldInternalize(*I, ExternalNames))
+    if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
       continue;
 
     I->setLinkage(GlobalValue::InternalLinkage);
@@ -191,10 +196,9 @@ bool InternalizePass::runOnModule(Module &M) {
 
   // Mark all global variables with initializers that are not in the api as
   // internal as well.
-  // FIXME: maybe use private linkage?
   for (Module::global_iterator I = M.global_begin(), E = M.global_end();
        I != E; ++I) {
-    if (!shouldInternalize(*I, ExternalNames))
+    if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
       continue;
 
     I->setLinkage(GlobalValue::InternalLinkage);
@@ -206,7 +210,7 @@ bool InternalizePass::runOnModule(Module &M) {
   // Mark all aliases that are not in the api as internal as well.
   for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
        I != E; ++I) {
-    if (!shouldInternalize(*I, ExternalNames))
+    if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
       continue;
 
     I->setLinkage(GlobalValue::InternalLinkage);
@@ -218,10 +222,15 @@ bool InternalizePass::runOnModule(Module &M) {
   return Changed;
 }
 
-ModulePass *llvm::createInternalizePass() {
-  return new InternalizePass();
+ModulePass *llvm::createInternalizePass(bool OnlyHidden) {
+  return new InternalizePass(OnlyHidden);
 }
 
-ModulePass *llvm::createInternalizePass(ArrayRef<const char *> ExportList) {
-  return new InternalizePass(ExportList);
+ModulePass *llvm::createInternalizePass(ArrayRef<const char *> ExportList,
+                                        bool OnlyHidden) {
+  return new InternalizePass(ExportList, OnlyHidden);
+}
+
+ModulePass *llvm::createInternalizePass(const char *SingleExport) {
+  return createInternalizePass(ArrayRef<const char *>(SingleExport));
 }

tools/lto/lto.cpp

@@ -252,6 +252,13 @@ void lto_codegen_set_assembler_args(lto_code_gen_t cg, const char **args,
   // In here only for backwards compatibility. We use MC now.
 }
 
+/// lto_codegen_set_internalize_strategy - Sets the strategy to use during
+/// internalize.
+void lto_codegen_set_internalize_strategy(lto_code_gen_t cg,
+                                          lto_internalize_strategy strategy) {
+  cg->setInternalizeStrategy(strategy);
+}
+
 /// lto_codegen_add_must_preserve_symbol - Adds to a list of all global symbols
 /// that must exist in the final generated code. If a function is not listed
 /// there, it might be inlined into every usage and optimized away.