Skip to content

[DebugInfo][GlobalOpt] Preserve source locs for optimized loads #134828

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 24, 2025

Conversation

SLTozer
Copy link
Contributor

@SLTozer SLTozer commented Apr 8, 2025

Some optimizations in globalopt simplify uses of a global value to uses of a generated global bool value; in some cases where this happens, the newly-generated instructions would not have the original source location(s) of the instructions they replaced propagated to them; this patch properly preserves those source locations.

Found using #107279.

Some optimizations in globalopt simplify uses of a global value to uses
of a generated global bool value; in some cases where this happens, the
newly-generated instructions would not have the original source
location(s) of the instructions they replaced propagated to them; this
patch properly preserves those source locations.
@llvmbot
Copy link
Member

llvmbot commented Apr 8, 2025

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-debuginfo

Author: Stephen Tozer (SLTozer)

Changes

Some optimizations in globalopt simplify uses of a global value to uses of a generated global bool value; in some cases where this happens, the newly-generated instructions would not have the original source location(s) of the instructions they replaced propagated to them; this patch properly preserves those source locations.

Found using #107279.


Full diff: https://github.com/llvm/llvm-project/pull/134828.diff

2 Files Affected:

  • (modified) llvm/lib/Transforms/IPO/GlobalOpt.cpp (+7)
  • (added) llvm/test/Transforms/GlobalOpt/X86/preserve-load-of-bool-dbgloc.ll (+94)
diff --git a/llvm/lib/Transforms/IPO/GlobalOpt.cpp b/llvm/lib/Transforms/IPO/GlobalOpt.cpp
index 83cc1e5f04f3d..397435a4271a6 100644
--- a/llvm/lib/Transforms/IPO/GlobalOpt.cpp
+++ b/llvm/lib/Transforms/IPO/GlobalOpt.cpp
@@ -988,6 +988,11 @@ OptimizeGlobalAddressOfAllocation(GlobalVariable *GV, CallInst *CI,
                                InitBool->getName() + ".val", false, Align(1),
                                LI->getOrdering(), LI->getSyncScopeID(),
                                LI->getIterator());
+      // FIXME: Should we use the DebugLoc of the load used by the predicate, or
+      // the predicate? The load seems most appropriate, but there's an argument
+      // that the new load does not represent the old load, but is simply a
+      // component of recomputing the predicate.
+      cast<LoadInst>(LV)->setDebugLoc(LI->getDebugLoc());
       InitBoolUsed = true;
       switch (ICI->getPredicate()) {
       default: llvm_unreachable("Unknown ICmp Predicate!");
@@ -1000,6 +1005,7 @@ OptimizeGlobalAddressOfAllocation(GlobalVariable *GV, CallInst *CI,
       case ICmpInst::ICMP_ULE:
       case ICmpInst::ICMP_EQ:
         LV = BinaryOperator::CreateNot(LV, "notinit", ICI->getIterator());
+        cast<BinaryOperator>(LV)->setDebugLoc(ICI->getDebugLoc());
         break;
       case ICmpInst::ICMP_NE:
       case ICmpInst::ICMP_UGT:
@@ -1276,6 +1282,7 @@ static bool TryToShrinkGlobalToBoolean(GlobalVariable *GV, Constant *OtherVal) {
               new LoadInst(NewGV->getValueType(), NewGV, LI->getName() + ".b",
                            false, Align(1), LI->getOrdering(),
                            LI->getSyncScopeID(), LI->getIterator());
+          cast<LoadInst>(StoreVal)->setDebugLoc(LI->getDebugLoc());
         } else {
           assert((isa<CastInst>(StoredVal) || isa<SelectInst>(StoredVal)) &&
                  "This is not a form that we understand!");
diff --git a/llvm/test/Transforms/GlobalOpt/X86/preserve-load-of-bool-dbgloc.ll b/llvm/test/Transforms/GlobalOpt/X86/preserve-load-of-bool-dbgloc.ll
new file mode 100644
index 0000000000000..4a97a67092906
--- /dev/null
+++ b/llvm/test/Transforms/GlobalOpt/X86/preserve-load-of-bool-dbgloc.ll
@@ -0,0 +1,94 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -p=globalopt -S | FileCheck %s
+
+;; Test that when we are able to simplify uses of global variables with loads of
+;; newly generated bool values, we transfer debuglocs over correctly.
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@global = internal global i32 0
+@global.1 = internal unnamed_addr global ptr null, align 8
+
+define void @ham() !dbg !7 {
+; CHECK-LABEL: define void @ham(
+; CHECK-SAME: ) local_unnamed_addr !dbg [[DBG4:![0-9]+]] {
+; CHECK-NEXT:  [[BB:.*:]]
+; CHECK-NEXT:    [[LOAD_B:%.*]] = load i1, ptr @global, align 1, !dbg [[DBG6:![0-9]+]]
+; CHECK-NEXT:    [[LOAD_B1:%.*]] = load i1, ptr @global, align 1, !dbg [[DBG6]]
+; CHECK-NEXT:    [[LOAD:%.*]] = zext i1 [[LOAD_B1]] to i32, !dbg [[DBG6]]
+; CHECK-NEXT:    store i1 [[LOAD_B]], ptr @global, align 1
+; CHECK-NEXT:    ret void
+;
+bb:
+  %load = load i32, ptr @global, align 4, !dbg !4
+  store i32 %load, ptr @global, align 4
+  ret void
+}
+
+define void @hoge() {
+; CHECK-LABEL: define void @hoge() local_unnamed_addr {
+; CHECK-NEXT:  [[BB:.*:]]
+; CHECK-NEXT:    store i1 true, ptr @global, align 1
+; CHECK-NEXT:    ret void
+;
+bb:
+  store i32 1, ptr @global, align 4
+  ret void
+}
+
+define void @bar() {
+; CHECK-LABEL: define void @bar() local_unnamed_addr {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    store i1 true, ptr @global.1.init, align 1
+; CHECK-NEXT:    ret void
+;
+entry:
+  %call = tail call noalias nonnull dereferenceable(48) ptr @_Znwm(i64 48)
+  store ptr %call, ptr @global.1, align 8
+  ret void
+}
+
+define void @pluto() !dbg !10 {
+; CHECK-LABEL: define void @pluto(
+; CHECK-SAME: ) local_unnamed_addr !dbg [[DBG8:![0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[GLOBAL_1_INIT_VAL:%.*]] = load i1, ptr @global.1.init, align 1, !dbg [[DBG9:![0-9]+]]
+; CHECK-NEXT:    [[NOTINIT:%.*]] = xor i1 [[GLOBAL_1_INIT_VAL]], true, !dbg [[DBG10:![0-9]+]]
+; CHECK-NEXT:    unreachable
+;
+entry:
+  %0 = load ptr, ptr @global.1, align 8, !dbg !11
+  %.not = icmp eq ptr %0, null, !dbg !12
+  unreachable
+}
+
+declare ptr @_Znwm(i64)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!3}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C11, file: !1, producer: "clang version 20.0.0git", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2, globals: !2, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "preserve-load-dbgloc.c", directory: "/tmp")
+!2 = !{}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!4 = !DILocation(line: 10, column: 1, scope: !5)
+!5 = distinct !DILexicalBlock(scope: !7, file: !6, line: 1524, column: 3)
+!6 = !DIFile(filename: "preserve-load-dbgloc.c", directory: "/tmp")
+!7 = distinct !DISubprogram(name: "ham", scope: !6, file: !6, line: 10, type: !8, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+!8 = distinct !DISubroutineType(types: !2)
+!10 = distinct !DISubprogram(name: "pluto", scope: !6, file: !6, line: 20, type: !8, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+!11 = !DILocation(line: 20, column: 2, scope: !10)
+!12 = !DILocation(line: 21, column: 3, scope: !10)
+;.
+; CHECK: [[META0:![0-9]+]] = distinct !DICompileUnit(language: DW_LANG_C11, file: [[META1:![0-9]+]], producer: "{{.*}}clang version {{.*}}", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: [[META2:![0-9]+]], retainedTypes: [[META2]], globals: [[META2]], splitDebugInlining: false, nameTableKind: None)
+; CHECK: [[META1]] = !DIFile(filename: "preserve-load-dbgloc.c", directory: {{.*}})
+; CHECK: [[META2]] = !{}
+; CHECK: [[DBG4]] = distinct !DISubprogram(name: "ham", scope: [[META1]], file: [[META1]], line: 10, type: [[META5:![0-9]+]], flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: [[META0]], retainedNodes: [[META2]])
+; CHECK: [[META5]] = distinct !DISubroutineType(types: [[META2]])
+; CHECK: [[DBG6]] = !DILocation(line: 10, column: 1, scope: [[META7:![0-9]+]])
+; CHECK: [[META7]] = distinct !DILexicalBlock(scope: [[DBG4]], file: [[META1]], line: 1524, column: 3)
+; CHECK: [[DBG8]] = distinct !DISubprogram(name: "pluto", scope: [[META1]], file: [[META1]], line: 20, type: [[META5]], flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: [[META0]], retainedNodes: [[META2]])
+; CHECK: [[DBG9]] = !DILocation(line: 20, column: 2, scope: [[DBG8]])
+; CHECK: [[DBG10]] = !DILocation(line: 21, column: 3, scope: [[DBG8]])
+;.

@@ -988,6 +988,11 @@ OptimizeGlobalAddressOfAllocation(GlobalVariable *GV, CallInst *CI,
InitBool->getName() + ".val", false, Align(1),
LI->getOrdering(), LI->getSyncScopeID(),
LI->getIterator());
// FIXME: Should we use the DebugLoc of the load used by the predicate, or
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see new StoreInst above which looks like it does similar straightforward replacement, should we carry over the sloc there?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It certainly looks like it - I'll find a test case and add it (and any other nearby cases if I spot any).

Copy link
Contributor

@OCHyams OCHyams left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM with inline question

ret void
}

define void @hoge() {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What's this function testing?

Copy link
Contributor Author

@SLTozer SLTozer Apr 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This function is necessary for GlobalOpt to make the decisions that trigger the bug.

Copy link
Collaborator

@rnk rnk left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks!

@SLTozer SLTozer merged commit 224cd50 into llvm:main Apr 24, 2025
11 checks passed
IanWood1 pushed a commit to IanWood1/llvm-project that referenced this pull request May 6, 2025
…#134828)

Some optimizations in globalopt simplify uses of a global value to uses
of a generated global bool value; in some cases where this happens, the
newly-generated instructions would not have the original source
location(s) of the instructions they replaced propagated to them; this
patch properly preserves those source locations.

Found using llvm#107279.
IanWood1 pushed a commit to IanWood1/llvm-project that referenced this pull request May 6, 2025
…#134828)

Some optimizations in globalopt simplify uses of a global value to uses
of a generated global bool value; in some cases where this happens, the
newly-generated instructions would not have the original source
location(s) of the instructions they replaced propagated to them; this
patch properly preserves those source locations.

Found using llvm#107279.
IanWood1 pushed a commit to IanWood1/llvm-project that referenced this pull request May 6, 2025
…#134828)

Some optimizations in globalopt simplify uses of a global value to uses
of a generated global bool value; in some cases where this happens, the
newly-generated instructions would not have the original source
location(s) of the instructions they replaced propagated to them; this
patch properly preserves those source locations.

Found using llvm#107279.
Ankur-0429 pushed a commit to Ankur-0429/llvm-project that referenced this pull request May 9, 2025
…#134828)

Some optimizations in globalopt simplify uses of a global value to uses
of a generated global bool value; in some cases where this happens, the
newly-generated instructions would not have the original source
location(s) of the instructions they replaced propagated to them; this
patch properly preserves those source locations.

Found using llvm#107279.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants