[DebugInfo] Generate .debug_names section when it makes sense

[labath]

Summary

This patch makes us generate the debug_names section in response to some
user-facing commands (previously it was only generated if explicitly
selected via the -accel-tables option).

My goal was to make this work for DWARF>=5 (as it's an official part of
that standard), and also, as an extension, for DWARF<5 if one is
explicitly tuning for lldb as a debugger (because it brings a large
performance improvement there).

This is slightly complicated by the fact that the debug_names tables are
incompatible with the DWARF v4 type units (they assume that the type
units are in the debug_info section), and unfortunately, right now we
generate DWARF v4-style type units even for -gdwarf-5. For this reason,
I disable all accelerator tables if the user requested type unit
generation. I do this even for apple tables, as they have the same
problem (in fact generating type units for apple targets makes us crash
even before we get around to emitting the accelerator tables).

[JDevlieghere]

LGTM!

[probinson]

This is slightly complicated by the fact that the debug_names tables are
incompatible with the DWARF v4 type units (they assume that the type
units are in the debug_info section), and unfortunately, right now we
generate DWARF v4-style type units even for -gdwarf-5.

yeah, sorry, I am working on it. However the dumper has the v4 style embedded very deeply, and the refactoring is taking a long time.
(I mean, *emitting* v5 type units in .debug_info would be pretty easy; it's the dumper that's taking all the effort, and I'm unwilling to emit DWARF that we can't dump.)

[probinson]

test/DebugInfo/X86/string-offsets-table.ll

96

I take it there's another string to check here? Please add that check.

[labath]

Updated the test to include the two extra strings that are used from the
debug_names table.

[labath]

test/DebugInfo/X86/string-offsets-table.ll

96

Yes, there are two more strings now which are used in the debug_names section.

BTW, this reminded me: We currently put all debug_str strings into the debug_str_offsets table. However, the strings, which are only used from the debug_names section (which can be most of them in split-dwarf), don't need/use the debug_str_offsets indirection as the debug_names section contains an embedded mini-offset-table.

Do you think it would make sense/be valid DWARF to restrict the debug_str_offsets table only to those strings which actually are referenced via the DW_FORM_strx and likes (in a separate patch, of course)?

[probinson]

test/DebugInfo/X86/string-offsets-table.ll

96

Yes, debug_str_offsets should ideally have only those strings referenced via DW_FORM_strx etc.

lib/CodeGen/AsmPrinter/DwarfDebug.h

class DwarfDebug : public DebugHandlerBase {
 
   /// True if the sections itself must be used as references and don't create
   /// temp symbols inside DWARF sections.
   bool UseSectionsAsReferences = false;
 
   ///Allow emission of the .debug_loc section.
   bool UseLocSection = true;
 
+  /// Generate DWARF v4 type units.
+  bool GenerateTypeUnits;
+
   /// DWARF5 Experimental Options
   /// @{
   AccelTableKind TheAccelTableKind;
   bool HasAppleExtensionAttributes;
   bool HasSplitDwarf;
 
   /// Whether to generate the DWARF v5 string offsets table.
   /// It consists of a series of contributions, each preceded by a header.
public:
   /// Returns whether to use sections as labels rather than temp symbols.
   bool useSectionsAsReferences() const {
     return UseSectionsAsReferences;
   }
 
   /// Returns whether .debug_loc section should be emitted.
   bool useLocSection() const { return UseLocSection; }
 
+  /// Returns whether to generate DWARF v4 type units.
+  bool generateTypeUnits() const { return GenerateTypeUnits; }
+
   // Experimental DWARF5 features.
 
   /// Returns what kind (if any) of accelerator tables to emit.
   AccelTableKind getAccelTableKind() const { return TheAccelTableKind; }
 
   bool useAppleExtensionAttributes() const {
     return HasAppleExtensionAttributes;
   }

lib/CodeGen/AsmPrinter/DwarfDebug.cpp

static cl::opt<bool> UseDwarfRangesBaseAddressSpecifier(
     "use-dwarf-ranges-base-address-specifier", cl::Hidden,
     cl::desc("Use base address specifiers in debug_ranges"), cl::init(false));
 
 static cl::opt<bool> GenerateARangeSection("generate-arange-section",
                                            cl::Hidden,
                                            cl::desc("Generate dwarf aranges"),
                                            cl::init(false));
 
+static cl::opt<bool>
+    GenerateDwarfTypeUnits("generate-type-units", cl::Hidden,
+                           cl::desc("Generate DWARF4 type units."),
+                           cl::init(false));
+
 static cl::opt<bool> SplitDwarfCrossCuReferences(
     "split-dwarf-cross-cu-references", cl::Hidden,
     cl::desc("Enable cross-cu references in DWO files"), cl::init(false));
 
 enum DefaultOnOff { Default, Enable, Disable };
 
 static cl::opt<DefaultOnOff> UnknownLocations(
     "use-unknown-locations", cl::Hidden,
void DbgVariable::addMMIEntry(const DbgVariable &V) {
   assert((FrameIndexExprs.size() == 1 ||
           llvm::all_of(FrameIndexExprs,
                        [](FrameIndexExpr &FIE) {
                          return FIE.Expr && FIE.Expr->isFragment();
                        })) &&
          "conflicting locations for variable");
 }
 
+static AccelTableKind computeAccelTableKind(unsigned DwarfVersion,
+                                            bool GenerateTypeUnits,
+                                            DebuggerKind Tuning,
+                                            const Triple &TT) {
+  // Honor an explicit request.
+  if (AccelTables != AccelTableKind::Default)
+    return AccelTables;
+
+  // Accelerator tables with type units are currently not supported.
+  if (GenerateTypeUnits)
+    return AccelTableKind::None;
+
+  // Accelerator tables get emitted if targetting DWARF v5 or LLDB.  DWARF v5
+  // always implies debug_names. For lower standard versions we use apple
+  // accelerator tables on apple platforms and debug_names elsewhere.
+  if (DwarfVersion >= 5)
+    return AccelTableKind::Dwarf;
+  if (Tuning == DebuggerKind::LLDB)
+    return TT.isOSBinFormatMachO() ? AccelTableKind::Apple
+                                   : AccelTableKind::Dwarf;
+  return AccelTableKind::None;
+}
+
 DwarfDebug::DwarfDebug(AsmPrinter *A, Module *M)
     : DebugHandlerBase(A), DebugLocs(A->OutStreamer->isVerboseAsm()),
       InfoHolder(A, "info_string", DIEValueAllocator),
       SkeletonHolder(A, "skel_string", DIEValueAllocator),
       IsDarwin(A->TM.getTargetTriple().isOSDarwin()) {
   const Triple &TT = Asm->TM.getTargetTriple();
 
   // Make sure we know our "debugger tuning."  The target option takes
   // precedence; fall back to triple-based defaults.
   if (Asm->TM.Options.DebuggerTuning != DebuggerKind::Default)
     DebuggerTuning = Asm->TM.Options.DebuggerTuning;
   else if (IsDarwin)
     DebuggerTuning = DebuggerKind::LLDB;
   else if (TT.isPS4CPU())
     DebuggerTuning = DebuggerKind::SCE;
   else
     DebuggerTuning = DebuggerKind::GDB;
 
-  // Turn on accelerator tables by default, if tuning for LLDB and the target is
-  // supported.
-  if (AccelTables == AccelTableKind::Default) {
-    if (tuneForLLDB() && A->TM.getTargetTriple().isOSBinFormatMachO())
-      TheAccelTableKind = AccelTableKind::Apple;
-    else
-      TheAccelTableKind = AccelTableKind::None;
-  } else
-    TheAccelTableKind = AccelTables;
-
   if (DwarfInlinedStrings == Default)
     UseInlineStrings = TT.isNVPTX();
   else
     UseInlineStrings = DwarfInlinedStrings == Enable;
 
   UseLocSection = !TT.isNVPTX();
 
   HasAppleExtensionAttributes = tuneForLLDB();
DwarfDebug::DwarfDebug(AsmPrinter *A, Module *M)
   UseRangesSection = !NoDwarfRangesSection && !TT.isNVPTX();
 
   // Use sections as references. Force for NVPTX.
   if (DwarfSectionsAsReferences == Default)
     UseSectionsAsReferences = TT.isNVPTX();
   else
     UseSectionsAsReferences = DwarfSectionsAsReferences == Enable;
 
+  GenerateTypeUnits = GenerateDwarfTypeUnits;
+
+  TheAccelTableKind = computeAccelTableKind(
+      DwarfVersion, GenerateTypeUnits, DebuggerTuning, A->TM.getTargetTriple());
+
   // Work around a GDB bug. GDB doesn't support the standard opcode;
   // SCE doesn't support GNU's; LLDB prefers the standard opcode, which
   // is defined as of DWARF 3.
   // See GDB bug 11616 - DW_OP_form_tls_address is unimplemented
   // https://sourceware.org/bugzilla/show_bug.cgi?id=11616
   UseGNUTLSOpcode = tuneForGDB() || DwarfVersion < 3;
 
   // GDB does not fully support the DWARF 4 representation for bitfields.

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

 #include <cstdint>
 #include <string>
 #include <utility>
 
 using namespace llvm;
 
 #define DEBUG_TYPE "dwarfdebug"
 
-static cl::opt<bool>
-GenerateDwarfTypeUnits("generate-type-units", cl::Hidden,
-                       cl::desc("Generate DWARF4 type units."),
-                       cl::init(false));
-
 DIEDwarfExpression::DIEDwarfExpression(const AsmPrinter &AP, DwarfUnit &DU,
                                        DIELoc &DIE)
     : DwarfExpression(AP.getDwarfVersion()), AP(AP), DU(DU),
       DIE(DIE) {}
 
 void DIEDwarfExpression::emitOp(uint8_t Op, const char* Comment) {
   DU.addUInt(DIE, dwarf::DW_FORM_data1, Op);
 }
bool DwarfUnit::isShareableAcrossCUs(const DINode *D) const {
   // cross-CU DIE sharing is used in LTO and removes type redundancy at that
   // level already) but may be implementable for some value in projects
   // building multiple independent libraries with LTO and then linking those
   // together.
   if (isDwoUnit() && !DD->shareAcrossDWOCUs())
     return false;
   return (isa<DIType>(D) ||
           (isa<DISubprogram>(D) && !cast<DISubprogram>(D)->isDefinition())) &&
-         !GenerateDwarfTypeUnits;
+         !DD->generateTypeUnits();
 }
 
 DIE *DwarfUnit::getDIE(const DINode *D) const {
   if (isShareableAcrossCUs(D))
     return DU->getDIE(D);
   return MDNodeToDieMap.lookup(D);
 }
 
DIE *DwarfUnit::getOrCreateTypeDIE(const MDNode *TyNode) {
 
   updateAcceleratorTables(Context, Ty, TyDIE);
 
   if (auto *BT = dyn_cast<DIBasicType>(Ty))
     constructTypeDIE(TyDIE, BT);
   else if (auto *STy = dyn_cast<DISubroutineType>(Ty))
     constructTypeDIE(TyDIE, STy);
   else if (auto *CTy = dyn_cast<DICompositeType>(Ty)) {
-    if (GenerateDwarfTypeUnits && !Ty->isForwardDecl())
+    if (DD->generateTypeUnits() && !Ty->isForwardDecl())
       if (MDString *TypeId = CTy->getRawIdentifier()) {
         DD->addDwarfTypeUnitType(getCU(), TypeId->getString(), TyDIE, CTy);
         // Skip updating the accelerator tables since this is not the full type.
         return &TyDIE;
       }
     constructTypeDIE(TyDIE, CTy);
   } else {
     constructTypeDIE(TyDIE, cast<DIDerivedType>(Ty));

test/DebugInfo/X86/accel-tables-dwarf5.ll

+; Verify the emission of accelerator tables for the DWARF v5 case.
+
+; debug_names should be emitted regardless of the target and debugger tuning
+; RUN: llc -mtriple=x86_64-apple-darwin12 -filetype=obj < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=DEBUG_NAMES %s
+; RUN: llc -mtriple=x86_64-apple-darwin12 -filetype=obj -debugger-tune=gdb < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=DEBUG_NAMES %s
+; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=DEBUG_NAMES %s
+; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj -debugger-tune=lldb < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=DEBUG_NAMES %s
+
+; But not if also type units are enabled.
+; TODO: This is the case because we currently don't generate DWARF v5-compatible
+; type units. Change this once DWARF v5 type units are implemented.
+; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj -generate-type-units -debugger-tune=lldb < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=NONE %s
+; RUN: llc -mtriple=x86_64-apple-darwin12 -generate-type-units -filetype=obj < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=NONE %s
+
+; NONE-NOT: apple_names
+; NONE-NOT: debug_names
+
+; DEBUG_NAMES-NOT: apple_names
+; DEBUG_NAMES: debug_names
+; DEBUG_NAMES-NOT: apple_names
+
+
+@x = common dso_local global i32 0, align 4, !dbg !0
+
+!llvm.dbg.cu = !{!2}
+!llvm.module.flags = !{!8, !9, !10}
+!llvm.ident = !{!11}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(name: "x", scope: !2, file: !6, line: 1, type: !7, isLocal: false, isDefinition: true)
+!2 = distinct !DICompileUnit(language: DW_LANG_C99, file: !3, producer: "clang version 7.0.0 (trunk 337260) (llvm/trunk 337262)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !4, globals: !5)
+!3 = !DIFile(filename: "-", directory: "/tmp", checksumkind: CSK_MD5, checksum: "06c25fe0c80b8959051a62f8f034710a")
+!4 = !{}
+!5 = !{!0}
+!6 = !DIFile(filename: "<stdin>", directory: "/tmp", checksumkind: CSK_MD5, checksum: "06c25fe0c80b8959051a62f8f034710a")
+!7 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!8 = !{i32 2, !"Dwarf Version", i32 5}
+!9 = !{i32 2, !"Debug Info Version", i32 3}
+!10 = !{i32 1, !"wchar_size", i32 4}
+!11 = !{!"clang version 7.0.0 (trunk 337260) (llvm/trunk 337262)"}

test/DebugInfo/X86/accel-tables.ll

-; Verify the emission of accelerator tables for various targets.
+; Verify the emission of accelerator tables for various targets for the DWARF<=4 case
 
-; Darwin has the tables unless we specifically tune for gdb
-; RUN: llc -mtriple=x86_64-apple-darwin12 -filetype=obj < %s | llvm-readobj -sections - | FileCheck --check-prefix=CHECK1 %s
-; RUN: llc -mtriple=x86_64-apple-darwin12 -filetype=obj -debugger-tune=gdb < %s | llvm-readobj -sections - | FileCheck --check-prefix=CHECK2 %s
-
-; Linux does not have the tables even if we explicitly tune for lldb
-; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj < %s | llvm-readobj -sections - | FileCheck --check-prefix=CHECK2 %s
-; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj -debugger-tune=lldb < %s | llvm-readobj -sections - | FileCheck --check-prefix=CHECK2 %s
-
-; CHECK1: apple_names
-
-; CHECK2-NOT: apple_names
+; Darwin has the apple tables unless we specifically tune for gdb
+; RUN: llc -mtriple=x86_64-apple-darwin12 -filetype=obj < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=APPLE %s
+; RUN: llc -mtriple=x86_64-apple-darwin12 -filetype=obj -debugger-tune=gdb < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=NONE %s
+
+; Linux does has debug_names tables only if we explicitly tune for lldb
+; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=NONE %s
+; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj -debugger-tune=lldb < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=DEBUG_NAMES %s
+
+; Neither target has accelerator tables if type units are enabled, as DWARF v4
+; type units are not compatible with accelerator tables.
+; RUN: llc -mtriple=x86_64-pc-linux -filetype=obj -generate-type-units -debugger-tune=lldb < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=NONE %s
+; RUN: llc -mtriple=x86_64-apple-darwin12 -generate-type-units -filetype=obj < %s \
+; RUN:   | llvm-readobj -sections - | FileCheck --check-prefix=NONE %s
+
+; APPLE-NOT: debug_names
+; APPLE: apple_names
+; APPLE-NOT: debug_names
+
+; NONE-NOT: apple_names
+; NONE-NOT: debug_names
+
+; DEBUG_NAMES-NOT: apple_names
+; DEBUG_NAMES: debug_names
+; DEBUG_NAMES-NOT: apple_names
 
 @var = thread_local global i32 0, align 4, !dbg !0
 
 ; Function Attrs: norecurse nounwind readnone uwtable
 define void @_Z3funv() local_unnamed_addr #0 !dbg !11 {
   ret void, !dbg !14
 }
 

test/DebugInfo/X86/string-offsets-table.ll

 ; SPLIT:      DW_TAG_enumerator
 ; SPLIT-NOT:  {{DW_TAG|NULL}}
 ; SPLIT:      DW_AT_name [DW_FORM_strx1]    ( indexed (00000005) string = "b")
 ;
 ; Extract the string offsets referenced in the main file by the skeleton unit.
 ; SPLIT:      .debug_str contents:
 ; SPLIT-NEXT: 0x00000000:{{.*}}
 ; SPLIT-NEXT: 0x[[STRING2SPLIT:[0-9a-f]*]]{{.*}}
+; SPLIT-NEXT: 0x[[STRING3SPLIT:[0-9a-f]*]]{{.*}}
+; SPLIT-NEXT: 0x[[STRING4SPLIT:[0-9a-f]*]]{{.*}}
 ;
 ; Extract the string offsets referenced in the .dwo file by the split unit.
 ; SPLIT:      .debug_str.dwo contents:
 ; SPLIT-NEXT: 0x00000000:{{.*}}
 ; SPLIT-NEXT: 0x[[STRING2DWO:[0-9a-f]*]]{{.*}}
 ; SPLIT-NEXT: 0x[[STRING3DWO:[0-9a-f]*]]{{.*}}
 ;
 ; Check the string offsets sections in both the main and the .dwo files and
 ; verify that the extracted string offsets are referenced correctly.
 ; SPLIT:      .debug_str_offsets contents:
-; SPLIT-NEXT: 0x00000000: Contribution size = 12, Format = DWARF32, Version = 5
+; SPLIT-NEXT: 0x00000000: Contribution size = 20, Format = DWARF32, Version = 5
 ; SPLIT-NEXT: 0x00000008: 00000000{{.*}}
 ; SPLIT-NEXT: 0x0000000c: [[STRING2SPLIT]]
+; SPLIT-NEXT: 0x00000010: [[STRING3SPLIT]]
+; SPLIT-NEXT: 0x00000014: [[STRING4SPLIT]]
 ; SPLIT:      .debug_str_offsets.dwo contents:
 ; SPLIT-NEXT: 0x00000000: Contribution size = 36, Format = DWARF32, Version = 5
 ; SPLIT-NEXT: 0x00000008: 00000000{{.*}}
 ; SPLIT-NEXT: 0x0000000c: [[STRING2DWO]]{{.*}}
 ; SPLIT-NEXT: 0x00000010: [[STRING3DWO]]
 
 @glob = global i32 0, align 4, !dbg !0