[llvm] [MergeFunctions] Fix thunks for non-instruction debug info (PR #82080)

[Shoaib Meenai via llvm-commits]

https://github.com/smeenai created https://github.com/llvm/llvm-project/pull/82080

When MergeFunctions creates new thunk functions, it needs to copy over
the debug info format kind from the original function, otherwise we'll
mix debug info formats and run into assertions. This was exposed by a
downstream change that runs MergeFunctions before inlining, which caused
assertions when inlining attempted to inline thunks created by merging,
and the added test covers both scenarios where merging creates thunks.


>From 498ccd009cc5e8dbe8cb1f24ec5e0ca13f74e355 Mon Sep 17 00:00:00 2001
From: Shoaib Meenai <smeenai at fb.com>
Date: Fri, 16 Feb 2024 16:35:10 -0800
Subject: [PATCH] [MergeFunctions] Fix thunks for non-instruction debug info

When MergeFunctions creates new thunk functions, it needs to copy over
the debug info format kind from the original function, otherwise we'll
mix debug info formats and run into assertions. This was exposed by a
downstream change that runs MergeFunctions before inlining, which caused
assertions when inlining attempted to inline thunks created by merging,
and the added test covers both scenarios where merging creates thunks.
---
 llvm/lib/Transforms/IPO/MergeFunctions.cpp    |  2 +
 .../MergeFunc/debuginfo-iterators.ll          | 54 +++++++++++++++++++
 2 files changed, 56 insertions(+)
 create mode 100644 llvm/test/Transforms/MergeFunc/debuginfo-iterators.ll

diff --git a/llvm/lib/Transforms/IPO/MergeFunctions.cpp b/llvm/lib/Transforms/IPO/MergeFunctions.cpp
index 4af9b2206745e2..a704694198b898 100644
--- a/llvm/lib/Transforms/IPO/MergeFunctions.cpp
+++ b/llvm/lib/Transforms/IPO/MergeFunctions.cpp
@@ -747,6 +747,7 @@ void MergeFunctions::writeThunk(Function *F, Function *G) {
     NewG = Function::Create(G->getFunctionType(), G->getLinkage(),
                             G->getAddressSpace(), "", G->getParent());
     NewG->setComdat(G->getComdat());
+    NewG->IsNewDbgInfoFormat = G->IsNewDbgInfoFormat;
     BB = BasicBlock::Create(F->getContext(), "", NewG);
   }
 
@@ -874,6 +875,7 @@ void MergeFunctions::mergeTwoFunctions(Function *F, Function *G) {
                                       F->getAddressSpace(), "", F->getParent());
     NewF->copyAttributesFrom(F);
     NewF->takeName(F);
+    NewF->IsNewDbgInfoFormat = F->IsNewDbgInfoFormat;
     // Ensure CFI type metadata is propagated to the new function.
     copyMetadataIfPresent(F, NewF, "type");
     copyMetadataIfPresent(F, NewF, "kcfi_type");
diff --git a/llvm/test/Transforms/MergeFunc/debuginfo-iterators.ll b/llvm/test/Transforms/MergeFunc/debuginfo-iterators.ll
new file mode 100644
index 00000000000000..7f259159ce9d33
--- /dev/null
+++ b/llvm/test/Transforms/MergeFunc/debuginfo-iterators.ll
@@ -0,0 +1,54 @@
+;; Ensure that the MergeFunctions pass creates thunks with the appropriate debug
+;; info format set (which would otherwise assert when inlining those thunks).
+; RUN: opt -S -passes=mergefunc,inline --try-experimental-debuginfo-iterators < %s | FileCheck %s
+
+declare void @f1()
+declare void @f2()
+
+define void @f3() {
+  call void @f1()
+  call void @f2()
+  ret void
+}
+
+;; MergeFunctions will replace f4 with a thunk that calls f3. Inlining will
+;; inline f3 into that thunk, which would assert if the thunk had the incorrect
+;; debug info format.
+define void @f4() {
+  call void @f1()
+  call void @f2()
+  ret void
+}
+
+; CHECK-LABEL: define void @f4() {
+; CHECK-NEXT:    call void @f1()
+; CHECK-NEXT:    call void @f2()
+; CHECK-NEXT:    ret void
+; CHECK-NEXT: }
+
+;; Both of these are interposable, so MergeFunctions will create a common thunk
+;; that both will call. Inlining will inline that thunk back, which would assert
+;; if the thunk had the incorrect debug info format.
+define weak void @f5() {
+  call void @f2()
+  call void @f1()
+  ret void
+}
+
+define weak void @f6() {
+  call void @f2()
+  call void @f1()
+  ret void
+}
+
+; CHECK-LABEL: define weak void @f6() {
+; CHECK-NEXT:    call void @f2()
+; CHECK-NEXT:    call void @f1()
+; CHECK-NEXT:    ret void
+; CHECK-NEXT:  }
+
+; CHECK-LABEL: define weak void @f5() {
+; CHECK-NEXT:    call void @f2()
+; CHECK-NEXT:    call void @f1()
+; CHECK-NEXT:    ret void
+; CHECK-NEXT:  }