Source Level Debugging with LLVM

Introduction

This document is the central repository for all information pertaining to debuginformation in LLVM. It describes the actual format that the LLVM debuginformation takes, which is useful for those interested in creatingfront-ends or dealing directly with the information. Further, this documentprovides specific examples of what debug information for C/C++ looks like.

Philosophy behind LLVM debugging information

The idea of the LLVM debugging information is to capture how the importantpieces of the source-language’s Abstract Syntax Tree map onto LLVM code.Several design aspects have shaped the solution that appears here. Theimportant ones are:

• Debugging information should have very little impact on the rest of thecompiler. No transformations, analyses, or code generators should need tobe modified because of debugging information.
• LLVM optimizations should interact in well-defined and easily describedways with the debugging information.
• Because LLVM is designed to support arbitrary programming languages,LLVM-to-LLVM tools should not need to know anything about the semantics ofthe source-level-language.
• Source-level languages are often widely different from one another.LLVM should not put any restrictions of the flavor of the source-language,and the debugging information should work with any language.
• With code generator support, it should be possible to use an LLVM compilerto compile a program to native machine code and standard debuggingformats. This allows compatibility with traditional machine-code leveldebuggers, like GDB or DBX.

The approach used by the LLVM implementation is to use a small set ofintrinsic functions to define a mappingbetween LLVM program objects and the source-level objects. The description ofthe source-level program is maintained in LLVM metadata in animplementation-defined format (the C/C++ front-endcurrently uses working draft 7 of the DWARF 3 standard).

When a program is being debugged, a debugger interacts with the user and turnsthe stored debug information into source-language specific information. Assuch, a debugger must be aware of the source-language, and is thus tied to aspecific language or family of languages.

Debug information consumers

The role of debug information is to provide meta information normally strippedaway during the compilation process. This meta information provides an LLVMuser a relationship between generated code and the original program sourcecode.

Currently, there are two backend consumers of debug info: DwarfDebug andCodeViewDebug. DwarfDebug produces DWARF suitable for use with GDB, LLDB, andother DWARF-based debuggers. CodeViewDebug produces CodeView,the Microsoft debug info format, which is usable with Microsoft debuggers suchas Visual Studio and WinDBG. LLVM’s debug information format is mostly derivedfrom and inspired by DWARF, but it is feasible to translate into other targetdebug info formats such as STABS.

It would also be reasonable to use debug information to feed profiling toolsfor analysis of generated code, or, tools for reconstructing the originalsource from generated code.

Debug information and optimizations

An extremely high priority of LLVM debugging information is to make it interactwell with optimizations and analysis. In particular, the LLVM debuginformation provides the following guarantees:

• LLVM debug information always provides information to accurately readthe source-level state of the program, regardless of which LLVMoptimizations have been run, and without any modification to theoptimizations themselves. However, some optimizations may impact theability to modify the current state of the program with a debugger, suchas setting program variables, or calling functions that have beendeleted.
• As desired, LLVM optimizations can be upgraded to be aware of debugginginformation, allowing them to update the debugging information as theyperform aggressive optimizations. This means that, with effort, the LLVMoptimizers could optimize debug code just as well as non-debug code.
• LLVM debug information does not prevent optimizations fromhappening (for example inlining, basic block reordering/merging/cleanup,tail duplication, etc).
• LLVM debug information is automatically optimized along with the rest ofthe program, using existing facilities. For example, duplicateinformation is automatically merged by the linker, and unused informationis automatically removed.

Basically, the debug information allows you to compile a program with“-O0 -g” and get full debug information, allowing you to arbitrarily modifythe program as it executes from a debugger. Compiling a program with“-O3 -g” gives you full debug information that is always available andaccurate for reading (e.g., you get accurate stack traces despite tail callelimination and inlining), but you might lose the ability to modify the programand call functions which were optimized out of the program, or inlined awaycompletely.

The LLVM test-suite provides a framework totest the optimizer’s handling of debugging information. It can be run likethis:

% cd llvm/projects/test-suite/MultiSource/Benchmarks  # or some other level
% make TEST=dbgopt

This will test impact of debugging information on optimization passes. Ifdebugging information influences optimization passes then it will be reportedas a failure. See LLVM Testing Infrastructure Guide for more information on LLVM testinfrastructure and how to run various tests.

Debugging information format

LLVM debugging information has been carefully designed to make it possible forthe optimizer to optimize the program and debugging information withoutnecessarily having to know anything about debugging information. Inparticular, the use of metadata avoids duplicated debugging information fromthe beginning, and the global dead code elimination pass automatically deletesdebugging information for a function if it decides to delete the function.

To do this, most of the debugging information (descriptors for types,variables, functions, source files, etc) is inserted by the language front-endin the form of LLVM metadata.

Debug information is designed to be agnostic about the target debugger anddebugging information representation (e.g. DWARF/Stabs/etc). It uses a genericpass to decode the information that represents variables, types, functions,namespaces, etc: this allows for arbitrary source-language semantics andtype-systems to be used, as long as there is a module written for the targetdebugger to interpret the information.

To provide basic functionality, the LLVM debugger does have to make someassumptions about the source-level language being debugged, though it keepsthese to a minimum. The only common features that the LLVM debugger assumesexist are source files, and program objects. These abstract objects are used by adebugger to form stack traces, show information about local variables, etc.

This section of the documentation first describes the representation aspectscommon to any source-language. C/C++ front-end specific debug information describes the data layoutconventions used by the C and C++ front-ends.

Debug information descriptors are specialized metadata nodes, first-class subclasses of Metadata.

Debugger intrinsic functions

LLVM uses several intrinsic functions (name prefixed with “llvm.dbg”) totrack source local variables through optimization and code generation.

llvm.dbg.addr

void @llvm.dbg.addr(metadata, metadata, metadata)

This intrinsic provides information about a local element (e.g., variable).The first argument is metadata holding the address of variable, typically astatic alloca in the function entry block. The second argument is alocal variable containing a description ofthe variable. The third argument is a complex expression. An llvm.dbg.addr intrinsic describes theaddress of a source variable.

%i.addr = alloca i32, align 4
call void @llvm.dbg.addr(metadata i32* %i.addr, metadata !1,
                         metadata !DIExpression()), !dbg !2
!1 = !DILocalVariable(name: "i", ...) ; int i
!2 = !DILocation(...)
...
%buffer = alloca [256 x i8], align 8
; The address of i is buffer+64.
call void @llvm.dbg.addr(metadata [256 x i8]* %buffer, metadata !3,
                         metadata !DIExpression(DW_OP_plus, 64)), !dbg !4
!3 = !DILocalVariable(name: "i", ...) ; int i
!4 = !DILocation(...)

A frontend should generate exactly one call to llvm.dbg.addr at the pointof declaration of a source variable. Optimization passes that fully promote thevariable from memory to SSA values will replace this call with possiblymultiple calls to llvm.dbg.value. Passes that delete stores are effectivelypartial promotion, and they will insert a mix of calls to llvm.dbg.valueand llvm.dbg.addr to track the source variable value when it is available.After optimization, there may be multiple calls to llvm.dbg.addr describingthe program points where the variables lives in memory. All calls for the sameconcrete source variable must agree on the memory location.

llvm.dbg.declare

void @llvm.dbg.declare(metadata, metadata, metadata)

This intrinsic is identical to llvm.dbg.addr, except that there can only beone call to llvm.dbg.declare for a given concrete local variable. It is not control-dependent, meaning that ifa call to llvm.dbg.declare exists and has a valid location argument, thataddress is considered to be the true home of the variable across its entirelifetime. This makes it hard for optimizations to preserve accurate debug infoin the presence of llvm.dbg.declare, so we are transitioning away from it,and we plan to deprecate it in future LLVM releases.

llvm.dbg.value

void @llvm.dbg.value(metadata, metadata, metadata)

This intrinsic provides information when a user source variable is set to a newvalue. The first argument is the new value (wrapped as metadata). The secondargument is a local variable containing adescription of the variable. The third argument is a complex expression.

An llvm.dbg.value intrinsic describes the value of a source variabledirectly, not its address. Note that the value operand of this intrinsic maybe indirect (i.e, a pointer to the source variable), provided that interpretingthe complex expression derives the direct value.

Object lifetimes and scoping

In many languages, the local variables in functions can have their lifetimes orscopes limited to a subset of a function. In the C family of languages, forexample, variables are only live (readable and writable) within the sourceblock that they are defined in. In functional languages, values are onlyreadable after they have been defined. Though this is a very obvious concept,it is non-trivial to model in LLVM, because it has no notion of scoping in thissense, and does not want to be tied to a language’s scoping rules.

In order to handle this, the LLVM debug format uses the metadata attached tollvm instructions to encode line number and scoping information. Consider thefollowing C fragment, for example:

1.  void foo() {
2.    int X = 21;
3.    int Y = 22;
4.    {
5.      int Z = 23;
6.      Z = X;
7.    }
8.    X = Y;
9.  }

Compiled to LLVM, this function would be represented like this:

; Function Attrs: nounwind ssp uwtable
define void @foo() #0 !dbg !4 {
entry:
  %X = alloca i32, align 4
  %Y = alloca i32, align 4
  %Z = alloca i32, align 4
  call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14
  store i32 21, i32* %X, align 4, !dbg !14
  call void @llvm.dbg.declare(metadata i32* %Y, metadata !15, metadata !13), !dbg !16
  store i32 22, i32* %Y, align 4, !dbg !16
  call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19
  store i32 23, i32* %Z, align 4, !dbg !19
  %0 = load i32, i32* %X, align 4, !dbg !20
  store i32 %0, i32* %Z, align 4, !dbg !21
  %1 = load i32, i32* %Y, align 4, !dbg !22
  store i32 %1, i32* %X, align 4, !dbg !23
  ret void, !dbg !24
}
 
; Function Attrs: nounwind readnone
declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
 
attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { nounwind readnone }
 
!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!7, !8, !9}
!llvm.ident = !{!10}
 
!0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2, subprograms: !3, globals: !2, imports: !2)
!1 = !DIFile(filename: "/dev/stdin", directory: "/Users/dexonsmith/data/llvm/debug-info")
!2 = !{}
!3 = !{!4}
!4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5, isLocal: false, isDefinition: true, scopeLine: 1, isOptimized: false, variables: !2)
!5 = !DISubroutineType(types: !6)
!6 = !{null}
!7 = !{i32 2, !"Dwarf Version", i32 2}
!8 = !{i32 2, !"Debug Info Version", i32 3}
!9 = !{i32 1, !"PIC Level", i32 2}
!10 = !{!"clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)"}
!11 = !DILocalVariable(name: "X", scope: !4, file: !1, line: 2, type: !12)
!12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
!13 = !DIExpression()
!14 = !DILocation(line: 2, column: 9, scope: !4)
!15 = !DILocalVariable(name: "Y", scope: !4, file: !1, line: 3, type: !12)
!16 = !DILocation(line: 3, column: 9, scope: !4)
!17 = !DILocalVariable(name: "Z", scope: !18, file: !1, line: 5, type: !12)
!18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
!19 = !DILocation(line: 5, column: 11, scope: !18)
!20 = !DILocation(line: 6, column: 11, scope: !18)
!21 = !DILocation(line: 6, column: 9, scope: !18)
!22 = !DILocation(line: 8, column: 9, scope: !4)
!23 = !DILocation(line: 8, column: 7, scope: !4)
!24 = !DILocation(line: 9, column: 3, scope: !4)

This example illustrates a few important details about LLVM debugginginformation. In particular, it shows how the llvm.dbg.declare intrinsic andlocation information, which are attached to an instruction, are appliedtogether to allow a debugger to analyze the relationship between statements,variable definitions, and the code used to implement the function.

call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14
  ; [debug line = 2:7] [debug variable = X]

The first intrinsic %llvm.dbg.declare encodes debugging information for thevariable X. The metadata !dbg !14 attached to the intrinsic providesscope information for the variable X.

!14 = !DILocation(line: 2, column: 9, scope: !4)
!4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5,
                            isLocal: false, isDefinition: true, scopeLine: 1,
                            isOptimized: false, variables: !2)

Here !14 is metadata providing location information. In this example, scope is encoded by !4, asubprogram descriptor. This way the locationinformation attached to the intrinsics indicates that the variable X isdeclared at line number 2 at a function level scope in function foo.

Now lets take another example.

call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19
  ; [debug line = 5:9] [debug variable = Z]

The third intrinsic %llvm.dbg.declare encodes debugging information forvariable Z. The metadata !dbg !19 attached to the intrinsic providesscope information for the variable Z.

!18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
!19 = !DILocation(line: 5, column: 11, scope: !18)

Here !19 indicates that Z is declared at line number 5 and columnnumber 11 inside of lexical scope !18. The lexical scope itself residesinside of subprogram !4 described above.

The scope information attached with each instruction provides a straightforwardway to find instructions covered by a scope.

Object lifetime in optimized code

In the example above, every variable assignment uniquely corresponds to amemory store to the variable’s position on the stack. However in heavilyoptimized code LLVM promotes most variables into SSA values, which caneventually be placed in physical registers or memory locations. To track SSAvalues through compilation, when objects are promoted to SSA values anllvm.dbg.value intrinsic is created for each assignment, recording thevariable’s new location. Compared with the llvm.dbg.declare intrinsic:

• A dbg.value terminates the effect of any preceding dbg.values for (anyoverlapping fragments of) the specified variable.
• The dbg.value’s position in the IR defines where in the instruction streamthe variable’s value changes.
• Operands can be constants, indicating the variable is assigned aconstant value.

Care must be taken to update llvm.dbg.value intrinsics when optimizationpasses alter or move instructions and blocks – the developer could observe suchchanges reflected in the value of variables when debugging the program. For anyexecution of the optimized program, the set of variable values presented to thedeveloper by the debugger should not show a state that would never have existedin the execution of the unoptimized program, given the same input. Doing sorisks misleading the developer by reporting a state that does not exist,damaging their understanding of the optimized program and undermining theirtrust in the debugger.

Sometimes perfectly preserving variable locations is not possible, often when aredundant calculation is optimized out. In such cases, a llvm.dbg.valuewith operand undef should be used, to terminate earlier variable locationsand let the debugger present optimized out to the developer. Withholdingthese potentially stale variable values from the developer diminishes theamount of available debug information, but increases the reliability of theremaining information.

To illustrate some potential issues, consider the following example:

define i32 @foo(i32 %bar, i1 %cond) {
entry:
  call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2)
  br i1 %cond, label %truebr, label %falsebr
truebr:
  %tval = add i32 %bar, 1
  call @llvm.dbg.value(metadata i32 %tval, metadata !1, metadata !2)
  %g1 = call i32 @gazonk()
  br label %exit
falsebr:
  %fval = add i32 %bar, 2
  call @llvm.dbg.value(metadata i32 %fval, metadata !1, metadata !2)
  %g2 = call i32 @gazonk()
  br label %exit
exit:
  %merge = phi [ %tval, %truebr ], [ %fval, %falsebr ]
  %g = phi [ %g1, %truebr ], [ %g2, %falsebr ]
  call @llvm.dbg.value(metadata i32 %merge, metadata !1, metadata !2)
  call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2)
  %plusten = add i32 %merge, 10
  %toret = add i32 %plusten, %g
  call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2)
  ret i32 %toret
}

Containing two source-level variables in !1 and !3. The function could,perhaps, be optimized into the following code:

define i32 @foo(i32 %bar, i1 %cond) {
entry:
  %g = call i32 @gazonk()
  %addoper = select i1 %cond, i32 11, i32 12
  %plusten = add i32 %bar, %addoper
  %toret = add i32 %plusten, %g
  ret i32 %toret
}

What llvm.dbg.value intrinsics should be placed to represent the original variablelocations in this code? Unfortunately the second, third and fourthdbg.values for !1 in the source function have had their operands(%tval, %fval, %merge) optimized out. Assuming we cannot recover them, wemight consider this placement of dbg.values:

define i32 @foo(i32 %bar, i1 %cond) {
entry:
  call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2)
  %g = call i32 @gazonk()
  call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2)
  %addoper = select i1 %cond, i32 11, i32 12
  %plusten = add i32 %bar, %addoper
  %toret = add i32 %plusten, %g
  call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2)
  ret i32 %toret
}

However, this will cause !3 to have the return value of @gazonk() atthe same time as !1 has the constant value zero – a pair of assignmentsthat never occurred in the unoptimized program. To avoid this, we must terminatethe range that !1 has the constant value assignment by inserting an undefdbg.value before the dbg.value for !3:

define i32 @foo(i32 %bar, i1 %cond) {
entry:
  call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2)
  %g = call i32 @gazonk()
  call @llvm.dbg.value(metadata i32 undef, metadata !1, metadata !2)
  call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2)
  %addoper = select i1 %cond, i32 11, i32 12
  %plusten = add i32 %bar, %addoper
  %toret = add i32 %plusten, %g
  call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2)
  ret i32 %toret
}

In general, if any dbg.value has its operand optimized out and cannot berecovered, then an undef dbg.value is necessary to terminate earlier variablelocations. Additional undef dbg.values may be necessary when the debugger canobserve re-ordering of assignments.

How variable location metadata is transformed during CodeGen

LLVM preserves debug information throughout mid-level and backend passes,ultimately producing a mapping between source-level information andinstruction ranges. Thisis relatively straightforwards for line number information, as mappinginstructions to line numbers is a simple association. For variable locationshowever the story is more complex. As each llvm.dbg.value intrinsicrepresents a source-level assignment of a value to a source variable, thevariable location intrinsics effectively embed a small imperative programwithin the LLVM IR. By the end of CodeGen, this becomes a mapping from eachvariable to their machine locations over ranges of instructions.From IR to object emission, the major transformations which affect variablelocation fidelity are:

• Instruction Selection
• Register allocation
• Block layouteach of which are discussed below. In addition, instruction scheduling cansignificantly change the ordering of the program, and occurs in a number ofdifferent passes.

Some variable locations are not transformed during CodeGen. Stack locationsspecified by llvm.dbg.declare are valid and unchanging for the entireduration of the function, and are recorded in a simple MachineFunction table.Location changes in the prologue and epilogue of a function are also ignored:frame setup and destruction may take several instructions, require adisproportionate amount of debugging information in the output binary todescribe, and should be stepped over by debuggers anyway.

Variable locations in Instruction Selection and MIR

Instruction selection creates a MIR function from an IR function, and just asit transforms intermediate instructions into machine instructions, so mustintermediate variable locations become machine variable locations.Within IR, variable locations are always identified by a Value, but in MIRthere can be different types of variable locations. In addition, some IRlocations become unavailable, for example if the operation of multiple IRinstructions are combined into one machine instruction (such asmultiply-and-accumulate) then intermediate Values are lost. To track variablelocations through instruction selection, they are first separated intolocations that do not depend on code generation (constants, stack locations,allocated virtual registers) and those that do. For those that do, debugmetadata is attached to SDNodes in SelectionDAGs. After instruction selectionhas occurred and a MIR function is created, if the SDNode associated with debugmetadata is allocated a virtual register, that virtual register is used as thevariable location. If the SDNode is folded into a machine instruction orotherwise transformed into a non-register, the variable location becomesunavailable.

Locations that are unavailable are treated as if they have been optimized out:in IR the location would be assigned undef by a debug intrinsic, and in MIRthe equivalent location is used.

After MIR locations are assigned to each variable, machine pseudo-instructionscorresponding to each llvm.dbg.value and llvm.dbg.addr intrinsic areinserted. These DBG_VALUE instructions appear thus:

DBG_VALUE %1, $noreg, !123, !DIExpression()

• And have the following operands:
• • The first operand can record the variable location as a register,a frame index, an immediate, or the base address register if the originaldebug intrinsic referred to memory. $noreg indicates the variablelocation is undefined, equivalent to an undef dbg.value operand.
  • The type of the second operand indicates whether the variable location isdirectly referred to by the DBG_VALUE, or whether it is indirect. The$noreg register signifies the former, an immediate operand (0) thelatter.
  • Operand 3 is the Variable field of the original debug intrinsic.
  • Operand 4 is the Expression field of the original debug intrinsic.

The position at which the DBG_VALUEs are inserted should correspond to thepositions of their matching llvm.dbg.value intrinsics in the IR block. Aswith optimization, LLVM aims to preserve the order in which variableassignments occurred in the source program. However SelectionDAG performs someinstruction scheduling, which can reorder assignments (discussed below).Function parameter locations are moved to the beginning of the function ifthey’re not already, to ensure they’re immediately available on function entry.

To demonstrate variable locations during instruction selection, considerthe following example:

define i32 @foo(i32* %addr) {
entry:
  call void @llvm.dbg.value(metadata i32 0, metadata !3, metadata !DIExpression()), !dbg !5
  br label %bb1, !dbg !5
 
bb1:                                              ; preds = %bb1, %entry
  %bar.0 = phi i32 [ 0, %entry ], [ %add, %bb1 ]
  call void @llvm.dbg.value(metadata i32 %bar.0, metadata !3, metadata !DIExpression()), !dbg !5
  %addr1 = getelementptr i32, i32 *%addr, i32 1, !dbg !5
  call void @llvm.dbg.value(metadata i32 *%addr1, metadata !3, metadata !DIExpression()), !dbg !5
  %loaded1 = load i32, i32* %addr1, !dbg !5
  %addr2 = getelementptr i32, i32 *%addr, i32 %bar.0, !dbg !5
  call void @llvm.dbg.value(metadata i32 *%addr2, metadata !3, metadata !DIExpression()), !dbg !5
  %loaded2 = load i32, i32* %addr2, !dbg !5
  %add = add i32 %bar.0, 1, !dbg !5
  call void @llvm.dbg.value(metadata i32 %add, metadata !3, metadata !DIExpression()), !dbg !5
  %added = add i32 %loaded1, %loaded2
  %cond = icmp ult i32 %added, %bar.0, !dbg !5
  br i1 %cond, label %bb1, label %bb2, !dbg !5
 
bb2:                                              ; preds = %bb1
  ret i32 0, !dbg !5
}

If one compiles this IR with llc -o - -start-after=codegen-prepare -stop-after=expand-isel-pseudos -mtriple=x86_64—, the following MIR is produced:

bb.0.entry:
  successors: %bb.1(0x80000000)
  liveins: $rdi
 
  %2:gr64 = COPY $rdi
  %3:gr32 = MOV32r0 implicit-def dead $eflags
  DBG_VALUE 0, $noreg, !3, !DIExpression(), debug-location !5
 
bb.1.bb1:
  successors: %bb.1(0x7c000000), %bb.2(0x04000000)
 
  %0:gr32 = PHI %3, %bb.0, %1, %bb.1
  DBG_VALUE %0, $noreg, !3, !DIExpression(), debug-location !5
  DBG_VALUE %2, $noreg, !3, !DIExpression(DW_OP_plus_uconst, 4, DW_OP_stack_value), debug-location !5
  %4:gr32 = MOV32rm %2, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
  %5:gr64_nosp = MOVSX64rr32 %0, debug-location !5
  DBG_VALUE $noreg, $noreg, !3, !DIExpression(), debug-location !5
  %1:gr32 = INC32r %0, implicit-def dead $eflags, debug-location !5
  DBG_VALUE %1, $noreg, !3, !DIExpression(), debug-location !5
  %6:gr32 = ADD32rm %4, %2, 4, killed %5, 0, $noreg, implicit-def dead $eflags :: (load 4 from %ir.addr2)
  %7:gr32 = SUB32rr %6, %0, implicit-def $eflags, debug-location !5
  JB_1 %bb.1, implicit $eflags, debug-location !5
  JMP_1 %bb.2, debug-location !5
 
bb.2.bb2:
  %8:gr32 = MOV32r0 implicit-def dead $eflags
  $eax = COPY %8, debug-location !5
  RET 0, $eax, debug-location !5

Observe first that there is a DBG_VALUE instruction for every llvm.dbg.valueintrinsic in the source IR, ensuring no source level assignments go missing.Then consider the different ways in which variable locations have been recorded:

• For the first dbg.value an immediate operand is used to record a zero value.
• The dbg.value of the PHI instruction leads to a DBG_VALUE of virtual register%0.
• The first GEP has its effect folded into the first load instruction(as a 4-byte offset), but the variable location is salvaged by foldingthe GEPs effect into the DIExpression.
• The second GEP is also folded into the corresponding load. However, it isinsufficiently simple to be salvaged, and is emitted as a $noregDBG_VALUE, indicating that the variable takes on an undefined location.
• The final dbg.value has its Value placed in virtual register %1.

Instruction Scheduling

A number of passes can reschedule instructions, notably instruction selectionand the pre-and-post RA machine schedulers. Instruction scheduling cansignificantly change the nature of the program – in the (very unlikely) worstcase the instruction sequence could be completely reversed. In suchcircumstances LLVM follows the principle applied to optimizations, that it isbetter for the debugger not to display any state than a misleading state.Thus, whenever instructions are advanced in order of execution, anycorresponding DBG_VALUE is kept in its original position, and if an instructionis delayed then the variable is given an undefined location for the durationof the delay. To illustrate, consider this pseudo-MIR:

%1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
DBG_VALUE %1, $noreg, !1, !2
%4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags
DBG_VALUE %4, $noreg, !3, !4
%7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags
DBG_VALUE %7, $noreg, !5, !6

Imagine that the SUB32rr were moved forward to give us the following MIR:

%7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags
%1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
DBG_VALUE %1, $noreg, !1, !2
%4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags
DBG_VALUE %4, $noreg, !3, !4
DBG_VALUE %7, $noreg, !5, !6

In this circumstance LLVM would leave the MIR as shown above. Were we to movethe DBG_VALUE of virtual register %7 upwards with the SUB32rr, we would re-orderassignments and introduce a new state of the program. Whereas with the solutionabove, the debugger will see one fewer combination of variable values, because!3 and !5 will change value at the same time. This is preferred overmisrepresenting the original program.

In comparison, if one sunk the MOV32rm, LLVM would produce the following:

DBG_VALUE $noreg, $noreg, !1, !2
%4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags
DBG_VALUE %4, $noreg, !3, !4
%7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags
DBG_VALUE %7, $noreg, !5, !6
%1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
DBG_VALUE %1, $noreg, !1, !2

Here, to avoid presenting a state in which the first assignment to !1disappears, the DBG_VALUE at the top of the block assigns the variable theundefined location, until its value is available at the end of the block wherean additional DBG_VALUE is added. Were any other DBG_VALUE for !1 to occurin the instructions that the MOV32rm was sunk past, the DBG_VALUE for %1would be dropped and the debugger would never observe it in the variable. Thisaccurately reflects that the value is not available during the correspondingportion of the original program.

Variable locations during Register Allocation

To avoid debug instructions interfering with the register allocator, theLiveDebugVariables pass extracts variable locations from a MIR function anddeletes the corresponding DBG_VALUE instructions. Some localized copypropagation is performed within blocks. After register allocation, theVirtRegRewriter pass re-inserts DBG_VALUE instructions in their originalpositions, translating virtual register references into their physicalmachine locations. To avoid encoding incorrect variable locations, in thispass any DBG_VALUE of a virtual register that is not live, is replaced bythe undefined location.

LiveDebugValues expansion of variable locations

After all optimizations have run and shortly before emission, theLiveDebugValues pass runs to achieve two aims:

• To propagate the location of variables through copies and register spills,
• For every block, to record every valid variable location in that block.

After this pass the DBG_VALUE instruction changes meaning: rather thancorresponding to a source-level assignment where the variable may change value,it asserts the location of a variable in a block, and loses effect outside theblock. Propagating variable locations through copies and spills isstraightforwards: determining the variable location in every basic blockrequires the consideration of control flow. Consider the following IR, whichpresents several difficulties:

define dso_local i32 @foo(i1 %cond, i32 %input) !dbg !12 {
entry:
  br i1 %cond, label %truebr, label %falsebr
 
bb1:
  %value = phi i32 [ %value1, %truebr ], [ %value2, %falsebr ]
  br label %exit, !dbg !26
 
truebr:
  call void @llvm.dbg.value(metadata i32 %input, metadata !30, metadata !DIExpression()), !dbg !24
  call void @llvm.dbg.value(metadata i32 1, metadata !23, metadata !DIExpression()), !dbg !24
  %value1 = add i32 %input, 1
  br label %bb1
 
falsebr:
  call void @llvm.dbg.value(metadata i32 %input, metadata !30, metadata !DIExpression()), !dbg !24
  call void @llvm.dbg.value(metadata i32 2, metadata !23, metadata !DIExpression()), !dbg !24
  %value = add i32 %input, 2
  br label %bb1
 
exit:
  ret i32 %value, !dbg !30
}

Here the difficulties are:

• The control flow is roughly the opposite of basic block order
• The value of the !23 variable merges into %bb1, but there is no PHInode

As mentioned above, the llvm.dbg.value intrinsics essentially form animperative program embedded in the IR, with each intrinsic defining a variablelocation. This could be converted to an SSA form by mem2reg, in the same waythat it uses use-def chains to identify control flow merges and insert phinodes for IR Values. However, because debug variable locations are defined forevery machine instruction, in effect every IR instruction uses every variablelocation, which would lead to a large number of debugging intrinsics beinggenerated.

Examining the example above, variable !30 is assigned %input on bothconditional paths through the function, while !23 is assigned differingconstant values on either path. Where control flow merges in %bb1 we wouldwant !30 to keep its location (%input), but !23 to become undefinedas we cannot determine at runtime what value it should have in %bb1 withoutinserting a PHI node. mem2reg does not insert the PHI node to avoid changingcodegen when debugging is enabled, and does not insert the other dbg.valuesto avoid adding very large numbers of intrinsics.

Instead, LiveDebugValues determines variable locations when controlflow merges. A dataflow analysis is used to propagate locations between blocks:when control flow merges, if a variable has the same location in allpredecessors then that location is propagated into the successor. If thepredecessor locations disagree, the location becomes undefined.

Once LiveDebugValues has run, every block should have all valid variablelocations described by DBG_VALUE instructions within the block. Very littleeffort is then required by supporting classes (such asDbgEntityHistoryCalculator) to build a map of each instruction to everyvalid variable location, without the need to consider control flow. Fromthe example above, it is otherwise difficult to determine that the locationof variable !30 should flow “up” into block %bb1, but that the locationof variable !23 should not flow “down” into the %exit block.

C/C++ front-end specific debug information

The C and C++ front-ends represent information about the program in aformat that is effectively identical to DWARFin terms of information content. This allows code generators totrivially support native debuggers by generating standard dwarfinformation, and contains enough information for non-dwarf targets totranslate it as needed.

This section describes the forms used to represent C and C++ programs. Otherlanguages could pattern themselves after this (which itself is tuned torepresenting programs in the same way that DWARF does), or they could chooseto provide completely different forms if they don’t fit into the DWARF model.As support for debugging information gets added to the various LLVMsource-language front-ends, the information used should be documented here.

The following sections provide examples of a few C/C++ constructs andthe debug information that would best describe those constructs. Thecanonical references are the DINode classes defined ininclude/llvm/IR/DebugInfoMetadata.h and the implementations of thehelper functions in lib/IR/DIBuilder.cpp.

C/C++ source file information

llvm::Instruction provides easy access to metadata attached with aninstruction. One can extract line number information encoded in LLVM IR usingInstruction::getDebugLoc() and DILocation::getLine().

if (DILocation *Loc = I->getDebugLoc()) { // Here I is an LLVM instruction
  unsigned Line = Loc->getLine();
  StringRef File = Loc->getFilename();
  StringRef Dir = Loc->getDirectory();
  bool ImplicitCode = Loc->isImplicitCode();
}

When the flag ImplicitCode is true then it means that the Instruction has beenadded by the front-end but doesn’t correspond to source code written by the user. For example

if (MyBoolean) {
  MyObject MO;
  ...
}

At the end of the scope the MyObject’s destructor is called but it isn’t writtenexplicitly. This information is useful to avoid to have counters on brackets whenmaking code coverage.

C/C++ global variable information

Given an integer global variable declared as follows:

_Alignas(8) int MyGlobal = 100;

a C/C++ front-end would generate the following descriptors:

;;
;; Define the global itself.
;;
@MyGlobal = global i32 100, align 8, !dbg !0
 
;;
;; List of debug info of globals
;;
!llvm.dbg.cu = !{!1}
 
;; Some unrelated metadata.
!llvm.module.flags = !{!6, !7}
!llvm.ident = !{!8}
 
;; Define the global variable itself
!0 = distinct !DIGlobalVariable(name: "MyGlobal", scope: !1, file: !2, line: 1, type: !5, isLocal: false, isDefinition: true, align: 64)
 
;; Define the compile unit.
!1 = distinct !DICompileUnit(language: DW_LANG_C99, file: !2,
                             producer: "clang version 4.0.0",
                             isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug,
                             enums: !3, globals: !4)
 
;;
;; Define the file
;;
!2 = !DIFile(filename: "/dev/stdin",
             directory: "/Users/dexonsmith/data/llvm/debug-info")
 
;; An empty array.
!3 = !{}
 
;; The Array of Global Variables
!4 = !{!0}
 
;;
;; Define the type
;;
!5 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
 
;; Dwarf version to output.
!6 = !{i32 2, !"Dwarf Version", i32 4}
 
;; Debug info schema version.
!7 = !{i32 2, !"Debug Info Version", i32 3}
 
;; Compiler identification
!8 = !{!"clang version 4.0.0"}

The align value in DIGlobalVariable description specifies variable alignment incase it was forced by C11 Alignas(), C++11 alignas() keywords or compilerattribute _attribute((aligned ())). In other case (when this field is missing)alignment is considered default. This is used when producing DWARF outputfor DW_AT_alignment value.

C/C++ function information

Given a function declared as follows:

int main(int argc, char *argv[]) {
  return 0;
}

a C/C++ front-end would generate the following descriptors:

;;
;; Define the anchor for subprograms.
;;
!4 = !DISubprogram(name: "main", scope: !1, file: !1, line: 1, type: !5,
                   isLocal: false, isDefinition: true, scopeLine: 1,
                   flags: DIFlagPrototyped, isOptimized: false,
                   variables: !2)
 
;;
;; Define the subprogram itself.
;;
define i32 @main(i32 %argc, i8** %argv) !dbg !4 {
...
}

C++ specific debug information

C++ special member functions information

DWARF v5 introduces attributes defined to enhance debugging information of C++ programs. LLVM can generate (or omit) these appropriate DWARF attributes. In C++ a special member function Ctors, Dtors, Copy/Move Ctors, assignment operators can be declared with C++11 keyword deleted. This is represented in LLVM using spFlags value DISPFlagDeleted.

Given a class declaration with copy constructor declared as deleted:

class foo {
 public:
   foo(const foo&) = deleted;
};

A C++ frontend would generate following:

!17 = !DISubprogram(name: "foo", scope: !11, file: !1, line: 5, type: !18, scopeLine: 5, flags: DIFlagPublic | DIFlagPrototyped, spFlags: DISPFlagDeleted)

and this will produce an additional DWARF attribute as:

DW_TAG_subprogram [7] *
  DW_AT_name [DW_FORM_strx1]    (indexed (00000006) string = "foo")
  DW_AT_decl_line [DW_FORM_data1]       (5)
  ...
  DW_AT_deleted [DW_FORM_flag_present]  (true)

Fortran specific debug information

Fortran function information

There are a few DWARF attributes defined to support client debugging of Fortran programs. LLVM can generate (or omit) the appropriate DWARF attributes for the prefix-specs of ELEMENTAL, PURE, IMPURE, RECURSIVE, and NON_RECURSIVE. This is done by using the spFlags values: DISPFlagElemental, DISPFlagPure, and DISPFlagRecursive.

elemental function elem_func(a)

a Fortran front-end would generate the following descriptors:

!11 = distinct !DISubprogram(name: "subroutine2", scope: !1, file: !1,
        line: 5, type: !8, scopeLine: 6,
        spFlags: DISPFlagDefinition | DISPFlagElemental, unit: !0,
        retainedNodes: !2)

and this will materialize an additional DWARF attribute as:

DW_TAG_subprogram [3]
   DW_AT_low_pc [DW_FORM_addr]     (0x0000000000000010 ".text")
   DW_AT_high_pc [DW_FORM_data4]   (0x00000001)
   ...
   DW_AT_elemental [DW_FORM_flag_present]  (true)

Debugging information format

Debugging Information Extension for Objective C Properties

Introduction

Objective C provides a simpler way to declare and define accessor methods usingdeclared properties. The language provides features to declare a property andto let compiler synthesize accessor methods.

The debugger lets developer inspect Objective C interfaces and their instancevariables and class variables. However, the debugger does not know anythingabout the properties defined in Objective C interfaces. The debugger consumesinformation generated by compiler in DWARF format. The format does not supportencoding of Objective C properties. This proposal describes DWARF extensions toencode Objective C properties, which the debugger can use to let developersinspect Objective C properties.

Proposal

Objective C properties exist separately from class members. A property can bedefined only by “setter” and “getter” selectors, and be calculated anew on eachaccess. Or a property can just be a direct access to some declared ivar.Finally it can have an ivar “automatically synthesized” for it by the compiler,in which case the property can be referred to in user code directly using thestandard C dereference syntax as well as through the property “dot” syntax, butthere is no entry in the @interface declaration corresponding to this ivar.

To facilitate debugging, these properties we will add a new DWARF TAG into theDW_TAG_structure_type definition for the class to hold the description of agiven property, and a set of DWARF attributes that provide said description.The property tag will also contain the name and declared type of the property.

If there is a related ivar, there will also be a DWARF property attribute placedin the DW_TAG_member DIE for that ivar referring back to the property TAGfor that property. And in the case where the compiler synthesizes the ivardirectly, the compiler is expected to generate a DW_TAG_member for thativar (with the DW_AT_artificial set to 1), whose name will be the name usedto access this ivar directly in code, and with the property attribute pointingback to the property it is backing.

The following examples will serve as illustration for our discussion:

@interface I1 {  int n2;}
@property int p1;@property int p2;@end
@implementation I1@synthesize p1;@synthesize p2 = n2;@end

This produces the following DWARF (this is a “pseudo dwarfdump” output):

0x00000100:  TAG_structure_type [7] *
               AT_APPLE_runtime_class( 0x10 )
               AT_name( "I1" )
               AT_decl_file( "Objc_Property.m" )
               AT_decl_line( 3 )
 
0x00000110    TAG_APPLE_property
                AT_name ( "p1" )
                AT_type ( {0x00000150} ( int ) )
 
0x00000120:   TAG_APPLE_property
                AT_name ( "p2" )
                AT_type ( {0x00000150} ( int ) )
 
0x00000130:   TAG_member [8]
                AT_name( "_p1" )
                AT_APPLE_property ( {0x00000110} "p1" )
                AT_type( {0x00000150} ( int ) )
                AT_artificial ( 0x1 )
 
0x00000140:    TAG_member [8]
                 AT_name( "n2" )
                 AT_APPLE_property ( {0x00000120} "p2" )
                 AT_type( {0x00000150} ( int ) )
 
0x00000150:  AT_type( ( int ) )

Note, the current convention is that the name of the ivar for anauto-synthesized property is the name of the property from which it deriveswith an underscore prepended, as is shown in the example. But we actuallydon’t need to know this convention, since we are given the name of the ivardirectly.

Also, it is common practice in ObjC to have different property declarations inthe @interface and @implementation - e.g. to provide a read-only property inthe interface, and a read-write interface in the implementation. In that case,the compiler should emit whichever property declaration will be in force in thecurrent translation unit.

Developers can decorate a property with attributes which are encoded usingDW_AT_APPLE_property_attribute.

@property (readonly, nonatomic) int pr;

TAG_APPLE_property [8]
  AT_name( "pr" )
  AT_type ( {0x00000147} (int) )
  AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)

The setter and getter method names are attached to the property usingDW_AT_APPLE_property_setter and DW_AT_APPLE_property_getter attributes.

@interface I1@property (setter=myOwnP3Setter:) int p3;-(void)myOwnP3Setter:(int)a;@end
@implementation I1@synthesize p3;-(void)myOwnP3Setter:(int)a{ }@end

The DWARF for this would be:

0x000003bd: TAG_structure_type [7] *
              AT_APPLE_runtime_class( 0x10 )
              AT_name( "I1" )
              AT_decl_file( "Objc_Property.m" )
              AT_decl_line( 3 )
 
0x000003cd      TAG_APPLE_property
                  AT_name ( "p3" )
                  AT_APPLE_property_setter ( "myOwnP3Setter:" )
                  AT_type( {0x00000147} ( int ) )
 
0x000003f3:     TAG_member [8]
                  AT_name( "_p3" )
                  AT_type ( {0x00000147} ( int ) )
                  AT_APPLE_property ( {0x000003cd} )
                  AT_artificial ( 0x1 )

New DWARF Tags

New DWARF Attributes

New DWARF Constants

Name Accelerator Tables

Introduction

The “.debug_pubnames” and “.debug_pubtypes” formats are not what adebugger needs. The “pub” in the section name indicates that the entriesin the table are publicly visible names only. This means no static or hiddenfunctions show up in the “.debug_pubnames”. No static variables or privateclass variables are in the “.debug_pubtypes”. Many compilers add differentthings to these tables, so we can’t rely upon the contents between gcc, icc, orclang.

The typical query given by users tends not to match up with the contents ofthese tables. For example, the DWARF spec states that “In the case of the nameof a function member or static data member of a C++ structure, class or union,the name presented in the “.debug_pubnames” section is not the simple namegiven by the DW_AT_name attribute of the referenced debugging informationentry, but rather the fully qualified name of the data or function member.”So the only names in these tables for complex C++ entries is a fullyqualified name. Debugger users tend not to enter their search strings as“a::b::c(int,const Foo&) const”, but rather as “c”, “b::c” , or“a::b::c”. So the name entered in the name table must be demangled inorder to chop it up appropriately and additional names must be manually enteredinto the table to make it effective as a name lookup table for debuggers touse.

All debuggers currently ignore the “.debug_pubnames” table as a result ofits inconsistent and useless public-only name content making it a waste ofspace in the object file. These tables, when they are written to disk, are notsorted in any way, leaving every debugger to do its own parsing and sorting.These tables also include an inlined copy of the string values in the tableitself making the tables much larger than they need to be on disk, especiallyfor large C++ programs.

Can’t we just fix the sections by adding all of the names we need to thistable? No, because that is not what the tables are defined to contain and wewon’t know the difference between the old bad tables and the new good tables.At best we could make our own renamed sections that contain all of the data weneed.

These tables are also insufficient for what a debugger like LLDB needs. LLDBuses clang for its expression parsing where LLDB acts as a PCH. LLDB is thenoften asked to look for type “foo” or namespace “bar”, or list items innamespace “baz”. Namespaces are not included in the pubnames or pubtypestables. Since clang asks a lot of questions when it is parsing an expression,we need to be very fast when looking up names, as it happens a lot. Having newaccelerator tables that are optimized for very quick lookups will benefit thistype of debugging experience greatly.

We would like to generate name lookup tables that can be mapped into memoryfrom disk, and used as is, with little or no up-front parsing. We would alsobe able to control the exact content of these different tables so they containexactly what we need. The Name Accelerator Tables were designed to fix theseissues. In order to solve these issues we need to:

• Have a format that can be mapped into memory from disk and used as is
• Lookups should be very fast
• Extensible table format so these tables can be made by many producers
• Contain all of the names needed for typical lookups out of the box
• Strict rules for the contents of tables

Table size is important and the accelerator table format should allow the reuseof strings from common string tables so the strings for the names are notduplicated. We also want to make sure the table is ready to be used as-is bysimply mapping the table into memory with minimal header parsing.

The name lookups need to be fast and optimized for the kinds of lookups thatdebuggers tend to do. Optimally we would like to touch as few parts of themapped table as possible when doing a name lookup and be able to quickly findthe name entry we are looking for, or discover there are no matches. In thecase of debuggers we optimized for lookups that fail most of the time.

Each table that is defined should have strict rules on exactly what is in theaccelerator tables and documented so clients can rely on the content.

Hash Tables

Standard Hash Tables

Typical hash tables have a header, buckets, and each bucket points to thebucket contents:

.------------.
|  HEADER    |
|------------|
|  BUCKETS   |
|------------|
|  DATA      |
`------------'

The BUCKETS are an array of offsets to DATA for each hash:

.------------.
| 0x00001000 | BUCKETS[0]
| 0x00002000 | BUCKETS[1]
| 0x00002200 | BUCKETS[2]
| 0x000034f0 | BUCKETS[3]
|            | ...
| 0xXXXXXXXX | BUCKETS[n_buckets]
'------------'

So for bucket[3] in the example above, we have an offset into the table0x000034f0 which points to a chain of entries for the bucket. Each bucket mustcontain a next pointer, full 32 bit hash value, the string itself, and the datafor the current string value.

            .------------.
0x000034f0: | 0x00003500 | next pointer
            | 0x12345678 | 32 bit hash
            | "erase"    | string value
            | data[n]    | HashData for this bucket
            |------------|
0x00003500: | 0x00003550 | next pointer
            | 0x29273623 | 32 bit hash
            | "dump"     | string value
            | data[n]    | HashData for this bucket
            |------------|
0x00003550: | 0x00000000 | next pointer
            | 0x82638293 | 32 bit hash
            | "main"     | string value
            | data[n]    | HashData for this bucket
            `------------'

The problem with this layout for debuggers is that we need to optimize for thenegative lookup case where the symbol we’re searching for is not present. Soif we were to lookup “printf” in the table above, we would make a 32-bithash for “printf”, it might match bucket[3]. We would need to go tothe offset 0x000034f0 and start looking to see if our 32 bit hash matches. Todo so, we need to read the next pointer, then read the hash, compare it, andskip to the next bucket. Each time we are skipping many bytes in memory andtouching new pages just to do the compare on the full 32 bit hash. All ofthese accesses then tell us that we didn’t have a match.

Name Hash Tables

To solve the issues mentioned above we have structured the hash tables a bitdifferently: a header, buckets, an array of all unique 32 bit hash values,followed by an array of hash value data offsets, one for each hash value, thenthe data for all hash values:

.-------------.
|  HEADER     |
|-------------|
|  BUCKETS    |
|-------------|
|  HASHES     |
|-------------|
|  OFFSETS    |
|-------------|
|  DATA       |
`-------------'

The BUCKETS in the name tables are an index into the HASHES array. Bymaking all of the full 32 bit hash values contiguous in memory, we allowourselves to efficiently check for a match while touching as little memory aspossible. Most often checking the 32 bit hash values is as far as the lookupgoes. If it does match, it usually is a match with no collisions. So for atable with “n_buckets” buckets, and “n_hashes” unique 32 bit hashvalues, we can clarify the contents of the BUCKETS, HASHES andOFFSETS as:

.-------------------------.
|  HEADER.magic           | uint32_t
|  HEADER.version         | uint16_t
|  HEADER.hash_function   | uint16_t
|  HEADER.bucket_count    | uint32_t
|  HEADER.hashes_count    | uint32_t
|  HEADER.header_data_len | uint32_t
|  HEADER_DATA            | HeaderData
|-------------------------|
|  BUCKETS                | uint32_t[n_buckets] // 32 bit hash indexes
|-------------------------|
|  HASHES                 | uint32_t[n_hashes] // 32 bit hash values
|-------------------------|
|  OFFSETS                | uint32_t[n_hashes] // 32 bit offsets to hash value data
|-------------------------|
|  ALL HASH DATA          |
`-------------------------'

So taking the exact same data from the standard hash example above we end upwith:

            .------------.
            | HEADER     |
            |------------|
            |          0 | BUCKETS[0]
            |          2 | BUCKETS[1]
            |          5 | BUCKETS[2]
            |          6 | BUCKETS[3]
            |            | ...
            |        ... | BUCKETS[n_buckets]
            |------------|
            | 0x........ | HASHES[0]
            | 0x........ | HASHES[1]
            | 0x........ | HASHES[2]
            | 0x........ | HASHES[3]
            | 0x........ | HASHES[4]
            | 0x........ | HASHES[5]
            | 0x12345678 | HASHES[6]    hash for BUCKETS[3]
            | 0x29273623 | HASHES[7]    hash for BUCKETS[3]
            | 0x82638293 | HASHES[8]    hash for BUCKETS[3]
            | 0x........ | HASHES[9]
            | 0x........ | HASHES[10]
            | 0x........ | HASHES[11]
            | 0x........ | HASHES[12]
            | 0x........ | HASHES[13]
            | 0x........ | HASHES[n_hashes]
            |------------|
            | 0x........ | OFFSETS[0]
            | 0x........ | OFFSETS[1]
            | 0x........ | OFFSETS[2]
            | 0x........ | OFFSETS[3]
            | 0x........ | OFFSETS[4]
            | 0x........ | OFFSETS[5]
            | 0x000034f0 | OFFSETS[6]   offset for BUCKETS[3]
            | 0x00003500 | OFFSETS[7]   offset for BUCKETS[3]
            | 0x00003550 | OFFSETS[8]   offset for BUCKETS[3]
            | 0x........ | OFFSETS[9]
            | 0x........ | OFFSETS[10]
            | 0x........ | OFFSETS[11]
            | 0x........ | OFFSETS[12]
            | 0x........ | OFFSETS[13]
            | 0x........ | OFFSETS[n_hashes]
            |------------|
            |            |
            |            |
            |            |
            |            |
            |            |
            |------------|
0x000034f0: | 0x00001203 | .debug_str ("erase")
            | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
            | 0x........ | HashData[0]
            | 0x........ | HashData[1]
            | 0x........ | HashData[2]
            | 0x........ | HashData[3]
            | 0x00000000 | String offset into .debug_str (terminate data for hash)
            |------------|
0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
            | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
            | 0x........ | HashData[0]
            | 0x........ | HashData[1]
            | 0x00001203 | String offset into .debug_str ("dump")
            | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
            | 0x........ | HashData[0]
            | 0x........ | HashData[1]
            | 0x........ | HashData[2]
            | 0x00000000 | String offset into .debug_str (terminate data for hash)
            |------------|
0x00003550: | 0x00001203 | String offset into .debug_str ("main")
            | 0x00000009 | A 32 bit array count - number of HashData with name "main"
            | 0x........ | HashData[0]
            | 0x........ | HashData[1]
            | 0x........ | HashData[2]
            | 0x........ | HashData[3]
            | 0x........ | HashData[4]
            | 0x........ | HashData[5]
            | 0x........ | HashData[6]
            | 0x........ | HashData[7]
            | 0x........ | HashData[8]
            | 0x00000000 | String offset into .debug_str (terminate data for hash)
            `------------'

So we still have all of the same data, we just organize it more efficiently fordebugger lookup. If we repeat the same “printf” lookup from above, wewould hash “printf” and find it matches BUCKETS[3] by taking the 32 bithash value and modulo it by n_buckets. BUCKETS[3] contains “6” whichis the index into the HASHES table. We would then compare any consecutive32 bit hashes values in the HASHES array as long as the hashes would be inBUCKETS[3]. We do this by verifying that each subsequent hash value modulon_buckets is still 3. In the case of a failed lookup we would access thememory for BUCKETS[3], and then compare a few consecutive 32 bit hashesbefore we know that we have no match. We don’t end up marching throughmultiple words of memory and we really keep the number of processor data cachelines being accessed as small as possible.

The string hash that is used for these lookup tables is the Daniel J.Bernstein hash which is also used in the ELF GNU_HASH sections. It is avery good hash for all kinds of names in programs with very few hashcollisions.

Empty buckets are designated by using an invalid hash index of UINT32_MAX.

Details

These name hash tables are designed to be generic where specializations of thetable get to define additional data that goes into the header (“HeaderData”),how the string value is stored (“KeyType”) and the content of the data for eachhash value.

Header Layout

The header has a fixed part, and the specialized part. The exact format of theheader is:

struct Header
{
  uint32_t   magic;           // 'HASH' magic value to allow endian detection
  uint16_t   version;         // Version number
  uint16_t   hash_function;   // The hash function enumeration that was used
  uint32_t   bucket_count;    // The number of buckets in this hash table
  uint32_t   hashes_count;    // The total number of unique hash values and hash data offsets in this table
  uint32_t   header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
                              // Specifically the length of the following HeaderData field - this does not
                              // include the size of the preceding fields
  HeaderData header_data;     // Implementation specific header data
};

The header starts with a 32 bit “magic” value which must be 'HASH'encoded as an ASCII integer. This allows the detection of the start of thehash table and also allows the table’s byte order to be determined so the tablecan be correctly extracted. The “magic” value is followed by a 16 bitversion number which allows the table to be revised and modified in thefuture. The current version number is 1. hash_function is a uint16_tenumeration that specifies which hash function was used to produce this table.The current values for the hash function enumerations include:

enum HashFunctionType
{
  eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
};

bucket_count is a 32 bit unsigned integer that represents how many bucketsare in the BUCKETS array. hashes_count is the number of unique 32 bithash values that are in the HASHES array, and is the same number of offsetsare contained in the OFFSETS array. header_data_len specifies the sizein bytes of the HeaderData that is filled in by specialized versions ofthis table.

Fixed Lookup

The header is followed by the buckets, hashes, offsets, and hash value data.

struct FixedTable
{
  uint32_t buckets[Header.bucket_count];  // An array of hash indexes into the "hashes[]" array below
  uint32_t hashes [Header.hashes_count];  // Every unique 32 bit hash for the entire table is in this table
  uint32_t offsets[Header.hashes_count];  // An offset that corresponds to each item in the "hashes[]" array above
};

buckets is an array of 32 bit indexes into the hashes array. Thehashes array contains all of the 32 bit hash values for all names in thehash table. Each hash in the hashes table has an offset in the offsetsarray that points to the data for the hash value.

This table setup makes it very easy to repurpose these tables to containdifferent data, while keeping the lookup mechanism the same for all tables.This layout also makes it possible to save the table to disk and map it inlater and do very efficient name lookups with little or no parsing.

DWARF lookup tables can be implemented in a variety of ways and can store a lotof information for each name. We want to make the DWARF tables extensible andable to store the data efficiently so we have used some of the DWARF featuresthat enable efficient data storage to define exactly what kind of data we storefor each name.

The HeaderData contains a definition of the contents of each HashData chunk.We might want to store an offset to all of the debug information entries (DIEs)for each name. To keep things extensible, we create a list of items, orAtoms, that are contained in the data for each name. First comes the type ofthe data in each atom:

enum AtomType
{
  eAtomTypeNULL       = 0u,
  eAtomTypeDIEOffset  = 1u,   // DIE offset, check form for encoding
  eAtomTypeCUOffset   = 2u,   // DIE offset of the compiler unit header that contains the item in question
  eAtomTypeTag        = 3u,   // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
  eAtomTypeNameFlags  = 4u,   // Flags from enum NameFlags
  eAtomTypeTypeFlags  = 5u,   // Flags from enum TypeFlags
};

The enumeration values and their meanings are:

eAtomTypeNULL       - a termination atom that specifies the end of the atom list
eAtomTypeDIEOffset  - an offset into the .debug_info section for the DWARF DIE for this name
eAtomTypeCUOffset   - an offset into the .debug_info section for the CU that contains the DIE
eAtomTypeDIETag     - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
eAtomTypeNameFlags  - Flags for functions and global variables (isFunction, isInlined, isExternal...)
eAtomTypeTypeFlags  - Flags for types (isCXXClass, isObjCClass, ...)

Then we allow each atom type to define the atom type and how the data for eachatom type data is encoded:

struct Atom
{
  uint16_t type;  // AtomType enum value
  uint16_t form;  // DWARF DW_FORM_XXX defines
};

The form type above is from the DWARF specification and defines the exactencoding of the data for the Atom type. See the DWARF specification for theDWFORM definitions.

struct HeaderData
{
  uint32_t die_offset_base;
  uint32_t atom_count;
  Atoms    atoms[atom_count0];
};

HeaderData defines the base DIE offset that should be added to any atomsthat are encoded using the DW_FORM_ref1, DW_FORM_ref2,DW_FORM_ref4, DW_FORM_ref8 or DW_FORM_ref_udata. It also defineswhat is contained in each HashData object – Atom.form tells us how largeeach field will be in the HashData and the Atom.type tells us how this datashould be interpreted.

For the current implementations of the “.apple_names” (all functions +globals), the “.apple_types” (names of all types that are defined), andthe “.apple_namespaces” (all namespaces), we currently set the Atomarray to be:

HeaderData.atom_count = 1;
HeaderData.atoms[0].type = eAtomTypeDIEOffset;
HeaderData.atoms[0].form = DW_FORM_data4;

This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that isencoded as a 32 bit value (DW_FORM_data4). This allows a single name to havemultiple matching DIEs in a single file, which could come up with an inlinedfunction for instance. Future tables could include more information about theDIE such as flags indicating if the DIE is a function, method, block,or inlined.

The KeyType for the DWARF table is a 32 bit string table offset into the“.debug_str” table. The “.debug_str” is the string table for the DWARF whichmay already contain copies of all of the strings. This helps make sure, withhelp from the compiler, that we reuse the strings between all of the DWARFsections and keeps the hash table size down. Another benefit to having thecompiler generate all strings as DW_FORM_strp in the debug info, is thatDWARF parsing can be made much faster.

After a lookup is made, we get an offset into the hash data. The hash dataneeds to be able to deal with 32 bit hash collisions, so the chunk of dataat the offset in the hash data consists of a triple:

uint32_t str_offset
uint32_t hash_data_count
HashData[hash_data_count]

If “str_offset” is zero, then the bucket contents are done. 99.9% of thehash data chunks contain a single item (no 32 bit hash collision):

.------------.
| 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
| 0x00000004 | uint32_t HashData count
| 0x........ | uint32_t HashData[0] DIE offset
| 0x........ | uint32_t HashData[1] DIE offset
| 0x........ | uint32_t HashData[2] DIE offset
| 0x........ | uint32_t HashData[3] DIE offset
| 0x00000000 | uint32_t KeyType (end of hash chain)
`------------'

If there are collisions, you will have multiple valid string offsets:

.------------.
| 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
| 0x00000004 | uint32_t HashData count
| 0x........ | uint32_t HashData[0] DIE offset
| 0x........ | uint32_t HashData[1] DIE offset
| 0x........ | uint32_t HashData[2] DIE offset
| 0x........ | uint32_t HashData[3] DIE offset
| 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
| 0x00000002 | uint32_t HashData count
| 0x........ | uint32_t HashData[0] DIE offset
| 0x........ | uint32_t HashData[1] DIE offset
| 0x00000000 | uint32_t KeyType (end of hash chain)
`------------'

Current testing with real world C++ binaries has shown that there is around 132 bit hash collision per 100,000 name entries.

Contents

As we said, we want to strictly define exactly what is included in thedifferent tables. For DWARF, we have 3 tables: “.apple_names”,“.apple_types”, and “.apple_namespaces”.

“.apple_names” sections should contain an entry for each DWARF DIE whoseDW_TAG is a DW_TAG_label, DW_TAG_inlined_subroutine, orDW_TAG_subprogram that has address attributes: DW_AT_low_pc,DW_AT_high_pc, DW_AT_ranges or DW_AT_entry_pc. It also containsDW_TAG_variable DIEs that have a DW_OP_addr in the location (global andstatic variables). All global and static variables should be included,including those scoped within functions and classes. For example using thefollowing code:

static int var = 0;
 
void f ()
{
  static int var = 0;
}

Both of the static var variables would be included in the table. Allfunctions should emit both their full names and their basenames. For C or C++,the full name is the mangled name (if available) which is usually in theDW_AT_MIPS_linkage_name attribute, and the DW_AT_name contains thefunction basename. If global or static variables have a mangled name in aDW_AT_MIPS_linkage_name attribute, this should be emitted along with thesimple name found in the DW_AT_name attribute.

“.apple_types” sections should contain an entry for each DWARF DIE whosetag is one of:

• DW_TAG_array_type
• DW_TAG_class_type
• DW_TAG_enumeration_type
• DW_TAG_pointer_type
• DW_TAG_reference_type
• DW_TAG_string_type
• DW_TAG_structure_type
• DW_TAG_subroutine_type
• DW_TAG_typedef
• DW_TAG_union_type
• DW_TAG_ptr_to_member_type
• DW_TAG_set_type
• DW_TAG_subrange_type
• DW_TAG_base_type
• DW_TAG_const_type
• DW_TAG_file_type
• DW_TAG_namelist
• DW_TAG_packed_type
• DW_TAG_volatile_type
• DW_TAG_restrict_type
• DW_TAG_atomic_type
• DW_TAG_interface_type
• DW_TAG_unspecified_type
• DW_TAG_shared_type

Only entries with a DW_AT_name attribute are included, and the entry mustnot be a forward declaration (DW_AT_declaration attribute with a non-zerovalue). For example, using the following code:

int main ()
{
  int *b = 0;
  return *b;
}

We get a few type DIEs:

0x00000067:     TAG_base_type [5]
                AT_encoding( DW_ATE_signed )
                AT_name( "int" )
                AT_byte_size( 0x04 )
 
0x0000006e:     TAG_pointer_type [6]
                AT_type( {0x00000067} ( int ) )
                AT_byte_size( 0x08 )

The DW_TAG_pointer_type is not included because it does not have a DW_AT_name.

“.apple_namespaces” section should contain all DW_TAG_namespace DIEs.If we run into a namespace that has no name this is an anonymous namespace, andthe name should be output as “(anonymous namespace)” (without the quotes).Why? This matches the output of the abi::cxa_demangle() that is in thestandard C++ library that demangles mangled names.

Language Extensions and File Format Changes

Objective-C Extensions

“.apple_objc” section should contain all DW_TAG_subprogram DIEs for anObjective-C class. The name used in the hash table is the name of theObjective-C class itself. If the Objective-C class has a category, then anentry is made for both the class name without the category, and for the classname with the category. So if we have a DIE at offset 0x1234 with a name ofmethod “-[NSString(my_additions) stringWithSpecialString:]”, we would addan entry for “NSString” that points to DIE 0x1234, and an entry for“NSString(my_additions)” that points to 0x1234. This allows us to quicklytrack down all Objective-C methods for an Objective-C class when doingexpressions. It is needed because of the dynamic nature of Objective-C whereanyone can add methods to a class. The DWARF for Objective-C methods is alsoemitted differently from C++ classes where the methods are not usuallycontained in the class definition, they are scattered about across one or morecompile units. Categories can also be defined in different shared libraries.So we need to be able to quickly find all of the methods and class functionsgiven the Objective-C class name, or quickly find all methods and classfunctions for a class + category name. This table does not contain anyselector names, it just maps Objective-C class names (or class names +category) to all of the methods and class functions. The selectors are addedas function basenames in the “.debug_names” section.

In the “.apple_names” section for Objective-C functions, the full name isthe entire function name with the brackets (“-[NSStringstringWithCString:]”) and the basename is the selector only(“stringWithCString:”).

Mach-O Changes

The sections names for the apple hash tables are for non-mach-o files. Formach-o files, the sections should be contained in the __DWARF segment withnames as follows:

• “.apple_names” -> “__apple_names”
• “.apple_types” -> “__apple_types”
• “.apple_namespaces” -> “__apple_namespac” (16 character limit)
• “.apple_objc” -> “__apple_objc”

CodeView Debug Info Format

LLVM supports emitting CodeView, the Microsoft debug info format, and thissection describes the design and implementation of that support.

Format Background

CodeView as a format is clearly oriented around C++ debugging, and in C++, themajority of debug information tends to be type information. Therefore, theoverriding design constraint of CodeView is the separation of type informationfrom other “symbol” information so that type information can be efficientlymerged across translation units. Both type information and symbol information isgenerally stored as a sequence of records, where each record begins with a16-bit record size and a 16-bit record kind.

Type information is usually stored in the .debug$T section of the objectfile. All other debug info, such as line info, string table, symbol info, andinlinee info, is stored in one or more .debug$S sections. There may only beone .debug$T section per object file, since all other debug info refers toit. If a PDB (enabled by the /Zi MSVC option) was used during compilation,the .debug$T section will contain only an LF_TYPESERVER2 record pointingto the PDB. When using PDBs, symbol information appears to remain in the objectfile .debug$S sections.

Type records are referred to by their index, which is the number of records inthe stream before a given record plus 0x1000. Many common basic types, suchas the basic integral types and unqualified pointers to them, are representedusing type indices less than 0x1000. Such basic types are built in toCodeView consumers and do not require type records.

Each type record may only contain type indices that are less than its own typeindex. This ensures that the graph of type stream references is acyclic. Whilethe source-level type graph may contain cycles through pointer types (consider alinked list struct), these cycles are removed from the type stream by alwaysreferring to the forward declaration record of user-defined record types. Only“symbol” records in the .debug$S streams may refer to complete,non-forward-declaration type records.

Working with CodeView

These are instructions for some common tasks for developers working to improveLLVM’s CodeView support. Most of them revolve around using the CodeView dumperembedded in llvm-readobj.

• Testing MSVC’s output:

$ cl -c -Z7 foo.cpp # Use /Z7 to keep types in the object file
$ llvm-readobj --codeview foo.obj

• Getting LLVM IR debug info out of Clang:

$ clang -g -gcodeview --target=x86_64-windows-msvc foo.cpp -S -emit-llvm

Use this to generate LLVM IR for LLVM test cases.

• Generate and dump CodeView from LLVM IR metadata:

$ llc foo.ll -filetype=obj -o foo.obj
$ llvm-readobj --codeview foo.obj > foo.txt

Use this pattern in lit test cases and FileCheck the output of llvm-readobj

Improving LLVM’s CodeView support is a process of finding interesting typerecords, constructing a C++ test case that makes MSVC emit those records,dumping the records, understanding them, and then generating equivalent recordsin LLVM’s backend.

Testing Debug Info Preservation in Optimizations

The following paragraphs are an introduction to the debugify utilityand examples of how to use it in regression tests to check debug infopreservation after optimizations.

The debugify utility

The debugify synthetic debug info testing utility consists of twomain parts. The debugify pass and the check-debugify one. They aremeant to be used with opt for development purposes.

The first applies synthetic debug information to every instruction of the module,while the latter checks that this DI is still available after an optimizationhas occurred, reporting any errors/warnings while doing so.

The instructions are assigned sequentially increasing line locations,and are immediately used by debug value intrinsics when possible.

For example, here is a module before:

define void @f(i32* %x) {
entry:
  %x.addr = alloca i32*, align 8
  store i32* %x, i32** %x.addr, align 8
  %0 = load i32*, i32** %x.addr, align 8
  store i32 10, i32* %0, align 4
  ret void
}