[llvm] CloneModule: Map global initializers after mapping the function (PR #134082)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/134082

Global initializers may contain references to blockaddresses, which
won't have their VMap entries until after the function body is processed.
Fixes emitting broken initializers in llvm-reduce when functions using
blockaddress are deleted.

>From 28a896dfb3536d4d44063273314f2d612ff9bac4 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 2 Apr 2025 20:09:47 +0700
Subject: [PATCH] CloneModule: Map global initializers after mapping the
 function

Global initializers may contain references to blockaddresses, which
won't have their VMap entries until after the function body is processed.
Fixes emitting broken initializers in llvm-reduce when functions using
blockaddress are deleted.
---
 llvm/lib/Transforms/Utils/CloneModule.cpp     | 52 +++++++++----------
 ...e-functions-blockaddress-wrong-function.ll |  8 +--
 2 files changed, 28 insertions(+), 32 deletions(-)

diff --git a/llvm/lib/Transforms/Utils/CloneModule.cpp b/llvm/lib/Transforms/Utils/CloneModule.cpp
index cabc2ab7933a4..88e2bfe45d2cb 100644
--- a/llvm/lib/Transforms/Utils/CloneModule.cpp
+++ b/llvm/lib/Transforms/Utils/CloneModule.cpp
@@ -124,32 +124,6 @@ std::unique_ptr<Module> llvm::CloneModule(
     VMap[&I] = GI;
   }
 
-  // Now that all of the things that global variable initializer can refer to
-  // have been created, loop through and copy the global variable referrers
-  // over...  We also set the attributes on the global now.
-  //
-  for (const GlobalVariable &G : M.globals()) {
-    GlobalVariable *GV = cast<GlobalVariable>(VMap[&G]);
-
-    SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
-    G.getAllMetadata(MDs);
-    for (auto MD : MDs)
-      GV->addMetadata(MD.first, *MapMetadata(MD.second, VMap));
-
-    if (G.isDeclaration())
-      continue;
-
-    if (!ShouldCloneDefinition(&G)) {
-      // Skip after setting the correct linkage for an external reference.
-      GV->setLinkage(GlobalValue::ExternalLinkage);
-      continue;
-    }
-    if (G.hasInitializer())
-      GV->setInitializer(MapValue(G.getInitializer(), VMap));
-
-    copyComdat(GV, &G);
-  }
-
   // Similarly, copy over function bodies now...
   //
   for (const Function &I : M) {
@@ -212,6 +186,32 @@ std::unique_ptr<Module> llvm::CloneModule(
       NewNMD->addOperand(MapMetadata(N, VMap));
   }
 
+  // Now that all of the things that global variable initializer can refer to
+  // have been created, loop through and copy the global variable referrers
+  // over...  We also set the attributes on the global now.
+  //
+  for (const GlobalVariable &G : M.globals()) {
+    GlobalVariable *GV = cast<GlobalVariable>(VMap[&G]);
+
+    SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
+    G.getAllMetadata(MDs);
+    for (auto MD : MDs)
+      GV->addMetadata(MD.first, *MapMetadata(MD.second, VMap));
+
+    if (G.isDeclaration())
+      continue;
+
+    if (!ShouldCloneDefinition(&G)) {
+      // Skip after setting the correct linkage for an external reference.
+      GV->setLinkage(GlobalValue::ExternalLinkage);
+      continue;
+    }
+    if (G.hasInitializer())
+      GV->setInitializer(MapValue(G.getInitializer(), VMap));
+
+    copyComdat(GV, &G);
+  }
+
   return New;
 }
 
diff --git a/llvm/test/tools/llvm-reduce/reduce-functions-blockaddress-wrong-function.ll b/llvm/test/tools/llvm-reduce/reduce-functions-blockaddress-wrong-function.ll
index f296553759f6b..a757cac0d2bbe 100644
--- a/llvm/test/tools/llvm-reduce/reduce-functions-blockaddress-wrong-function.ll
+++ b/llvm/test/tools/llvm-reduce/reduce-functions-blockaddress-wrong-function.ll
@@ -1,14 +1,10 @@
 ; RUN: llvm-reduce --abort-on-invalid-reduction --delta-passes=functions --test FileCheck --test-arg --check-prefixes=INTERESTING --test-arg %s --test-arg --input-file %s -o %t
 ; RUN: FileCheck --check-prefixes=RESULT --input-file=%t %s
 
-; FIXME: This testcase exhibits nonsensical behavior. The first
-; function has blockaddress references. When the second function is
-; deleted, it causes the blockreferences from the first to be replaced
-; with inttoptr.
-
 ; INTERESTING: @blockaddr.table.other
 
-; RESULT: @blockaddr.table.other = private unnamed_addr constant [2 x ptr] [ptr inttoptr (i32 1 to ptr), ptr inttoptr (i32 1 to ptr)]
+; RESULT: @blockaddr.table.other = private unnamed_addr constant [2 x ptr] [ptr blockaddress(@bar, %L1), ptr blockaddress(@bar, %L2)]
+
 
 @blockaddr.table.other = private unnamed_addr constant [2 x ptr] [ptr blockaddress(@bar, %L1), ptr blockaddress(@bar, %L2)]