[KeyInstr] Inline atom info

[OCHyams]

Source atom groups are identified by an atom group number and inlined-at pair,
so we simply can copy the atom numbers into the caller when inlining.

[OCHyams]

• [NFC][KeyInstr][Clang] Add some additional tests #134654
• [KeyInstr][Clang] Coerced store atoms #134653
• [KeyInstr][Clang] Ret atom #134652
• [KeyIntsr][Clang] Builtins alloca auto-init atom #134651
• [KeyInstr][Clang] Assign matrix element atom #134650
• [KeyInstr][Clang] Assign vector element atom #134649
• [KeyInstr][Clang] Bitfield atom #134648
• [KeyInstr][Clang] For range stmt atoms #134647
• [KeyInstr][Clang] For stmt atom #134646
• [KeyInstr][Clang] While stmt atom #134645
• [KeyInstr][Clang] Do stmt atom #134644
• [KeyInstr][Clang] Switch stmt atom #134643
• [KeyInstr][Clang] If stmt atom #134642
• [KeyInstr][Clang] Catch variable init atom #134641
• [KeyInstr][Clang] Member initalization atom #134640
• [KeyInstr][Clang] Aggregate init + copy #134639
• [KeyInstr] Complex assignment atoms #134638
• [KeyInstr][Clang] Assignment atom group #134637
• [KeyInstr][Clang] Static variable init atom #134636
• [KeyInstr][Clang] Agg init atom #134635
• [Clang][NFC] Move some static functions into CodeGenFunction #134634
• [KeyInstr][Clang] Scalar init atom #134633
• [KeyInstr][Clang] Add ApplyAtomGroup #134632
• [KeyInstr][Clang] Add Clang option -g[no-]key-instructions #134627
• [KeyInstr][DwarfDebug] Add is_stmt emission support #133495
• [KeyInstr] Add MIR parser support #133494
• [KeyInstr] Remap cloned PHIs in BreakCriticalEdges #133493
• Propagate DebugLocs on phis in BreakCriticalEdges #133492
• [KeyInstr][LoopUnswitch] Remap cloned instructions' atoms #133491
• [KeyInstr][LoopRotate] Remap atoms of duplicated instructions #133490
• [KeyInstr][LoopUnroll] Remap atoms while unrolling #133489
• [KeyInstr][JumpThreading] Remap atoms duping bb with cond br on phi into pred #133488
• [KeyInstr][JumpThreading] Remap atoms after threading #133487
• [KeyInstr][JumpThreading] Remap atoms in blocks duplicated for threading #133486
• [KeyInstr][Inline] Don't propagate atoms to inlined nodebug instructions #133485
• [KeyInstr][SimplifyCFG] Remap atoms after duplication for threading #133484
• [KeyInstr][debugify] Add --debugify-atoms to add key instructions metadata #133483
• [KeyInstr][SimplifyCFG] Remap atoms when folding br to common succ into pred #133482
• [KeyInstr] Inline atom info #133481 👈 (View in Graphite)
• [KeyInstr] Merge atoms in DILocation::getMergedLocation #133480
• [NFC][KeyInstr] Add Atom Group (re)mapping #133479
• [KeyInstr] Add Atom Group waterline to LLVMContext #133478
• [KeyInstr] Add fields to DILocation behind compile time flag #133477
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-llvm-ir

@llvm/pr-subscribers-debuginfo

Author: Orlando Cazalet-Hyams (OCHyams)

Changes

Source atom groups are identified by an atom group number and inlined-at pair,
so we simply can copy the atom numbers into the caller when inlining.

---

Full diff: https://github.com/llvm/llvm-project/pull/133481.diff

3 Files Affected:

• (modified) llvm/lib/IR/DebugLoc.cpp (+1)
• (modified) llvm/lib/Transforms/Utils/InlineFunction.cpp (+2-1)
• (added) llvm/test/DebugInfo/KeyInstructions/Generic/inline.ll (+53)

diff --git a/llvm/lib/IR/DebugLoc.cpp b/llvm/lib/IR/DebugLoc.cpp
index bdea52180f74a..61f5bd8e9b0a2 100644
--- a/llvm/lib/IR/DebugLoc.cpp
+++ b/llvm/lib/IR/DebugLoc.cpp
@@ -129,6 +129,7 @@ DebugLoc DebugLoc::appendInlinedAt(const DebugLoc &DL, DILocation *InlinedAt,
   // Starting from the top, rebuild the nodes to point to the new inlined-at
   // location (then rebuilding the rest of the chain behind it) and update the
   // map of already-constructed inlined-at nodes.
+  // Key Instructions: InlinedAt fields don't need atom info.
   for (const DILocation *MD : reverse(InlinedAtLocations))
     Cache[MD] = Last = DILocation::getDistinct(
         Ctx, MD->getLine(), MD->getColumn(), MD->getScope(), Last);
diff --git a/llvm/lib/Transforms/Utils/InlineFunction.cpp b/llvm/lib/Transforms/Utils/InlineFunction.cpp
index 1404867fda6bc..9202451f5726d 100644
--- a/llvm/lib/Transforms/Utils/InlineFunction.cpp
+++ b/llvm/lib/Transforms/Utils/InlineFunction.cpp
@@ -1813,7 +1813,8 @@ static DebugLoc inlineDebugLoc(DebugLoc OrigDL, DILocation *InlinedAt,
                                DenseMap<const MDNode *, MDNode *> &IANodes) {
   auto IA = DebugLoc::appendInlinedAt(OrigDL, InlinedAt, Ctx, IANodes);
   return DILocation::get(Ctx, OrigDL.getLine(), OrigDL.getCol(),
-                         OrigDL.getScope(), IA);
+                         OrigDL.getScope(), IA, OrigDL.isImplicitCode(),
+                         OrigDL->getAtomGroup(), OrigDL->getAtomRank());
 }
 
 /// Update inlined instructions' line numbers to
diff --git a/llvm/test/DebugInfo/KeyInstructions/Generic/inline.ll b/llvm/test/DebugInfo/KeyInstructions/Generic/inline.ll
new file mode 100644
index 0000000000000..c2663ee51a77d
--- /dev/null
+++ b/llvm/test/DebugInfo/KeyInstructions/Generic/inline.ll
@@ -0,0 +1,53 @@
+; RUN: opt %s -passes=inline -S -o - | FileCheck %s
+
+;; Inline `f` into `g`. The inlined assignment store and add should retain
+;; their atom info.
+
+; CHECK: _Z1gi
+; CHECK-NOT: _Z1fi
+; CHECK: %add.i = add nsw i32 %mul.i, 1, !dbg [[G1R2:!.*]]
+; CHECK-NEXT: store i32 %add.i, ptr %x.i, align 4, !dbg [[G1R1:!.*]]
+
+; CHECK: [[G1R2]] = !DILocation({{.*}}, atomGroup: 1, atomRank: 2)
+; CHECK: [[G1R1]] = !DILocation({{.*}}, atomGroup: 1, atomRank: 1)
+
+define hidden void @_Z1fi(i32 noundef %a) !dbg !11 {
+entry:
+  %a.addr = alloca i32, align 4
+  %x = alloca i32, align 4
+  store i32 %a, ptr %a.addr, align 4
+  %0 = load i32, ptr %a.addr, align 4, !dbg !18
+  %mul = mul nsw i32 %0, 2, !dbg !18
+  %add = add nsw i32 %mul, 1, !dbg !19
+  store i32 %add, ptr %x, align 4, !dbg !20
+  ret void, !dbg !22
+}
+
+define hidden void @_Z1gi(i32 noundef %b) !dbg !23 {
+entry:
+  %b.addr = alloca i32, align 4
+  store i32 %b, ptr %b.addr, align 4
+  %0 = load i32, ptr %b.addr, align 4, !dbg !24
+  call void @_Z1fi(i32 noundef %0), !dbg !24
+  ret void, !dbg !25
+}
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3}
+!llvm.ident = !{!10}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_17, file: !1, producer: "clang version 19.0.0", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "test.cpp", directory: "/")
+!2 = !{i32 7, !"Dwarf Version", i32 5}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!10 = !{!"clang version 19.0.0"}
+!11 = distinct !DISubprogram(name: "f", scope: !1, file: !1, line: 1, type: !12, scopeLine: 1, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!12 = !DISubroutineType(types: !13)
+!13 = !{}
+!18 = !DILocation(line: 2, scope: !11)
+!19 = !DILocation(line: 2, scope: !11, atomGroup: 1, atomRank: 2)
+!20 = !DILocation(line: 2, scope: !11, atomGroup: 1, atomRank: 1)
+!22 = !DILocation(line: 3, scope: !11, atomGroup: 2, atomRank: 1)
+!23 = distinct !DISubprogram(name: "g", scope: !1, file: !1, line: 4, type: !12, scopeLine: 4, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!24 = !DILocation(line: 5, scope: !23)
+!25 = !DILocation(line: 6, scope: !23, atomGroup: 1, atomRank: 1)

[SLTozer]

LGTM, besides a couple inline comments.

[SLTozer]

Similar to my comment on a different review, propagating OrigDL's IsImplicitCode field is a change in behaviour.

[OCHyams]

(note to others: this was discussed offline as fine to ignore)

[SLTozer]

Nit, could remove DILocations from the instructions that aren't relevant to the test.

[jmorse]

LGTM