[WebAssembly] Mangle the argc/argv `main` as `__wasm_argc_argv`.

WebAssembly enforces a rule that caller and callee signatures must
match. This means that the traditional technique of passing `main`
`argc` and `argv` even when it doesn't need them doesn't work.

Currently the backend renames `main` to `__original_main`, however this
doesn't interact well with LTO'ing libc, and the name isn't intuitive.
This patch allows us to transition to `__main_argc_argv` instead.

This implements the proposal in
https://github.com/WebAssembly/tool-conventions/pull/134
with a flag to disable it when targeting Emscripten, though this is
expected to be temporary, as discussed in the proposal comments.

Differential Revision: https://reviews.llvm.org/D70700

clang/lib/AST/Mangle.cpp

@@ -50,7 +50,8 @@ enum CCMangling {
   CCM_Fast,
   CCM_RegCall,
   CCM_Vector,
-  CCM_Std
+  CCM_Std,
+  CCM_WasmMainArgcArgv
 };
 
 static bool isExternC(const NamedDecl *ND) {
@@ -63,6 +64,16 @@ static CCMangling getCallingConvMangling(const ASTContext &Context,
                                          const NamedDecl *ND) {
   const TargetInfo &TI = Context.getTargetInfo();
   const llvm::Triple &Triple = TI.getTriple();
+
+  // On wasm, the argc/argv form of "main" is renamed so that the startup code
+  // can call it with the correct function signature.
+  // On Emscripten, users may be exporting "main" and expecting to call it
+  // themselves, so we can't mangle it.
+  if (Triple.isWasm() && !Triple.isOSEmscripten())
+    if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND))
+      if (FD->isMain() && FD->hasPrototype() && FD->param_size() == 2)
+        return CCM_WasmMainArgcArgv;
+
   if (!Triple.isOSWindows() || !Triple.isX86())
     return CCM_Other;
 
@@ -143,6 +154,12 @@ void MangleContext::mangleName(const NamedDecl *D, raw_ostream &Out) {
 
   const ASTContext &ASTContext = getASTContext();
   CCMangling CC = getCallingConvMangling(ASTContext, D);
+
+  if (CC == CCM_WasmMainArgcArgv) {
+    Out << "__main_argc_argv";
+    return;
+  }
+
   bool MCXX = shouldMangleCXXName(D);
   const TargetInfo &TI = Context.getTargetInfo();
   if (CC == CCM_Other || (MCXX && TI.getCXXABI() == TargetCXXABI::Microsoft)) {

clang/lib/CodeGen/CodeGenModule.cpp

@@ -448,6 +448,10 @@ void CodeGenModule::Release() {
     CodeGenFunction(*this).EmitCfiCheckStub();
   }
   emitAtAvailableLinkGuard();
+  if (Context.getTargetInfo().getTriple().isWasm() &&
+      !Context.getTargetInfo().getTriple().isOSEmscripten()) {
+    EmitMainVoidAlias();
+  }
   emitLLVMUsed();
   if (SanStats)
     SanStats->finish();
@@ -5600,6 +5604,17 @@ void CodeGenModule::EmitDeferredUnusedCoverageMappings() {
   }
 }
 
+void CodeGenModule::EmitMainVoidAlias() {
+  // In order to transition away from "__original_main" gracefully, emit an
+  // alias for "main" in the no-argument case so that libc can detect when
+  // new-style no-argument main is in used.
+  if (llvm::Function *F = getModule().getFunction("main")) {
+    if (!F->isDeclaration() && F->arg_size() == 0 && !F->isVarArg() &&
+        F->getReturnType()->isIntegerTy(Context.getTargetInfo().getIntWidth()))
+      addUsedGlobal(llvm::GlobalAlias::create("__main_void", F));
+  }
+}
+
 /// Turns the given pointer into a constant.
 static llvm::Constant *GetPointerConstant(llvm::LLVMContext &Context,
                                           const void *Ptr) {

clang/lib/CodeGen/CodeGenModule.h

@@ -1023,6 +1023,9 @@ public:
   /// for the uninstrumented functions.
   void EmitDeferredUnusedCoverageMappings();
 
+  /// Emit an alias for "main" if it has no arguments (needed for wasm).
+  void EmitMainVoidAlias();
+
   /// Tell the consumer that this variable has been instantiated.
   void HandleCXXStaticMemberVarInstantiation(VarDecl *VD);
 

clang/lib/Frontend/InitHeaderSearch.cpp

@@ -433,8 +433,7 @@ void InitHeaderSearch::AddDefaultIncludePaths(const LangOptions &Lang,
     break;
 
   case llvm::Triple::UnknownOS:
-    if (triple.getArch() == llvm::Triple::wasm32 ||
+    if (triple.isWasm())
-        triple.getArch() == llvm::Triple::wasm64)
       return;
     break;
   }

clang/test/CodeGen/wasm-call-main.c

@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple wasm32 -o - -emit-llvm %s | FileCheck %s
+
+// Mangle argc/argv main even when it's not defined in this TU.
+
+#include <stddef.h>
+
+int main(int argc, char *argv[]);
+
+int foo(void) {
+    return main(0, NULL);
+}
+
+// CHECK: call i32 @__main_argc_argv(

clang/test/CodeGen/wasm-main.c

@@ -0,0 +1,9 @@
+// RUN: %clang_cc1 -triple wasm32 -o - -emit-llvm %s | FileCheck %s
+
+// Don't mangle the no-arg form of main.
+
+int main(void) {
+  return 0;
+}
+
+// CHECK-LABEL: define i32 @main()

clang/test/CodeGen/wasm-main_argc_argv.c

@@ -0,0 +1,9 @@
+// RUN: %clang_cc1 -triple wasm32 -o - -emit-llvm %s | FileCheck %s
+
+// Mangle the argc/argv form of main.
+
+int main(int argc, char **argv) {
+  return 0;
+}
+
+// CHECK-LABEL: define i32 @__main_argc_argv(i32 %argc, i8** %argv)