[IROutliner] Disallow functions that return twice

[AndrewLitteken]

Summary

Functions that return twice can cause the IR Outliner to miscompile the given program. These function rely on information about the stack to be the same, and this may not necessarily be the case if called from an outlined function. So, we simply call these instructions illegal for the outliner to remove.

[paquette]

LGTM

llvm/include/llvm/Transforms/IPO/IROutliner.h

struct InstructionAllowed : public InstVisitor<InstructionAllowed, bool> {
     // handled.
     bool IntrinsicInst(IntrinsicInst &II) { return false; }
     // We only handle CallInsts that are not indirect, since we cannot guarantee
     // that they have a name in these cases.
     bool visitCallInst(CallInst &CI) {
       Function *F = CI.getCalledFunction();
       if (!F || CI.isIndirectCall() || !F->hasName())
         return false;
+      // Returning twice can cause issues with the state of the function call
+      // that were not expected when the function was used, so we do not include
+      // the call in outlined functions.
+      if (CI.canReturnTwice())
+        return false;
       return true;
     }
     // TODO: Handle FreezeInsts.  Since a frozen value could be frozen inside
     // the outlined region, and then returned as an output, this will have to be
     // handled differently.
     bool visitFreezeInst(FreezeInst &CI) { return false; }
     // TODO: We do not current handle similarity that changes the control flow.
     bool visitInvokeInst(InvokeInst &II) { return false; }

llvm/test/Transforms/IROutliner/illegal-returns-twice.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S -verify -iroutliner -ir-outlining-no-cost < %s | FileCheck %s
+
+; This test checks that we do not outline functions that are marked as returns
+; twice, since these can alter the frame of the function and affect how the
+; outliner behaves, causing miscompiles.
+
+; Function Attrs: optsize returns_twice
+declare i32 @setjmp(i32*) local_unnamed_addr #1
+@tmp_jmpb = global [37 x i32] zeroinitializer, align 16
+
+define void @function1() {
+; CHECK-LABEL: @function1(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[B:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[C:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    store i32 2, i32* [[A]], align 4
+; CHECK-NEXT:    store i32 3, i32* [[B]], align 4
+; CHECK-NEXT:    store i32 4, i32* [[C]], align 4
+; CHECK-NEXT:    [[CALL:%.*]] = call i32 @setjmp(i32* getelementptr inbounds ([37 x i32], [37 x i32]* @tmp_jmpb, i64 0, i64 0))
+; CHECK-NEXT:    call void @outlined_ir_func_0(i32* [[A]], i32* [[B]], i32* [[C]])
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = alloca i32, align 4
+  %b = alloca i32, align 4
+  %c = alloca i32, align 4
+  store i32 2, i32* %a, align 4
+  store i32 3, i32* %b, align 4
+  store i32 4, i32* %c, align 4
+  %call = call i32 @setjmp(i32* getelementptr inbounds ([37 x i32], [37 x i32]* @tmp_jmpb, i64 0, i64 0))
+  %al = load i32, i32* %a
+  %bl = load i32, i32* %b
+  %cl = load i32, i32* %c
+  ret void
+}
+
+define void @function2() {
+; CHECK-LABEL: @function2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[B:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[C:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    store i32 2, i32* [[A]], align 4
+; CHECK-NEXT:    store i32 3, i32* [[B]], align 4
+; CHECK-NEXT:    store i32 4, i32* [[C]], align 4
+; CHECK-NEXT:    [[CALL:%.*]] = call i32 @setjmp(i32* getelementptr inbounds ([37 x i32], [37 x i32]* @tmp_jmpb, i64 0, i64 0))
+; CHECK-NEXT:    call void @outlined_ir_func_0(i32* [[A]], i32* [[B]], i32* [[C]])
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = alloca i32, align 4
+  %b = alloca i32, align 4
+  %c = alloca i32, align 4
+  store i32 2, i32* %a, align 4
+  store i32 3, i32* %b, align 4
+  store i32 4, i32* %c, align 4
+  %call = call i32 @setjmp(i32* getelementptr inbounds ([37 x i32], [37 x i32]* @tmp_jmpb, i64 0, i64 0))
+  %al = load i32, i32* %a
+  %bl = load i32, i32* %b
+  %cl = load i32, i32* %c
+  ret void
+}
+
+attributes #1 = { optsize returns_twice }