[DebugInfo] Stop undef fragments from closing non-overlapping fragments

[dstenb]

Summary

When DwarfDebug::buildLocationList() encountered an undef debug value,
it would truncate all open values, regardless if they were overlapping or
not. This patch fixes so that it only does that for overlapping fragments.

This change unearthed a bug that I had introduced in D57511,
which I have fixed in this patch. The code in DebugHandlerBase that
changes labels for parameter debug values could break DwarfDebug's
assumption that the labels for the entries in the debug value history
are monotonically increasing. Before this patch, that bug could result
in location list entries whose ending address was lower than the
beginning address, and with the changes for undef debug values that this
patch introduces it could trigger an assertion, due to attempting to
emit location list entries with empty ranges. A reproducer for the bug
is added in param-reg-const-mix.mir.

[jmorse]

LGTM with a couple of questions, will leave it for further comments.

test/DebugInfo/X86/undef-fragment.ll

28–32

Is it OK to ditch the absolute addresses here? (Assuming that this isn't testing the address monoticity). I don't have any strong feeling, just the impression that this might be brittle to future codegen changes. (No big deal).

52–54

Can we ditch the attributes?

[dstenb]

Update test cases after review comments.

[dstenb]

test/DebugInfo/X86/undef-fragment.ll

28–32

Good idea! I think it is okay to test the monoticity in the other attached test case. I generalized the address checks so that they just verify that the entries are placed directly after each other.

52–54

It should be okay to remove them I think. I kept dbg.value's attributes as I was not sure if it is customary to remove attributes from intrinsics.

[aprantl]

That looks reasonable, thanks!

lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1178

This could also become MachineInstr::isUndefDbgValue().

[dstenb]

Address comments (add MachineInstr:isUndefDebugValue query function)

[dstenb]

Add a comment to the query function (undef debug values are mentioned in SourceLevelDebugging.rst, but I did not find any central explanations of such debug values in the code base).

I'm planning on landing this shortly, as I assume that the changes after the LGTM are small and non-controversial.

lib/CodeGen/AsmPrinter/DebugHandlerBase.cpp

if (DIVar->isParameter() &&
           const DIExpression *Fragment = I->getInstr()->getDebugExpression();
           if (std::any_of(Entries.begin(), I,
                           [&](DbgValueHistoryMap::Entry Pred) {
                             return Pred.isDbgValue() &&
                                    Fragment->fragmentsOverlap(
                                        Pred.getInstr()->getDebugExpression());
                           }))
             break;
-          if (!IsDescribedByReg(I->getInstr()))
-            LabelsBeforeInsn[I->getInstr()] = Asm->getFunctionBegin();
+          // The code that generates location lists for DWARF assumes that the
+          // entries' start labels are monotonically increasing, and since we
+          // don't change the label for fragments that are described by
+          // registers, we must bail out when encountering such a fragment.
+          if (IsDescribedByReg(I->getInstr()))
+            break;
         }
       }
     }
 
     for (const auto &Entry : Entries) {
       if (Entry.isDbgValue())
         requestLabelBeforeInsn(Entry.getInstr());
       else

lib/CodeGen/AsmPrinter/DwarfDebug.cpp

void DwarfDebug::buildLocationList(SmallVectorImpl<DebugLocEntry> &DebugLoc,
                                    const DbgValueHistoryMap::Entries &Entries) {
   using OpenRange =
       std::pair<DbgValueHistoryMap::EntryIndex, DebugLocEntry::Value>;
   SmallVector<OpenRange, 4> OpenRanges;
 
   for (auto EB = Entries.begin(), EI = EB, EE = Entries.end(); EI != EE; ++EI) {
     const MachineInstr *Instr = EI->getInstr();
 
-    if (EI->isDbgValue()) {
-      // Check if a variable is inaccessible in this range.
-      // TODO: This should only truncate open ranges that are overlapping.
-      if (Instr->getNumOperands() > 1 &&
-          Instr->getOperand(0).isReg() && !Instr->getOperand(0).getReg()) {
-        OpenRanges.clear();
-        continue;
-      }
-    }
-
     // Remove all values that are no longer live.
     size_t Index = std::distance(EB, EI);
     auto Last =
         remove_if(OpenRanges, [&](OpenRange &R) { return R.first <= Index; });
     OpenRanges.erase(Last, OpenRanges.end());
 
     // If we are dealing with a clobbering entry, this iteration will result in
     // a location list entry starting after the clobbering instruction.
for (auto EB = Entries.begin(), EI = EB, EE = Entries.end(); EI != EE; ++EI) {
     assert(EndLabel && "Forgot label after instruction ending a range!");
 
     if (EI->isDbgValue())
       LLVM_DEBUG(dbgs() << "DotDebugLoc: " << *Instr << "\n");
 
     // If this history map entry has a debug value, add that to the list of
     // open ranges.
     if (EI->isDbgValue()) {
+      // Do not add undef debug values, as they are redundant information in
+      // the location list entries. An undef debug results in an empty location
+      // description. If there are any non-undef fragments then padding pieces
+      // with empty location descriptions will automatically be inserted, and if
+      // all fragments are undef then the whole location list entry is
+      // redundant.
+      bool IsUndef = Instr->getNumOperands() > 1 &&
+                     Instr->getOperand(0).isReg() &&
+                     !Instr->getOperand(0).getReg();
+
+      if (!IsUndef) {
         auto Value = getDebugLocValue(Instr);
         OpenRanges.emplace_back(EI->getEndIndex(), Value);
       }
+    }
 
     // Location list entries with empty location descriptions are redundant
     // information in DWARF, so do not emit those.
     if (OpenRanges.empty())
       continue;
 
     // Omit entries with empty ranges as they do not have any effect in DWARF.
     if (StartLabel == EndLabel) {

test/DebugInfo/MIR/ARM/param-reg-const-mix.mir

+# RUN: llc -mtriple=armv4t-unknown-unknown -start-after=livedebugvalues -filetype=obj -o - %s | llvm-dwarfdump - | FileCheck %s
+
+# This reproducer is based on the following C code:
+#
+# struct S0 { int f1; int f2; int f3; };
+#
+# int a;
+#
+# void fn1(struct S0 p1) {
+#   a = p1.f1 >= fn2(p1.f2);
+# }
+#
+# and was generated using the following commands:
+# $ clang -O1 -g --target=armv4t -S -emit-llvm
+# $ llc -O1 -stop-after=livedebugvalues
+
+--- |
+  target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+  target triple = "armv4t-unknown-unknown"
+
+  ; Function Attrs: nounwind
+  define arm_aapcscc i32 @fn1([3 x i32] %p1.coerce) !dbg !7 {
+  entry:
+    %p1.coerce.fca.0.extract = extractvalue [3 x i32] %p1.coerce, 0
+    call void @llvm.dbg.value(metadata i32 %p1.coerce.fca.0.extract, metadata !17, metadata !DIExpression(DW_OP_LLVM_fragment, 0, 32)), !dbg !18
+    %p1.coerce.fca.1.extract = extractvalue [3 x i32] %p1.coerce, 1
+    call void @llvm.dbg.value(metadata i32 %p1.coerce.fca.1.extract, metadata !17, metadata !DIExpression(DW_OP_LLVM_fragment, 32, 32)), !dbg !18
+    call void @llvm.dbg.value(metadata i32 undef, metadata !17, metadata !DIExpression(DW_OP_LLVM_fragment, 64, 32)), !dbg !18
+    %call = tail call arm_aapcscc i32 bitcast (i32 (...)* @fn2 to i32 (i32)*)(i32 %p1.coerce.fca.1.extract), !dbg !19
+    %cmp = icmp sge i32 %p1.coerce.fca.0.extract, %call, !dbg !19
+    %conv = zext i1 %cmp to i32, !dbg !19
+    ret i32 %conv, !dbg !19
+  }
+
+  declare arm_aapcscc i32 @fn2(...)
+
+  ; Function Attrs: nounwind readnone speculatable
+  declare void @llvm.dbg.value(metadata, metadata, metadata) #0
+
+  attributes #0 = { nounwind readnone speculatable }
+
+  !llvm.dbg.cu = !{!0}
+  !llvm.module.flags = !{!3, !4, !5}
+  !llvm.ident = !{!6}
+
+  !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 9.0.0", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, nameTableKind: None)
+  !1 = !DIFile(filename: "test.c", directory: "/")
+  !2 = !{}
+  !3 = !{i32 2, !"Dwarf Version", i32 4}
+  !4 = !{i32 2, !"Debug Info Version", i32 3}
+  !5 = !{i32 1, !"min_enum_size", i32 4}
+  !6 = !{!"clang version 9.0.0"}
+  !7 = distinct !DISubprogram(name: "fn1", scope: !1, file: !1, line: 3, type: !8, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !16)
+  !8 = !DISubroutineType(types: !9)
+  !9 = !{!10, !11}
+  !10 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+  !11 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "S0", file: !1, line: 1, size: 96, elements: !12)
+  !12 = !{!13, !14, !15}
+  !13 = !DIDerivedType(tag: DW_TAG_member, name: "f1", scope: !11, file: !1, line: 1, baseType: !10, size: 32)
+  !14 = !DIDerivedType(tag: DW_TAG_member, name: "f2", scope: !11, file: !1, line: 1, baseType: !10, size: 32, offset: 32)
+  !15 = !DIDerivedType(tag: DW_TAG_member, name: "f3", scope: !11, file: !1, line: 1, baseType: !10, size: 32, offset: 64)
+  !16 = !{!17}
+  !17 = !DILocalVariable(name: "p1", arg: 1, scope: !7, file: !1, line: 3, type: !11)
+  !18 = !DILocation(line: 3, scope: !7)
+  !19 = !DILocation(line: 4, scope: !7)
+
+...
+---
+name:            fn1
+tracksRegLiveness: false
+body:             |
+  bb.0.entry:
+    $sp = frame-setup STMDB_UPD $sp, 14, $noreg, killed $r4, killed $lr
+    $r4 = MOVr $r0, 14, $noreg, $noreg
+    DBG_VALUE $r1, $noreg, !17, !DIExpression(DW_OP_LLVM_fragment, 32, 32), debug-location !18
+    DBG_VALUE $noreg, $noreg, !17, !DIExpression(DW_OP_LLVM_fragment, 64, 32), debug-location !18
+    DBG_VALUE $r4, $noreg, !17, !DIExpression(DW_OP_LLVM_fragment, 0, 32), debug-location !18
+    $r0 = MOVr killed $r1, 14, $noreg, $noreg, debug-location !19
+    BL @fn2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0, debug-location !19
+    renamable $r1 = MOVi 0, 14, $noreg, $noreg
+    CMPrr killed renamable $r4, killed renamable $r0, 14, $noreg, implicit-def $cpsr, debug-location !19
+    $r1 = MOVi 1, 10, killed $cpsr, $noreg, implicit killed renamable $r1, debug-location !19
+    $r0 = MOVr killed $r1, 14, $noreg, $noreg, debug-location !19
+    $sp = LDMIA_UPD $sp, 14, $noreg, def $r4, def $lr, debug-location !19
+    BX_RET 14, $noreg, implicit $r0, debug-location !19
+
+...
+
+# Verify that the addresses in the location list for the parameter are
+# monotonically increasing, and that the undef debug value fragment does not
+# terminate the non-overlapping fragment that is described by $r1.
+
+# CHECK: DW_AT_location (0x00000000
+# CHECK-NEXT: [0x00000008, 0x00000010): DW_OP_reg4 R4, DW_OP_piece 0x4, DW_OP_reg1 R1, DW_OP_piece 0x4
+# CHECK-NEXT: [0x00000010, 0x00000024): DW_OP_reg4 R4, DW_OP_piece 0x4)
+# CHECK-NEXT: DW_AT_name ("p1")

test/DebugInfo/X86/undef-fragment.ll

+; RUN: llc -mtriple=x86_64-pc-linux-gnu -O2 -filetype=obj < %s | llvm-dwarfdump - | FileCheck %s
+
+; This reproducer is based on the following C code:
+;
+; typedef struct { int a; int b; } S;
+;
+; extern int ext(int);
+;
+; int foo() {
+;   S s = {123, 456};
+;   ext(1);
+;   s.a = ext(2);
+;   ext(3);
+;   s.a = 789;
+;   return s.b;
+; }
+;
+; and was generated using -O2 -g -fno-inline.
+;
+; As a small note, the third dbg.value's value has been changed from %call1 to
+; undef (it would have become undef either way, but this was done to make the
+; intention of the test a bit more clear).
+
+; Verify that a location list entry describing s.b is started at the
+; non-overlapping undef debug value.
+
+; CHECK: DW_AT_location (0x00000000
+; CHECK-NEXT: {{0x[0-9a-f]+}}, [[ADDR1:0x[0-9a-f]+]]): DW_OP_constu 0x7b, DW_OP_stack_value, DW_OP_piece 0x4, DW_OP_constu 0x1c8, DW_OP_stack_value, DW_OP_piece 0x4
+; CHECK-NEXT: [[ADDR1]], [[ADDR2:0x[0-9a-f]+]]): DW_OP_piece 0x4, DW_OP_constu 0x1c8, DW_OP_stack_value, DW_OP_piece 0x4
+; CHECK-NEXT: [[ADDR2]], {{0x[0-9a-f]+}}): DW_OP_constu 0x315, DW_OP_stack_value, DW_OP_piece 0x4, DW_OP_constu 0x1c8, DW_OP_stack_value, DW_OP_piece 0x4)
+; CHECK-NEXT: DW_AT_name    ("s")
+
+; Function Attrs: noinline nounwind uwtable
+define i32 @main() !dbg !7 {
+entry:
+  call void @llvm.dbg.value(metadata i32 123, metadata !12, metadata !DIExpression(DW_OP_LLVM_fragment, 0, 32)), !dbg !18
+  call void @llvm.dbg.value(metadata i32 456, metadata !12, metadata !DIExpression(DW_OP_LLVM_fragment, 32, 32)), !dbg !18
+  %call = tail call i32 @ext(i32 1), !dbg !19
+  %call1 = tail call i32 @ext(i32 2), !dbg !20
+  call void @llvm.dbg.value(metadata i32 undef, metadata !12, metadata !DIExpression(DW_OP_LLVM_fragment, 0, 32)), !dbg !18
+  %call2 = tail call i32 @ext(i32 3), !dbg !21
+  call void @llvm.dbg.value(metadata i32 789, metadata !12, metadata !DIExpression(DW_OP_LLVM_fragment, 0, 32)), !dbg !18
+  ret i32 456, !dbg !22
+}
+
+declare i32 @ext(i32)
+
+; Function Attrs: nounwind readnone speculatable
+declare void @llvm.dbg.value(metadata, metadata, metadata) #0
+
+attributes #0 = { nounwind readnone speculatable }
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!3, !4}
+!llvm.ident = !{!6}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 9.0.0", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, nameTableKind: None)
+!1 = !DIFile(filename: "undef.c", directory: "/")
+!2 = !{}
+!3 = !{i32 2, !"Dwarf Version", i32 4}
+!4 = !{i32 2, !"Debug Info Version", i32 3}
+!6 = !{!"clang version 9.0.0"}
+!7 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 5, type: !8, scopeLine: 5, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !11)
+!8 = !DISubroutineType(types: !9)
+!9 = !{!10}
+!10 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!11 = !{!12}
+!12 = !DILocalVariable(name: "s", scope: !7, file: !1, line: 6, type: !13)
+!13 = !DIDerivedType(tag: DW_TAG_typedef, name: "S", file: !1, line: 1, baseType: !14)
+!14 = distinct !DICompositeType(tag: DW_TAG_structure_type, file: !1, line: 1, size: 64, elements: !15)
+!15 = !{!16, !17}
+!16 = !DIDerivedType(tag: DW_TAG_member, name: "a", scope: !14, file: !1, line: 1, baseType: !10, size: 32)
+!17 = !DIDerivedType(tag: DW_TAG_member, name: "b", scope: !14, file: !1, line: 1, baseType: !10, size: 32, offset: 32)
+!18 = !DILocation(line: 6, scope: !7)
+!19 = !DILocation(line: 7, scope: !7)
+!20 = !DILocation(line: 8, scope: !7)
+!21 = !DILocation(line: 9, scope: !7)
+!22 = !DILocation(line: 11, scope: !7)