Don't insert stackrestore on deoptimizing returns

They're not necessary (since the stack pointer is trivially restored on
return), and the way LLVM inserts the stackrestore calls breaks the
IR (we get a stackrestore between the deoptimize call and the return).

llvm-svn: 265101

2 files changed, 20 insertions, 2 deletions

diff --git a/llvm/lib/Transforms/Utils/InlineFunction.cpp b/llvm/lib/Transforms/Utils/InlineFunction.cpp
index 42179851718..646afd8cd61 100644
--- a/llvm/lib/Transforms/Utils/InlineFunction.cpp
+++ b/llvm/lib/Transforms/Utils/InlineFunction.cpp
@@ -1731,10 +1731,12 @@ bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI,
     // Insert a call to llvm.stackrestore before any return instructions in the
     // inlined function.
     for (ReturnInst *RI : Returns) {
-      // Don't insert llvm.stackrestore calls between a musttail call and a
-      // return.  The return will restore the stack pointer.
+      // Don't insert llvm.stackrestore calls between a musttail or deoptimize
+      // call and a return.  The return will restore the stack pointer.
       if (InlinedMustTailCalls && RI->getParent()->getTerminatingMustTailCall())
         continue;
+      if (InlinedDeoptimizeCalls && RI->getParent()->getTerminatingDeoptimizeCall())
+        continue;
       IRBuilder<>(RI).CreateCall(StackRestore, SavedPtr);
     }
   }
diff --git a/llvm/test/Transforms/Inline/deoptimize-intrinsic.ll b/llvm/test/Transforms/Inline/deoptimize-intrinsic.ll
index d9b88985551..800ace49e15 100644
--- a/llvm/test/Transforms/Inline/deoptimize-intrinsic.ll
+++ b/llvm/test/Transforms/Inline/deoptimize-intrinsic.ll
@@ -104,3 +104,19 @@ entry:
   call i8 @callee_with_alloca();
   ret void
 }
+
+define i8 @callee_with_dynamic_alloca(i32 %n) alwaysinline {
+  %p = alloca i8, i32 %n
+  %v = call i8(...) @llvm.experimental.deoptimize.i8(i32 1) [ "deopt"(i8* %p) ]
+  ret i8 %v
+}
+
+define void @caller_with_stacksaverestore(i32 %n) {
+; CHECK-LABEL: void @caller_with_stacksaverestore(
+; CHECK:  call void (...) @llvm.experimental.deoptimize.isVoid(i32 1) [ "deopt"(i8* %p.i) ]
+; CHECK-NEXT:  ret void
+
+  %p = alloca i32, i32 %n
+  call i8 @callee_with_dynamic_alloca(i32 %n)
+  ret void
+}