diff --git a/cl/_testgo/goroutine/in.go b/cl/_testgo/goroutine/in.go
new file mode 100644
index 00000000..85def841
--- /dev/null
+++ b/cl/_testgo/goroutine/in.go
@@ -0,0 +1,12 @@
+package main
+
+func main() {
+	done := false
+	go func() {
+		println("Hello, World!")
+		done = true
+	}()
+	for !done {
+		print(".")
+	}
+}
diff --git a/cl/_testgo/goroutine/out.ll b/cl/_testgo/goroutine/out.ll
new file mode 100644
index 00000000..00c29c48
--- /dev/null
+++ b/cl/_testgo/goroutine/out.ll
@@ -0,0 +1,89 @@
+; ModuleID = 'main'
+source_filename = "main"
+
+%"github.com/goplus/llgo/internal/runtime.String" = type { ptr, i64 }
+
+@"main.init$guard" = global ptr null
+@__llgo_argc = global ptr null
+@__llgo_argv = global ptr null
+@0 = private unnamed_addr constant [2 x i8] c".\00", align 1
+@1 = private unnamed_addr constant [14 x i8] c"Hello, World!\00", align 1
+
+define void @main.init() {
+_llgo_0:
+  %0 = load i1, ptr @"main.init$guard", align 1
+  br i1 %0, label %_llgo_2, label %_llgo_1
+
+_llgo_1:                                          ; preds = %_llgo_0
+  store i1 true, ptr @"main.init$guard", align 1
+  br label %_llgo_2
+
+_llgo_2:                                          ; preds = %_llgo_1, %_llgo_0
+  ret void
+}
+
+define i32 @main(i32 %0, ptr %1) {
+_llgo_0:
+  store i32 %0, ptr @__llgo_argc, align 4
+  store ptr %1, ptr @__llgo_argv, align 8
+  call void @"github.com/goplus/llgo/internal/runtime.init"()
+  call void @main.init()
+  %2 = call ptr @"github.com/goplus/llgo/internal/runtime.AllocZ"(i64 1)
+  store i1 false, ptr %2, align 1
+  %3 = call ptr @"github.com/goplus/llgo/internal/runtime.AllocU"(i64 8)
+  %4 = getelementptr inbounds { ptr }, ptr %3, i32 0, i32 0
+  store ptr %2, ptr %4, align 8
+  %5 = alloca { ptr, ptr }, align 8
+  %6 = getelementptr inbounds { ptr, ptr }, ptr %5, i32 0, i32 0
+  store ptr @"main.main$1", ptr %6, align 8
+  %7 = getelementptr inbounds { ptr, ptr }, ptr %5, i32 0, i32 1
+  store ptr %3, ptr %7, align 8
+  %8 = load { ptr, ptr }, ptr %5, align 8
+  %9 = extractvalue { ptr, ptr } %8, 1
+  %10 = extractvalue { ptr, ptr } %8, 0
+  call void %10(ptr %9)
+  br label %_llgo_3
+
+_llgo_1:                                          ; preds = %_llgo_3
+  %11 = alloca %"github.com/goplus/llgo/internal/runtime.String", align 8
+  %12 = getelementptr inbounds %"github.com/goplus/llgo/internal/runtime.String", ptr %11, i32 0, i32 0
+  store ptr @0, ptr %12, align 8
+  %13 = getelementptr inbounds %"github.com/goplus/llgo/internal/runtime.String", ptr %11, i32 0, i32 1
+  store i64 1, ptr %13, align 4
+  %14 = load %"github.com/goplus/llgo/internal/runtime.String", ptr %11, align 8
+  call void @"github.com/goplus/llgo/internal/runtime.PrintString"(%"github.com/goplus/llgo/internal/runtime.String" %14)
+  br label %_llgo_3
+
+_llgo_2:                                          ; preds = %_llgo_3
+  ret i32 0
+
+_llgo_3:                                          ; preds = %_llgo_1, %_llgo_0
+  %15 = load i1, ptr %2, align 1
+  br i1 %15, label %_llgo_2, label %_llgo_1
+}
+
+declare void @"github.com/goplus/llgo/internal/runtime.init"()
+
+declare ptr @"github.com/goplus/llgo/internal/runtime.AllocZ"(i64)
+
+define void @"main.main$1"(ptr %0) {
+_llgo_0:
+  %1 = alloca %"github.com/goplus/llgo/internal/runtime.String", align 8
+  %2 = getelementptr inbounds %"github.com/goplus/llgo/internal/runtime.String", ptr %1, i32 0, i32 0
+  store ptr @1, ptr %2, align 8
+  %3 = getelementptr inbounds %"github.com/goplus/llgo/internal/runtime.String", ptr %1, i32 0, i32 1
+  store i64 13, ptr %3, align 4
+  %4 = load %"github.com/goplus/llgo/internal/runtime.String", ptr %1, align 8
+  call void @"github.com/goplus/llgo/internal/runtime.PrintString"(%"github.com/goplus/llgo/internal/runtime.String" %4)
+  call void @"github.com/goplus/llgo/internal/runtime.PrintByte"(i8 10)
+  %5 = load { ptr }, ptr %0, align 8
+  %6 = extractvalue { ptr } %5, 0
+  store i1 true, ptr %6, align 1
+  ret void
+}
+
+declare ptr @"github.com/goplus/llgo/internal/runtime.AllocU"(i64)
+
+declare void @"github.com/goplus/llgo/internal/runtime.PrintString"(%"github.com/goplus/llgo/internal/runtime.String")
+
+declare void @"github.com/goplus/llgo/internal/runtime.PrintByte"(i8)
diff --git a/cl/_testgo/print/in.go b/cl/_testgo/print/in.go
new file mode 100644
index 00000000..bf63848a
--- /dev/null
+++ b/cl/_testgo/print/in.go
@@ -0,0 +1,5 @@
+package main
+
+func main() {
+	println('.', byte('.'))
+}
diff --git a/cl/_testgo/print/out.ll b/cl/_testgo/print/out.ll
new file mode 100644
index 00000000..dd4a2524
--- /dev/null
+++ b/cl/_testgo/print/out.ll
@@ -0,0 +1,40 @@
+; ModuleID = 'main'
+source_filename = "main"
+
+@"main.init$guard" = global ptr null
+@__llgo_argc = global ptr null
+@__llgo_argv = global ptr null
+
+define void @main.init() {
+_llgo_0:
+  %0 = load i1, ptr @"main.init$guard", align 1
+  br i1 %0, label %_llgo_2, label %_llgo_1
+
+_llgo_1:                                          ; preds = %_llgo_0
+  store i1 true, ptr @"main.init$guard", align 1
+  br label %_llgo_2
+
+_llgo_2:                                          ; preds = %_llgo_1, %_llgo_0
+  ret void
+}
+
+define i32 @main(i32 %0, ptr %1) {
+_llgo_0:
+  store i32 %0, ptr @__llgo_argc, align 4
+  store ptr %1, ptr @__llgo_argv, align 8
+  call void @"github.com/goplus/llgo/internal/runtime.init"()
+  call void @main.init()
+  call void @"github.com/goplus/llgo/internal/runtime.PrintInt"(i64 46)
+  call void @"github.com/goplus/llgo/internal/runtime.PrintByte"(i8 32)
+  call void @"github.com/goplus/llgo/internal/runtime.PrintUint"(i64 46)
+  call void @"github.com/goplus/llgo/internal/runtime.PrintByte"(i8 10)
+  ret i32 0
+}
+
+declare void @"github.com/goplus/llgo/internal/runtime.init"()
+
+declare void @"github.com/goplus/llgo/internal/runtime.PrintInt"(i64)
+
+declare void @"github.com/goplus/llgo/internal/runtime.PrintByte"(i8)
+
+declare void @"github.com/goplus/llgo/internal/runtime.PrintUint"(i64)
diff --git a/cl/compile.go b/cl/compile.go
index f5708851..05d78220 100644
--- a/cl/compile.go
+++ b/cl/compile.go
@@ -548,6 +548,72 @@ func (p *context) compilePhi(b llssa.Builder, v *ssa.Phi) (ret llssa.Expr) {
 	return
 }
 
+func (p *context) call(b llssa.Builder, act llssa.DoAction, call *ssa.CallCommon) (ret llssa.Expr) {
+	cv := call.Value
+	if mthd := call.Method; mthd != nil {
+		o := p.compileValue(b, cv)
+		fn := b.Imethod(o, mthd)
+		args := p.compileValues(b, call.Args, fnNormal)
+		ret = b.Do(act, fn, args...)
+		return
+	}
+	kind := p.funcKind(cv)
+	if kind == fnIgnore {
+		return
+	}
+	args := call.Args
+	if debugGoSSA {
+		log.Println(">>> Do", act, cv, args)
+	}
+	switch cv := cv.(type) {
+	case *ssa.Builtin:
+		fn := cv.Name()
+		if fn == "ssa:wrapnilchk" { // TODO(xsw): check nil ptr
+			arg := args[0]
+			ret = p.compileValue(b, arg)
+			// log.Println("wrapnilchk:", ret.TypeOf())
+		} else {
+			args := p.compileValues(b, args, kind)
+			ret = b.BuiltinDo(act, fn, args...)
+		}
+	case *ssa.Function:
+		aFn, pyFn, ftype := p.compileFunction(cv)
+		// TODO(xsw): check ca != llssa.Call
+		switch ftype {
+		case goFunc, cFunc:
+			args := p.compileValues(b, args, kind)
+			ret = b.Do(act, aFn.Expr, args...)
+		case pyFunc:
+			args := p.compileValues(b, args, kind)
+			ret = b.Do(act, pyFn.Expr, args...)
+		case llgoPyList:
+			args := p.compileValues(b, args, fnHasVArg)
+			ret = b.PyList(args...)
+		case llgoCstr:
+			ret = cstr(b, args)
+		case llgoAdvance:
+			ret = p.advance(b, args)
+		case llgoIndex:
+			ret = p.index(b, args)
+		case llgoAlloca:
+			ret = p.alloca(b, args)
+		case llgoAllocaCStr:
+			ret = p.allocaCStr(b, args)
+		case llgoStringData:
+			ret = p.stringData(b, args)
+		case llgoUnreachable: // func unreachable()
+			b.Unreachable()
+		default:
+			log.Panicln("unknown ftype:", ftype)
+		}
+	default:
+		fn := p.compileValue(b, cv)
+		args := p.compileValues(b, args, kind)
+		ret = b.Do(act, fn, args...)
+	}
+	return
+}
+
 func (p *context) compileInstrOrValue(b llssa.Builder, iv instrOrValue, asValue bool) (ret llssa.Expr) {
 	if asValue {
 		if v, ok := p.bvals[iv]; ok {
@@ -557,67 +623,7 @@ func (p *context) compileInstrOrValue(b llssa.Builder, iv instrOrValue, asValue
 	}
 	switch v := iv.(type) {
 	case *ssa.Call:
-		cv := v.Call.Value
-		if mthd := v.Call.Method; mthd != nil {
-			o := p.compileValue(b, cv)
-			fn := b.Imethod(o, v.Call.Method)
-			args := p.compileValues(b, v.Call.Args, fnNormal)
-			ret = b.Call(fn, args...)
-			break
-		}
-		kind := p.funcKind(cv)
-		if kind == fnIgnore {
-			return
-		}
-		args := v.Call.Args
-		if debugGoSSA {
-			log.Println(">>> Call", cv, args)
-		}
-		switch cv := cv.(type) {
-		case *ssa.Builtin:
-			fn := cv.Name()
-			if fn == "ssa:wrapnilchk" { // TODO(xsw): check nil ptr
-				arg := args[0]
-				ret = p.compileValue(b, arg)
-				// log.Println("wrapnilchk:", ret.TypeOf())
-			} else {
-				args := p.compileValues(b, args, kind)
-				ret = b.BuiltinCall(fn, args...)
-			}
-		case *ssa.Function:
-			aFn, pyFn, ftype := p.compileFunction(cv)
-			switch ftype {
-			case goFunc, cFunc:
-				args := p.compileValues(b, args, kind)
-				ret = b.Call(aFn.Expr, args...)
-			case pyFunc:
-				args := p.compileValues(b, args, kind)
-				ret = b.Call(pyFn.Expr, args...)
-			case llgoPyList:
-				args := p.compileValues(b, args, fnHasVArg)
-				ret = b.PyList(args...)
-			case llgoCstr:
-				ret = cstr(b, args)
-			case llgoAdvance:
-				ret = p.advance(b, args)
-			case llgoIndex:
-				ret = p.index(b, args)
-			case llgoAlloca:
-				ret = p.alloca(b, args)
-			case llgoAllocaCStr:
-				ret = p.allocaCStr(b, args)
-			case llgoStringData:
-				ret = p.stringData(b, args)
-			case llgoUnreachable: // func unreachable()
-				b.Unreachable()
-			default:
-				log.Panicln("unknown ftype:", ftype)
-			}
-		default:
-			fn := p.compileValue(b, cv)
-			args := p.compileValues(b, args, kind)
-			ret = b.Call(fn, args...)
-		}
+		ret = p.call(b, llssa.Call, &v.Call)
 	case *ssa.BinOp:
 		x := p.compileValue(b, v.X)
 		y := p.compileValue(b, v.Y)
@@ -791,6 +797,8 @@ func (p *context) compileInstr(b llssa.Builder, instr ssa.Instruction) {
 		key := p.compileValue(b, v.Key)
 		val := p.compileValue(b, v.Value)
 		b.MapUpdate(m, key, val)
+	case *ssa.Go:
+		p.call(b, llssa.Go, &v.Call)
 	case *ssa.Panic:
 		arg := p.compileValue(b, v.X)
 		b.Panic(arg)
diff --git a/ssa/expr.go b/ssa/expr.go
index 1a6836cd..9c3cfae0 100644
--- a/ssa/expr.go
+++ b/ssa/expr.go
@@ -892,13 +892,26 @@ func (b Builder) InlineCall(fn Expr, args ...Expr) (ret Expr) {
 //	t2 = println(t0, t1)
 //	t4 = t3()
 func (b Builder) Call(fn Expr, args ...Expr) (ret Expr) {
+	return b.Do(Call, fn, args...)
+}
+
+type DoAction int
+
+const (
+	Call DoAction = iota
+	Go
+	Defer
+)
+
+// Do call a function with an action.
+func (b Builder) Do(da DoAction, fn Expr, args ...Expr) (ret Expr) {
 	if debugInstr {
 		var b bytes.Buffer
 		name := fn.impl.Name()
 		if name == "" {
 			name = "closure"
 		}
-		fmt.Fprint(&b, "Call ", fn.kind, " ", fn.raw.Type, " ", name)
+		fmt.Fprint(&b, "Do ", da, " ", fn.kind, " ", fn.raw.Type, " ", name)
 		sep := ": "
 		for _, arg := range args {
 			fmt.Fprint(&b, sep, arg.impl)
@@ -987,6 +1000,11 @@ func (b Builder) Next(iter Expr, isString bool) (ret Expr) {
 // `fn` indicates the function: one of the built-in functions from the
 // Go spec (excluding "make" and "new").
 func (b Builder) BuiltinCall(fn string, args ...Expr) (ret Expr) {
+	return b.BuiltinDo(Call, fn, args...)
+}
+
+// BuiltinDo call a builtin function with an action.
+func (b Builder) BuiltinDo(da DoAction, fn string, args ...Expr) (ret Expr) {
 	switch fn {
 	case "len":
 		if len(args) == 1 {