[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 ↗

(On Diff #326203)

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 ↗

(On Diff #333421)

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/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,
if (!CI->getCalledFunction())
     PGO.valueProfile(Builder, llvm::IPVK_IndirectCallTarget,
                      CI, CalleePtr);
 
   // In ObjC ARC mode with no ObjC ARC exception safety, tell the ARC
   // optimizer it can aggressively ignore unwind edges.
   if (CGM.getLangOpts().ObjCAutoRefCount)
     AddObjCARCExceptionMetadata(CI);
 
-  // Suppress tail calls if requested.
+  // Adjust tail call behavior based on TargetDecl's attributes and CallInfo.
   if (llvm::CallInst *Call = dyn_cast<llvm::CallInst>(CI)) {
     if (TargetDecl && TargetDecl->hasAttr<NotTailCalledAttr>())
       Call->setTailCallKind(llvm::CallInst::TCK_NoTail);
+    else if (CallInfo.getASTCallingConvention() == CallingConv::CC_SwiftAsync &&
+             CurFnInfo->getASTCallingConvention() == CallingConv::CC_SwiftAsync)
+      Call->setTailCallKind(llvm::CallInst::TCK_Tail);
   }
 
   // Add metadata for calls to MSAllocator functions
   if (getDebugInfo() && TargetDecl &&
       TargetDecl->hasAttr<MSAllocatorAttr>())
     getDebugInfo()->addHeapAllocSiteMetadata(CI, RetTy->getPointeeType(), Loc);
 
   // 4. Finish the call.

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
+
+// 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;
+}
+
+// CHECK-LABEL: swifttailcc void @async_branch
+// CHECK: tail call swifttailcc void @async_leaf1
+// CHECK: tail call swifttailcc void @async_leaf2
+SWIFTASYNCCALL void async_branch(_Bool b, char * ASYNC_CONTEXT ctx) {
+  if (b) {
+    return async_leaf1(ctx);
+  } else {
+    return async_leaf2(ctx);
+  }
+}
+
+// CHECK-LABEL: swifttailcc void @async_loop
+// CHECK: tail call swifttailcc void @async_leaf1
+// CHECK: tail call swifttailcc void @async_leaf2
+// CHECK: tail call swifttailcc void @async_loop
+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: swifttailcc void @async_mutual_loop
+// CHECK: tail call swifttailcc void @async_leaf
+// CHECK: tail call swifttailcc void @async_leaf
+// CHECK: tail call swifttailcc void @async_mutual_loop
+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: swifttailcc void @async_mutual_loop
+// CHECK: tail call swifttailcc void @async_leaf1
+// CHECK: tail call swifttailcc void @async_leaf2
+// CHECK: tail call swifttailcc void @async_mutual_loop1
+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(_Bool 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(_Bool b, unsigned u) {
+  char x = 'a';
+  async_branch(b, &x);
+  async_loop(u, &x);
+  return x;
+}
+

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 thumbv7-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()