[CIR] Initial support for emitting coroutine body

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -2341,6 +2341,12 @@ def CIR_FuncOp : CIR_Op<"func", [
     The function linkage information is specified by `linkage`, as defined by
     `GlobalLinkageKind` attribute.
 
+    A compiler builtin function must be marked as `builtin` for further
+    processing when lowering from CIR.
+
+    The `coroutine` keyword is used to mark coroutine function, which requires
+    at least one `cir.await` instruction to be used in its body.
+
     The `lambda` translates to a C++ `operator()` that implements a lambda, this
     allow callsites to make certain assumptions about the real function nature
     when writing analysis.
@@ -2362,11 +2368,22 @@ def CIR_FuncOp : CIR_Op<"func", [
     // Linkage information
     cir.func linkonce_odr @some_method(...)
     ```
+    // Builtin function
+    cir.func builtin @__builtin_coro_end(!cir.ptr<i8>, !cir.bool) -> !cir.bool
+    // Coroutine
+    cir.func coroutine @_Z10silly_taskv() -> !CoroTask {
+      ...
+      cir.await(...)
+      ...
+    }
+    ```
   }];
 
   let arguments = (ins SymbolNameAttr:$sym_name,
                        CIR_VisibilityAttr:$global_visibility,
                        TypeAttrOf<CIR_FuncType>:$function_type,
+                       UnitAttr:$builtin,
+                       UnitAttr:$coroutine,
                        UnitAttr:$lambda,
                        UnitAttr:$no_proto,
                        UnitAttr:$dso_local,

clang/include/clang/CIR/MissingFeatures.h

@@ -137,6 +137,13 @@ struct MissingFeatures {
   static bool recordZeroInitPadding() { return false; }
   static bool zeroSizeRecordMembers() { return false; }
 
+  // Coroutines
+  static bool coroAllocBuiltinCall() { return false; }
+  static bool coroBeginBuiltinCall() { return false; }
+  static bool coroEndBuiltinCall() { return false; }
+  static bool coroSizeBuiltinCall() { return false; }
+  static bool coroutineFrame() { return false; }
+
   // Various handling of deferred processing in CIRGenModule.
   static bool cgmRelease() { return false; }
   static bool deferredVtables() { return false; }

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -428,6 +428,31 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl &gd, unsigned builtinID,
     return emitUnaryFPBuiltin<cir::ATanOp>(*this, *e);
   case Builtin::BI__builtin_elementwise_cos:
     return emitUnaryFPBuiltin<cir::CosOp>(*this, *e);
+  case Builtin::BI__builtin_coro_id:
+  case Builtin::BI__builtin_coro_promise:
+  case Builtin::BI__builtin_coro_resume:
+  case Builtin::BI__builtin_coro_noop:
+  case Builtin::BI__builtin_coro_destroy:
+  case Builtin::BI__builtin_coro_done:
+  case Builtin::BI__builtin_coro_alloc:
+  case Builtin::BI__builtin_coro_begin:
+  case Builtin::BI__builtin_coro_end:
+  case Builtin::BI__builtin_coro_suspend:
+  case Builtin::BI__builtin_coro_align:
+    llvm_unreachable("BI__builtin_coro_id like NYI");
+
+  case Builtin::BI__builtin_coro_frame: {
+    cgm.errorNYI(e->getSourceRange(), "BI__builtin_coro_frame NYI");
+    assert(!cir::MissingFeatures::coroutineFrame());
+    return getUndefRValue(e->getType());
+  }
+  case Builtin::BI__builtin_coro_free:
+  case Builtin::BI__builtin_coro_size: {
+    cgm.errorNYI(e->getSourceRange(),
+                 "BI__builtin_coro_free, BI__builtin_coro_size NYI");
+    assert(!cir::MissingFeatures::coroSizeBuiltinCall());
+    return getUndefRValue(e->getType());
+  }
   }
 
   // If this is an alias for a lib function (e.g. __builtin_sin), emit

clang/lib/CIR/CodeGen/CIRGenCoroutine.cpp

@@ -0,0 +1,85 @@
+//===----- CGCoroutine.cpp - Emit CIR Code for C++ coroutines -------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This contains code dealing with C++ code generation of coroutines.
+//
+//===----------------------------------------------------------------------===//
+
+#include "CIRGenFunction.h"
+#include "mlir/Support/LLVM.h"
+#include "clang/AST/StmtCXX.h"
+#include "clang/Basic/TargetInfo.h"
+#include "clang/CIR/Dialect/IR/CIRTypes.h"
+
+using namespace clang;
+using namespace clang::CIRGen;
+
+struct clang::CIRGen::CGCoroData {
+  // Stores the __builtin_coro_id emitted in the function so that we can supply
+  // it as the first argument to other builtins.
+  cir::CallOp coroId = nullptr;
+};
+
+// Defining these here allows to keep CGCoroData private to this file.
+CIRGenFunction::CGCoroInfo::CGCoroInfo() {}
+CIRGenFunction::CGCoroInfo::~CGCoroInfo() {}
+
+static void createCoroData(CIRGenFunction &cgf,
+                           CIRGenFunction::CGCoroInfo &curCoro,
+                           cir::CallOp coroId) {
+  if (curCoro.data) {
+    llvm_unreachable("EmitCoroutineBodyStatement called twice?");
+    return;
+  }
+
+  curCoro.data = std::unique_ptr<CGCoroData>(new CGCoroData);
+  curCoro.data->coroId = coroId;
+}
+
+cir::CallOp CIRGenFunction::emitCoroIDBuiltinCall(mlir::Location loc,
+                                                  mlir::Value nullPtr) {
+  cir::IntType int32Ty = builder.getUInt32Ty();
+
+  const TargetInfo &ti = cgm.getASTContext().getTargetInfo();
+  unsigned newAlign = ti.getNewAlign() / ti.getCharWidth();
+
+  mlir::Operation *builtin = cgm.getGlobalValue(cgm.builtinCoroId);
+
+  cir::FuncOp fnOp;
+  if (!builtin) {
+    fnOp = cgm.createCIRBuiltinFunction(
+        loc, cgm.builtinCoroId,
+        cir::FuncType::get({int32Ty, VoidPtrTy, VoidPtrTy, VoidPtrTy}, int32Ty),
+        /*FD=*/nullptr);
+    assert(fnOp && "should always succeed");
+  } else
+    fnOp = cast<cir::FuncOp>(builtin);
+
+  return builder.createCallOp(loc, fnOp,
+                              mlir::ValueRange{builder.getUInt32(newAlign, loc),
+                                               nullPtr, nullPtr, nullPtr});
+}
+
+mlir::LogicalResult
+CIRGenFunction::emitCoroutineBody(const CoroutineBodyStmt &s) {
+  mlir::Location openCurlyLoc = getLoc(s.getBeginLoc());
+  cir::ConstantOp nullPtrCst = builder.getNullPtr(VoidPtrTy, openCurlyLoc);
+
+  auto fn = dyn_cast<cir::FuncOp>(curFn);
+  assert(fn && "other callables are NYI");
+  fn.setCoroutine(true);
+  cir::CallOp coroId = emitCoroIDBuiltinCall(openCurlyLoc, nullPtrCst);
+  createCoroData(*this, curCoro, coroId);
+
+  assert(!cir::MissingFeatures::coroAllocBuiltinCall());
+
+  assert(!cir::MissingFeatures::coroBeginBuiltinCall());
+
+  assert(!cir::MissingFeatures::generateDebugInfo());
+  return mlir::success();
+}

clang/lib/CIR/CodeGen/CIRGenFunction.cpp

@@ -342,6 +342,9 @@ void CIRGenFunction::LexicalScope::cleanup() {
 cir::ReturnOp CIRGenFunction::LexicalScope::emitReturn(mlir::Location loc) {
   CIRGenBuilderTy &builder = cgf.getBuilder();
 
+  // If we are on a coroutine, add the coro_end builtin call.
+  assert(!cir::MissingFeatures::coroEndBuiltinCall());
+
   auto fn = dyn_cast<cir::FuncOp>(cgf.curFn);
   assert(fn && "emitReturn from non-function");
   if (!fn.getFunctionType().hasVoidReturn()) {

clang/lib/CIR/CodeGen/CIRGenFunction.h

@@ -47,6 +47,8 @@ class LoopOp;
 
 namespace clang::CIRGen {
 
+struct CGCoroData;
+
 class CIRGenFunction : public CIRGenTypeCache {
 public:
   CIRGenModule &cgm;
@@ -66,6 +68,18 @@ class CIRGenFunction : public CIRGenTypeCache {
   /// The compiler-generated variable that holds the return value.
   std::optional<mlir::Value> fnRetAlloca;
 
+  // Holds coroutine data if the current function is a coroutine. We use a
+  // wrapper to manage its lifetime, so that we don't have to define CGCoroData
+  // in this header.
+  struct CGCoroInfo {
+    std::unique_ptr<CGCoroData> data;
+    CGCoroInfo();
+    ~CGCoroInfo();
+  };
+  CGCoroInfo curCoro;
+
+  bool isCoroutine() const { return curCoro.data != nullptr; }
+
   /// The temporary alloca to hold the return value. This is
   /// invalid iff the function has no return value.
   Address returnValue = Address::invalid();
@@ -1174,6 +1188,10 @@ class CIRGenFunction : public CIRGenTypeCache {
 
   void emitConstructorBody(FunctionArgList &args);
 
+  mlir::LogicalResult emitCoroutineBody(const CoroutineBodyStmt &s);
+  cir::CallOp emitCoroEndBuiltinCall(mlir::Location loc, mlir::Value nullPtr);
+  cir::CallOp emitCoroIDBuiltinCall(mlir::Location loc, mlir::Value nullPtr);
+
   void emitDestroy(Address addr, QualType type, Destroyer *destroyer);
 
   void emitDestructorBody(FunctionArgList &args);

clang/lib/CIR/CodeGen/CIRGenModule.cpp

@@ -2065,6 +2065,15 @@ CIRGenModule::createCIRFunction(mlir::Location loc, StringRef name,
   return func;
 }
 
+cir::FuncOp
+CIRGenModule::createCIRBuiltinFunction(mlir::Location loc, StringRef name,
+                                       cir::FuncType ty,
+                                       const clang::FunctionDecl *fd) {
+  cir::FuncOp fnOp = createCIRFunction(loc, name, ty, fd);
+  fnOp.setBuiltin(true);
+  return fnOp;
+}
+
 mlir::SymbolTable::Visibility
 CIRGenModule::getMLIRVisibility(cir::GlobalOp op) {
   // MLIR doesn't accept public symbols declarations (only

clang/lib/CIR/CodeGen/CIRGenModule.h

@@ -473,6 +473,13 @@ class CIRGenModule : public CIRGenTypeCache {
                                 cir::FuncType funcType,
                                 const clang::FunctionDecl *funcDecl);
 
+  /// Create a CIR function with builtin attribute set.
+  cir::FuncOp createCIRBuiltinFunction(mlir::Location loc, llvm::StringRef name,
+                                       cir::FuncType ty,
+                                       const clang::FunctionDecl *fd);
+
+  static constexpr const char *builtinCoroId = "__builtin_coro_id";
+
   /// Given a builtin id for a function like "__builtin_fabsf", return a
   /// Function* for "fabsf".
   cir::FuncOp getBuiltinLibFunction(const FunctionDecl *fd, unsigned builtinID);

clang/lib/CIR/CodeGen/CIRGenStmt.cpp

@@ -197,6 +197,7 @@ mlir::LogicalResult CIRGenFunction::emitStmt(const Stmt *s,
   case Stmt::SEHLeaveStmtClass:
   case Stmt::SYCLKernelCallStmtClass:
   case Stmt::CoroutineBodyStmtClass:
+    return emitCoroutineBody(cast<CoroutineBodyStmt>(*s));
   case Stmt::CoreturnStmtClass:
   case Stmt::CXXTryStmtClass:
   case Stmt::IndirectGotoStmtClass:

clang/lib/CIR/CodeGen/CMakeLists.txt

@@ -14,6 +14,7 @@ add_clang_library(clangCIR
   CIRGenCall.cpp
   CIRGenClass.cpp
   CIRGenCleanup.cpp
+  CIRGenCoroutine.cpp
   CIRGenCXX.cpp
   CIRGenCXXABI.cpp
   CIRGenBuiltin.cpp

clang/lib/CIR/Dialect/IR/CIRDialect.cpp

@@ -1632,12 +1632,19 @@ ParseResult cir::FuncOp::parse(OpAsmParser &parser, OperationState &state) {
   llvm::SMLoc loc = parser.getCurrentLocation();
   mlir::Builder &builder = parser.getBuilder();
 
+  mlir::StringAttr builtinNameAttr = getBuiltinAttrName(state.name);
+  mlir::StringAttr coroutineNameAttr = getCoroutineAttrName(state.name);
   mlir::StringAttr lambdaNameAttr = getLambdaAttrName(state.name);
   mlir::StringAttr noProtoNameAttr = getNoProtoAttrName(state.name);
   mlir::StringAttr visNameAttr = getSymVisibilityAttrName(state.name);
   mlir::StringAttr visibilityNameAttr = getGlobalVisibilityAttrName(state.name);
   mlir::StringAttr dsoLocalNameAttr = getDsoLocalAttrName(state.name);
 
+  if (::mlir::succeeded(parser.parseOptionalKeyword(builtinNameAttr.strref())))
+    state.addAttribute(builtinNameAttr, parser.getBuilder().getUnitAttr());
+  if (::mlir::succeeded(
+          parser.parseOptionalKeyword(coroutineNameAttr.strref())))
+    state.addAttribute(coroutineNameAttr, parser.getBuilder().getUnitAttr());
   if (::mlir::succeeded(parser.parseOptionalKeyword(lambdaNameAttr.strref())))
     state.addAttribute(lambdaNameAttr, parser.getBuilder().getUnitAttr());
   if (parser.parseOptionalKeyword(noProtoNameAttr).succeeded())
@@ -1747,6 +1754,12 @@ mlir::Region *cir::FuncOp::getCallableRegion() {
 }
 
 void cir::FuncOp::print(OpAsmPrinter &p) {
+  if (getBuiltin())
+    p << " builtin";
+
+  if (getCoroutine())
+    p << " coroutine";
+
   if (getLambda())
     p << " lambda";