Merge pull request #48 from alexcrichton/update-wasm-bits

Update wasm handling of embedded bitcode

Author: cuviper

clang/test/Driver/embed-bitcode-wasm.c

@@ -0,0 +1,6 @@
+// REQUIRES: webassembly-registered-target
+
+// RUN: %clang -c -target wasm32-unknown-unknown %s -fembed-bitcode -o %t.o
+// RUN: llvm-readobj -S %t.o | FileCheck --check-prefix=CHECK %s
+// CHECK:   Name: .llvmbc
+// CHECK:   Name: .llvmcmd

clang/test/Driver/fembed-bitcode.c

@@ -34,3 +34,12 @@
 // RUN:     | FileCheck --check-prefix=CHECK-HEXAGON %s
 // CHECK-HEXAGON: "-target-feature"
 // CHECK-HEXAGON: "+reserved-r19"
+//
+// RUN: %clang -target wasm32-unknown-unknown -fembed-bitcode=all -pthread -c %s -o /dev/null -### 2>&1 \
+// RUN:     | FileCheck --check-prefix=CHECK-WASM %s
+
+// CHECK-WASM: "-cc1"
+// CHECK-WASM: "-target-feature" "+atomics"
+
+// CHECK-WASM: "-cc1"
+// CHECK-WASM: "-target-feature" "+atomics"

lld/wasm/Writer.cpp

@@ -111,13 +111,21 @@ void Writer::calculateCustomSections() {
     for (InputSection *section : file->customSections) {
       StringRef name = section->getName();
       // These custom sections are known the linker and synthesized rather than
-      // blindly copied
+      // blindly copied.
       if (name == "linking" || name == "name" || name == "producers" ||
           name == "target_features" || name.startswith("reloc."))
         continue;
-      // .. or it is a debug section
+      // These custom sections are generated by `clang -fembed-bitcode`.
+      // These are used by the rust toolchain to ship LTO data along with
+      // compiled object code, but they don't want this included in the linker
+      // output.
+      if (name == ".llvmbc" || name == ".llvmcmd")
+        continue;
+      // Strip debug section in that option was specified.
       if (stripDebug && name.startswith(".debug_"))
         continue;
+      // Otherwise include custom sections by default and concatenate their
+      // contents.
       customSectionMapping[name].push_back(section);
     }
   }

llvm/include/llvm/MC/MCSymbolWasm.h

@@ -18,6 +18,7 @@ class MCSymbolWasm : public MCSymbol {
   bool IsWeak = false;
   bool IsHidden = false;
   bool IsComdat = false;
+  mutable bool IsUsedInInitArray = false;
   mutable bool IsUsedInGOT = false;
   Optional<std::string> ImportModule;
   Optional<std::string> ImportName;
@@ -82,6 +83,9 @@ class MCSymbolWasm : public MCSymbol {
   void setUsedInGOT() const { IsUsedInGOT = true; }
   bool isUsedInGOT() const { return IsUsedInGOT; }
 
+  void setUsedInInitArray() const { IsUsedInInitArray = true; }
+  bool isUsedInInitArray() const { return IsUsedInInitArray; }
+
   const wasm::WasmSignature *getSignature() const { return Signature; }
   void setSignature(wasm::WasmSignature *Sig) { Signature = Sig; }
 

llvm/include/llvm/Object/Wasm.h

@@ -183,7 +183,6 @@ class WasmObjectFile : public ObjectFile {
   bool isSectionData(DataRefImpl Sec) const override;
   bool isSectionBSS(DataRefImpl Sec) const override;
   bool isSectionVirtual(DataRefImpl Sec) const override;
-  bool isSectionBitcode(DataRefImpl Sec) const override;
   relocation_iterator section_rel_begin(DataRefImpl Sec) const override;
   relocation_iterator section_rel_end(DataRefImpl Sec) const override;
 

llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp

@@ -1688,6 +1688,14 @@ static const Comdat *getWasmComdat(const GlobalValue *GV) {
 }
 
 static SectionKind getWasmKindForNamedSection(StringRef Name, SectionKind K) {
+  // Certain data sections we treat as named custom sections rather than
+  // segments within the data section.
+  // This could be avoided if all data segements (the wasm sense) were
+  // represented as thier own sections (in the llvm sense).
+  // TODO(sbc): https://github.com/WebAssembly/tool-conventions/issues/138
+  if (Name == ".llvmcmd" || Name == ".llvmbc")
+    return SectionKind::getMetadata();
+
   // If we're told we have function data, then use that.
   if (K.isText())
     return SectionKind::getText();

llvm/lib/MC/WasmObjectWriter.cpp

@@ -434,10 +434,6 @@ void WasmObjectWriter::recordRelocation(MCAssembler &Asm,
   uint64_t FixupOffset = Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
   MCContext &Ctx = Asm.getContext();
 
-  // The .init_array isn't translated as data, so don't do relocations in it.
-  if (FixupSection.getSectionName().startswith(".init_array"))
-    return;
-
   if (const MCSymbolRefExpr *RefB = Target.getSymB()) {
     assert(RefB->getKind() == MCSymbolRefExpr::VK_None &&
            "Should not have constructed this");
@@ -483,6 +479,12 @@ void WasmObjectWriter::recordRelocation(MCAssembler &Asm,
   const MCSymbolRefExpr *RefA = Target.getSymA();
   const auto *SymA = RefA ? cast<MCSymbolWasm>(&RefA->getSymbol()) : nullptr;
 
+  // The .init_array isn't translated as data, so don't do relocations in it.
+  if (FixupSection.getSectionName().startswith(".init_array")) {
+    SymA->setUsedInInitArray();
+    return;
+  }
+
   if (SymA && SymA->isVariable()) {
     const MCExpr *Expr = SymA->getVariableValue();
     const auto *Inner = cast<MCSymbolRefExpr>(Expr);
@@ -1113,16 +1115,13 @@ void WasmObjectWriter::registerEventType(const MCSymbolWasm &Symbol) {
 }
 
 static bool isInSymtab(const MCSymbolWasm &Sym) {
-  if (Sym.isUsedInReloc())
+  if (Sym.isUsedInReloc() || Sym.isUsedInInitArray())
     return true;
 
   if (Sym.isComdat() && !Sym.isDefined())
     return false;
 
-  if (Sym.isTemporary() && Sym.getName().empty())
-    return false;
-
-  if (Sym.isTemporary() && Sym.isData() && !Sym.getSize())
+  if (Sym.isTemporary())
     return false;
 
   if (Sym.isSection())
@@ -1578,7 +1577,7 @@ uint64_t WasmObjectWriter::writeObject(MCAssembler &Asm,
         report_fatal_error("fixups in .init_array should be symbol references");
       const auto &TargetSym = cast<const MCSymbolWasm>(SymRef->getSymbol());
       if (TargetSym.getIndex() == InvalidIndex)
-        report_fatal_error("symbols in .init_array should exist in symbtab");
+        report_fatal_error("symbols in .init_array should exist in symtab");
       if (!TargetSym.isFunction())
         report_fatal_error("symbols in .init_array should be for functions");
       InitFuncs.push_back(

llvm/lib/Object/IRObjectFile.cpp

@@ -94,6 +94,7 @@ IRObjectFile::findBitcodeInMemBuffer(MemoryBufferRef Object) {
     return Object;
   case file_magic::elf_relocatable:
   case file_magic::macho_object:
+  case file_magic::wasm_object:
   case file_magic::coff_object: {
     Expected<std::unique_ptr<ObjectFile>> ObjFile =
         ObjectFile::createObjectFile(Object, Type);

llvm/lib/Object/WasmObjectFile.cpp

@@ -1455,8 +1455,6 @@ bool WasmObjectFile::isSectionBSS(DataRefImpl Sec) const { return false; }
 
 bool WasmObjectFile::isSectionVirtual(DataRefImpl Sec) const { return false; }
 
-bool WasmObjectFile::isSectionBitcode(DataRefImpl Sec) const { return false; }
-
 relocation_iterator WasmObjectFile::section_rel_begin(DataRefImpl Ref) const {
   DataRefImpl RelocRef;
   RelocRef.d.a = Ref.d.a;