[CodeGen] Fix codegen for __attribute__((swiftasynccall)).

[varungandhi-apple]

Summary

1. It should be mapped to LLVM's swifttailcc, which is now available.
2. We should make sure we generate a tail call instead of an ordinary call.

Fixes rdar://73762895.

[varungandhi-apple]

I am planning to add more tests, figured it is better to put up the patch sooner rather than later.

[varungandhi-apple]

Generate tail call when doing codegen for return statement.

[rjmccall]

Okay, seems fine to me.

clang/lib/CodeGen/CGStmt.cpp

1156

Hmm. I guess this should work in any situation where we can successfully actually do a tail call.

[rjmccall]

Please add some C++ tests, just in case.

[varungandhi-apple]

Added checks for C++.

[rjmccall]

LGTM

[varungandhi-apple]

Add null check for callee; instead of using getFunctionType() directly.

Otherwise, it leads to a crash when compiling gtest-all.cc.

[varungandhi-apple]

Use musttail call instead of tail call. (This patch needs to be marked as dependent on another patch which implements proper musttail support, but that is not available yet.)

[varungandhi-apple]

Fix codegen + add tests for methods and function pointers.

[rjmccall]

clang/lib/CodeGen/CGStmt.cpp

1240

What's with the explicit :: here?

[varungandhi-apple]

1. Remove '::' when calling static function.
2. Fix bug in function merging around missing musttail.
3. Add off-by-default check in verifier for swifttailcc->swifttailcc tail calls that are not marked musttail.

[varungandhi-apple]

Combined changes into D95561.

clang/lib/AST/ExprCXX.cpp

QualType CXXMemberCallExpr::getObjectType() const {
   return Ty;
 }
 
 CXXMethodDecl *CXXMemberCallExpr::getMethodDecl() const {
   if (const auto *MemExpr = dyn_cast<MemberExpr>(getCallee()->IgnoreParens()))
     return cast<CXXMethodDecl>(MemExpr->getMemberDecl());
 
   // FIXME: Will eventually need to cope with member pointers.
+  // NOTE: Update makeTailCallIfSwiftAsync on fixing this.
   return nullptr;
 }
 
 CXXRecordDecl *CXXMemberCallExpr::getRecordDecl() const {
   Expr* ThisArg = getImplicitObjectArgument();
   if (!ThisArg)
     return nullptr;
 

clang/lib/CodeGen/CGCall.cpp

unsigned CodeGenTypes::ClangCallConvToLLVMCallConv(CallingConv CC) {
   // TODO: Add support for __vectorcall to LLVM.
   case CC_X86VectorCall: return llvm::CallingConv::X86_VectorCall;
   case CC_AArch64VectorCall: return llvm::CallingConv::AArch64_VectorCall;
   case CC_SpirFunction: return llvm::CallingConv::SPIR_FUNC;
   case CC_OpenCLKernel: return CGM.getTargetCodeGenInfo().getOpenCLKernelCallingConv();
   case CC_PreserveMost: return llvm::CallingConv::PreserveMost;
   case CC_PreserveAll: return llvm::CallingConv::PreserveAll;
   case CC_Swift: return llvm::CallingConv::Swift;
-  // [FIXME: swiftasynccc] Update to SwiftAsync once LLVM support lands.
-  case CC_SwiftAsync:
-    return llvm::CallingConv::Swift;
+  case CC_SwiftAsync: return llvm::CallingConv::SwiftTail;
   }
 }
 
 /// Derives the 'this' type for codegen purposes, i.e. ignoring method CVR
 /// qualification. Either or both of RD and MD may be null. A null RD indicates
 /// that there is no meaningful 'this' type, and a null MD can occur when
 /// calling a method pointer.
 CanQualType CodeGenTypes::DeriveThisType(const CXXRecordDecl *RD,

clang/lib/CodeGen/CGStmt.cpp

SaveRetExprRAII(const Expr *RetExpr, CodeGenFunction &CGF)
     CGF.RetExpr = RetExpr;
   }
   ~SaveRetExprRAII() { CGF.RetExpr = OldRetExpr; }
   const Expr *OldRetExpr;
   CodeGenFunction &CGF;
 };
 } // namespace
 
+/// If we have 'return f(...);', where both caller and callee are SwiftAsync,
+/// codegen it as 'tail call ...; ret void;'.
+static void makeTailCallIfSwiftAsync(const CallExpr *CE, CGBuilderTy &Builder,
+                                     const CGFunctionInfo *CurFnInfo) {
+  auto calleeQualType = CE->getCallee()->getType();
+  const FunctionType *calleeType = nullptr;
+  if (calleeQualType->isFunctionPointerType() ||
+      calleeQualType->isFunctionReferenceType() ||
+      calleeQualType->isBlockPointerType() ||
+      calleeQualType->isMemberFunctionPointerType()) {
+    calleeType = calleeQualType->getPointeeType()->castAs<FunctionType>();
+  } else if (auto *ty = dyn_cast<FunctionType>(calleeQualType)) {
+    calleeType = ty;
+  } else if (auto CMCE = dyn_cast<CXXMemberCallExpr>(CE)) {
+    if (auto methodDecl = CMCE->getMethodDecl()) {
+      // getMethodDecl() doesn't handle member pointers at the moment.
+      calleeType = methodDecl->getType()->castAs<FunctionType>();
+    } else {
+      return;
+    }
+  } else {
+    return;
+  if (calleeType->getCallConv() == CallingConv::CC_SwiftAsync &&
+      (CurFnInfo->getASTCallingConvention() == CallingConv::CC_SwiftAsync)) {
+    auto CI = cast<llvm::CallInst>(&Builder.GetInsertBlock()->back());
+    CI->setTailCallKind(llvm::CallInst::TCK_MustTail);
+    Builder.CreateRetVoid();
+    Builder.ClearInsertionPoint();
+  }
+}
+
 /// EmitReturnStmt - Note that due to GCC extensions, this can have an operand
 /// if the function returns void, or may be missing one if the function returns
 /// non-void.  Fun stuff :).
 void CodeGenFunction::EmitReturnStmt(const ReturnStmt &S) {
   if (requiresReturnValueCheck()) {
     llvm::Constant *SLoc = EmitCheckSourceLocation(S.getBeginLoc());
     auto *SLocPtr =
         new llvm::GlobalVariable(CGM.getModule(), SLoc->getType(), false,
if (getLangOpts().ElideConstructors && S.getNRVOCandidate() &&
 
     // If there is an NRVO flag for this variable, set it to 1 into indicate
     // that the cleanup code should not destroy the variable.
     if (llvm::Value *NRVOFlag = NRVOFlags[S.getNRVOCandidate()])
       Builder.CreateFlagStore(Builder.getTrue(), NRVOFlag);
   } else if (!ReturnValue.isValid() || (RV && RV->getType()->isVoidType())) {
     // Make sure not to return anything, but evaluate the expression
     // for side effects.
-    if (RV)
+    if (RV) {
       EmitAnyExpr(RV);
+      if (auto *CE = dyn_cast<CallExpr>(RV))
+        ::makeTailCallIfSwiftAsync(CE, Builder, CurFnInfo);
+    }
   } else if (!RV) {
     // Do nothing (return value is left uninitialized)
   } else if (FnRetTy->isReferenceType()) {
     // If this function returns a reference, take the address of the expression
     // rather than the value.
     RValue Result = EmitReferenceBindingToExpr(RV);
     Builder.CreateStore(Result.getScalarVal(), ReturnValue);
   } else {

clang/test/CodeGen/64bit-swiftcall.c

 // RUN: %clang_cc1 -triple x86_64-apple-darwin10 -target-cpu core2 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=X86-64
 // RUN: %clang_cc1 -triple arm64-apple-ios9 -target-cpu cyclone -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=ARM64
 
 // REQUIRES: aarch64-registered-target,x86-registered-target
 
 #define SWIFTCALL __attribute__((swiftcall))
+#define SWIFTASYNCCALL __attribute__((swiftasynccall))
 #define OUT __attribute__((swift_indirect_result))
 #define ERROR __attribute__((swift_error_result))
 #define CONTEXT __attribute__((swift_context))
+#define ASYNC_CONTEXT __attribute__((swift_async_context))
 
 // CHECK-DAG: %struct.atomic_padded = type { { %struct.packed, [7 x i8] } }
 // CHECK-DAG: %struct.packed = type <{ i64, i8 }>
 //
 // CHECK: [[STRUCT2_RESULT:@.*]] = private {{.*}} constant [[STRUCT2_TYPE:%.*]] { i32 0, i8 0, i8 undef, i8 0, i32 0, i32 0 }
 
 /*****************************************************************************/
 /****************************** PARAMETER ABIS *******************************/
 /*****************************************************************************/
 
 SWIFTCALL void indirect_result_1(OUT int *arg0, OUT float *arg1) {}
 // CHECK-LABEL: define {{.*}} void @indirect_result_1(i32* noalias sret(i32*) align 4 dereferenceable(4){{.*}}, float* noalias align 4 dereferenceable(4){{.*}})
 
 // TODO: maybe this shouldn't suppress sret.
 SWIFTCALL int indirect_result_2(OUT int *arg0, OUT float *arg1) {  __builtin_unreachable(); }
 // CHECK-LABEL: define {{.*}} i32 @indirect_result_2(i32* noalias align 4 dereferenceable(4){{.*}}, float* noalias align 4 dereferenceable(4){{.*}})
 
 typedef struct { char array[1024]; } struct_reallybig;
 SWIFTCALL struct_reallybig indirect_result_3(OUT int *arg0, OUT float *arg1) { __builtin_unreachable(); }
 // CHECK-LABEL: define {{.*}} void @indirect_result_3({{.*}}* noalias sret(%struct.struct_reallybig) {{.*}}, i32* noalias align 4 dereferenceable(4){{.*}}, float* noalias align 4 dereferenceable(4){{.*}})
 
 SWIFTCALL void context_1(CONTEXT void *self) {}
 // CHECK-LABEL: define {{.*}} void @context_1(i8* swiftself
 
+SWIFTASYNCCALL void async_context_1(ASYNC_CONTEXT void *ctx) {}
+// CHECK-LABEL: define {{.*}} void @async_context_1(i8* swiftasync
+
 SWIFTCALL void context_2(void *arg0, CONTEXT void *self) {}
 // CHECK-LABEL: define {{.*}} void @context_2(i8*{{.*}}, i8* swiftself
 
+SWIFTASYNCCALL void async_context_2(void *arg0, ASYNC_CONTEXT void *ctx) {}
+// CHECK-LABEL: define {{.*}} void @async_context_2(i8*{{.*}}, i8* swiftasync
+
 SWIFTCALL void context_error_1(CONTEXT int *self, ERROR float **error) {}
 // CHECK-LABEL: define {{.*}} void @context_error_1(i32* swiftself{{.*}}, float** swifterror %0)
 // CHECK:       [[TEMP:%.*]] = alloca float*, align 8
 // CHECK:       [[T0:%.*]] = load float*, float** [[ERRORARG:%.*]], align 8
 // CHECK:       store float* [[T0]], float** [[TEMP]], align 8
 // CHECK:       [[T0:%.*]] = load float*, float** [[TEMP]], align 8
 // CHECK:       store float* [[T0]], float** [[ERRORARG]], align 8
 void test_context_error_1() {

clang/test/CodeGen/arm-swiftcall.c

 
 typedef struct { char array[1024]; } struct_reallybig;
 SWIFTCALL struct_reallybig indirect_result_3(OUT int *arg0, OUT float *arg1) { __builtin_unreachable(); }
 // CHECK-LABEL: define{{.*}} void @indirect_result_3({{.*}}* noalias sret(%struct.struct_reallybig) {{.*}}, i32* noalias align 4 dereferenceable(4){{.*}}, float* noalias align 4 dereferenceable(4){{.*}})
 
 SWIFTCALL void context_1(CONTEXT void *self) {}
 // CHECK-LABEL: define{{.*}} void @context_1(i8* swiftself
 
+SWIFTASYNCCALL void async_context_1(ASYNC_CONTEXT void *self) {}
+// CHECK-LABEL: define{{.*}} void @async_context_1(i8* swiftasync
+
 SWIFTCALL void context_2(void *arg0, CONTEXT void *self) {}
 // CHECK-LABEL: define{{.*}} void @context_2(i8*{{.*}}, i8* swiftself
 
+SWIFTASYNCCALL void async_context_2(void *arg0, ASYNC_CONTEXT void *self) {}
+// CHECK-LABEL: define{{.*}} void @async_context_2(i8*{{.*}}, i8* swiftasync
+
 SWIFTCALL void context_error_1(CONTEXT int *self, ERROR float **error) {}
 // CHECK-LABEL: define{{.*}} void @context_error_1(i32* swiftself{{.*}}, float** swifterror %0)
 // CHECK:       [[TEMP:%.*]] = alloca float*, align 4
 // CHECK:       [[T0:%.*]] = load float*, float** [[ERRORARG:%.*]], align 4
 // CHECK:       store float* [[T0]], float** [[TEMP]], align 4
 // CHECK:       [[T0:%.*]] = load float*, float** [[TEMP]], align 4
 // CHECK:       store float* [[T0]], float** [[ERRORARG]], align 4
 void test_context_error_1() {

 // CHECK:       store float* [[T0]], float** [[TEMP]], align 4
 // CHECK:       call [[SWIFTCC:swiftcc]] void @context_error_1(i32* swiftself [[X]], float** swifterror [[TEMP]])
 // CHECK:       [[T0:%.*]] = load float*, float** [[TEMP]], align 4
 // CHECK:       store float* [[T0]], float** [[ERROR]], align 4
 
 SWIFTCALL void context_error_2(short s, CONTEXT int *self, ERROR float **error) {}
 // CHECK-LABEL: define{{.*}} void @context_error_2(i16{{.*}}, i32* swiftself{{.*}}, float** swifterror %0)
 
-SWIFTASYNCCALL void async_context_1(ASYNC_CONTEXT void *self) {}
-// CHECK-LABEL: define {{.*}} void @async_context_1(i8* swiftasync
-
 /*****************************************************************************/
 /********************************** LOWERING *********************************/
 /*****************************************************************************/
 
 typedef float float3 __attribute__((ext_vector_type(3)));
 typedef float float4 __attribute__((ext_vector_type(4)));
 typedef float float8 __attribute__((ext_vector_type(8)));
 typedef double double2 __attribute__((ext_vector_type(2)));

clang/test/CodeGen/swift-async-call-conv.c

+// RUN: %clang_cc1 -triple x86_64-apple-darwin10 -target-cpu core2 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple arm64-apple-ios9 -target-cpu cyclone -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple armv7-apple-darwin9 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple armv7s-apple-ios9 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple armv7k-apple-ios9 -emit-llvm -o - %s | FileCheck %s
+
+// RUN: %clang_cc1 -x c++ -triple x86_64-apple-darwin10 -target-cpu core2 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=CPPONLY
+// RUN: %clang_cc1 -x c++ -triple arm64-apple-ios9 -target-cpu cyclone -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=CPPONLY
+// RUN: %clang_cc1 -x c++ -triple armv7-apple-darwin9 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=CPPONLY
+// RUN: %clang_cc1 -x c++ -triple armv7s-apple-ios9 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=CPPONLY
+// RUN: %clang_cc1 -x c++ -triple armv7k-apple-ios9 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK --check-prefix=CPPONLY
+
+// Test tail call behavior when a swiftasynccall function is called
+// from another swiftasynccall function.
+
+#define SWIFTCALL __attribute__((swiftcall))
+#define SWIFTASYNCCALL __attribute__((swiftasynccall))
+#define ASYNC_CONTEXT __attribute__((swift_async_context))
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_leaf1{{.*}}(i8* swiftasync
+SWIFTASYNCCALL void async_leaf1(char * ASYNC_CONTEXT ctx) {
+  *ctx += 1;
+}
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_leaf2{{.*}}(i8* swiftasync
+SWIFTASYNCCALL void async_leaf2(char * ASYNC_CONTEXT ctx) {
+  *ctx += 2;
+}
+
+#if __cplusplus
+  #define MYBOOL bool
+#else
+  #define MYBOOL _Bool
+#endif
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_branch{{.*}}i8* swiftasync
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf1
+// CHECK-NEXT: ret void
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf2
+// CHECK-NEXT: ret void
+SWIFTASYNCCALL void async_branch(MYBOOL b, char * ASYNC_CONTEXT ctx) {
+  if (b) {
+    return async_leaf1(ctx);
+  } else {
+    return async_leaf2(ctx);
+  }
+}
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_not_all_tail
+// CHECK-NOT:  musttail call swifttailcc void @{{.*}}async_leaf1
+// CHECK:      call swifttailcc void @{{.*}}async_leaf1
+// CHECK-NOT:  ret void
+// CHECK:      musttail call swifttailcc void @{{.*}}async_leaf2
+// CHECK-NEXT: ret void
+SWIFTASYNCCALL void async_not_all_tail(char * ASYNC_CONTEXT ctx) {
+  async_leaf1(ctx);
+  return async_leaf2(ctx);
+}
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_loop
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf1
+// CHECK-NEXT: ret void
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf2
+// CHECK-NEXT: ret void
+// CHECK: musttail call swifttailcc void @{{.*}}async_loop
+// CHECK-NEXT: ret void
+SWIFTASYNCCALL void async_loop(unsigned u, char * ASYNC_CONTEXT ctx) {
+  if (u == 0) {
+    return async_leaf1(ctx);
+  } else if (u == 1) {
+    return async_leaf2(ctx);
+  }
+  return async_loop(u - 2, ctx);
+}
+
+// Forward-declaration + mutual recursion is okay.
+
+SWIFTASYNCCALL void async_mutual_loop2(unsigned u, char * ASYNC_CONTEXT ctx);
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_mutual_loop1
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf1
+// CHECK-NEXT: ret void
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf2
+// CHECK-NEXT: ret void
+// There is some bugginess around FileCheck's greediness/matching,
+// so skipping the check for async_mutual_loop2 here.
+SWIFTASYNCCALL void async_mutual_loop1(unsigned u, char * ASYNC_CONTEXT ctx) {
+  if (u == 0) {
+    return async_leaf1(ctx);
+  } else if (u == 1) {
+    return async_leaf2(ctx);
+  }
+  return async_mutual_loop2(u - 2, ctx);
+}
+
+// CHECK-LABEL: swifttailcc void {{.*}}async_mutual_loop2
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf1
+// CHECK-NEXT: ret void
+// CHECK: musttail call swifttailcc void @{{.*}}async_leaf2
+// CHECK-NEXT: ret void
+// CHECK: musttail call swifttailcc void @{{.*}}async_mutual_loop1
+// CHECK-NEXT: ret void
+SWIFTASYNCCALL void async_mutual_loop2(unsigned u, char * ASYNC_CONTEXT ctx) {
+  if (u == 0) {
+    return async_leaf1(ctx);
+  } else if (u == 1) {
+    return async_leaf2(ctx);
+  }
+  return async_mutual_loop1(u - 2, ctx);
+}
+
+// When swiftasynccall functions are called by non-swiftasynccall functions,
+// the call isn't marked as a tail call.
+
+// CHECK-LABEL: swiftcc i8 {{.*}}sync_calling_async
+// CHECK-NOT: tail call
+// CHECK: call swifttailcc void @{{.*}}async_branch
+// CHECK-NOT: tail call
+// CHECK: call swifttailcc void @{{.*}}async_loop
+SWIFTCALL char sync_calling_async(MYBOOL b, unsigned u) {
+  char x = 'a';
+  async_branch(b, &x);
+  async_loop(u, &x);
+  return x;
+}
+
+// CHECK-LABEL: i8 {{.*}}c_calling_async
+// CHECK-NOT: tail call
+// CHECK: call swifttailcc void @{{.*}}async_branch
+// CHECK-NOT: tail call
+// CHECK: call swifttailcc void @{{.*}}async_loop
+char c_calling_async(MYBOOL b, unsigned u) {
+  char x = 'a';
+  async_branch(b, &x);
+  async_loop(u, &x);
+  return x;
+}
+
+#if __cplusplus
+struct S {
+  SWIFTASYNCCALL void (*fptr)(char * ASYNC_CONTEXT);
+
+  SWIFTASYNCCALL void async_leaf_method(char * ASYNC_CONTEXT ctx) {
+    *ctx += 1;
+  }
+  SWIFTASYNCCALL void async_nonleaf_method1(char * ASYNC_CONTEXT ctx) {
+    return async_leaf_method(ctx);
+  }
+  SWIFTASYNCCALL void async_nonleaf_method2(char * ASYNC_CONTEXT ctx) {
+    return this->async_leaf_method(ctx);
+  }
+};
+
+SWIFTASYNCCALL void (S::*async_leaf_method_ptr)(char * ASYNC_CONTEXT) = &S::async_leaf_method;
+
+// CPPONLY-LABEL: swifttailcc void {{.*}}async_struct_field_and_methods
+// CPPONLY: musttail call swifttailcc void %{{[0-9]+}}
+// CPPONLY: musttail call swifttailcc void @{{.*}}async_nonleaf_method1
+// CPPONLY: musttail call swifttailcc void %{{[0-9]+}}
+// CPPONLY: musttail call swifttailcc void @{{.*}}async_nonleaf_method2
+// CPPONLY-NOT: musttail call swifttailcc void @{{.*}}async_leaf_method
+// ^ TODO: Member pointers should also work.
+SWIFTASYNCCALL void async_struct_field_and_methods(int i, S &sref, S *sptr) {
+  char x = 'a';
+  if (i == 0) {
+    return (*sref.fptr)(&x);
+  } else if (i == 1) {
+    return sref.async_nonleaf_method1(&x);
+  } else if (i == 2) {
+    return (*(sptr->fptr))(&x);
+  } else if (i == 3) {
+    return sptr->async_nonleaf_method2(&x);
+  } else if (i == 4) {
+    return (sref.*async_leaf_method_ptr)(&x);
+  }
+  return (sptr->*async_leaf_method_ptr)(&x);
+}
+
+// CPPONLY-LABEL: define{{.*}} swifttailcc void @{{.*}}async_nonleaf_method1
+// CPPONLY: musttail call swifttailcc void @{{.*}}async_leaf_method
+
+// CPPONLY-LABEL: define{{.*}} swifttailcc void @{{.*}}async_nonleaf_method2
+// CPPONLY: musttail call swifttailcc void @{{.*}}async_leaf_method
+#endif

clang/test/CodeGen/swift-call-conv.c

 // RUN: %clang_cc1 -triple aarch64-unknown-windows-msvc -emit-llvm %s -o - | FileCheck %s
 // RUN: %clang_cc1 -triple x86_64-unknown-windows-msvc -emit-llvm %s -o - | FileCheck %s
 
 // REQUIRES: aarch64-registered-target,arm-registered-target,x86-registered-target
 
 void __attribute__((__swiftcall__)) f(void) {}
 // CHECK-LABEL: define dso_local swiftcc void @f()
 
+void __attribute__((__swiftasynccall__)) f_async(void) {}
+// CHECK-LABEL: define dso_local swifttailcc void @f_async()