[DebugInfo][GVNSink] Merging debug locations of sinked instructions

[Apochens]

Fix #85069 .

[llvmbot]

@llvm/pr-subscribers-debuginfo

@llvm/pr-subscribers-llvm-transforms

Author: Shan Huang (Apochens)

Changes

Fix #85069 .

---

Full diff: https://github.com/llvm/llvm-project/pull/92859.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Scalar/GVNSink.cpp (+3-1)
• (modified) llvm/test/Transforms/GVNSink/sink-ignore-dbg-intrinsics.ll (+5-2)

diff --git a/llvm/lib/Transforms/Scalar/GVNSink.cpp b/llvm/lib/Transforms/Scalar/GVNSink.cpp
index b0f716cb17598..3dfa2dd9df27f 100644
--- a/llvm/lib/Transforms/Scalar/GVNSink.cpp
+++ b/llvm/lib/Transforms/Scalar/GVNSink.cpp
@@ -921,8 +921,10 @@ void GVNSink::sinkLastInstruction(ArrayRef<BasicBlock *> Blocks,
     }
 
   for (auto *I : Insts)
-    if (I != I0)
+    if (I != I0) {
       I->replaceAllUsesWith(I0);
+      I0->applyMergedLocation(I0->getDebugLoc(), I->getDebugLoc());
+    }
   foldPointlessPHINodes(BBEnd);
 
   // Finally nuke all instructions apart from the common instruction.
diff --git a/llvm/test/Transforms/GVNSink/sink-ignore-dbg-intrinsics.ll b/llvm/test/Transforms/GVNSink/sink-ignore-dbg-intrinsics.ll
index 8aafd689e781d..e54a17a155b8c 100644
--- a/llvm/test/Transforms/GVNSink/sink-ignore-dbg-intrinsics.ll
+++ b/llvm/test/Transforms/GVNSink/sink-ignore-dbg-intrinsics.ll
@@ -1,6 +1,8 @@
 ; RUN: opt < %s -passes=gvn-sink -S | FileCheck %s
 
 ; Test that GVNSink correctly performs the sink optimization in the presence of debug information
+; Test that GVNSink correctly merges the debug locations of sinked instruction, eg, propagating
+; the merged debug location of `%add` and `%add1` to the sinked add instruction.
 
 ; Function Attrs: noinline nounwind uwtable
 define dso_local i32 @fun(i32 noundef %a, i32 noundef %b) #0 !dbg !10 {
@@ -8,8 +10,9 @@ define dso_local i32 @fun(i32 noundef %a, i32 noundef %b) #0 !dbg !10 {
 ; CHECK-SAME:    i32 noundef [[A:%.*]], i32 noundef [[B:%.*]])
 ; CHECK:       if.end:
 ; CHECK:         [[B_SINK:%.*]] = phi i32 [ [[B]], %if.else ], [ [[A]], %if.then ]
-; CHECK:         [[ADD1:%.*]] = add nsw i32 [[B_SINK]], 1
-; CHECK:         [[XOR2:%.*]] = xor i32 [[ADD1]], 1
+; CHECK:         [[ADD1:%.*]] = add nsw i32 [[B_SINK]], 1, !dbg [[DBG:![0-9]+]]
+; CHECK:         [[XOR2:%.*]] = xor i32 [[ADD1]], 1, !dbg [[DBG:![0-9]+]]
+; CHECK:       [[DBG]] = !DILocation(line: 0,
 ;
 entry:
   tail call void @llvm.dbg.value(metadata i32 %a, metadata !15, metadata !DIExpression()), !dbg !16

[Apochens]

@hiraditya Hi, please review this! (I don't have the permission to assign reviewers)

[hiraditya]

LGTM, it'll be nice to have two more tests

1. sinking three way (test18 in llvm/test/Transforms/GVNSink/sink-common-code.ll
2. sinking to a newly created post dominator (@test_pr30244 in llvm/test/Transforms/GVNSink/sink-common-code.ll)

[hiraditya]

LGTM, please give it a couple of days in case other reviewers have comments.