[RFC][DwarfDebug][AsmPrinter] Support emitting function-local declaration for a lexical block

[krisb]

Summary

This is another attempt to make function-local declarations
(like static variables, structs/classes, and other) be correctly emitted
within a lexical (bracketed) block.

Some ideas were taken from previous attempts [0] and [1].

Limitations:

Since we skip lexical blocks if we couldn't find non-scope children for them,
we need a way to force the emission of blocks that have static locals or local
type declarations but don't have local variables. As a workaround, I
collect such scopes by collectLocalScopesWithDeclsFromCU() and
store them in a set. This set is used later to check whether we need to emit
a lexical block. It, however, lacks information about used records
and typedefs because there is no access to them from a CU.
In this case, we will put the entity to a first available parent scope
(with subprogram as a bound).

Fixes https://bugs.llvm.org/show_bug.cgi?id=19238.

clang's part is at D113743.

[0] https://reviews.llvm.org/D11180
[1] https://reviews.llvm.org/D15976
[2] https://bugs.llvm.org/show_bug.cgi?id=27579

[ellis]

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp

1625

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.h

64

Does this need to map from DIScope (rather than DILocalScope) because it could take a DICommonBlock? If so maybe we should rename this map to remove the "Local".

Or could we possibly make this a map from DISubprogram?

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1074

Can a DICommonBlock have a scope that is also a DICommonBlock?

1187–1192

This is really great. Previously in https://reviews.llvm.org/D113736 I recorded concrete functions, but that is limited because we aren't sure if there will be an abstract origin. Here we record exactly which subprograms have abstract origins. Also, thanks for including my tests from that diff!

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.h

848

NIT: Since this is recursive we could record multiple SPs from a single Loc.

852

859

[krisb]

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.h

64

There should be subclasses of DILocalScope only. Will fix the key. Thank you for pointing to this!

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1074

I'm not a Fortran expert, but as far as I can understand, this shouldn't be possible according to the common block's semantic (https://j3-fortran.org/doc/year/18/18-007r1.pdf, p. 8.10.2 COMMON statement).
I didn't find any examples of usage of createCommonBlock(), but I guess, that DICommonBlock should only have a DISubprogram as a scope. If so, DICommonBlock can be made a DILocalScope subclass, but I'm not sure this is semantically correct from Fortran language perspective.

1187–1192

It's nice to know whether the given SP will have an abstract origin or not, but unfortunately, this way of obtaining such information isn't fully reliable. The problem is that having some DILocations with inlinedAt field doesn't guarantee that the inlined/abstract LexicalScope will be created (especially in optimized code). So, I'm not sure this is the right way to go. Maybe I should avoid doing this and collect all the scope DIEs instead to have more chances to emit accurate dwarf.

Also, thanks for including my tests from that diff!

It's me who should thank you for the tests :) By the way, do you plan to commit them separately?

[ellis]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1187–1192

By the way, do you plan to commit them separately?

Let's just commit them here assuming they pass.

1187–1192

It's nice to know whether the given SP will have an abstract origin or not, but unfortunately, this way of obtaining such information isn't fully reliable. The problem is that having some DILocations with inlinedAt field doesn't guarantee that the inlined/abstract LexicalScope will be created (especially in optimized code). So, I'm not sure this is the right way to go. Maybe I should avoid doing this and collect all the scope DIEs instead to have more chances to emit accurate dwarf.

If I understand correctly, that means HasInlinedInstances() returns true for inlined SPs even if their DIEs will never be emitted (because they were optimized away for example). In that case we won't construct the scope anyway, so I think that's ok.

[krisb]

Applied ellis's comments.
Stored all abstract scopes into a single map.
Stopped trying to guess which functions have inlined instances.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1187–1192

Well, yet I decided to try a different approach. Instead of guessing which SPs have inlined instances and which have not in order to decide do we need to store the given DIE to the map, we simply can store all the DIEs (but those that are known to belong inlined instances). So later we can choose which tree to use: an abstract (if it exists) or a concrete one. It looks a bit better to me and we can use a single map for all abstract scopes and avoid duplcation. What do you think?

[ellis]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1187–1192

I think any solution needs to be able to determine if an SP will have an abstract origin before processing any functions, i.e., this needs to be computed in beginModule(). Suppose a function f() has a concrete instance and is also inlined at the end of the module. It should have an abstract origin and all references should be made to the abstract origin DIE instead of the concrete DIE. That means getOrCreateSubprogramContextDIE() needs to know if it should get/create an abstract or concrete DIE even before the abstract origin is created.

I guess the import-inlined-declaration.ll test was not complete because it doesn't test if ns::foo() references the abstract origin if the concrete DIE exists. Could you update this test to handle that? Here is what I'm thinking, but I haven't actually tested this on your code yet.
https://godbolt.org/z/j1xdv3KWG

This would also be a bug when an imported entity tries to reference an SP before its abstract origin was created, but I don't have a concrete example of that.

Let's follow up on https://groups.google.com/g/llvm-dev/c/mlNUyj-Q5To if you have any more questions. Thanks!

[krisb]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1187–1192

Suppose a function f() has a concrete instance and is also inlined at the end of the module. It should have an abstract origin and all references should be made to the abstract origin DIE instead of the concrete DIE.

This is why I finally decided to move the emission of imported entities (together with types and global variables) to DwarfDebug::endModule() instead of trying to handle these cases earlier. By endModule() we have all the local context emitted, so we don't need to force emitting abstract DIEs ahead of normal flow.

Could you update this test to handle that?

Sure, I will do. But I believe this case should be handled properly cause we never try to emit imported declarations before all the subprograms get processed.

[ellis]

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp

1641

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1187–1192

This is why I finally decided to move the emission of imported entities (together with types and global variables) to DwarfDebug::endModule() instead of trying to handle these cases earlier. By endModule() we have all the local context emitted, so we don't need to force emitting abstract DIEs ahead of normal flow.

Ah ok makes sense!

llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp

630–633

Where are we creating DIEs without giving them a parent. In a previous attempt @dblaikie mentioned we should avoid even temporarily creating DIEs without a parent because of the getUnit() function.

[krisb]

Rebase on D114350.

llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp

630–633

Right, this is a problem. I've opened another review request with refactoring for createScopeChildrenDIE() and createScopeChildrenDIE() that makes parent set right after creation of a DIE for local entity. Please, take a look if you have a chance.

[krisb]

Minor fixups.

[krisb]

Extracted https://reviews.llvm.org/D114705 (with more details on the change), rebased.

[dblaikie]

Generally on board with this direction - perhaps it doesn't make anything /worse/ (if the cases where types are not reachable through the CU's retainedTypes list get the same behavior they've always got - can you confirm that's the case? If that's the case, perhaps we can go forward with this, and separately/in addition, if you have time/interest/motivnation, you could work on adding local entities to the DISubprogram's retainedNodes list and search that to ensure those entities are put in the right scopes)

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1091–1094

This is fairly problematic to the strategy - generally we don't put types in the "retained types" list because the point is to lazily reference types/other content as much as possible (so if the middle end optimizes away a function and that's the only place where the type is referenced, then the type gets optimized away too).

Perhaps this points to some other direction (perhaps in addition) being needed? Perhaps local entities could go on the DISubprogram's retainedNodes attribute?

(this would address a test failure/FIXME you mentioned in one of the test cases - that a local type was missing from the intended scope)

llvm/lib/DebugInfo/DWARF/DWARFDie.cpp

675–676 ↗

(On Diff #390332)

This change could probably be committed separately with a llvm-dwarfdump+assembly test case demonstrating this fix. Then the LLVM code generation changes that depend on this would come in the second patch. Not a huge deal, though.

[krisb]

@dblaikie thank you for looking at this!

if the cases where types are not reachable through the CU's retainedTypes list get the same behavior they've always got - can you confirm that's the case?

For local types that are not reachable through the CU's retainedTypes list there are two cases:

• if type's direct parent scope was emitted (because it contains some other declarations), this type will be placed to this scope (as expected),
• if type's direct parent scope was omitted (since there are no other decls in it), the type will go to a first available (emitted) parent scope (in the worst case - to the parent subprogram).

So, with this patch, local types will have the closest/most suitable parent from available. Before this patch, local types were placed directly to subprogram no matter which scope they really have (and actually, frontend never emits types with lexical block scope), so the old behavior is only kept as a fallback for the worst case.
For all non function-local types, this patch changes nothing.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1091–1094

Perhaps this points to some other direction (perhaps in addition) being needed? Perhaps local entities could go on the DISubprogram's retainedNodes attribute?

Unless I'm missing something retainedNodes semantic is to keep entities even if they can be optimized out (as of retainedTypes is for types, but only if we have corresponding compile option passed). So, I'm not sure we should use those fields to keep some local entities that are okay to be completely removed, if they unused/ get optimized away.
I was thinking about a kind of a new field for DISubprogram (and maybe for DILexicalBlock), something like localDecls which would keep references for static locals, local imported entities, and records/typedefs that belongs to this local scope (and stop referencing local entities from globals and imports fields of a compile unit), but have not come to a comprehensive solution yet.

llvm/lib/DebugInfo/DWARF/DWARFDie.cpp

675–676 ↗

(On Diff #390332)

Before this patch, AsmPrinter would crash on types scoped within a lexical block (cause DwarfCompileUnit would not able to create a context for such types), so I'm not sure how to reproduce the case. Do you have some ideas in mind?

[dblaikie]

Sounds like this generally moves things forward, thanks for your patience!

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1091–1094

Yep, complicated tradeoffs. By having the list we'd avoid miscoping these entities (because they're otherwise intractible to find - they could appear in any part of the DWARF) - but any attempt to keep them in a list risks them being preserved too much too.

keeping the list on the DILexicalBlock would be pretty good - that way if the lexical block became unreferenced (all instructions in it were optimized away) the lexical block could be removed and the list along with it - hmm, but there are still cases where the type could be alive despite the function being totally optimized away (eg: a trivial function that returns a stateless local type, like a lambda - so the local type escapes the scope, and the function gets inlined and optimized away - no scope, but the type is still out there in the rest of the DWARF) - in which case we'd still scope the type incorrectly...

llvm/lib/DebugInfo/DWARF/DWARFDie.cpp

675–676 ↗

(On Diff #390332)

What you could do is using this patch, generate the assembly for a type scoped in a lexical block - but then commit that assembly test case ahead of the rest of the patch. It's still a good/general improvement to llvm-dwarfdump/libDebugInfoDWARF that applies even if LLVM never generates this DWARF (because llvm-dwarfdump is intended as a tool to investigate any DWARF, not just the DWARF that llvm generates).

This is true for most llvm-dwarfdump/libDebugInfoDWARF changes - they can come before the LLVM change that generates them, because they should/can be tested without any LLVM support at all, through assembly test cases .

[krisb]

llvm/lib/DebugInfo/DWARF/DWARFDie.cpp

675–676 ↗

(On Diff #390332)

Ah, got it! Thank you for all the details! I've opened a separate review for this change https://reviews.llvm.org/D114892. Please, take a look if you have a chance.

[krisb]

Extracted https://reviews.llvm.org/D114892 + minor fixups for code style and comments.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1091–1094

Another problem is even if for every scope there would be a list of types declared within it wasn't guaranteed that these types would be placed in the correct scope. Lexical blocks DIEs are emitted based on LexicalScopes which may (and likely) not be created for optimized out scopes. Thus it still would be possible to misplace a type DIE. But as far as I know, this should not be a problem for debuggers (at least for those that I'm aware of).
We need a more tricky solution (or more changes in DWARF emission design) to make everything come together.

because they're otherwise intractible to find - they could appear in any part of the DWARF

I hope they will be at least within a parent subprogram. We may do the things in a more conservative way and in case there is no suitable parent scope DIE emitted for this type (or any other local declaration in the subject) place these entities directly into a subprogram DIE, so we will have 2 possible places for local decls:

• direct scope,
• subprogram scope if the direct scope is missed.

Concerning the case with lambdas, I've checked a few examples and found that clang produces scopeless DIs for such types, and this seems to be one more problem here.

[thakis]

I see this is already reverted (thanks for watching the bots after committing!), but fyi for the relabel: this also broke tests on macOS, e.g. http://45.33.8.238/mac/39897/step_11.txt

[MaskRay]

llvm/test/DebugInfo/Generic/lexical_block_static.ll

1 ↗

(On Diff #391839)

Somewhat strange the patch introduced new tests with both - and _ filename separators...

In most components of llvm, - is more popular though I can see directories where _ is more popular.

[dblaikie]

FWIW this also caused some ThinLTO build failures (linker errors specifically) along the lines of:

(.debug_gnu_pubnames): name lookup table at offset 0x9a5 has a terminator at offset 0x9b3 before the expected end at 0x9f2

I don't have a simple/small reproduction at the moment. But would be good to work together/reach out to me if it's not easily discoverable and figure out this issue before recommitting.

[dblaikie]

In D113741#3190544, @dblaikie wrote:

FWIW this also caused some ThinLTO build failures (linker errors specifically) along the lines of:

(.debug_gnu_pubnames): name lookup table at offset 0x9a5 has a terminator at offset 0x9b3 before the expected end at 0x9f2

I don't have a simple/small reproduction at the moment. But would be good to work together/reach out to me if it's not easily discoverable and figure out this issue before recommitting.

Oh, I misread - that this issue was seen on the recommit & hasn't ben reverted yet.

I'll try to reduce a test case, but might be worth reverting in the interim regardless?

[krisb]

In D113741#3190810, @dblaikie wrote:

In D113741#3190544, @dblaikie wrote:

FWIW this also caused some ThinLTO build failures (linker errors specifically) along the lines of:

(.debug_gnu_pubnames): name lookup table at offset 0x9a5 has a terminator at offset 0x9b3 before the expected end at 0x9f2

I don't have a simple/small reproduction at the moment. But would be good to work together/reach out to me if it's not easily discoverable and figure out this issue before recommitting.

Oh, I misread - that this issue was seen on the recommit & hasn't ben reverted yet.

I'll try to reduce a test case, but might be worth reverting in the interim regardless?

If you think it'd be better to revert until the issue gets resolved then okay, let me know and I'll revert it. I can't say more without a reproducer.
But could you please share a way to reproduce the issue? Not small/simple is also okay.

llvm/test/DebugInfo/Generic/lexical_block_static.ll

1 ↗

(On Diff #391839)

This happened cause I borrowed some tests from ellis's patch, those ones with '-'.
Sorry, I somehow didn't notice this difference. I'll either fix it in an NFC or in the next recommit if it'll be decided it's better to revert this patch until LTO issues get resolved.

[dblaikie]

In D113741#3195668, @krisb wrote:

In D113741#3190810, @dblaikie wrote:

In D113741#3190544, @dblaikie wrote:

FWIW this also caused some ThinLTO build failures (linker errors specifically) along the lines of:

(.debug_gnu_pubnames): name lookup table at offset 0x9a5 has a terminator at offset 0x9b3 before the expected end at 0x9f2

I don't have a simple/small reproduction at the moment. But would be good to work together/reach out to me if it's not easily discoverable and figure out this issue before recommitting.

Oh, I misread - that this issue was seen on the recommit & hasn't ben reverted yet.

I'll try to reduce a test case, but might be worth reverting in the interim regardless?

If you think it'd be better to revert until the issue gets resolved then okay, let me know and I'll revert it. I can't say more without a reproducer.
But could you please share a way to reproduce the issue? Not small/simple is also okay.

Unfortunately due to ThinLTO it's a bit hard for me to extract (& not sure if there are bits I need to anonymize, etc - which is easier when reducing just a single file) and share at the moment :/

[dblaikie]

Some DIEs are coming through with a 0 offset, which prematurely terminates the pubnames/types list. I'm working on figuring out in what situation that happens.

[dblaikie]

Ah, seems to be related to "isDebugDirectivesOnly" - ie: we have debug info but only for middle end diagnostics, not to produce DWARF in the output...

[dblaikie]

hmm, wait, no, not that...

[dblaikie]

Hmm, think I'm honing in on it.

It looks like it's related to a unit that defines a type that's already defined by another unit - usually that first unit could be skipped entirely by https://github.com/llvm/llvm-project/blob/6683099a0d0a17fcde3576733e9c85e3b5f71de5/llvm/lib/CodeGen/AsmPrinter/DwarfFile.cpp#L47 - but since the namespace now gets created first, then the type is skipped...

Hm, no, maybe I'm chasing down a different manifestation caused by 78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2 - perhaps still with this patch (D113741) as a root cause? I'm not sure yet.

[dblaikie]

In D113741#3205175, @dblaikie wrote:

Hmm, think I'm honing in on it.

It looks like it's related to a unit that defines a type that's already defined by another unit - usually that first unit could be skipped entirely by https://github.com/llvm/llvm-project/blob/6683099a0d0a17fcde3576733e9c85e3b5f71de5/llvm/lib/CodeGen/AsmPrinter/DwarfFile.cpp#L47 - but since the namespace now gets created first, then the type is skipped...

Correct link is actually: https://github.com/llvm/llvm-project/blob/6683099a0d0a17fcde3576733e9c85e3b5f71de5/llvm/lib/CodeGen/AsmPrinter/DwarfFile.cpp#L71

Hm, no, maybe I'm chasing down a different manifestation caused by 78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2 - perhaps still with this patch (D113741) as a root cause? I'm not sure yet.

[dblaikie]

Alrighty, got an A/B comparison with/without this patch (but still not reduced to share) - seems we end up creating a second split unit for one of the imported units under ThinLTO, which is incorrect (because Split DWARF only really supports one CU per split unit - so having two in a single dwo file is broken - at least if you're going to package the file with dwp, which we do).

Looking into it further/trying to get a reduced example.

[krisb]

In D113741#3205306, @dblaikie wrote:

Alrighty, got an A/B comparison with/without this patch (but still not reduced to share) - seems we end up creating a second split unit for one of the imported units under ThinLTO, which is incorrect (because Split DWARF only really supports one CU per split unit - so having two in a single dwo file is broken - at least if you're going to package the file with dwp, which we do).

Looking into it further/trying to get a reduced example.

(trying to guess the issue by keywords...)

Is the issue about a function-local import?

The only diff in imported entities handling I see is that we no longer skip emission of local imports under 'includeMinimalInlineScopes()' (which was the case before this patch, but it seems I missed this fact).
Can it be a hidden issue revealed by this mistake?

P.S. I'll fix 'includeMinimalInlineScopes()' for local import.

[dblaikie]

In D113741#3206529, @krisb wrote:

In D113741#3205306, @dblaikie wrote:

Alrighty, got an A/B comparison with/without this patch (but still not reduced to share) - seems we end up creating a second split unit for one of the imported units under ThinLTO, which is incorrect (because Split DWARF only really supports one CU per split unit - so having two in a single dwo file is broken - at least if you're going to package the file with dwp, which we do).

Looking into it further/trying to get a reduced example.

(trying to guess the issue by keywords...)

Is the issue about a function-local import?

The only diff in imported entities handling I see is that we no longer skip emission of local imports under 'includeMinimalInlineScopes()' (which was the case before this patch, but it seems I missed this fact).
Can it be a hidden issue revealed by this mistake?

P.S. I'll fix 'includeMinimalInlineScopes()' for local import.

Yep, I'm certainly seeing some function-local imports in the test case I'm investigating, and that sounds related to be sure. Can't guarantee it's the precise issue, but I'll keep poking around at it & having a fix for the local import under minimal-inline-scopes would be good to have in any case.

[dblaikie]

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp

1055–1075

Adding back these two includeMinimalInlineScopes checks doesn't seem to address the failure I'm investigating, at least.

But they may still represent a bug (either an independent one, or as part of the issue I'm investigating).

[dblaikie]

OK, think I've got it now.

It requires splitDebugInlining=false in the source CU, and `true' in the destination CU. Needs a local using decl in the source CU and some function import from source to destination CU.

This example, compiled at -O1 does the trick:

; ModuleID = 'ab.bc'
source_filename = "llvm-link"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

%"struct.ns::t1" = type { i8 }

; Function Attrs: mustprogress uwtable
define available_externally dso_local void @_Z2f1v() #0 !dbg !4 {
entry:
  %v1 = alloca %"struct.ns::t1", align 1
  call void @llvm.dbg.declare(metadata %"struct.ns::t1"* %v1, metadata !21, metadata !DIExpression()), !dbg !22
  call void @_Z3pinv(), !dbg !23
  ret void, !dbg !24
}

; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
declare void @llvm.dbg.declare(metadata, metadata, metadata) #1

declare dso_local void @_Z3pinv() #2

; Function Attrs: mustprogress norecurse uwtable
define dso_local i32 @main() #3 !dbg !25 {
entry:
  call void @_Z2f1v(), !dbg !29
  ret i32 0, !dbg !30
}

attributes #0 = { mustprogress uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #1 = { nofree nosync nounwind readnone speculatable willreturn }
attributes #2 = { "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #3 = { mustprogress norecurse uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }

!llvm.dbg.cu = !{!0, !10}
!llvm.ident = !{!12, !12}
!llvm.module.flags = !{!13, !14, !15, !16, !17}

!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 14.0.0 ([e-mail]:llvm/llvm-project.git 75b622a7959479ccdb758ebe0973061b7b4fcacd)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, imports: !2, splitDebugInlining: false, nameTableKind: GNU)
!1 = !DIFile(filename: "a.cpp", directory: "/usr/local/google/home/blaikie/dev/scratch")
!2 = !{!3}
!3 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !4, entity: !8, file: !1, line: 3)
!4 = distinct !DISubprogram(name: "f1", linkageName: "_Z2f1v", scope: !1, file: !1, line: 3, type: !5, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
!5 = !DISubroutineType(types: !6)
!6 = !{null}
!7 = !{}
!8 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "t1", scope: !9, file: !1, line: 1, size: 8, flags: DIFlagTypePassByValue, elements: !7, identifier: "_ZTSN2ns2t1E")
!9 = !DINamespace(name: "ns", scope: null)
!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, producer: "clang version 14.0.0 ([e-mail]:llvm/llvm-project.git 75b622a7959479ccdb758ebe0973061b7b4fcacd)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: true, nameTableKind: GNU)
!11 = !DIFile(filename: "b.cpp", directory: "/usr/local/google/home/blaikie/dev/scratch")
!12 = !{!"clang version 14.0.0 ([e-mail]:llvm/llvm-project.git 75b622a7959479ccdb758ebe0973061b7b4fcacd)"}
!13 = !{i32 7, !"Dwarf Version", i32 4}
!14 = !{i32 2, !"Debug Info Version", i32 3}
!15 = !{i32 1, !"wchar_size", i32 4}
!16 = !{i32 7, !"uwtable", i32 1}
!17 = !{i32 7, !"frame-pointer", i32 2}
!18 = distinct !DISubprogram(name: "f2", linkageName: "_Z2f2v", scope: !1, file: !1, line: 4, type: !5, scopeLine: 4, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
!19 = !DILocation(line: 4, column: 13, scope: !18)
!20 = !DILocation(line: 4, column: 19, scope: !18)
!21 = !DILocalVariable(name: "v1", scope: !4, file: !1, line: 3, type: !8)
!22 = !DILocation(line: 3, column: 37, scope: !4)
!23 = !DILocation(line: 3, column: 41, scope: !4)
!24 = !DILocation(line: 3, column: 48, scope: !4)
!25 = distinct !DISubprogram(name: "main", scope: !11, file: !11, line: 2, type: !26, scopeLine: 2, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !10, retainedNodes: !7)
!26 = !DISubroutineType(types: !27)
!27 = !{!28}
!28 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
!29 = !DILocation(line: 3, column: 3, scope: !25)
!30 = !DILocation(line: 4, column: 1, scope: !25)

I'll look into reverting this (& dependent patches) after lunch to unblock the main branch - but happy to work with you/continue to work on this to figure out how to address the issue. It may be a latent bug of my own creation regarding testing the splitDwarfInilning property on the wrong CU or something.

[dblaikie]

oh, and in another example I found that this patch caused LLVM to produce two dwo CUs, which is incorrect (but unlike the other example, doesn't crash). (this is the same as the first example, but with splitDedugInlining enabled on both CUs)

; ModuleID = 'ab.bc'
source_filename = "llvm-link"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

%"struct.ns::t1" = type { i8 }

; Function Attrs: mustprogress uwtable
define available_externally dso_local void @_Z2f1v() #0 !dbg !4 {
entry:
  %v1 = alloca %"struct.ns::t1", align 1
  call void @llvm.dbg.declare(metadata %"struct.ns::t1"* %v1, metadata !21, metadata !DIExpression()), !dbg !22
  call void @_Z3pinv(), !dbg !23
  ret void, !dbg !24
}

; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
declare void @llvm.dbg.declare(metadata, metadata, metadata) #1

declare dso_local void @_Z3pinv() #2

; Function Attrs: mustprogress norecurse uwtable
define dso_local i32 @main() #3 !dbg !25 {
entry:
  call void @_Z2f1v(), !dbg !29
  ret i32 0, !dbg !30
}

attributes #0 = { mustprogress uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #1 = { nofree nosync nounwind readnone speculatable willreturn }
attributes #2 = { "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #3 = { mustprogress norecurse uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }

!llvm.dbg.cu = !{!0, !10}
!llvm.ident = !{!12, !12}
!llvm.module.flags = !{!13, !14, !15, !16, !17}

!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 14.0.0 ([e-mail]:llvm/llvm-project.git 75b622a7959479ccdb758ebe0973061b7b4fcacd)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, imports: !2, splitDebugInlining: true, nameTableKind: GNU)
!1 = !DIFile(filename: "a.cpp", directory: "/usr/local/google/home/blaikie/dev/scratch")
!2 = !{!3}
!3 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !4, entity: !8, file: !1, line: 3)
!4 = distinct !DISubprogram(name: "f1", linkageName: "_Z2f1v", scope: !1, file: !1, line: 3, type: !5, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
!5 = !DISubroutineType(types: !6)
!6 = !{null}
!7 = !{}
!8 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "t1", scope: !9, file: !1, line: 1, size: 8, flags: DIFlagTypePassByValue, elements: !7, identifier: "_ZTSN2ns2t1E")
!9 = !DINamespace(name: "ns", scope: null)
!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, producer: "clang version 14.0.0 ([e-mail]:llvm/llvm-project.git 75b622a7959479ccdb758ebe0973061b7b4fcacd)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: true, nameTableKind: GNU)
!11 = !DIFile(filename: "b.cpp", directory: "/usr/local/google/home/blaikie/dev/scratch")
!12 = !{!"clang version 14.0.0 ([e-mail]:llvm/llvm-project.git 75b622a7959479ccdb758ebe0973061b7b4fcacd)"}
!13 = !{i32 7, !"Dwarf Version", i32 4}
!14 = !{i32 2, !"Debug Info Version", i32 3}
!15 = !{i32 1, !"wchar_size", i32 4}
!16 = !{i32 7, !"uwtable", i32 1}
!17 = !{i32 7, !"frame-pointer", i32 2}
!18 = distinct !DISubprogram(name: "f2", linkageName: "_Z2f2v", scope: !1, file: !1, line: 4, type: !5, scopeLine: 4, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
!19 = !DILocation(line: 4, column: 13, scope: !18)
!20 = !DILocation(line: 4, column: 19, scope: !18)
!21 = !DILocalVariable(name: "v1", scope: !4, file: !1, line: 3, type: !8)
!22 = !DILocation(line: 3, column: 37, scope: !4)
!23 = !DILocation(line: 3, column: 41, scope: !4)
!24 = !DILocation(line: 3, column: 48, scope: !4)
!25 = distinct !DISubprogram(name: "main", scope: !11, file: !11, line: 2, type: !26, scopeLine: 2, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !10, retainedNodes: !7)
!26 = !DISubroutineType(types: !27)
!27 = !{!28}
!28 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
!29 = !DILocation(line: 3, column: 3, scope: !25)
!30 = !DILocation(line: 4, column: 1, scope: !25)

[krisb]

@dblaikie, thank you for the reproducers!

I've tried to compile&examine them, but haven't found any issues. Likely, I did something wrong.
Could you, please, give more details on how to reproduce the issue(s)? I mean, exact compilation command line and/or crash details / wrong dwarf output? Thank you!

[dblaikie]

In D113741#3207109, @krisb wrote:

@dblaikie, thank you for the reproducers!

I've tried to compile&examine them, but haven't found any issues. Likely, I did something wrong.
Could you, please, give more details on how to reproduce the issue(s)? I mean, exact compilation command line and/or crash details / wrong dwarf output? Thank you!

Oh, right, sorry:

For the first one, you should observe a crash (in a +Asserts build) when running this command:

$ clang++-tot ab-crash.ll -gsplit-dwarf -c -O1

The backtrace looks something like this:

$ clang++-tot ab-crash.ll -gsplit-dwarf -c -O1 -S 
clang++-tot: /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:660: llvm::DIE *llvm::DwarfCompileUnit::constructInlinedScopeDIE(llvm::LexicalScope *): Assertion `OriginDIE && "Unable to find original DIE for an inlined subprogram."' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: clang++-tot ab-crash.ll -gsplit-dwarf -c -O1 -S
1.      Code generation
2.      Running pass 'Function Pass Manager' on module 'ab-crash.ll'.
3.      Running pass 'X86 Assembly Printer' on function '@main'
 #0 0x0000000009e604ea llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Unix/Signals.inc:565:11
 #1 0x0000000009e6069b PrintStackTraceSignalHandler(void*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Unix/Signals.inc:632:1
 #2 0x0000000009e5ed63 llvm::sys::RunSignalHandlers() /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Signals.cpp:97:5
 #3 0x0000000009e5fdfe llvm::sys::CleanupOnSignal(unsigned long) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Unix/Signals.inc:362:1
 #4 0x0000000009d6fa4e (anonymous namespace)::CrashRecoveryContextImpl::HandleCrash(int, unsigned long) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/CrashRecoveryContext.cpp:0:7
 #5 0x0000000009d6fe03 CrashRecoverySignalHandler(int) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/CrashRecoveryContext.cpp:390:1
 #6 0x00007f60568c08e0 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x138e0)
 #7 0x00007f6056389e71 raise ./signal/../sysdeps/unix/sysv/linux/raise.c:50:1
 #8 0x00007f6056373536 abort ./stdlib/abort.c:81:7
 #9 0x00007f605637341f get_sysdep_segment_value ./intl/loadmsgcat.c:509:8
#10 0x00007f605637341f _nl_load_domain ./intl/loadmsgcat.c:970:34
#11 0x00007f60563827f2 (/lib/x86_64-linux-gnu/libc.so.6+0x357f2)
#12 0x000000000b57ce6e llvm::DwarfCompileUnit::constructInlinedScopeDIE(llvm::LexicalScope*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:0:3
#13 0x000000000b57cb6c llvm::DwarfCompileUnit::constructScopeDIE(llvm::LexicalScope*, llvm::DIE&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:538:10
#14 0x000000000b57d3ae llvm::DwarfCompileUnit::createAndAddScopeChildren(llvm::LexicalScope*, llvm::DIE&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:1072:7
#15 0x000000000b57efa1 llvm::DwarfCompileUnit::constructSubprogramScopeDIE(llvm::DISubprogram const*, llvm::LexicalScope*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:1023:14
#16 0x000000000b52bc0d llvm::DwarfDebug::endFunctionImpl(llvm::MachineFunction const*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp:2301:50
#17 0x000000000b505ab2 llvm::DebugHandlerBase::endFunction(llvm::MachineFunction const*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DebugHandlerBase.cpp:0:5
#18 0x000000000b4e1f4c llvm::AsmPrinter::emitFunctionBody() /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp:1505:3
#19 0x0000000007ed7113 llvm::X86AsmPrinter::runOnMachineFunction(llvm::MachineFunction&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Target/X86/X86AsmPrinter.cpp:82:3
#20 0x0000000008a96477 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/MachineFunctionPass.cpp:72:8
#21 0x0000000009165d5e llvm::FPPassManager::runOnFunction(llvm::Function&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1439:23
#22 0x000000000916abb2 llvm::FPPassManager::runOnModule(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1485:16
#23 0x0000000009166649 (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1554:23
#24 0x00000000091661bd llvm::legacy::PassManagerImpl::run(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:542:16
#25 0x000000000916ae91 llvm::legacy::PassManager::run(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1681:3

But that's actually not the original situation I was investigating, so maybe it's another bug? (possibly same root cause, though)

Though the original issue I was investigating was from invalid .debug_gnu_pubnames (DIE offsets of 0, which prematurely terminates the .debug_gnu_pubnames list and produces a warning from llvm-dwarfdump), and by adding assert(Offset) in llvm::DIE::getOffset (initially I added it in the pubnames emission code, but generally it should be true that the offset of a DIE should not be queried until it's been computed, and no DIE has offset zero, because the offset is from the start of the CU contribution - so the headers come first and mean the DIE offset is always non-zero). It seems I still don't have a small reproduction for that exact issue, though.

The second test case provided above, that produces multiple DWO CUs can be reproduced as follows:

$ clang++-tot ab-multi-cu.ll -gsplit-dwarf -c -O1 && llvm-dwarfdump-tot ab-multi-cu.o
ab-multi-cu.o:  file format elf64-x86-64

.debug_info contents:
0x00000000: Compile Unit: length = 0x00000053, format = DWARF32, version = 0x0004, abbr_offset = 0x0000, addr_size = 0x08 (next unit at 0x00000057)

0x0000000b: DW_TAG_compile_unit
              DW_AT_stmt_list   (0x00000000)
              DW_AT_comp_dir    ("/usr/local/google/home/blaikie/dev/scratch")
              DW_AT_GNU_pubnames        (true)
              DW_AT_GNU_dwo_name        ("ab-multi-cu.dwo")
              DW_AT_GNU_dwo_id  (0xe9d1660e0395035d)
              DW_AT_low_pc      (0x0000000000000000)
              DW_AT_high_pc     (0x000000000000000d)
              DW_AT_GNU_addr_base       (0x00000000)

0x00000030:   DW_TAG_subprogram
                DW_AT_low_pc    (0x0000000000000000)
                DW_AT_high_pc   (0x000000000000000d)
                DW_AT_name      ("main")

0x00000041:     DW_TAG_inlined_subroutine
                  DW_AT_abstract_origin (0x000000000000007b "f1")
                  DW_AT_low_pc  (0x0000000000000004)
                  DW_AT_high_pc (0x0000000000000009)
                  DW_AT_call_file       ("/usr/local/google/home/blaikie/dev/scratch/b.cpp")
                  DW_AT_call_line       (3)
                  DW_AT_call_column     (0x03)

0x00000055:     NULL

0x00000056:   NULL
0x00000057: Compile Unit: length = 0x00000027, format = DWARF32, version = 0x0004, abbr_offset = 0x0000, addr_size = 0x08 (next unit at 0x00000082)

0x00000062: DW_TAG_compile_unit
              DW_AT_stmt_list   (0x0000004c)
              DW_AT_comp_dir    ("/usr/local/google/home/blaikie/dev/scratch")
              DW_AT_GNU_pubnames        (true)
              DW_AT_GNU_dwo_name        ("ab-multi-cu.dwo")
              DW_AT_GNU_dwo_id  (0xeb227ecffcd242c1)
              DW_AT_GNU_addr_base       (0x00000000)

0x0000007b:   DW_TAG_subprogram
                DW_AT_name      ("f1")
                DW_AT_inline    (DW_INL_inlined)

0x00000081:   NULL

In this case the second CU is OK (for the cross-CU inlining that's occurred), but it shouldn't have any DWO_* attributes, and the .dwo file shouldn't have a second CU in it (which it does after this patch) because Split DWARF doesn't have a good way to describe cross-CU references correctly once the CU is packaged in a dwp file. (this behavior is meant to be controlled by the SplitDwarfCrossCUReferences flag.

[krisb]

@dblaikie added another attempt to revert this as a review: D116225. Will commit once pre-commit tests + local lldb tests passed.

About the issues.
First of all, thank you for all the details! I was able to reproduce both issues.
(1) The crash doesn't seem related to the patch and reverting it doesn't make things better. It seems there is an inconsistency between when/where an abstract origin gets emitted and when/where it's requested.
(2) The issue with wrong dwo attributes is indeed caused by the imported declaration that gets emitted where it shouldn't (at least comparing with the behavior before this patch).
Before this patch 'includeMinimalInlineScopes()' makes emission of imported entity skipped and the second CU assumed empty, so we did not emit dwo attributes for it.
But I'm not sure I fully understand how DwarfDebug handles split dwarf. Why do we need the 'includeMinimalInlineScopes()' check for correctness here?

[dblaikie]

In D113741#3208256, @krisb wrote:

@dblaikie added another attempt to revert this as a review: D116225. Will commit once pre-commit tests + local lldb tests passed.

About the issues.
First of all, thank you for all the details! I was able to reproduce both issues.
(1) The crash doesn't seem related to the patch and reverting it doesn't make things better. It seems there is an inconsistency between when/where an abstract origin gets emitted and when/where it's requested.

Hmm, bother - guess I failed to correctly reduce the issue. I'll try again.

(2) The issue with wrong dwo attributes is indeed caused by the imported declaration that gets emitted where it shouldn't (at least comparing with the behavior before this patch).
Before this patch 'includeMinimalInlineScopes()' makes emission of imported entity skipped and the second CU assumed empty, so we did not emit dwo attributes for it.
But I'm not sure I fully understand how DwarfDebug handles split dwarf. Why do we need the 'includeMinimalInlineScopes()' check for correctness here?

It gets a bit complicated because a DWO file can only contain a single CU, but the .o file can contain multiple CUs - so to preserve as much fidelity as possible, if "split dwarf inlining" is enabled (include inline stack frames in the CU in the .o file, so the program can be symbolized even without the dwo/dwp files) then the .o file will get multiple CUs with -g1-like debug info, and one of those CUs will also have the dwo attributes and be a skeleton for the .dwo CU (but the dwo file, instead of having multiple CUs to describe the cross-CU inlining, it'll degrade the debug info quality by moving the DIEs that should be in other CUs in the dwo into the single CU there).

[dblaikie]

Turned out we had two different patches causing the same assert (should've bisected them both to be sure - added some unnecessary confusion to this whole situation). One of my reduced test cases actually was for https://reviews.llvm.org/D115325#3209087 - working on another for this patch (D113741) now...

[shchenz]

To continue with a more accurate fix, one example for your reference. (The behavior is changed with/without this reversion):

int main() {      // <====== function scope
  int x = 3;
  int num = 6;
  printf("%d\n", x);
  if (num) {   // <======= lexical block 1
    static int x = 45;
    printf("%d\n", x);
    if (x) {   // <======= lexical block 2
      int x = 145;
      printf("%d\n", x);
    }
  }
  return 0;
}

Without this patch:

<1><2a>: Abbrev Number: 2 (DW_TAG_subprogram)
   <2b>   DW_AT_low_pc      : 0x10010680
   <33>   DW_AT_high_pc     : 0xd4
   <37>   DW_AT_frame_base  : 1 byte block: 6f         (DW_OP_reg31 (r31))
   <39>   DW_AT_name        : (indirect string, offset: 0xcb): main
   <3f>   DW_AT_type        : <0x94>
   <43>   DW_AT_external    : 1
<2><43>: Abbrev Number: 3 (DW_TAG_variable)
   <44>   DW_AT_location    : 3 byte block: 91 f0 0    (DW_OP_fbreg: 112)
   <48>   DW_AT_name        : (indirect string, offset: 0xd4): x
   <4e>   DW_AT_type        : <0x94>
<2><52>: Abbrev Number: 3 (DW_TAG_variable)
   <53>   DW_AT_location    : 3 byte block: 91 ec 0    (DW_OP_fbreg: 108)
   <57>   DW_AT_name        : (indirect string, offset: 0xd0): num
   <5d>   DW_AT_type        : <0x94>
<2><61>: Abbrev Number: 4 (DW_TAG_lexical_block)     <<<<<<<<<<this should be for lexical block 2, so its DIE level should be 3
   <62>   DW_AT_low_pc      : 0x1001070c
   <6a>   DW_AT_high_pc     : 0x20
<3><6e>: Abbrev Number: 3 (DW_TAG_variable)             <<<<<<<<<this is for x=145 and nested in lexical block2, so its DIE level should be 4
   <6f>   DW_AT_location    : 3 byte block: 91 e8 0    (DW_OP_fbreg: 104)
   <73>   DW_AT_name        : (indirect string, offset: 0xd4): x
   <79>   DW_AT_type        : <0x94>
<3><7d>: Abbrev Number: 0
<2><7e>: Abbrev Number: 5 (DW_TAG_variable)              <<<<<<<<<<this is for static int x = 45, so there should be another lexical block DIE for lexical block 1, and its DIE level should be 2.
   <7f>   DW_AT_name        : (indirect string, offset: 0xd4): x
   <83>   DW_AT_type        : <0x94>
   <89>   DW_AT_location    : 9 byte block: 3 e8 a 3 10 0 0 0 0        (DW_OP_addr: 10030ae8)
<2><93>: Abbrev Number: 0
<1><94>: Abbrev Number: 6 (DW_TAG_base_type)
   <95>   DW_AT_name        : (indirect string, offset: 0x0): int
   <99>   DW_AT_encoding    : 5        (signed)
   <9a>   DW_AT_byte_size   : 4
<1><9b>: Abbrev Number: 0

And we get wrong result for static number x at line 7. In debugger, it is printed as 3.

With this patch:

0x0000002a:   DW_TAG_subprogram
                DW_AT_low_pc    (0x0000000010010680)
                DW_AT_high_pc   (0x0000000010010754)
                DW_AT_frame_base        (DW_OP_reg31 X31)
                DW_AT_name      ("main")
                DW_AT_type      (0x00000094 "int")
                DW_AT_external  (true)

0x00000043:     DW_TAG_variable
                  DW_AT_name    ("x")          <<<<<<<<<<<<we need a lexical block for this definition
                  DW_AT_type    (0x00000094 "int")
                  DW_AT_location        (DW_OP_addr 0x10030ae8)

0x00000058:     DW_TAG_variable
                  DW_AT_location        (DW_OP_fbreg +112)
                  DW_AT_name    ("x")
                  DW_AT_type    (0x00000094 "int")

0x00000067:     DW_TAG_variable
                  DW_AT_location        (DW_OP_fbreg +108)
                  DW_AT_name    ("num")
                  DW_AT_type    (0x00000094 "int")

0x00000076:     DW_TAG_lexical_block
                  DW_AT_low_pc  (0x000000001001070c)
                  DW_AT_high_pc (0x000000001001072c)

0x00000083:       DW_TAG_variable
                    DW_AT_location      (DW_OP_fbreg +104)
                    DW_AT_name  ("x")
                    DW_AT_type  (0x00000094 "int")

0x00000092:       NULL

Now we get wrong result for local variable x at line 3. In debugger, it is printed as 45. (The DIE for the local variable x is overridden by the static variable x in lexical block 1).

[dblaikie]

In D113741#3210175, @shchenz wrote:

To continue with a more accurate fix, one example for your reference. (The behavior is changed with/without this reversion):

Sorry, I'm not following this comment ^ (& the rest) - could you rephrase if there's a question/concern you have I can help clarify/describe/work with you on?

[shchenz]

In D113741#3210838, @dblaikie wrote:

In D113741#3210175, @shchenz wrote:

To continue with a more accurate fix, one example for your reference. (The behavior is changed with/without this reversion):

Sorry, I'm not following this comment ^ (& the rest) - could you rephrase if there's a question/concern you have I can help clarify/describe/work with you on?

I guess there will be a follow-up patch for the function-local declaration for lexical block? This patch seems like a reversion of a previous functionality change?
In our internal test, we also found an issue related to variables in the lexical block, so I pasted it here for reference : )

[krisb]

In D113741#3211123, @shchenz wrote:

In D113741#3210838, @dblaikie wrote:

In D113741#3210175, @shchenz wrote:

To continue with a more accurate fix, one example for your reference. (The behavior is changed with/without this reversion):

Sorry, I'm not following this comment ^ (& the rest) - could you rephrase if there's a question/concern you have I can help clarify/describe/work with you on?

I guess there will be a follow-up patch for the function-local declaration for lexical block? This patch seems like a reversion of a previous functionality change?
In our internal test, we also found an issue related to variables in the lexical block, so I pasted it here for reference : )

This is the functional change itself, reverting patch is at https://reviews.llvm.org/D116225.

Without this patch:

I guess you meant without the revert, and actually _with_ this patch, right?
If so, you'd need a clang side change https://reviews.llvm.org/D113743 (which has been reverted as well) to see static locals placed correctly.
If to apply all the reverted patches (D114705, D113741, and D113743) I believe you should see a correct DWARF compiled for your example.

[shchenz]

If to apply all the reverted patches (D114705, D113741, and D113743) I believe you should see a correct DWARF compiled for your example.

Yeah, verify with manually applying all the above patches, it works as expected. Looking forward to seeing these patches merged in again.

[krisb]

• Rebase on ToT & minor fixes.
• Fix the issue when an local imported entity caused emission of multiple CUs in a dwo when split dwarf is enabled.
• Add a test for the aforementioned issue.

[krisb]

@dblaikie I've updated the patch with a fix for split dwarf issue (the test is included). Could I ask you to check the full build on your side?

The issue was in wrong assumptions about in which CUs we need to create those local entities. Basically, they [CUs] should match the ones where the parent subprogram was emitted.
It's not an easy thing to find the particular list of suited CUs while we are in endModule() (especially, because split-dwarf and split-dwarf inlining introduce a tricky logic around), so I chose to collect this information during subprograms emission.
The solution still looks a bit hacky, but I hope this is a step forward.

What about the second issue, unfortunately (or fortunately? :)), I'm still able to reproduce the crash on ToT of main, so I'm not sure this patch may cause the issue. I'll appreciate if could check it once again.

Thank you and sorry for the long delay.

[dblaikie]

In D113741#3343268, @krisb wrote:

@dblaikie I've updated the patch with a fix for split dwarf issue (the test is included). Could I ask you to check the full build on your side?

The issue was in wrong assumptions about in which CUs we need to create those local entities. Basically, they [CUs] should match the ones where the parent subprogram was emitted.
It's not an easy thing to find the particular list of suited CUs while we are in endModule() (especially, because split-dwarf and split-dwarf inlining introduce a tricky logic around), so I chose to collect this information during subprograms emission.
The solution still looks a bit hacky, but I hope this is a step forward.

What about the second issue, unfortunately (or fortunately? :)), I'm still able to reproduce the crash on ToT of main, so I'm not sure this patch may cause the issue. I'll appreciate if could check it once again.

Thank you and sorry for the long delay.

All good.

I'm not immediately following why a list of CUs is needed/why we'd create the variable in more than one CU - could you explain that?

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1405–1406

This looks buggy - it's returning a pointer to a local variable (the TheCU pointer) which means it'll be dangling once the function returns, yeah? (I'd expect msan to catch this, for instance)

[krisb]

Minor fix.

[krisb]

In D113741#3343496, @dblaikie wrote:

In D113741#3343268, @krisb wrote:

@dblaikie I've updated the patch with a fix for split dwarf issue (the test is included). Could I ask you to check the full build on your side?

The issue was in wrong assumptions about in which CUs we need to create those local entities. Basically, they [CUs] should match the ones where the parent subprogram was emitted.
It's not an easy thing to find the particular list of suited CUs while we are in endModule() (especially, because split-dwarf and split-dwarf inlining introduce a tricky logic around), so I chose to collect this information during subprograms emission.
The solution still looks a bit hacky, but I hope this is a step forward.

What about the second issue, unfortunately (or fortunately? :)), I'm still able to reproduce the crash on ToT of main, so I'm not sure this patch may cause the issue. I'll appreciate if could check it once again.

Thank you and sorry for the long delay.

All good.

I'm not immediately following why a list of CUs is needed/why we'd create the variable in more than one CU - could you explain that?

Well, this is because we may have a parent subprogram emitted in more than one CU (in the case of split dwarf), and we do not know in which one.
Here are the situations I see (hope I got them right):
(1) If split dwarf disabled, we have a single list of subprogram's abstract origins and can reference them between CUs. So, we'll emit a concrete definition and its abstract origin (any existing) in the same CU. In this case, we have only one CU.
(2) If we have split dwarf enabled, and shareAcrossDWOCUs() is false (it's false by default), every CU will maintain its own abstract origin's list. In this case, we cannot have references between CUs, and if we have cross-CU inlining, we'll create:

• concrete def -> in the CU the subprogram is defined,
• abstract origin -> in every CU the subprogram is inlined. For all local entities we should follow the same logic.

(3) SplitDwarfInlining adds skeleton units to this list. Basically, we do not need to emit any local entities for skeleton units in this case, but I just reused already existing logic around includeMinimalInlineScopes().

The reproducer you provided is about the case (2). We tried to emit the imported entity in a CU, where its parent subprogram was defined, but the subprogram itself was emitted in the CU it was inlined.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1405–1406

Uuups, bad experiments. Fixed.

[dblaikie]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1405–1406

This code still looks buggy, it's still returning the address of a local variable (the address of "TheU" pointer) so it'd still be UB/dangling/msan-unclean, I think?

[krisb]

Attempt 2.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1405–1406

Hurry isn't a good way to do smth good. Sorry. One more try and I'll go to check this with msan.

[dblaikie]

In D113741#3343683, @krisb wrote:

In D113741#3343496, @dblaikie wrote:

In D113741#3343268, @krisb wrote:

@dblaikie I've updated the patch with a fix for split dwarf issue (the test is included). Could I ask you to check the full build on your side?

The issue was in wrong assumptions about in which CUs we need to create those local entities. Basically, they [CUs] should match the ones where the parent subprogram was emitted.
It's not an easy thing to find the particular list of suited CUs while we are in endModule() (especially, because split-dwarf and split-dwarf inlining introduce a tricky logic around), so I chose to collect this information during subprograms emission.
The solution still looks a bit hacky, but I hope this is a step forward.

What about the second issue, unfortunately (or fortunately? :)), I'm still able to reproduce the crash on ToT of main, so I'm not sure this patch may cause the issue. I'll appreciate if could check it once again.

Thank you and sorry for the long delay.

All good.

I'm not immediately following why a list of CUs is needed/why we'd create the variable in more than one CU - could you explain that?

Well, this is because we may have a parent subprogram emitted in more than one CU (in the case of split dwarf), and we do not know in which one.
Here are the situations I see (hope I got them right):
(1) If split dwarf disabled, we have a single list of subprogram's abstract origins and can reference them between CUs. So, we'll emit a concrete definition and its abstract origin (any existing) in the same CU. In this case, we have only one CU.
(2) If we have split dwarf enabled, and shareAcrossDWOCUs() is false (it's false by default), every CU will maintain its own abstract origin's list. In this case, we cannot have references between CUs, and if we have cross-CU inlining, we'll create:

• concrete def -> in the CU the subprogram is defined,
• abstract origin -> in every CU the subprogram is inlined. For all local entities we should follow the same logic.

Do you have a test case for this (sorry, I realize it might be one of the test cases in the patch - I haven't looked/would appreciate some pointers)? My recollection is that without shareAcrossDWOCUs we should probably be only emitting one CU if using Split DWARF - Split DWARF doesn't really support multiple CUs in a single .dwo file (& we only support emitting a single .dwo for a given .o file) - but I guess maybe we do in full LTO mode, even if we don't do it in ThinLTO mode (ThinLTO's the only thing I've been especially concerned with/the only mode we use at Google)?

(3) SplitDwarfInlining adds skeleton units to this list. Basically, we do not need to emit any local entities for skeleton units in this case, but I just reused already existing logic around includeMinimalInlineScopes().

The reproducer you provided is about the case (2). We tried to emit the imported entity in a CU, where its parent subprogram was defined, but the subprogram itself was emitted in the CU it was inlined.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1405–1406

That doesn't seem like a valid range either - that range would be from "TheCU" until nullptr - so a huge chunk of the address space?

I think now you need/could have "TheCU + 1" as the end of the range and that'd be correct, now that it's passed in a pointer-to-pointer.

[krisb]

In D113741#3343849, @dblaikie wrote:

In D113741#3343683, @krisb wrote:

In D113741#3343496, @dblaikie wrote:

All good.

I'm not immediately following why a list of CUs is needed/why we'd create the variable in more than one CU - could you explain that?

Well, this is because we may have a parent subprogram emitted in more than one CU (in the case of split dwarf), and we do not know in which one.
Here are the situations I see (hope I got them right):
(1) If split dwarf disabled, we have a single list of subprogram's abstract origins and can reference them between CUs. So, we'll emit a concrete definition and its abstract origin (any existing) in the same CU. In this case, we have only one CU.
(2) If we have split dwarf enabled, and shareAcrossDWOCUs() is false (it's false by default), every CU will maintain its own abstract origin's list. In this case, we cannot have references between CUs, and if we have cross-CU inlining, we'll create:

• concrete def -> in the CU the subprogram is defined,
• abstract origin -> in every CU the subprogram is inlined. For all local entities we should follow the same logic.

Do you have a test case for this (sorry, I realize it might be one of the test cases in the patch - I haven't looked/would appreciate some pointers)? My recollection is that without shareAcrossDWOCUs we should probably be only emitting one CU if using Split DWARF - Split DWARF doesn't really support multiple CUs in a single .dwo file (& we only support emitting a single .dwo for a given .o file) - but I guess maybe we do in full LTO mode, even if we don't do it in ThinLTO mode (ThinLTO's the only thing I've been especially concerned with/the only mode we use at Google)?

I was inspired by these two tests:

LLVM :: DebugInfo/X86/split-dwarf-cross-cu-gmlt-g.ll
LLVM :: DebugInfo/X86/split-dwarf-cross-unit-reference.ll

They test the case I described in (2), with 2 CUs in a single .dwo.

[krisb]

Apply David's comment and add 'REQUIRES: object-emission' to tests.

[krisb]

In D113741#3343849, @dblaikie wrote:

In D113741#3343683, @krisb wrote:

In D113741#3343496, @dblaikie wrote:

I'm not immediately following why a list of CUs is needed/why we'd create the variable in more than one CU - could you explain that?

Well, this is because we may have a parent subprogram emitted in more than one CU (in the case of split dwarf), and we do not know in which one.
Here are the situations I see (hope I got them right):
(1) If split dwarf disabled, we have a single list of subprogram's abstract origins and can reference them between CUs. So, we'll emit a concrete definition and its abstract origin (any existing) in the same CU. In this case, we have only one CU.
(2) If we have split dwarf enabled, and shareAcrossDWOCUs() is false (it's false by default), every CU will maintain its own abstract origin's list. In this case, we cannot have references between CUs, and if we have cross-CU inlining, we'll create:

• concrete def -> in the CU the subprogram is defined,
• abstract origin -> in every CU the subprogram is inlined. For all local entities we should follow the same logic.

Do you have a test case for this (sorry, I realize it might be one of the test cases in the patch - I haven't looked/would appreciate some pointers)? My recollection is that without shareAcrossDWOCUs we should probably be only emitting one CU if using Split DWARF - Split DWARF doesn't really support multiple CUs in a single .dwo file (& we only support emitting a single .dwo for a given .o file) - but I guess maybe we do in full LTO mode, even if we don't do it in ThinLTO mode (ThinLTO's the only thing I've been especially concerned with/the only mode we use at Google)?

Okay, let's summarize this a bit.

Statement 1. The problem with the example you provided above was because this patch made subprograms and their local "global" entities emitted in different CUs in the case of split DWARF. This is a bug, for sure.
Statement 2. Currently, there is nothing in the code that restricts emitting multiple CUs in a single .dwo (whether for FullLTO or for manually linked modules (as in the tests) we will generate more than one CU per .dwo).

All my assumptions and conclusions are based on the code and existing regression tests. Unfortunately, I'm not aware of the real situation of split DWARF or split DWARF inlining support (and I couldn't find anything related in the docs either).

Having this, I think we cannot assume that a subprogram is guaranteed to be emitted only in one CU, unless we explicitly disallow the case with multiple CUs per .dwo. DWARF standard seems to assume only one CU per .dwo (but I couldn't find any explicit wording about restrictions on the number of CUs in a .dwo). However, it looks like GDB is okay and works fine with more than one CU in a .dwo. It'd be good to have this situation clarified. But it seems to be out of the scope of this patch. Do you have any thoughs on how to proceed with this?

(3) SplitDwarfInlining adds skeleton units to this list. Basically, we do not need to emit any local entities for skeleton units in this case, but I just reused already existing logic around includeMinimalInlineScopes().

The reproducer you provided is about the case (2). We tried to emit the imported entity in a CU, where its parent subprogram was defined, but the subprogram itself was emitted in the CU it was inlined.

[krisb]

Ping

[krisb]

Friendly ping :)

[dblaikie]

In D113741#3380215, @krisb wrote:

Friendly ping :)

Sorry I haven't got to this yet - it is on my list.

[dblaikie]

In D113741#3348795, @krisb wrote:

In D113741#3343849, @dblaikie wrote:

In D113741#3343683, @krisb wrote:

In D113741#3343496, @dblaikie wrote:

I'm not immediately following why a list of CUs is needed/why we'd create the variable in more than one CU - could you explain that?

Well, this is because we may have a parent subprogram emitted in more than one CU (in the case of split dwarf), and we do not know in which one.
Here are the situations I see (hope I got them right):
(1) If split dwarf disabled, we have a single list of subprogram's abstract origins and can reference them between CUs. So, we'll emit a concrete definition and its abstract origin (any existing) in the same CU. In this case, we have only one CU.
(2) If we have split dwarf enabled, and shareAcrossDWOCUs() is false (it's false by default), every CU will maintain its own abstract origin's list. In this case, we cannot have references between CUs, and if we have cross-CU inlining, we'll create:

• concrete def -> in the CU the subprogram is defined,
• abstract origin -> in every CU the subprogram is inlined. For all local entities we should follow the same logic.

Do you have a test case for this (sorry, I realize it might be one of the test cases in the patch - I haven't looked/would appreciate some pointers)? My recollection is that without shareAcrossDWOCUs we should probably be only emitting one CU if using Split DWARF - Split DWARF doesn't really support multiple CUs in a single .dwo file (& we only support emitting a single .dwo for a given .o file) - but I guess maybe we do in full LTO mode, even if we don't do it in ThinLTO mode (ThinLTO's the only thing I've been especially concerned with/the only mode we use at Google)?

Okay, let's summarize this a bit.

Statement 1. The problem with the example you provided above was because this patch made subprograms and their local "global" entities emitted in different CUs in the case of split DWARF. This is a bug, for sure.

Is that bug now addressed in the proposed patch & with tests to demonstrate it's fixed?

Statement 2. Currently, there is nothing in the code that restricts emitting multiple CUs in a single .dwo (whether for FullLTO or for manually linked modules (as in the tests) we will generate more than one CU per .dwo).

I believe DwarfDebug.cpp:SplitDwarfCrossCuReferences is intended to disallow this functionality, so far as I can tell/remember?

All my assumptions and conclusions are based on the code and existing regression tests. Unfortunately, I'm not aware of the real situation of split DWARF or split DWARF inlining support (and I couldn't find anything related in the docs either).

Having this, I think we cannot assume that a subprogram is guaranteed to be emitted only in one CU, unless we explicitly disallow the case with multiple CUs per .dwo. DWARF standard seems to assume only one CU per .dwo (but I couldn't find any explicit wording about restrictions on the number of CUs in a .dwo). However, it looks like GDB is okay and works fine with more than one CU in a .dwo. It'd be good to have this situation clarified. But it seems to be out of the scope of this patch. Do you have any thoughs on how to proceed with this?

One of the issues is that the dwp file format describes an index by CU hash (DWO ID) - it's unclear what it would mean to have multiple entries that overlap (because a single region contains multiple CUs) and where the region doesn't start with the CU for the lookup (because it's some region with multiple units).

[krisb]

In D113741#3414966, @dblaikie wrote:

In D113741#3348795, @krisb wrote:

Statement 1. The problem with the example you provided above was because this patch made subprograms and their local "global" entities emitted in different CUs in the case of split DWARF. This is a bug, for sure.

Is that bug now addressed in the proposed patch & with tests to demonstrate it's fixed?

I believe, it is. I created a test from the example you provided above, see llvm/test/DebugInfo/Generic/split-dwarf-local-import.ll.
It checks that the second CU doesn't have DW_AT_GNU_dwo_name and DW_AT_GNU_dwo_id and the imported entity has a proper parent.
But, if you have a chance to check the updated patch in the same conditions this bug originally reproduced, I'll highly appreciate it.

Statement 2. Currently, there is nothing in the code that restricts emitting multiple CUs in a single .dwo (whether for FullLTO or for manually linked modules (as in the tests) we will generate more than one CU per .dwo).

I believe DwarfDebug.cpp:SplitDwarfCrossCuReferences is intended to disallow this functionality, so far as I can tell/remember?

Hmm, I'm not sure fully I understood your point. SplitDwarfCrossCuReferences seems to _allow_ references between CUs, but it isn't enabled by default.
For example, if you enable split-dwarf with FullLTO, clang would emit a single .dwo per complied&linked binary with all the CUs in this .dwo. Neither clang nor backend would emit any errors/warnings.
I'm not insisting this is correct, or this may be useful for someone (I'm simply not aware), just want to mention that this is possible.

All my assumptions and conclusions are based on the code and existing regression tests. Unfortunately, I'm not aware of the real situation of split DWARF or split DWARF inlining support (and I couldn't find anything related in the docs either).

Having this, I think we cannot assume that a subprogram is guaranteed to be emitted only in one CU, unless we explicitly disallow the case with multiple CUs per .dwo. DWARF standard seems to assume only one CU per .dwo (but I couldn't find any explicit wording about restrictions on the number of CUs in a .dwo). However, it looks like GDB is okay and works fine with more than one CU in a .dwo. It'd be good to have this situation clarified. But it seems to be out of the scope of this patch. Do you have any thoughs on how to proceed with this?

One of the issues is that the dwp file format describes an index by CU hash (DWO ID) - it's unclear what it would mean to have multiple entries that overlap (because a single region contains multiple CUs) and where the region doesn't start with the CU for the lookup (because it's some region with multiple units).

So the question is: should we explicitly disallow emitting multiple CUs in .dwo (meaning, for example, split-dwarf with FullLTO w/o split-dwarf-cross-cu-references)? By reporting an error or at least specifying somewhere in docs that this case it not correct thus not supported?

[dblaikie]

I tried applying this internally & running the same build that tripped over something last time this was committed & I seem to be tripping over the same thing:

(.debug_gnu_pubnames): name lookup table at offset 0x38cd has a terminator at offset 0x38db before the expected end at 0x3a3d

As well as a bunch of unresolved symbols during linking... :/

I'll have to work on reducing a test case for this.

[krisb]

In D113741#3437182, @dblaikie wrote:

I tried applying this internally & running the same build that tripped over something last time this was committed & I seem to be tripping over the same thing:

(.debug_gnu_pubnames): name lookup table at offset 0x38cd has a terminator at offset 0x38db before the expected end at 0x3a3d

As well as a bunch of unresolved symbols during linking... :/

I'll have to work on reducing a test case for this.

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

[jmorse]

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

[dblaikie]

In D113741#3439940, @krisb wrote:

In D113741#3437182, @dblaikie wrote:

I tried applying this internally & running the same build that tripped over something last time this was committed & I seem to be tripping over the same thing:

(.debug_gnu_pubnames): name lookup table at offset 0x38cd has a terminator at offset 0x38db before the expected end at 0x3a3d

As well as a bunch of unresolved symbols during linking... :/

I'll have to work on reducing a test case for this.

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

That's correct. Sorry I lost context on this again and don't think I have a reduced test case/something to work from right now, but I'll keep in on my radar to work on.

[krisb]

There is an alternative implementation that relies on addition field of DISubprogram/DILexicalScope that tracks static locals, local imports and types D125693.

---

In D113741#3448685, @jmorse wrote:

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

I've tried to reproduce the issue using the example from D94976 (and with this patch), it compiled fine and didn't show the mentioned problem with pubnames.
Another question is about the combination of FullLTO and split-dwarf which has been already discussed in this thread and which isn't (fully) supported by LLVM mainline (yet, at least in a way that doesn't violate DWARF specification).

So, I'm still looking for a reproducer for the issue with pubnames. Any help will be appreciated.

[dblaikie]

In D113741#3516038, @krisb wrote:

There is an alternative implementation that relies on addition field of DISubprogram/DILexicalScope that tracks static locals, local imports and types D125693.

Ah, OK - so that 5-part sequence is an alternative to this one?

Might be good to figure out the issue here before debating which direction to go in, etc. (hmm, one of the patches in that series I've already approved? So am I reviewing alternatives here? Given the complexity here that's a bit costly for me/end up feeling like I'm drowning under trying to figure out all these moving parts... - a linear sequence is generally OK/good, I can review the easy parts and focus on the difficult ones, but multiple large alternatives can be a bit daunting)

---

In D113741#3448685, @jmorse wrote:

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

I've tried to reproduce the issue using the example from D94976 (and with this patch), it compiled fine and didn't show the mentioned problem with pubnames.
Another question is about the combination of FullLTO and split-dwarf which has been already discussed in this thread and which isn't (fully) supported by LLVM mainline (yet, at least in a way that doesn't violate DWARF specification).

So, I'm still looking for a reproducer for the issue with pubnames. Any help will be appreciated.

Yeah, I still need to figure out how to do that/get you that. :/

(is this at all/possibly related to https://reviews.llvm.org/D124892 ?)

[ayermolo]

In D113741#3532850, @dblaikie wrote:

In D113741#3516038, @krisb wrote:

There is an alternative implementation that relies on addition field of DISubprogram/DILexicalScope that tracks static locals, local imports and types D125693.

Ah, OK - so that 5-part sequence is an alternative to this one?

Might be good to figure out the issue here before debating which direction to go in, etc. (hmm, one of the patches in that series I've already approved? So am I reviewing alternatives here? Given the complexity here that's a bit costly for me/end up feeling like I'm drowning under trying to figure out all these moving parts... - a linear sequence is generally OK/good, I can review the easy parts and focus on the difficult ones, but multiple large alternatives can be a bit daunting)

---

In D113741#3448685, @jmorse wrote:

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

I've tried to reproduce the issue using the example from D94976 (and with this patch), it compiled fine and didn't show the mentioned problem with pubnames.
Another question is about the combination of FullLTO and split-dwarf which has been already discussed in this thread and which isn't (fully) supported by LLVM mainline (yet, at least in a way that doesn't violate DWARF specification).

So, I'm still looking for a reproducer for the issue with pubnames. Any help will be appreciated.

Yeah, I still need to figure out how to do that/get you that. :/

(is this at all/possibly related to https://reviews.llvm.org/D124892 ?)

Seems like a discussion is more about multiple CUs in DWO with full LTO? Need to read all the history in more details.
Side note BOLT doesn't support multiple CUs in the dwo. There are certain assumptions about abbrev, and how it writes out new dwo files.
Does pubnames actually used for anything? My impression is that debuggers (or at least lldb) ignores it.

[dblaikie]

In D113741#3534761, @ayermolo wrote:

In D113741#3532850, @dblaikie wrote:

In D113741#3516038, @krisb wrote:

There is an alternative implementation that relies on addition field of DISubprogram/DILexicalScope that tracks static locals, local imports and types D125693.

Ah, OK - so that 5-part sequence is an alternative to this one?

Might be good to figure out the issue here before debating which direction to go in, etc. (hmm, one of the patches in that series I've already approved? So am I reviewing alternatives here? Given the complexity here that's a bit costly for me/end up feeling like I'm drowning under trying to figure out all these moving parts... - a linear sequence is generally OK/good, I can review the easy parts and focus on the difficult ones, but multiple large alternatives can be a bit daunting)

Ping on this ^

---

In D113741#3448685, @jmorse wrote:

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

I've tried to reproduce the issue using the example from D94976 (and with this patch), it compiled fine and didn't show the mentioned problem with pubnames.
Another question is about the combination of FullLTO and split-dwarf which has been already discussed in this thread and which isn't (fully) supported by LLVM mainline (yet, at least in a way that doesn't violate DWARF specification).

So, I'm still looking for a reproducer for the issue with pubnames. Any help will be appreciated.

Yeah, I still need to figure out how to do that/get you that. :/

(is this at all/possibly related to https://reviews.llvm.org/D124892 ?)

Seems like a discussion is more about multiple CUs in DWO with full LTO? Need to read all the history in more details.
Side note BOLT doesn't support multiple CUs in the dwo. There are certain assumptions about abbrev, and how it writes out new dwo files.
Does pubnames actually used for anything? My impression is that debuggers (or at least lldb) ignores it.

Yep, debug_gnu_pubnames are used by gold and lld to build gdb_index which improves gdb startup time.

[ayermolo]

In D113741#3536166, @dblaikie wrote:

In D113741#3534761, @ayermolo wrote:

In D113741#3532850, @dblaikie wrote:

In D113741#3516038, @krisb wrote:

There is an alternative implementation that relies on addition field of DISubprogram/DILexicalScope that tracks static locals, local imports and types D125693.

Ah, OK - so that 5-part sequence is an alternative to this one?

Might be good to figure out the issue here before debating which direction to go in, etc. (hmm, one of the patches in that series I've already approved? So am I reviewing alternatives here? Given the complexity here that's a bit costly for me/end up feeling like I'm drowning under trying to figure out all these moving parts... - a linear sequence is generally OK/good, I can review the easy parts and focus on the difficult ones, but multiple large alternatives can be a bit daunting)

Ping on this ^

---

In D113741#3448685, @jmorse wrote:

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

I've tried to reproduce the issue using the example from D94976 (and with this patch), it compiled fine and didn't show the mentioned problem with pubnames.
Another question is about the combination of FullLTO and split-dwarf which has been already discussed in this thread and which isn't (fully) supported by LLVM mainline (yet, at least in a way that doesn't violate DWARF specification).

So, I'm still looking for a reproducer for the issue with pubnames. Any help will be appreciated.

Yeah, I still need to figure out how to do that/get you that. :/

(is this at all/possibly related to https://reviews.llvm.org/D124892 ?)

Seems like a discussion is more about multiple CUs in DWO with full LTO? Need to read all the history in more details.
Side note BOLT doesn't support multiple CUs in the dwo. There are certain assumptions about abbrev, and how it writes out new dwo files.
Does pubnames actually used for anything? My impression is that debuggers (or at least lldb) ignores it.

Yep, debug_gnu_pubnames are used by gold and lld to build gdb_index which improves gdb startup time.

Ah right, forgot about it.

[krisb]

In D113741#3532850, @dblaikie wrote:

In D113741#3516038, @krisb wrote:

There is an alternative implementation that relies on addition field of DISubprogram/DILexicalScope that tracks static locals, local imports and types D125693.

Ah, OK - so that 5-part sequence is an alternative to this one?

Mostly. The alternative implementation reuses large part of work of this patch.
Basically, this first implementation has (had) 3 patches in sequence:

• D114705 (moving handling of globals, types and imports from beginModule() to endModule() in DwarfDebug.cpp)
• D113741 (this one, backend's part)
• D113743 (clang's part)

The new one has 5 patches:

• D125691 (adding new 'localDecls' field to DILexicalBlock and DISubprogram),
• D114705 (taken from previous implementation w/o changes),
• D125693 (backend's part, replacement for D113741)
• D125694, D125695 (clang's part, the same as D113743, but split on 2 patches and with some minor improvements).

Might be good to figure out the issue here before debating which direction to go in, etc. (hmm, one of the patches in that series I've already approved? So am I reviewing alternatives here? Given the complexity here that's a bit costly for me/end up feeling like I'm drowning under trying to figure out all these moving parts... - a linear sequence is generally OK/good, I can review the easy parts and focus on the difficult ones, but multiple large alternatives can be a bit daunting)

I'm ready to continue working with this patchset to find and fix the issue with pubnames, but I have no ideas how I can reproduce it, so I decided to move forward and try another approach which looks a bit better to me.
If you have some time to help me finding the reproducer or could give some hints on this, I'll be happy to fix the aforementioned problem first.

In D113741#3448685, @jmorse wrote:

Drive-by comment on the topic of the gnu pubnames error,

In D113741#3439940, @krisb wrote:

If I remember correctly, this issue with .debug_gnu_pubnames was caused by another patch (https://reviews.llvm.org/rG78d15a112cbd545fbb6e1aa37c221ef5aeffb3f2).
Did I understand you right that the same issue is reproduced with w/o that patch?

I've run into that error myself when trying to alter which unit a subprogram definition goes in -- see the comment at D94976#2543257 . That patch never landed in the end because I couldn't figure out exactly what was wrong. IMHO, there's some latent weirdness that triggers that error, exactly how is unclear though. IMO, it's not specific to rG78d15a112cbd5 .

[0] "(.debug_gnu_pubnames): name lookup table at offset (...) has a terminator at offset (...) before the expected end at (...)"

I've tried to reproduce the issue using the example from D94976 (and with this patch), it compiled fine and didn't show the mentioned problem with pubnames.
Another question is about the combination of FullLTO and split-dwarf which has been already discussed in this thread and which isn't (fully) supported by LLVM mainline (yet, at least in a way that doesn't violate DWARF specification).

So, I'm still looking for a reproducer for the issue with pubnames. Any help will be appreciated.

Yeah, I still need to figure out how to do that/get you that. :/

(is this at all/possibly related to https://reviews.llvm.org/D124892 ?)

One of the issues you mentioned here https://reviews.llvm.org/D113741#3207125 seems to relate to D124892, but as I said, it reproduces w/o this patchset and likely has different root cause.

[dblaikie]

OK, so if I want to reproduce the issue for you, I need to apply these three patches? D114705 D113741 D113743

[krisb]

In D113741#3538443, @dblaikie wrote:

OK, so if I want to reproduce the issue for you, I need to apply these three patches? D114705 D113741 D113743

You, likely, not need D113743 (it'd been reverted just after being committed, so it seems it couldn't be the guilty one).

[dblaikie]

In D113741#3538443, @dblaikie wrote:

OK, so if I want to reproduce the issue for you, I need to apply these three patches? D114705 D113741 D113743

OK, reapplied the patches and reproduced the issue with a simple test program using Control Flow Integrity. I'm not sure entirely how easy this will be to extract into something reproducible outside Google's internal build system (my understanding is that CFI is a whole-program (including the standard library) sort of thing, which might make satisfying those requirements outside difficult, though not impossible - the features do exist outside of Google/are usable/tested there, etc) but I'll get started.

[ayermolo]

In D113741#3207125, @dblaikie wrote:

In D113741#3207109, @krisb wrote:

@dblaikie, thank you for the reproducers!

I've tried to compile&examine them, but haven't found any issues. Likely, I did something wrong.
Could you, please, give more details on how to reproduce the issue(s)? I mean, exact compilation command line and/or crash details / wrong dwarf output? Thank you!

Oh, right, sorry:

For the first one, you should observe a crash (in a +Asserts build) when running this command:

$ clang++-tot ab-crash.ll -gsplit-dwarf -c -O1

The backtrace looks something like this:

$ clang++-tot ab-crash.ll -gsplit-dwarf -c -O1 -S 
clang++-tot: /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:660: llvm::DIE *llvm::DwarfCompileUnit::constructInlinedScopeDIE(llvm::LexicalScope *): Assertion `OriginDIE && "Unable to find original DIE for an inlined subprogram."' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: clang++-tot ab-crash.ll -gsplit-dwarf -c -O1 -S
1.      Code generation
2.      Running pass 'Function Pass Manager' on module 'ab-crash.ll'.
3.      Running pass 'X86 Assembly Printer' on function '@main'
 #0 0x0000000009e604ea llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Unix/Signals.inc:565:11
 #1 0x0000000009e6069b PrintStackTraceSignalHandler(void*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Unix/Signals.inc:632:1
 #2 0x0000000009e5ed63 llvm::sys::RunSignalHandlers() /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Signals.cpp:97:5
 #3 0x0000000009e5fdfe llvm::sys::CleanupOnSignal(unsigned long) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/Unix/Signals.inc:362:1
 #4 0x0000000009d6fa4e (anonymous namespace)::CrashRecoveryContextImpl::HandleCrash(int, unsigned long) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/CrashRecoveryContext.cpp:0:7
 #5 0x0000000009d6fe03 CrashRecoverySignalHandler(int) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Support/CrashRecoveryContext.cpp:390:1
 #6 0x00007f60568c08e0 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x138e0)
 #7 0x00007f6056389e71 raise ./signal/../sysdeps/unix/sysv/linux/raise.c:50:1
 #8 0x00007f6056373536 abort ./stdlib/abort.c:81:7
 #9 0x00007f605637341f get_sysdep_segment_value ./intl/loadmsgcat.c:509:8
#10 0x00007f605637341f _nl_load_domain ./intl/loadmsgcat.c:970:34
#11 0x00007f60563827f2 (/lib/x86_64-linux-gnu/libc.so.6+0x357f2)
#12 0x000000000b57ce6e llvm::DwarfCompileUnit::constructInlinedScopeDIE(llvm::LexicalScope*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:0:3
#13 0x000000000b57cb6c llvm::DwarfCompileUnit::constructScopeDIE(llvm::LexicalScope*, llvm::DIE&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:538:10
#14 0x000000000b57d3ae llvm::DwarfCompileUnit::createAndAddScopeChildren(llvm::LexicalScope*, llvm::DIE&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:1072:7
#15 0x000000000b57efa1 llvm::DwarfCompileUnit::constructSubprogramScopeDIE(llvm::DISubprogram const*, llvm::LexicalScope*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:1023:14
#16 0x000000000b52bc0d llvm::DwarfDebug::endFunctionImpl(llvm::MachineFunction const*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp:2301:50
#17 0x000000000b505ab2 llvm::DebugHandlerBase::endFunction(llvm::MachineFunction const*) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/DebugHandlerBase.cpp:0:5
#18 0x000000000b4e1f4c llvm::AsmPrinter::emitFunctionBody() /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp:1505:3
#19 0x0000000007ed7113 llvm::X86AsmPrinter::runOnMachineFunction(llvm::MachineFunction&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/Target/X86/X86AsmPrinter.cpp:82:3
#20 0x0000000008a96477 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/CodeGen/MachineFunctionPass.cpp:72:8
#21 0x0000000009165d5e llvm::FPPassManager::runOnFunction(llvm::Function&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1439:23
#22 0x000000000916abb2 llvm::FPPassManager::runOnModule(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1485:16
#23 0x0000000009166649 (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1554:23
#24 0x00000000091661bd llvm::legacy::PassManagerImpl::run(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:542:16
#25 0x000000000916ae91 llvm::legacy::PassManager::run(llvm::Module&) /usr/local/google/home/blaikie/dev/llvm/src/llvm/lib/IR/LegacyPassManager.cpp:1681:3

But that's actually not the original situation I was investigating, so maybe it's another bug? (possibly same root cause, though)

Though the original issue I was investigating was from invalid .debug_gnu_pubnames (DIE offsets of 0, which prematurely terminates the .debug_gnu_pubnames list and produces a warning from llvm-dwarfdump), and by adding assert(Offset) in llvm::DIE::getOffset (initially I added it in the pubnames emission code, but generally it should be true that the offset of a DIE should not be queried until it's been computed, and no DIE has offset zero, because the offset is from the start of the CU contribution - so the headers come first and mean the DIE offset is always non-zero). It seems I still don't have a small reproduction for that exact issue, though.

The second test case provided above, that produces multiple DWO CUs can be reproduced as follows:

$ clang++-tot ab-multi-cu.ll -gsplit-dwarf -c -O1 && llvm-dwarfdump-tot ab-multi-cu.o
ab-multi-cu.o:  file format elf64-x86-64

.debug_info contents:
0x00000000: Compile Unit: length = 0x00000053, format = DWARF32, version = 0x0004, abbr_offset = 0x0000, addr_size = 0x08 (next unit at 0x00000057)

0x0000000b: DW_TAG_compile_unit
              DW_AT_stmt_list   (0x00000000)
              DW_AT_comp_dir    ("/usr/local/google/home/blaikie/dev/scratch")
              DW_AT_GNU_pubnames        (true)
              DW_AT_GNU_dwo_name        ("ab-multi-cu.dwo")
              DW_AT_GNU_dwo_id  (0xe9d1660e0395035d)
              DW_AT_low_pc      (0x0000000000000000)
              DW_AT_high_pc     (0x000000000000000d)
              DW_AT_GNU_addr_base       (0x00000000)

0x00000030:   DW_TAG_subprogram
                DW_AT_low_pc    (0x0000000000000000)
                DW_AT_high_pc   (0x000000000000000d)
                DW_AT_name      ("main")

0x00000041:     DW_TAG_inlined_subroutine
                  DW_AT_abstract_origin (0x000000000000007b "f1")
                  DW_AT_low_pc  (0x0000000000000004)
                  DW_AT_high_pc (0x0000000000000009)
                  DW_AT_call_file       ("/usr/local/google/home/blaikie/dev/scratch/b.cpp")
                  DW_AT_call_line       (3)
                  DW_AT_call_column     (0x03)

0x00000055:     NULL

0x00000056:   NULL
0x00000057: Compile Unit: length = 0x00000027, format = DWARF32, version = 0x0004, abbr_offset = 0x0000, addr_size = 0x08 (next unit at 0x00000082)

0x00000062: DW_TAG_compile_unit
              DW_AT_stmt_list   (0x0000004c)
              DW_AT_comp_dir    ("/usr/local/google/home/blaikie/dev/scratch")
              DW_AT_GNU_pubnames        (true)
              DW_AT_GNU_dwo_name        ("ab-multi-cu.dwo")
              DW_AT_GNU_dwo_id  (0xeb227ecffcd242c1)
              DW_AT_GNU_addr_base       (0x00000000)

0x0000007b:   DW_TAG_subprogram
                DW_AT_name      ("f1")
                DW_AT_inline    (DW_INL_inlined)

0x00000081:   NULL

In this case the second CU is OK (for the cross-CU inlining that's occurred), but it shouldn't have any DWO_* attributes, and the .dwo file shouldn't have a second CU in it (which it does after this patch) because Split DWARF doesn't have a good way to describe cross-CU references correctly once the CU is packaged in a dwp file. (this behavior is meant to be controlled by the SplitDwarfCrossCUReferences flag.

This does look like a stack trace for D124892.

[dblaikie]

Right, so this does only reproduce with Split DWARF + Split DWARF inlining + ThinLTO, I think. The CUs for the imported units Split DWARF inlining are being designated as split units, but then split units aren't actually emitted (because of the single-unit limitation of Split DWARF) and so the pubnames are emitted for DIEs that were never layed out/emitted, by the looks of it, so they get an offset of zero, which looks like/creates a truncated pubnames contribution.

Trying to make a small reproducer.

[dblaikie]

In D113741#3538747, @dblaikie wrote:

Right, so this does only reproduce with Split DWARF + Split DWARF inlining + ThinLTO, I think. The CUs for the imported units Split DWARF inlining are being designated as split units, but then split units aren't actually emitted (because of the single-unit limitation of Split DWARF) and so the pubnames are emitted for DIEs that were never layed out/emitted, by the looks of it, so they get an offset of zero, which looks like/creates a truncated pubnames contribution.

Trying to make a small reproducer.

OK, think I've got something in IR at least. Looks like it could be reproduced from source + LTO probably. With an assertions-enabled build of LLVM/Clang:

define void @f1() !dbg !13 {
lbl:
  ret void, !dbg !16
}

define void @f2() !dbg !22 {
lbl:
  ret void, !dbg !23
}

!llvm.dbg.cu = !{!0, !2, !10}
!llvm.module.flags = !{!12}

!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, emissionKind: FullDebug)
!1 = !DIFile(filename: "a.cc", directory: "")
!2 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !3, emissionKind: FullDebug, imports: !4)
!3 = !DIFile(filename: "b.cc", directory: "")
!4 = !{!5}
!5 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !6, entity: !7)
!6 = !DISubprogram(scope: null, spFlags: DISPFlagOptimized)
!7 = !DINamespace(scope: !2)
!8 = !DISubroutineType(types: !9)
!9 = !{}
!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, emissionKind: FullDebug)
!11 = !DIFile(filename: "c.cc", directory: "")
!12 = !{i32 2, !"Debug Info Version", i32 3}
!13 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
!16 = !DILocation(line: 0, scope: !17, inlinedAt: !18)
!17 = distinct !DISubprogram(scope: null, unit: !10)
!18 = !DILocation(line: 0, scope: !21)
!21 = !DILexicalBlockFile(scope: !13, discriminator: 0)
!22 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
!23 = !DILocation(line: 0, scope: !24, inlinedAt: !25)
!24 = distinct !DISubprogram(scope: null, unit: !2)
!25 = !DILocation(line: 0, scope: !22)

clang -cc1 -emit-obj  -split-dwarf-file x.dwo x.ll

Does that work for you?

[krisb]

In D113741#3543156, @dblaikie wrote:

In D113741#3538747, @dblaikie wrote:

Right, so this does only reproduce with Split DWARF + Split DWARF inlining + ThinLTO, I think. The CUs for the imported units Split DWARF inlining are being designated as split units, but then split units aren't actually emitted (because of the single-unit limitation of Split DWARF) and so the pubnames are emitted for DIEs that were never layed out/emitted, by the looks of it, so they get an offset of zero, which looks like/creates a truncated pubnames contribution.

Trying to make a small reproducer.

OK, think I've got something in IR at least. Looks like it could be reproduced from source + LTO probably. With an assertions-enabled build of LLVM/Clang:

define void @f1() !dbg !13 {
lbl:
  ret void, !dbg !16
}

define void @f2() !dbg !22 {
lbl:
  ret void, !dbg !23
}

!llvm.dbg.cu = !{!0, !2, !10}
!llvm.module.flags = !{!12}

!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, emissionKind: FullDebug)
!1 = !DIFile(filename: "a.cc", directory: "")
!2 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !3, emissionKind: FullDebug, imports: !4)
!3 = !DIFile(filename: "b.cc", directory: "")
!4 = !{!5}
!5 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !6, entity: !7)
!6 = !DISubprogram(scope: null, spFlags: DISPFlagOptimized)
!7 = !DINamespace(scope: !2)
!8 = !DISubroutineType(types: !9)
!9 = !{}
!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, emissionKind: FullDebug)
!11 = !DIFile(filename: "c.cc", directory: "")
!12 = !{i32 2, !"Debug Info Version", i32 3}
!13 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
!16 = !DILocation(line: 0, scope: !17, inlinedAt: !18)
!17 = distinct !DISubprogram(scope: null, unit: !10)
!18 = !DILocation(line: 0, scope: !21)
!21 = !DILexicalBlockFile(scope: !13, discriminator: 0)
!22 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
!23 = !DILocation(line: 0, scope: !24, inlinedAt: !25)
!24 = distinct !DISubprogram(scope: null, unit: !2)
!25 = !DILocation(line: 0, scope: !22)

clang -cc1 -emit-obj  -split-dwarf-file x.dwo x.ll

Does that work for you?

Yes, I was able to reproduce the issue. Thank you, @dblaikie!

[krisb]

In D113741#3543156, @dblaikie wrote:

In D113741#3538747, @dblaikie wrote:

Right, so this does only reproduce with Split DWARF + Split DWARF inlining + ThinLTO, I think. The CUs for the imported units Split DWARF inlining are being designated as split units, but then split units aren't actually emitted (because of the single-unit limitation of Split DWARF) and so the pubnames are emitted for DIEs that were never layed out/emitted, by the looks of it, so they get an offset of zero, which looks like/creates a truncated pubnames contribution.

Trying to make a small reproducer.

OK, think I've got something in IR at least. Looks like it could be reproduced from source + LTO probably. With an assertions-enabled build of LLVM/Clang:

define void @f1() !dbg !13 {
lbl:
  ret void, !dbg !16
}

define void @f2() !dbg !22 {
lbl:
  ret void, !dbg !23
}

!llvm.dbg.cu = !{!0, !2, !10}
!llvm.module.flags = !{!12}

!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, emissionKind: FullDebug)
!1 = !DIFile(filename: "a.cc", directory: "")
!2 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !3, emissionKind: FullDebug, imports: !4)
!3 = !DIFile(filename: "b.cc", directory: "")
!4 = !{!5}
!5 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !6, entity: !7)
!6 = !DISubprogram(scope: null, spFlags: DISPFlagOptimized)
!7 = !DINamespace(scope: !2)
!8 = !DISubroutineType(types: !9)
!9 = !{}
!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, emissionKind: FullDebug)
!11 = !DIFile(filename: "c.cc", directory: "")
!12 = !{i32 2, !"Debug Info Version", i32 3}
!13 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
!16 = !DILocation(line: 0, scope: !17, inlinedAt: !18)
!17 = distinct !DISubprogram(scope: null, unit: !10)
!18 = !DILocation(line: 0, scope: !21)
!21 = !DILexicalBlockFile(scope: !13, discriminator: 0)
!22 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
!23 = !DILocation(line: 0, scope: !24, inlinedAt: !25)
!24 = distinct !DISubprogram(scope: null, unit: !2)
!25 = !DILocation(line: 0, scope: !22)

clang -cc1 -emit-obj  -split-dwarf-file x.dwo x.ll

Does that work for you?

Okay, I got what's wrong.

auto getCUs = [this](const DIScope *S, DwarfCompileUnit *const *TheCU) {
  if (auto *LScope = dyn_cast_or_null<DILocalScope>(S)) {
    auto CUIt = SPCUsMap.find(LScope->getSubprogram());
    if (CUIt != SPCUsMap.end()) {
      auto &CUs = CUIt->second;
      return llvm::make_range(CUs.begin(), CUs.end());
    }
  }
  return llvm::make_range(TheCU, TheCU + 1);
};

Doing this I was assuming that if SP is not in SPCUsMap, we should fall back to current CU and emit a local entity anyway.
But actually we should skip emitting any local entities if we haven't emitted their parent subprogram by this time. This corresponds to previous behavior of DwarfDebug/DwarfCompileUnit.

I'll update this patch with a fix and a test soon (at least for history).
D125693 doesn't suffer from the same issue, as we can decide whether we need to create a local entity while emitting its parent subprogram (so when we skip emission of a subprogram, we also skip emission of all its local entities).

[krisb]

• Skip emission of local entities if its parent is not in SPCUsMap.
• Skip emission of local entities for CUs if includeMinimalInlineScopes() == true.
• Add test llvm/test/DebugInfo/Generic/split-dwarf-local-import2.ll.
• Fix llvm/test/Instrumentation/InstrProfiling/debug-info-correlate.ll

(add DILocaltion to force emission of DISubprogram as we no longer emit local entities (including static locals)
for empty SPs (i.e. w/o location info).

[dblaikie]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1405

Rather than filtering here - could we avoid this work by not putting minimal-inline-scope CUs in the SPCUsMap values in the first place?

1408

Wouldn't this return a dangling reference, since CUs is local to this function?

[krisb]

• Apply review comments.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1408

Right. Second time 'make_range' makes me thinking it's creating a copy here. My bad. Fixed (might be not the best way & best naming, but should be ok for a patch that, likely, will be abandoned in favor of D125693).

[dblaikie]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1408

patch that, likely, will be abandoned in favor of D125693

Not sure, I think this patch might still be a good direction - without the IR changes necessary for D125693. Does mean local types that end up unreferenced (because their uses are optimized awya) could be eliminated, which is good and bad (it's something we favor for non-local types for size reasons, but arguably we try more to keep all the local names for scoping/shadowing reasons - though it's not a terribly consistent principle)

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.h

318

Why does this map from one subprogram to multiple CUs? Shouldn't we be only placing local types/static variables in one CU (on the abstract origin SP, which should be singular)?

[krisb]

Add tests:

• llvm/test/DebugInfo/Generic/split-dwarf-local-import-fulllto.ll (temporary)
• llvm/test/DebugInfo/Generic/split-dwarf-local-import3.ll

[krisb]

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

1408

The main problem with this patch is that in some cases function-local declarations may be placed incorrectly, and moreover, their placement may not be deterministic, depending on order of functions, as an example.
And unfortunately, I do not see good options to fix/improve this w/o IR changes.

The problem comes from collectLocalScopesWithDeclsFromCU() and a necessity to predict which lexical blocks are non-empty and require to be emitted (so later we can place local declarations to them).

collectLocalScopesWithDeclsFromCU() fills in a map that is a member of DwarfCompileUnit and it does this iff a DwarfCompileUnit is created. So, LocalScopesWithLocalDecls describes only local scopes belong to this CU.

But if to talk about split-dwarf, a subprogram from one CU (source CU) may be inlined into another CU (destination CU). So, when we emit such a subprogram we will check LocalScopesWithLocalDecls of destination CU which knows nothing about local entities for the subprogram. Moreover, source CU for this subprogram may be never emitted (added split-dwarf-local-import3.ll to demonstrate this).

We'd make LocalScopesWithLocalDecls a member of DwarfDebug and fill it in in beginModule(), but this looks a kind of weird.

It'd be good also to address the issue described in collectLocalScopesWithDeclsFromCU() for //used/ types, but it'll require additional computations (like, iterating over all the instructions, find all the used types and record their scopes to be sure later we can place that type correctly).

I do not like all this workarounds including the necessity to track mapping between emitted subprogram DIE and its parent CU. The approach splits emission of local declarations between 3 steps and it's hard to maintain everything in sync.

D125693 requires changes in IR, but proposes a far more cleaner emission flow.
If IR changes from D125691 is something that should be strongly avoided (but I found multiple places where comments mention function-local import should be tracked by a subprogram, not by a compile unit), I'll fix/improve whatever I can for this patch, but I'd like to consider D125693 as a next (second) step then.

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.h

318

We are returning to the question about multiple CUs in a DWO we've discussed a couple months ago. It still isn't clear whether this case is valid/useful for someone (e.g. split-dwarf+FullLTO).

I've temporarily added 'split-dwarf-local-import-fulllto.ll' test that demonstrates this case and backend's behavior before (see CHECK prefix) and after (CHECK-NEW prefix) this patch. (Basically, llvm/test/DebugInfo/X86/split-dwarf-cross-unit-reference.ll also tests this case).

[krisb]

Ping?

[ellis]

Seems fine to me, but I'm not quite familiar enough with this area to give this an accept.

llvm/test/Instrumentation/InstrProfiling/debug-info-correlate.ll

23

Why is this test changed?

[krisb]

llvm/test/Instrumentation/InstrProfiling/debug-info-correlate.ll

23

Answered at https://reviews.llvm.org/D125693#inline-1240997.

[krisb]

Abandon in favor of https://discourse.llvm.org/t/rfc-dwarfdebug-fix-and-improve-handling-imported-entities-types-and-static-local-in-subprogram-and-lexical-block-scopes/68544.

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.h

class DwarfCompileUnit final : public DwarfUnit {
   DwarfCompileUnit *Skeleton = nullptr;
 
   /// The start of the unit within its section.
   MCSymbol *LabelBegin = nullptr;
 
   /// The start of the unit macro info within macro section.
   MCSymbol *MacroLabelBegin;
 
-  using ImportedEntityList = SmallVector<const MDNode *, 8>;
-  using ImportedEntityMap = DenseMap<const MDNode *, ImportedEntityList>;
-
-  ImportedEntityMap ImportedEntities;
-
   /// GlobalNames - A map of globally visible named entities for this unit.
   StringMap<const DIE *> GlobalNames;
 
   /// GlobalTypes - A map of globally visible types for this unit.
   StringMap<const DIE *> GlobalTypes;
 
   // List of ranges for a given compile unit.
   SmallVector<RangeSpan, 2> CURanges;
 
   // The base address of this unit, if any. Used for relative references in
   // ranges/locs.
   const MCSymbol *BaseAddress = nullptr;
 
-  DenseMap<const MDNode *, DIE *> AbstractSPDies;
+  DenseMap<const DILocalScope *, DIE *> LocalScopeDIEs;
+  DenseMap<const DILocalScope *, DIE *> AbstractLocalScopeDIEs;
   DenseMap<const DINode *, std::unique_ptr<DbgEntity>> AbstractEntities;
 
+  /// LocalScopesWithLocalDecls - A list of non-empty local scopes
+  /// (with declaraions of static locals, function-local types, or imports).
+  SmallPtrSet<const DILocalScope *, 8> LocalScopesWithLocalDecls;
+
   /// DWO ID for correlating skeleton and split units.
   uint64_t DWOId = 0;
 
   const DIFile *LastFile = nullptr;
   unsigned LastFileID;
 
   /// Construct a DIE for the given DbgVariable without initializing the
   /// DbgVariable's DIE reference.
   DIE *constructVariableDIEImpl(const DbgVariable &DV, bool Abstract);
 
   bool isDwoUnit() const override;
 
-  DenseMap<const MDNode *, DIE *> &getAbstractSPDies() {
+  DenseMap<const DILocalScope *, DIE *> &getAbstractScopeDIEs() {
     if (isDwoUnit() && !DD->shareAcrossDWOCUs())
-      return AbstractSPDies;
-    return DU->getAbstractSPDies();
+      return AbstractLocalScopeDIEs;
+    return DU->getAbstractScopeDIEs();
   }
 
   DenseMap<const DINode *, std::unique_ptr<DbgEntity>> &getAbstractEntities() {
     if (isDwoUnit() && !DD->shareAcrossDWOCUs())
       return AbstractEntities;
     return DU->getAbstractEntities();
   }
 
public:
   /// DW_FORM_addr only.
   void addLocalLabelAddress(DIE &Die, dwarf::Attribute Attribute,
                             const MCSymbol *Label);
 
   DwarfCompileUnit &getCU() override { return *this; }
 
   unsigned getOrCreateSourceID(const DIFile *File) override;
 
-  void addImportedEntity(const DIImportedEntity* IE) {
-    DIScope *Scope = IE->getScope();
-    assert(Scope && "Invalid Scope encoding!");
-    if (!isa<DILocalScope>(Scope))
-      // No need to add imported enities that are not local declaration.
-      return;
-
-    auto *LocalScope = cast<DILocalScope>(Scope)->getNonLexicalBlockFileScope();
-    ImportedEntities[LocalScope].push_back(IE);
-  }
-
   /// addRange - Add an address range to the list of ranges for this unit.
   void addRange(RangeSpan Range);
 
   void attachLowHighPC(DIE &D, const MCSymbol *Begin, const MCSymbol *End);
 
   /// Find DIE for the given subprogram and attach appropriate
   /// DW_AT_low_pc and DW_AT_high_pc attributes. If there are global
   /// variables in this scope then create and insert DIEs for these
public:
   DIE *constructVariableDIE(DbgVariable &DV, const LexicalScope &Scope,
                             DIE *&ObjectPointer);
 
   /// Construct a DIE for the given DbgLabel.
   DIE *constructLabelDIE(DbgLabel &DL, const LexicalScope &Scope);
 
   void createBaseTypeDIEs();
 
+  DIE *findLocalScopeDIE(const DIScope *S);
+  DIE *getOrCreateContextDIE(const DIScope *Ty) override;
+
   /// Construct a DIE for this subprogram scope.
   DIE &constructSubprogramScopeDIE(const DISubprogram *Sub,
                                    LexicalScope *Scope);
 
   DIE *createAndAddScopeChildren(LexicalScope *Scope, DIE &ScopeDIE);
 
   void constructAbstractSubprogramScopeDIE(LexicalScope *Scope);
 
DIE &constructCallSiteEntryDIE(DIE &ScopeDIE, const DISubprogram *CalleeSP,
                                  const MCSymbol *CallAddr, unsigned CallReg);
   /// Construct call site parameter DIEs for the \p CallSiteDIE. The \p Params
   /// were collected by the \ref collectCallSiteParameters.
   /// Note: The order of parameters does not matter, since debuggers recognize
   ///       call site parameters by the DW_AT_location attribute.
   void constructCallSiteParmEntryDIEs(DIE &CallSiteDIE,
                                       SmallVector<DbgCallSiteParam, 4> &Params);
 
-  /// Construct import_module DIE.
-  DIE *constructImportedEntityDIE(const DIImportedEntity *Module);
+  /// Construct DIE for an imported entity.
+  DIE *createImportedEntityDIE(const DIImportedEntity *IE);
+  void createAndAddImportedEntityDIE(const DIImportedEntity *IE);
 
   void finishSubprogramDefinition(const DISubprogram *SP);
   void finishEntityDefinition(const DbgEntity *Entity);
 
   /// Find abstract variable associated with Var.
   using InlinedEntity = DbgValueHistoryMap::InlinedEntity;
   DbgEntity *getExistingAbstractEntity(const DINode *Node);
   void createAbstractEntity(const DINode *Node, LexicalScope *Scope);
public:
   const MCSymbol *getBaseAddress() const { return BaseAddress; }
 
   uint64_t getDWOId() const { return DWOId; }
   void setDWOId(uint64_t DwoId) { DWOId = DwoId; }
 
   bool hasDwarfPubSections() const;
 
   void addBaseTypeRef(DIEValueList &Die, int64_t Idx);
+
+  void recordLocalScopeWithDecls(const DILocalScope *S) {
+    LocalScopesWithLocalDecls.insert(S);
+  }
 };
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_CODEGEN_ASMPRINTER_DWARFCOMPILEUNIT_H

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp

void DwarfCompileUnit::constructScopeDIE(LexicalScope *Scope,
 
   // Early exit when we know the scope DIE is going to be null.
   if (DD->isLexicalScopeDIENull(Scope))
     return;
 
   // Emit lexical blocks.
   DIE *ScopeDIE = constructLexicalScopeDIE(Scope);
   assert(ScopeDIE && "Scope DIE should not be null.");
-
   ParentScopeDIE.addChild(ScopeDIE);
+
+  // Track abstract and concrete lexical block scopes.
+  if (Scope->isAbstractScope()) {
+    assert(!getAbstractScopeDIEs().count(DS) &&
+           "Abstract DIE for this scope exists!");
+    getAbstractScopeDIEs()[DS] = ScopeDIE;
+  } else if (!Scope->getInlinedAt()) {
+    assert(!LocalScopeDIEs.count(DS) && "Concrete DIE for this scope exists!");
+    LocalScopeDIEs[DS] = ScopeDIE;
+  }
+
   createAndAddScopeChildren(Scope, *ScopeDIE);
 }
 
 void DwarfCompileUnit::addScopeRangeList(DIE &ScopeDIE,
                                          SmallVector<RangeSpan, 2> Range) {
 
   HasRangeLists = true;
 

 // This scope represents inlined body of a function. Construct DIE to
 // represent this concrete inlined copy of the function.
 DIE *DwarfCompileUnit::constructInlinedScopeDIE(LexicalScope *Scope) {
   assert(Scope->getScopeNode());
   auto *DS = Scope->getScopeNode();
   auto *InlinedSP = getDISubprogram(DS);
   // Find the subprogram's DwarfCompileUnit in the SPMap in case the subprogram
   // was inlined from another compile unit.
-  DIE *OriginDIE = getAbstractSPDies()[InlinedSP];
+  DIE *OriginDIE = getAbstractScopeDIEs()[InlinedSP];
   assert(OriginDIE && "Unable to find original DIE for an inlined subprogram.");
 
   auto ScopeDIE = DIE::get(DIEValueAllocator, dwarf::DW_TAG_inlined_subroutine);
   addDIEEntry(*ScopeDIE, dwarf::DW_AT_abstract_origin, *OriginDIE);
 
   attachRangesOrLowHighPC(*ScopeDIE, Scope->getRanges());
 
   // Add the call site information to the DIE.

 
 DIE &DwarfCompileUnit::constructSubprogramScopeDIE(const DISubprogram *Sub,
                                                    LexicalScope *Scope) {
   DIE &ScopeDIE = updateSubprogramScopeDIE(Sub);
 
   if (Scope) {
     assert(!Scope->getInlinedAt());
     assert(!Scope->isAbstractScope());
+
+    // Track record in which CU this subprogram gets emitted.
+    DD->recordSPContainingCU(Sub, this);
+
+    // Record the subrogram scope before creating child entities.
+    assert(!LocalScopeDIEs.count(Sub) &&
+           "Concrete DIE for the subprogram exists!");
+    LocalScopeDIEs[Sub] = &ScopeDIE;
+
     // Collect lexical scope children first.
     // ObjectPointer might be a local (non-argument) local variable if it's a
     // block's synthetic this pointer.
     if (DIE *ObjectPointer = createAndAddScopeChildren(Scope, ScopeDIE))
       addDIEEntry(ScopeDIE, dwarf::DW_AT_object_pointer, *ObjectPointer);
   }
 
   // If this is a variadic function, add an unspecified parameter.
DIE *DwarfCompileUnit::createAndAddScopeChildren(LexicalScope *Scope,
   for (auto &DV : Vars.Args)
     ScopeDIE.addChild(constructVariableDIE(*DV.second, *Scope, ObjectPointer));
 
   // Emit local variables.
   auto Locals = sortLocalVars(Vars.Locals);
   for (DbgVariable *DV : Locals)
     ScopeDIE.addChild(constructVariableDIE(*DV, *Scope, ObjectPointer));
 
-  // Emit imported entities (skipped in gmlt-like data).
-  if (!includeMinimalInlineScopes()) {
-    for (const auto *IE : ImportedEntities[Scope->getScopeNode()])
-      ScopeDIE.addChild(constructImportedEntityDIE(cast<DIImportedEntity>(IE)));
-  }
-
   // Emit labels.
   for (DbgLabel *DL : DU->getScopeLabels().lookup(Scope))
     ScopeDIE.addChild(constructLabelDIE(*DL, *Scope));
 
   // Emit inner lexical scopes.
-  auto needToEmitLexicalScope = [this](LexicalScope *LS) {
+  auto needToEmitLexicalScope = [this](LexicalScope *LS) -> bool {
     if (isa<DISubprogram>(LS->getScopeNode()))
       return true;
     auto Vars = DU->getScopeVariables().lookup(LS);
     if (!Vars.Args.empty() || !Vars.Locals.empty())
       return true;
-    if (!includeMinimalInlineScopes() &&
-        !ImportedEntities[LS->getScopeNode()].empty())
-      return true;
-    return false;
+    return LocalScopesWithLocalDecls.count(LS->getScopeNode());
   };
   for (LexicalScope *LS : Scope->getChildren()) {
     // If the lexical block doesn't have non-scope children, skip
     // its emission and put its children directly to the parent scope.
     if (needToEmitLexicalScope(LS))
       constructScopeDIE(LS, ScopeDIE);
     else
       createAndAddScopeChildren(LS, ScopeDIE);
   }
 
   return ObjectPointer;
 }
 
 void DwarfCompileUnit::constructAbstractSubprogramScopeDIE(
     LexicalScope *Scope) {
-  DIE *&AbsDef = getAbstractSPDies()[Scope->getScopeNode()];
-  if (AbsDef)
-    return;
 
   auto *SP = cast<DISubprogram>(Scope->getScopeNode());
+  if (getAbstractScopeDIEs().count(SP))
+    return;
 
   DIE *ContextDIE;
   DwarfCompileUnit *ContextCU = this;
 
   if (includeMinimalInlineScopes())
     ContextDIE = &getUnitDie();
   // Some of this is duplicated from DwarfUnit::getOrCreateSubprogramDIE, with
   // the important distinction that the debug node is not associated with the
   // DIE (since the debug node will be associated with the concrete DIE, if
   // any). It could be refactored to some common utility function.
   else if (auto *SPDecl = SP->getDeclaration()) {
     ContextDIE = &getUnitDie();
     getOrCreateSubprogramDIE(SPDecl);
   } else {
     ContextDIE = getOrCreateContextDIE(SP->getScope());
     // The scope may be shared with a subprogram that has already been
     // constructed in another CU, in which case we need to construct this
     // subprogram in the same CU.
     ContextCU = DD->lookupCU(ContextDIE->getUnitDie());
   }
 
+  // Track record in which CU this subprogram is emitted.
+  DD->recordSPContainingCU(SP, ContextCU);
+
   // Passing null as the associated node because the abstract definition
   // shouldn't be found by lookup.
-  AbsDef = &ContextCU->createAndAddDIE(dwarf::DW_TAG_subprogram, *ContextDIE, nullptr);
-  ContextCU->applySubprogramAttributesToDefinition(SP, *AbsDef);
-  ContextCU->addSInt(*AbsDef, dwarf::DW_AT_inline,
+  DIE &AbsDef = ContextCU->createAndAddDIE(dwarf::DW_TAG_subprogram,
+                                           *ContextDIE, nullptr);
+
+  // Store the DIE before creating children.
+  getAbstractScopeDIEs()[SP] = &AbsDef;
+
+  ContextCU->applySubprogramAttributesToDefinition(SP, AbsDef);
+  ContextCU->addSInt(AbsDef, dwarf::DW_AT_inline,
                      DD->getDwarfVersion() <= 4 ? Optional<dwarf::Form>()
                                                 : dwarf::DW_FORM_implicit_const,
                      dwarf::DW_INL_inlined);
-  if (DIE *ObjectPointer = ContextCU->createAndAddScopeChildren(Scope, *AbsDef))
-    ContextCU->addDIEEntry(*AbsDef, dwarf::DW_AT_object_pointer, *ObjectPointer);
+  if (DIE *ObjectPointer = ContextCU->createAndAddScopeChildren(Scope, AbsDef))
+    ContextCU->addDIEEntry(AbsDef, dwarf::DW_AT_object_pointer, *ObjectPointer);
 }
 
 bool DwarfCompileUnit::useGNUAnalogForDwarf5Feature() const {
   return DD->getDwarfVersion() == 4 && !DD->tuneForLLDB();
 }
 
 dwarf::Tag DwarfCompileUnit::getDwarf5OrGNUTag(dwarf::Tag Tag) const {
   if (!useGNUAnalogForDwarf5Feature())
for (const auto &Param : Params) {
 
     addBlock(*CallSiteDieParam, getDwarf5OrGNUAttr(dwarf::DW_AT_call_value),
              DwarfExpr.finalize());
 
     CallSiteDIE.addChild(CallSiteDieParam);
   }
 }
 
-DIE *DwarfCompileUnit::constructImportedEntityDIE(
-    const DIImportedEntity *Module) {
-  DIE *IMDie = DIE::get(DIEValueAllocator, (dwarf::Tag)Module->getTag());
-  insertDIE(Module, IMDie);
+DIE *DwarfCompileUnit::createImportedEntityDIE(const DIImportedEntity *IE) {
+  DIE *IMDie = DIE::get(DIEValueAllocator, (dwarf::Tag)IE->getTag());
+  insertDIE(IE, IMDie);
+
   DIE *EntityDie;
-  auto *Entity = Module->getEntity();
+  auto *Entity = IE->getEntity();
   if (auto *NS = dyn_cast<DINamespace>(Entity))
     EntityDie = getOrCreateNameSpace(NS);
   else if (auto *M = dyn_cast<DIModule>(Entity))
     EntityDie = getOrCreateModule(M);
-  else if (auto *SP = dyn_cast<DISubprogram>(Entity))
+  else if (auto *SP = dyn_cast<DISubprogram>(Entity)) {
+    // If we have abstract subprogram created, refer it.
+    if (auto *AbsSPDie = getAbstractScopeDIEs().lookup(SP))
+      EntityDie = AbsSPDie;
+    else
       EntityDie = getOrCreateSubprogramDIE(SP);
-  else if (auto *T = dyn_cast<DIType>(Entity))
+  } else if (auto *T = dyn_cast<DIType>(Entity))
     EntityDie = getOrCreateTypeDIE(T);
   else if (auto *GV = dyn_cast<DIGlobalVariable>(Entity))
     EntityDie = getOrCreateGlobalVariableDIE(GV, {});
   else
     EntityDie = getDIE(Entity);
   assert(EntityDie);
-  addSourceLine(*IMDie, Module->getLine(), Module->getFile());
+
+  addSourceLine(*IMDie, IE->getLine(), IE->getFile());
   addDIEEntry(*IMDie, dwarf::DW_AT_import, *EntityDie);
-  StringRef Name = Module->getName();
+  StringRef Name = IE->getName();
   if (!Name.empty())
     addString(*IMDie, dwarf::DW_AT_name, Name);
 
   // This is for imported module with renamed entities (such as variables and
   // subprograms).
-  DINodeArray Elements = Module->getElements();
+  DINodeArray Elements = IE->getElements();
   for (const auto *Element : Elements) {
     if (!Element)
       continue;
-    IMDie->addChild(
-        constructImportedEntityDIE(cast<DIImportedEntity>(Element)));
+    IMDie->addChild(createImportedEntityDIE(cast<DIImportedEntity>(Element)));
   }
-
   return IMDie;
 }
 
+void DwarfCompileUnit::createAndAddImportedEntityDIE(
+    const DIImportedEntity *IE) {
+
+  DIE *ContextDIE = getOrCreateContextDIE(IE->getScope());
+  assert(ContextDIE &&
+         "Could not get or create scope for the imported entity!");
+  if (!ContextDIE)
+    return;
+
+  ContextDIE->addChild(createImportedEntityDIE(IE));
+}
+
 void DwarfCompileUnit::finishSubprogramDefinition(const DISubprogram *SP) {
   DIE *D = getDIE(SP);
-  if (DIE *AbsSPDIE = getAbstractSPDies().lookup(SP)) {
+  if (DIE *AbsSPDIE = getAbstractScopeDIEs().lookup(SP)) {
     if (D)
       // If this subprogram has an abstract definition, reference that
       addDIEEntry(*D, dwarf::DW_AT_abstract_origin, *AbsSPDIE);
   } else {
     assert(D || includeMinimalInlineScopes());
     if (D)
       // And attach the attributes
       applySubprogramAttributesToDefinition(SP, *D);
addString(Die, dwarf::DW_AT_name,
                     "_" + Twine(Btr.BitSize)).toStringRef(Str));
     addUInt(Die, dwarf::DW_AT_encoding, dwarf::DW_FORM_data1, Btr.Encoding);
     // Round up to smallest number of bytes that contains this number of bits.
     addUInt(Die, dwarf::DW_AT_byte_size, None, divideCeil(Btr.BitSize, 8));
 
     Btr.Die = &Die;
   }
 }
+
+static DIE *
+findLocalScopeDIE(const DILocalScope *LS,
+                  DenseMap<const DILocalScope *, DIE *> &ScopeDIEs) {
+  DIE *ScopeDIE = ScopeDIEs.lookup(LS);
+  if (isa<DISubprogram>(LS) && !ScopeDIE)
+    return nullptr;
+  if (!ScopeDIE)
+    return findLocalScopeDIE(cast<DILocalScope>(LS->getScope()), ScopeDIEs);
+  return ScopeDIE;
+}
+
+DIE *DwarfCompileUnit::findLocalScopeDIE(const DIScope *S) {
+  auto *LScope = dyn_cast_or_null<DILocalScope>(S);
+  if (!LScope)
+    return nullptr;
+
+  // Check if we have an abstract tree.
+  if (getAbstractScopeDIEs().count(LScope->getSubprogram()))
+    return ::findLocalScopeDIE(LScope, getAbstractScopeDIEs());
+
+  return ::findLocalScopeDIE(LScope, LocalScopeDIEs);
+}
+
+DIE *DwarfCompileUnit::getOrCreateContextDIE(const DIScope *Context) {
+  if (auto *LScope = dyn_cast_or_null<DILocalScope>(Context)) {
+    if (DIE *ScopeDIE = findLocalScopeDIE(LScope))
+      return ScopeDIE;
+
+    // If nothing was found, fall back to DISubprogram and let
+    // DwarfUnit::getOrCreateContextDIE() create a new DIE for it.
+    Context = LScope->getSubprogram();
+  }
+  return DwarfUnit::getOrCreateContextDIE(Context);
+}

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.h

class DwarfDebug : public DebugHandlerBase {
 
   /// Collection of abstract variables/labels.
   SmallVector<std::unique_ptr<DbgEntity>, 64> ConcreteEntities;
 
   /// Collection of DebugLocEntry. Stored in a linked list so that DIELocLists
   /// can refer to them in spite of insertions into this list.
   DebugLocStream DebugLocs;
 
+  /// Maps subprogram to its containing DwarfCompileUnit(s).
+  DenseMap<const DISubprogram *,
+           SetVector<DwarfCompileUnit *, SmallVector<DwarfCompileUnit *, 2>,
+                     SmallPtrSet<DwarfCompileUnit *, 2>>>
+      SPCUsMap;
+
+  using SubprogramSetVector =
+      SetVector<const DISubprogram *, SmallVector<const DISubprogram *, 16>,
+                SmallPtrSet<const DISubprogram *, 16>>;
+
   /// This is a collection of subprogram MDNodes that are processed to
   /// create DIEs.
-  SetVector<const DISubprogram *, SmallVector<const DISubprogram *, 16>,
-            SmallPtrSet<const DISubprogram *, 16>>
-      ProcessedSPNodes;
+  SubprogramSetVector ProcessedSPNodes;
 
   /// If nonnull, stores the current machine function we're processing.
   const MachineFunction *CurFn = nullptr;
 
   /// If nonnull, stores the CU in which the previous subprogram was contained.
   const DwarfCompileUnit *PrevCU = nullptr;
 
   /// As an optimization, there is no need to emit an entry in the directory
private:
   void addGnuPubAttributes(DwarfCompileUnit &U, DIE &D) const;
 
   /// Create new DwarfCompileUnit for the given metadata node with tag
   /// DW_TAG_compile_unit.
   DwarfCompileUnit &getOrCreateDwarfCompileUnit(const DICompileUnit *DIUnit);
   void finishUnitAttributes(const DICompileUnit *DIUnit,
                             DwarfCompileUnit &NewCU);
 
-  /// Construct imported_module or imported_declaration DIE.
-  void constructAndAddImportedEntityDIE(DwarfCompileUnit &TheCU,
-                                        const DIImportedEntity *N);
-
   /// Register a source line with debug info. Returns the unique
   /// label that was emitted and which provides correspondence to the
   /// source line list.
   void recordSourceLine(unsigned Line, unsigned Col, const MDNode *Scope,
                         unsigned Flags);
 
   /// Populate LexicalScope entries with variables' info.
   void collectEntityInfo(DwarfCompileUnit &TheCU, const DISubprogram *SP,
public:
 
   static void emitDebugLocValue(const AsmPrinter &AP, const DIBasicType *BT,
                                 const DbgValueLoc &Value,
                                 DwarfExpression &DwarfExpr);
 
   /// If the \p File has an MD5 checksum, return it as an MD5Result
   /// allocated in the MCContext.
   Optional<MD5::MD5Result> getMD5AsBytes(const DIFile *File) const;
+
+  /// Track records of CUs containg the given subprogram.
+  void recordSPContainingCU(const DISubprogram *SP, DwarfCompileUnit *CU);
 };
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_CODEGEN_ASMPRINTER_DWARFDEBUG_H

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

void DwarfDebug::addSubprogramNames(const DICompileUnit &CU,
 
   if (SP->getName() != "")
     addAccelName(CU, SP->getName(), Die);
 
   // If the linkage name is different than the name, go ahead and output that as
   // well into the name table. Only do that if we are going to actually emit
   // that name.
   if (SP->getLinkageName() != "" && SP->getName() != SP->getLinkageName() &&
-      (useAllLinkageNames() || InfoHolder.getAbstractSPDies().lookup(SP)))
+      (useAllLinkageNames() || InfoHolder.getAbstractScopeDIEs().lookup(SP)))
     addAccelName(CU, SP->getLinkageName(), Die);
 
   // If this is an Objective-C selector name add it to the ObjC accelerator
   // too.
   if (isObjCClass(SP->getName())) {
     StringRef Class, Category;
     getObjCClassCategory(SP->getName(), Class, Category);
     addAccelObjC(CU, Class, Die);
if (!DIUnit->getSplitDebugFilename().empty()) {
       // This is a prefabricated skeleton CU.
       dwarf::Attribute attrDWOName = getDwarfVersion() >= 5
                                          ? dwarf::DW_AT_dwo_name
                                          : dwarf::DW_AT_GNU_dwo_name;
       NewCU.addString(Die, attrDWOName, DIUnit->getSplitDebugFilename());
     }
   }
 }
+
+// Collect local scopes that contain any local declarations
+// (excluding local variables) to be sure they will be emitted
+// (see DwarfCompileUnit::createAndAddScopeChildren() for details).
+static void collectLocalScopesWithDeclsFromCU(const DICompileUnit *CUNode,
+                                              DwarfCompileUnit &CU) {
+  auto getLocalScope = [](const DIScope *S) -> const DILocalScope * {
+    if (!S)
+      return nullptr;
+    if (isa<DICommonBlock>(S))
+      S = S->getScope();
+    if (const auto *LScope = dyn_cast_or_null<DILocalScope>(S))
+      return LScope->getNonLexicalBlockFileScope();
+    return nullptr;
+  };
+
+  for (auto *GVE : CUNode->getGlobalVariables())
+    if (auto *LScope = getLocalScope(GVE->getVariable()->getScope()))
+      CU.recordLocalScopeWithDecls(LScope);
+
+  for (auto *Ty : CUNode->getEnumTypes())
+    if (auto *LScope = getLocalScope(Ty->getScope()))
+      CU.recordLocalScopeWithDecls(LScope);
+
+  for (auto *Ty : CUNode->getRetainedTypes())
+    if (DIType *RT = dyn_cast<DIType>(Ty))
+      if (auto *LScope = getLocalScope(RT->getScope()))
+        CU.recordLocalScopeWithDecls(LScope);
+
+  for (auto *IE : CUNode->getImportedEntities())
+    if (auto *LScope = getLocalScope(IE->getScope()))
+      CU.recordLocalScopeWithDecls(LScope);
+
+  // FIXME: We know nothing about local records and typedefs here,
+  // since nothing but local variables (and members of local records)
+  // references them. So that they will be emitted in a first available
+  // parent scope DIE.
+}
+
 // Create new DwarfCompileUnit for the given metadata node with tag
 // DW_TAG_compile_unit.
 DwarfCompileUnit &
 DwarfDebug::getOrCreateDwarfCompileUnit(const DICompileUnit *DIUnit) {
   if (auto *CU = CUMap.lookup(DIUnit))
     return *CU;
 
   CompilationDir = DIUnit->getDirectory();
 
   auto OwnedUnit = std::make_unique<DwarfCompileUnit>(
       InfoHolder.getUnits().size(), DIUnit, Asm, this, &InfoHolder);
   DwarfCompileUnit &NewCU = *OwnedUnit;
   InfoHolder.addUnit(std::move(OwnedUnit));
 
-  for (auto *IE : DIUnit->getImportedEntities())
-    NewCU.addImportedEntity(IE);
-
   // LTO with assembly output shares a single line table amongst multiple CUs.
   // To avoid the compilation directory being ambiguous, let the line table
   // explicitly describe the directory of all files, never relying on the
   // compilation directory.
   if (!Asm->OutStreamer->hasRawTextSupport() || SingleCU)
     Asm->OutStreamer->emitDwarfFile0Directive(
         CompilationDir, DIUnit->getFilename(), getMD5AsBytes(DIUnit->getFile()),
         DIUnit->getSource(), NewCU.getUniqueID());
 
   if (useSplitDwarf()) {
     NewCU.setSkeleton(constructSkeletonCU(NewCU));
     NewCU.setSection(Asm->getObjFileLowering().getDwarfInfoDWOSection());
   } else {
     finishUnitAttributes(DIUnit, NewCU);
     NewCU.setSection(Asm->getObjFileLowering().getDwarfInfoSection());
   }
 
   CUMap.insert({DIUnit, &NewCU});
   CUDieMap.insert({&NewCU.getUnitDie(), &NewCU});
-  return NewCU;
-}
 
-void DwarfDebug::constructAndAddImportedEntityDIE(DwarfCompileUnit &TheCU,
-                                                  const DIImportedEntity *N) {
-  if (isa<DILocalScope>(N->getScope()))
-    return;
-  if (DIE *D = TheCU.getOrCreateContextDIE(N->getScope()))
-    D->addChild(TheCU.constructImportedEntityDIE(N));
+  // Record local scopes, that have some globals (static locals),
+  // imports or types declared within.
+  collectLocalScopesWithDeclsFromCU(DIUnit, NewCU);
+
+  return NewCU;
 }
 
 // Emit all Dwarf sections that should come prior to the content. Create
 // global DIEs and emit initial debug info sections. This is invoked by
 // the target AsmPrinter.
 void DwarfDebug::beginModule(Module *M) {
   DebugHandlerBase::beginModule(M);
 
if (useSplitDwarf())
       InfoHolder.setRnglistsTableBaseSym(
           Asm->createTempSymbol("rnglists_dwo_table_base"));
   }
 
   // Create the symbol that points to the first entry following the debug
   // address table (.debug_addr) header.
   AddrPool.setLabel(Asm->createTempSymbol("addr_table_base"));
   DebugLocs.setSym(Asm->createTempSymbol("loclists_table_base"));
 }
 
 void DwarfDebug::finishEntityDefinitions() {
   for (const auto &Entity : ConcreteEntities) {
     DIE *Die = Entity->getDIE();
     assert(Die);
     // FIXME: Consider the time-space tradeoff of just storing the unit pointer
     // in the ConcreteEntities list, rather than looking it up again here.
     // DIE::getUnit isn't simple - it walks parent pointers, etc.
     DwarfCompileUnit *Unit = CUDieMap.lookup(Die->getUnitDie());
     assert(Unit);
     Unit->finishEntityDefinition(Entity.get());
   }
 }

 void DwarfDebug::endModule() {
   // Terminate the pending line table.
   if (PrevCU)
     terminateLineTable(PrevCU);
   PrevCU = nullptr;
   assert(CurFn == nullptr);
   assert(CurMI == nullptr);
 
+  // Collect global variables info.
   DenseMap<DIGlobalVariable *, SmallVector<DwarfCompileUnit::GlobalExpr, 1>>
       GVMap;
   for (const GlobalVariable &Global : MMI->getModule()->globals()) {
     SmallVector<DIGlobalVariableExpression *, 1> GVs;
     Global.getDebugInfo(GVs);
     for (auto *GVE : GVs)
       GVMap[GVE->getVariable()].push_back({&Global, GVE->getExpression()});
   }
 
+  auto getCUs = [this](const DIScope *S, DwarfCompileUnit **TheCU) {
+    if (auto *LScope = dyn_cast_or_null<DILocalScope>(S)) {
+      SmallVector<DwarfCompileUnit *, 1> CUs;
+      auto CUIt = SPCUsMap.find(LScope->getSubprogram());
+      if (CUIt != SPCUsMap.end()) {
+        // Filter out CUs for which we don't what to emit local entities.
+        std::copy_if(CUIt->second.begin(), CUIt->second.end(),
+                     std::back_inserter(CUs), [](DwarfCompileUnit *CU) {
+                       return !CU->includeMinimalInlineScopes();
+                     });
+      }
+      return llvm::make_range(CUs.begin(), CUs.end());
+    }
+    return llvm::make_range(TheCU, TheCU + 1);
+  };
+
   for (DICompileUnit *CUNode : MMI->getModule()->debug_compile_units()) {
     auto *CU = CUMap.lookup(CUNode);
 
     // If this CU hasn't been emitted yet, create it here unless it is empty.
     if (!CU) {
       // FIXME: Move local imported entities into a list attached to the
       // subprogram, then this search won't be needed and a
       // getImportedEntities().empty() test should go below with the rest.
       bool HasNonLocalImportedEntities = llvm::any_of(
           CUNode->getImportedEntities(), [](const DIImportedEntity *IE) {
             return !isa<DILocalScope>(IE->getScope());
           });
 
       if (!HasNonLocalImportedEntities && CUNode->getEnumTypes().empty() &&
           CUNode->getRetainedTypes().empty() &&
           CUNode->getGlobalVariables().empty() && CUNode->getMacros().empty())
         continue;
 
       CU = &getOrCreateDwarfCompileUnit(CUNode);
     }
 
     // Global Variables.
     for (auto *GVE : CUNode->getGlobalVariables()) {
       // Don't bother adding DIGlobalVariableExpressions listed in the CU if we
-      // already know about the variable and it isn't adding a constant
-      // expression.
+      // already know about the variable and it isn't adding a const expression.
       auto &GVMapEntry = GVMap[GVE->getVariable()];
       auto *Expr = GVE->getExpression();
       if (!GVMapEntry.size() || (Expr && Expr->isConstant()))
         GVMapEntry.push_back({nullptr, Expr});
     }
 
     DenseSet<DIGlobalVariable *> Processed;
     for (auto *GVE : CUNode->getGlobalVariables()) {
       DIGlobalVariable *GV = GVE->getVariable();
       if (Processed.insert(GV).second)
-        CU->getOrCreateGlobalVariableDIE(GV, sortGlobalExprs(GVMap[GV]));
+        for (auto *U : getCUs(GV->getScope(), &CU))
+          U->getOrCreateGlobalVariableDIE(GV, sortGlobalExprs(GVMap[GV]));
     }
 
     for (auto *Ty : CUNode->getEnumTypes())
-      CU->getOrCreateTypeDIE(cast<DIType>(Ty));
+      for (auto *U : getCUs(Ty->getScope(), &CU))
+        U->getOrCreateTypeDIE(cast<DIType>(Ty));
 
-    for (auto *Ty : CUNode->getRetainedTypes())
-      if (DIType *RT = dyn_cast<DIType>(Ty))
-        // There is no point in force-emitting a forward declaration.
-        CU->getOrCreateTypeDIE(RT);
+    for (auto *Ty : CUNode->getRetainedTypes()) {
+      if (DIType *RT = dyn_cast<DIType>(Ty))
+        for (auto *U : getCUs(RT->getScope(), &CU))
+          U->getOrCreateTypeDIE(cast<DIType>(RT));
+    }
 
-    // Emit imported entities last so that the relevant context
-    // is already available.
+    // Emit imported entities last so that
+    // the relevant context is already available.
     for (auto *IE : CUNode->getImportedEntities())
-      constructAndAddImportedEntityDIE(*CU, IE);
+      for (auto *U : getCUs(IE->getScope(), &CU))
+        U->createAndAddImportedEntityDIE(IE);
 
     CU->createBaseTypeDIEs();
   }
 
   // If we aren't actually generating debug info (check beginModule -
   // conditionalized on the presence of the llvm.dbg.cu metadata node)
   if (!Asm || !MMI->hasDebugInfo())
     return;
Optional<MD5::MD5Result> DwarfDebug::getMD5AsBytes(const DIFile *File) const {
   // Convert the string checksum to an MD5Result for the streamer.
   // The verifier validates the checksum so we assume it's okay.
   // An MD5 checksum is 16 bytes.
   std::string ChecksumString = fromHex(Checksum->Value);
   MD5::MD5Result CKMem;
   std::copy(ChecksumString.begin(), ChecksumString.end(), CKMem.data());
   return CKMem;
 }
+
+void DwarfDebug::recordSPContainingCU(const DISubprogram *SP,
+                                      DwarfCompileUnit *CU) {
+  SPCUsMap[SP].insert(CU);
+}

llvm/lib/CodeGen/AsmPrinter/DwarfFile.h

 
 namespace llvm {
 
 class AsmPrinter;
 class DbgEntity;
 class DbgVariable;
 class DbgLabel;
 class DINode;
+class DILocalScope;
 class DwarfCompileUnit;
 class DwarfUnit;
 class LexicalScope;
 class MCSection;
 class MDNode;
 
 // Data structure to hold a range for range lists.
 struct RangeSpan {
class DwarfFile {
   /// Collection of DbgVariables of each lexical scope.
   DenseMap<LexicalScope *, ScopeVars> ScopeVariables;
 
   /// Collection of DbgLabels of each lexical scope.
   using LabelList = SmallVector<DbgLabel *, 4>;
   DenseMap<LexicalScope *, LabelList> ScopeLabels;
 
   // Collection of abstract subprogram DIEs.
-  DenseMap<const MDNode *, DIE *> AbstractSPDies;
+  DenseMap<const DILocalScope *, DIE *> AbstractLocalScopeDIEs;
   DenseMap<const DINode *, std::unique_ptr<DbgEntity>> AbstractEntities;
 
   /// Maps MDNodes for type system with the corresponding DIEs. These DIEs can
   /// be shared across CUs, that is why we keep the map here instead
   /// of in DwarfCompileUnit.
   DenseMap<const MDNode *, DIE *> DITypeNodeToDieMap;
 
 public:
public:
   DenseMap<LexicalScope *, ScopeVars> &getScopeVariables() {
     return ScopeVariables;
   }
 
   DenseMap<LexicalScope *, LabelList> &getScopeLabels() {
     return ScopeLabels;
   }
 
-  DenseMap<const MDNode *, DIE *> &getAbstractSPDies() {
-    return AbstractSPDies;
+  DenseMap<const DILocalScope *, DIE *> &getAbstractScopeDIEs() {
+    return AbstractLocalScopeDIEs;
   }
 
   DenseMap<const DINode *, std::unique_ptr<DbgEntity>> &getAbstractEntities() {
     return AbstractEntities;
   }
 
   void insertDIE(const MDNode *TypeMD, DIE *Die) {
     DITypeNodeToDieMap.insert(std::make_pair(TypeMD, Die));

llvm/lib/CodeGen/AsmPrinter/DwarfUnit.h

public:
 
   /// Creates type DIE with specific context.
   DIE *createTypeDIE(const DIScope *Context, DIE &ContextDIE, const DIType *Ty);
 
   /// Find existing DIE or create new DIE for the given type.
   DIE *getOrCreateTypeDIE(const MDNode *TyNode);
 
   /// Get context owner's DIE.
-  DIE *getOrCreateContextDIE(const DIScope *Context);
+  virtual DIE *getOrCreateContextDIE(const DIScope *Context);
 
   /// Construct DIEs for types that contain vtables.
   void constructContainingTypeDIEs();
 
   /// Construct function argument DIEs.
   void constructSubprogramArguments(DIE &Buffer, DITypeRefArray Args);
 
   /// Create a DIE with the given Tag, add the DIE to its parent, and

llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp

DIE *DwarfUnit::getOrCreateTypeDIE(const MDNode *TyNode) {
   // such construction creates the DIE.
   auto *Context = Ty->getScope();
   DIE *ContextDIE = getOrCreateContextDIE(Context);
   assert(ContextDIE);
 
   if (DIE *TyDIE = getDIE(Ty))
     return TyDIE;
 
   return static_cast<DwarfUnit *>(ContextDIE->getUnit())
       ->createTypeDIE(Context, *ContextDIE, Ty);
 }
 
 void DwarfUnit::updateAcceleratorTables(const DIScope *Context,
                                         const DIType *Ty, const DIE &TyDIE) {
   if (!Ty->getName().empty() && !Ty->isForwardDecl()) {
     bool IsImplementation = false;
     if (auto *CT = dyn_cast<DICompositeType>(Ty)) {
       // A runtime language of 0 actually means C/C++ and that any
       // non-negative value is some version of Objective-C/C++.
       IsImplementation = CT->getRuntimeLang() == 0 || CT->isObjcClassComplete();
bool DwarfUnit::applySubprogramDefinitionAttributes(const DISubprogram *SP,
 
   // Add the linkage name if we have one and it isn't in the Decl.
   StringRef LinkageName = SP->getLinkageName();
   assert(((LinkageName.empty() || DeclLinkageName.empty()) ||
           LinkageName == DeclLinkageName) &&
          "decl has a linkage name and it is different");
   if (DeclLinkageName.empty() &&
       // Always emit it for abstract subprograms.
-      (DD->useAllLinkageNames() || DU->getAbstractSPDies().lookup(SP)))
+      (DD->useAllLinkageNames() || DU->getAbstractScopeDIEs().lookup(SP)))
     addLinkageName(SPDie, LinkageName);
 
   if (!DeclDie)
     return false;
 
   // Refer to the function declaration where all the other attributes will be
   // found.
   addDIEEntry(SPDie, dwarf::DW_AT_specification, *DeclDie);

llvm/test/DebugInfo/Generic/import-inlined-declaration.ll

+; RUN: %llc_dwarf -O0 -filetype=obj < %s | llvm-dwarfdump - | FileCheck --implicit-check-not "{{DW_TAG|NULL}}" %s
+; REQUIRES: object-emission
+
+; namespace ns {
+; inline __attribute__((always_inline))
+; void foo() { int a = 4; }
+; }
+;
+; void goo() {
+;   using ns::foo;
+;   foo();
+; }
+
+; Ensure that imported declarations reference the correct subprograms even if
+; those subprograms are inlined.
+
+; CHECK: DW_TAG_compile_unit
+; CHECK:   DW_TAG_namespace
+; CHECK:     DW_AT_name     ("ns")
+; CHECK: [[FOO:0x.*]]:     DW_TAG_subprogram
+; CHECK:       DW_AT_name   ("foo")
+; CHECK:       DW_TAG_variable
+; CHECK:       NULL
+; CHECK:     NULL
+; CHECK:   DW_TAG_base_type
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name     ("goo")
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:       DW_AT_abstract_origin ([[FOO]]
+; CHECK:       DW_TAG_variable
+; CHECK:       NULL
+; CHECK:     DW_TAG_imported_declaration
+; CHECK:       DW_AT_import ([[FOO]])
+; CHECK:     NULL
+; CHECK:   NULL
+
+; Function Attrs: mustprogress noinline optnone uwtable
+define dso_local void @_Z3goov() !dbg !4 {
+entry:
+  %a.i = alloca i32, align 4
+  call void @llvm.dbg.declare(metadata i32* %a.i, metadata !16, metadata !DIExpression()), !dbg !18
+  store i32 4, i32* %a.i, align 4, !dbg !18
+  ret void, !dbg !20
+}
+
+; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
+declare void @llvm.dbg.declare(metadata, metadata, metadata)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!10, !11, !12, !13, !14}
+!llvm.ident = !{!15}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 14.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, imports: !2, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "imported-inlined-declaration.cpp", directory: "")
+!2 = !{!3}
+!3 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !4, entity: !8, file: !1, line: 7)
+!4 = distinct !DISubprogram(name: "goo", linkageName: "_Z3goov", scope: !1, file: !1, line: 6, type: !5, scopeLine: 6, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
+!5 = !DISubroutineType(types: !6)
+!6 = !{null}
+!7 = !{}
+!8 = distinct !DISubprogram(name: "foo", linkageName: "_ZN2ns3fooEv", scope: !9, file: !1, line: 3, type: !5, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
+!9 = !DINamespace(name: "ns", scope: null)
+!10 = !{i32 7, !"Dwarf Version", i32 4}
+!11 = !{i32 2, !"Debug Info Version", i32 3}
+!12 = !{i32 1, !"wchar_size", i32 4}
+!13 = !{i32 7, !"uwtable", i32 1}
+!14 = !{i32 7, !"frame-pointer", i32 2}
+!15 = !{!"clang version 14.0.0"}
+!16 = !DILocalVariable(name: "a", scope: !8, file: !1, line: 3, type: !17)
+!17 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!18 = !DILocation(line: 3, column: 18, scope: !8, inlinedAt: !19)
+!19 = distinct !DILocation(line: 8, column: 2, scope: !4)
+!20 = !DILocation(line: 9, column: 1, scope: !4)

llvm/test/DebugInfo/Generic/imported-name-inlined.ll

 ; RUN: %llc_dwarf -O0 -filetype=obj < %s | llvm-dwarfdump -debug-info - | FileCheck --implicit-check-not "{{DW_TAG|NULL}}" %s
 
 ; Generated from the following source:
 ; namespace ns {
 ; void f();
 ; }
 ; inline __attribute__((always_inline)) void f1() {
 ;   using ns::f;
 ;   f();
 ; }
 ; void f2() { f1(); }
 
 ; Ensure that top level imported declarations don't produce an extra degenerate
 ; concrete subprogram definition.
 
-; FIXME: imported entities should only be emitted to the abstract origin if one is present
-
 ; CHECK: DW_TAG_compile_unit
 ; CHECK:   DW_TAG_subprogram
 ; CHECK:     DW_AT_name ("f1")
 ; CHECK:     DW_TAG_imported_declaration
 ; CHECK:     NULL
-; CHECK:   DW_TAG_namespace
-; CHECK:     DW_TAG_subprogram
-; CHECK:     NULL
 ; CHECK:   DW_TAG_subprogram
 ; CHECK:     DW_AT_name ("f2")
 ; CHECK:     DW_TAG_inlined_subroutine
-; CHECK:       DW_TAG_imported_declaration
-; CHECK:       NULL
+; CHECK:     NULL
+; CHECK:   DW_TAG_namespace
+; CHECK:     DW_TAG_subprogram
 ; CHECK:     NULL
 ; CHECK:   NULL
 
 ; Function Attrs: noinline
 define void @_Z2f2v() noinline !dbg !14 {
 entry:
   call void @_ZN2ns1fEv(), !dbg !15
   ret void, !dbg !17

llvm/test/DebugInfo/Generic/inlined-local-type.ll

+; RUN: %llc_dwarf -O0 -filetype=obj < %s | llvm-dwarfdump -debug-info - | FileCheck --implicit-check-not "{{DW_TAG|NULL}}" %s
+; REQUIRES: object-emission
+
+; inline __attribute__((always_inline))
+; int removed() { struct A {int i;}; struct A a; return a.i++; }
+;
+; __attribute__((always_inline))
+; int not_removed() { struct B {int i;}; struct B b; return b.i++; }
+;
+; int foo() { return removed() + not_removed(); }}
+
+; Ensure that function-local types have the correct subprogram parent even if
+; those subprograms are inlined.
+
+; CHECK: DW_TAG_compile_unit
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_abstract_origin  ({{0x.*}} "not_removed")
+; CHECK:     DW_TAG_variable
+; CHECK:     NULL
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("removed")
+; CHECK: [[A:0x.*]]: DW_TAG_structure_type
+; CHECK:       DW_AT_name ("A")
+; CHECK:       DW_TAG_member
+; CHECK:       NULL
+; CHECK:     DW_TAG_variable
+; CHECK:       DW_AT_type ([[A]] "A")
+; CHECK:     NULL
+; CHECK:   DW_TAG_base_type
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("not_removed")
+; CHECK: [[B:0x.*]]: DW_TAG_structure_type
+; CHECK:       DW_AT_name ("B")
+; CHECK:       DW_TAG_member
+; CHECK:       NULL
+; CHECK:     DW_TAG_variable
+; CHECK:       DW_AT_type ([[B]] "B")
+; CHECK:     NULL
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:       DW_TAG_variable
+; CHECK:       NULL
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:       DW_TAG_variable
+; CHECK:       NULL
+; CHECK:     NULL
+; CHECK:   NULL
+
+%struct.B = type { i32 }
+%struct.A = type { i32 }
+
+define dso_local i32 @not_removed() !dbg !12 {
+  %1 = alloca %struct.B, align 4
+  call void @llvm.dbg.declare(metadata %struct.B* %1, metadata !18, metadata !DIExpression()), !dbg !22
+  %2 = getelementptr inbounds %struct.B, %struct.B* %1, i32 0, i32 0, !dbg !23
+  %3 = load i32, i32* %2, align 4, !dbg !24
+  %4 = add nsw i32 %3, 1, !dbg !24
+  store i32 %4, i32* %2, align 4, !dbg !24
+  ret i32 %3, !dbg !25
+}
+
+declare void @llvm.dbg.declare(metadata, metadata, metadata)
+
+define dso_local i32 @foo() !dbg !26 {
+  %1 = alloca %struct.A, align 4
+  %2 = alloca %struct.B, align 4
+  call void @llvm.dbg.declare(metadata %struct.A* %1, metadata !27, metadata !DIExpression()), !dbg !32
+  %3 = getelementptr inbounds %struct.A, %struct.A* %1, i32 0, i32 0, !dbg !34
+  %4 = load i32, i32* %3, align 4, !dbg !35
+  %5 = add nsw i32 %4, 1, !dbg !35
+  store i32 %5, i32* %3, align 4, !dbg !35
+  call void @llvm.dbg.declare(metadata %struct.B* %2, metadata !18, metadata !DIExpression()), !dbg !36
+  %6 = getelementptr inbounds %struct.B, %struct.B* %2, i32 0, i32 0, !dbg !38
+  %7 = load i32, i32* %6, align 4, !dbg !39
+  %8 = add nsw i32 %7, 1, !dbg !39
+  store i32 %8, i32* %6, align 4, !dbg !39
+  %9 = add nsw i32 %4, %7, !dbg !40
+  ret i32 %9, !dbg !41
+}
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3, !4, !5, !6, !7, !8, !9, !10}
+!llvm.ident = !{!11}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 14.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "inlined-local-type.cpp", directory: "")
+!2 = !{i32 7, !"Dwarf Version", i32 4}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!4 = !{i32 1, !"wchar_size", i32 4}
+!5 = !{i32 1, !"branch-target-enforcement", i32 0}
+!6 = !{i32 1, !"sign-return-address", i32 0}
+!7 = !{i32 1, !"sign-return-address-all", i32 0}
+!8 = !{i32 1, !"sign-return-address-with-bkey", i32 0}
+!9 = !{i32 7, !"uwtable", i32 1}
+!10 = !{i32 7, !"frame-pointer", i32 1}
+!11 = !{!"clang version 14.0.0"}
+!12 = distinct !DISubprogram(name: "not_removed", scope: !13, file: !13, line: 5, type: !14, scopeLine: 5, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !17)
+!13 = !DIFile(filename: "inlined-local-type.cpp", directory: "")
+!14 = !DISubroutineType(types: !15)
+!15 = !{!16}
+!16 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!17 = !{}
+!18 = !DILocalVariable(name: "b", scope: !12, file: !13, line: 5, type: !19)
+!19 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "B", scope: !12, file: !13, line: 5, size: 32, elements: !20)
+!20 = !{!21}
+!21 = !DIDerivedType(tag: DW_TAG_member, name: "i", scope: !19, file: !13, line: 5, baseType: !16, size: 32)
+!22 = !DILocation(line: 5, column: 49, scope: !12)
+!23 = !DILocation(line: 5, column: 61, scope: !12)
+!24 = !DILocation(line: 5, column: 62, scope: !12)
+!25 = !DILocation(line: 5, column: 52, scope: !12)
+!26 = distinct !DISubprogram(name: "foo", scope: !13, file: !13, line: 7, type: !14, scopeLine: 7, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !17)
+!27 = !DILocalVariable(name: "a", scope: !28, file: !13, line: 2, type: !29)
+!28 = distinct !DISubprogram(name: "removed", scope: !13, file: !13, line: 2, type: !14, scopeLine: 2, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !17)
+!29 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "A", scope: !28, file: !13, line: 2, size: 32, elements: !30)
+!30 = !{!31}
+!31 = !DIDerivedType(tag: DW_TAG_member, name: "i", scope: !29, file: !13, line: 2, baseType: !16, size: 32)
+!32 = !DILocation(line: 2, column: 45, scope: !28, inlinedAt: !33)
+!33 = distinct !DILocation(line: 7, column: 20, scope: !26)
+!34 = !DILocation(line: 2, column: 57, scope: !28, inlinedAt: !33)
+!35 = !DILocation(line: 2, column: 58, scope: !28, inlinedAt: !33)
+!36 = !DILocation(line: 5, column: 49, scope: !12, inlinedAt: !37)
+!37 = distinct !DILocation(line: 7, column: 32, scope: !26)
+!38 = !DILocation(line: 5, column: 61, scope: !12, inlinedAt: !37)
+!39 = !DILocation(line: 5, column: 62, scope: !12, inlinedAt: !37)
+!40 = !DILocation(line: 7, column: 30, scope: !26)
+!41 = !DILocation(line: 7, column: 13, scope: !26)

llvm/test/DebugInfo/Generic/inlined-static-var.ll

+; RUN: %llc_dwarf -O0 -filetype=obj < %s | llvm-dwarfdump -debug-info - | FileCheck --implicit-check-not "{{DW_TAG|NULL}}" %s
+; REQUIRES: object-emission
+
+; inline __attribute__((always_inline))
+; int removed() { static int A; return A++; }
+;
+; __attribute__((always_inline))
+; int not_removed() { static int B; return B++; }
+;
+; int foo() { return removed() + not_removed(); }
+
+; Ensure that global variables belong to the correct subprograms even if those
+; subprograms are inlined.
+
+; CHECK: DW_TAG_compile_unit
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_abstract_origin {{.*}} "_Z11not_removedv"
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name       ("removed")
+; CHECK:     DW_TAG_variable
+; CHECK:       DW_AT_name     ("A")
+; CHECK:     NULL
+; CHECK:   DW_TAG_base_type
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name       ("not_removed")
+; CHECK:     DW_TAG_variable
+; CHECK:       DW_AT_name     ("B")
+; CHECK:     NULL
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name       ("foo")
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:     NULL
+; CHECK:   NULL
+
+@_ZZ11not_removedvE1A = internal global i32 0, align 4, !dbg !0
+@_ZZ7removedvE1A = linkonce_odr dso_local global i32 0, align 4, !dbg !10
+
+define dso_local i32 @_Z11not_removedv() !dbg !2 {
+  %1 = load i32, i32* @_ZZ11not_removedvE1A, align 4, !dbg !24
+  %2 = add nsw i32 %1, 1, !dbg !24
+  store i32 %2, i32* @_ZZ11not_removedvE1A, align 4, !dbg !24
+  ret i32 %1, !dbg !25
+}
+
+define dso_local i32 @_Z3foov() !dbg !26 {
+  %1 = load i32, i32* @_ZZ7removedvE1A, align 4, !dbg !27
+  %2 = add nsw i32 %1, 1, !dbg !27
+  store i32 %2, i32* @_ZZ7removedvE1A, align 4, !dbg !27
+  %3 = load i32, i32* @_ZZ11not_removedvE1A, align 4, !dbg !29
+  %4 = add nsw i32 %3, 1, !dbg !29
+  store i32 %4, i32* @_ZZ11not_removedvE1A, align 4, !dbg !29
+  %5 = add nsw i32 %1, %3, !dbg !31
+  ret i32 %5, !dbg !32
+}
+
+!llvm.dbg.cu = !{!7}
+!llvm.module.flags = !{!14, !15, !16, !17, !18, !19, !20, !21, !22}
+!llvm.ident = !{!23}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(name: "B", scope: !2, file: !3, line: 5, type: !6, isLocal: true, isDefinition: true)
+!2 = distinct !DISubprogram(name: "not_removed", linkageName: "_Z11not_removedv", scope: !3, file: !3, line: 5, type: !4, scopeLine: 5, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !7, retainedNodes: !13)
+!3 = !DIFile(filename: "example.cpp", directory: "")
+!4 = !DISubroutineType(types: !5)
+!5 = !{!6}
+!6 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!7 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !8, producer: "clang version 14.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, globals: !9, splitDebugInlining: false, nameTableKind: None)
+!8 = !DIFile(filename: "example.cpp", directory: "")
+!9 = !{!0, !10}
+!10 = !DIGlobalVariableExpression(var: !11, expr: !DIExpression())
+!11 = distinct !DIGlobalVariable(name: "A", scope: !12, file: !3, line: 2, type: !6, isLocal: false, isDefinition: true)
+!12 = distinct !DISubprogram(name: "removed", linkageName: "_Z7removedv", scope: !3, file: !3, line: 2, type: !4, scopeLine: 2, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !7, retainedNodes: !13)
+!13 = !{}
+!14 = !{i32 7, !"Dwarf Version", i32 4}
+!15 = !{i32 2, !"Debug Info Version", i32 3}
+!16 = !{i32 1, !"wchar_size", i32 4}
+!17 = !{i32 1, !"branch-target-enforcement", i32 0}
+!18 = !{i32 1, !"sign-return-address", i32 0}
+!19 = !{i32 1, !"sign-return-address-all", i32 0}
+!20 = !{i32 1, !"sign-return-address-with-bkey", i32 0}
+!21 = !{i32 7, !"uwtable", i32 1}
+!22 = !{i32 7, !"frame-pointer", i32 1}
+!23 = !{!"clang version 14.0.0"}
+!24 = !DILocation(line: 5, column: 43, scope: !2)
+!25 = !DILocation(line: 5, column: 35, scope: !2)
+!26 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !3, file: !3, line: 7, type: !4, scopeLine: 7, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !7, retainedNodes: !13)
+!27 = !DILocation(line: 2, column: 39, scope: !12, inlinedAt: !28)
+!28 = distinct !DILocation(line: 7, column: 20, scope: !26)
+!29 = !DILocation(line: 5, column: 43, scope: !2, inlinedAt: !30)
+!30 = distinct !DILocation(line: 7, column: 32, scope: !26)
+!31 = !DILocation(line: 7, column: 30, scope: !26)
+!32 = !DILocation(line: 7, column: 13, scope: !26)

llvm/test/DebugInfo/Generic/lexical-block-static-var.ll

+; RUN: %llc_dwarf -O0 -filetype=obj < %s | llvm-dwarfdump -debug-info - | FileCheck --implicit-check-not "{{DW_TAG|NULL}}" %s
+; REQUIRES: object-emission
+
+; inline __attribute__((always_inline))
+; int removed() {
+;   {
+;     static int A;
+;     return A++;
+;   }
+; }
+;
+; __attribute__((always_inline))
+; int not_removed() {
+;   {
+;     static int B;
+;     return B++;
+;   }
+; }
+;
+; int foo() {
+;   {
+;     static int C;
+;     return ++C + removed() + not_removed();
+;   }
+; }
+
+; CHECK: DW_TAG_compile_unit
+
+; Out-of-line definition of `not_removed()`.
+; The empty lexical block is created to match abstract origin.
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_abstract_origin {{.*}} "_Z11not_removedv"
+; CHECK:     DW_TAG_lexical_block
+; CHECK:     NULL
+
+; Abstract definition of `removed()`
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("removed")
+; CHECK:     DW_AT_inline (DW_INL_inlined)
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_name ("A")
+; CHECK:         DW_AT_location (DW_OP_addr 0x0)
+; CHECK:       NULL
+; CHECK:     NULL
+; CHECK:   DW_TAG_base_type
+
+; Abstract definition of `not_removed()`
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("not_removed")
+; CHECK:     DW_AT_inline (DW_INL_inlined)
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_name ("B")
+; CHECK:         DW_AT_location (DW_OP_addr 0x0)
+; CHECK:       NULL
+; CHECK:     NULL
+
+; Definition of foo().
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("foo")
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_inlined_subroutine
+; CHECK:         DW_TAG_lexical_block
+; CHECK:         NULL
+; CHECK:       DW_TAG_inlined_subroutine
+; CHECK:         DW_TAG_lexical_block
+; CHECK:         NULL
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_name ("C")
+; CHECK:         DW_AT_location (DW_OP_addr 0x0)
+; CHECK:       NULL
+; CHECK:     NULL
+; CHECK:   NULL
+
+$_ZZ7removedvE1A = comdat any
+
+@_ZZ11not_removedvE1B = internal global i32 0, align 4, !dbg !0
+@_ZZ3foovE1C = internal global i32 0, align 4, !dbg !10
+@_ZZ7removedvE1A = linkonce_odr dso_local global i32 0, comdat, align 4, !dbg !15
+
+; Function Attrs: alwaysinline mustprogress nounwind uwtable
+define dso_local i32 @_Z11not_removedv() #0 !dbg !4 {
+entry:
+  %0 = load i32, i32* @_ZZ11not_removedvE1B, align 4, !dbg !25
+  %inc = add nsw i32 %0, 1, !dbg !25
+  store i32 %inc, i32* @_ZZ11not_removedvE1B, align 4, !dbg !25
+  ret i32 %0, !dbg !26
+}
+
+; Function Attrs: mustprogress noinline optnone uwtable
+define dso_local i32 @_Z3foov() #1 !dbg !13 {
+entry:
+  %0 = load i32, i32* @_ZZ3foovE1C, align 4, !dbg !27
+  %inc = add nsw i32 %0, 1, !dbg !27
+  store i32 %inc, i32* @_ZZ3foovE1C, align 4, !dbg !27
+  %1 = load i32, i32* @_ZZ7removedvE1A, align 4, !dbg !28
+  %inc.i3 = add nsw i32 %1, 1, !dbg !28
+  store i32 %inc.i3, i32* @_ZZ7removedvE1A, align 4, !dbg !28
+  %add = add nsw i32 %inc, %1, !dbg !30
+  %2 = load i32, i32* @_ZZ11not_removedvE1B, align 4, !dbg !31
+  %inc.i = add nsw i32 %2, 1, !dbg !31
+  store i32 %inc.i, i32* @_ZZ11not_removedvE1B, align 4, !dbg !31
+  %add2 = add nsw i32 %add, %2, !dbg !33
+  ret i32 %add2, !dbg !34
+}
+
+attributes #0 = { alwaysinline mustprogress nounwind uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
+attributes #1 = { mustprogress noinline optnone uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
+
+!llvm.dbg.cu = !{!8}
+!llvm.module.flags = !{!19, !20, !21, !22, !23}
+!llvm.ident = !{!24}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(name: "B", scope: !2, file: !3, line: 13, type: !7, isLocal: true, isDefinition: true)
+!2 = distinct !DILexicalBlock(scope: !4, file: !3, line: 12, column: 3)
+!3 = !DIFile(filename: "test_static.cpp", directory: "/home/kbessonova/workspace/llvm/llvm-project/build")
+!4 = distinct !DISubprogram(name: "not_removed", linkageName: "_Z11not_removedv", scope: !3, file: !3, line: 11, type: !5, scopeLine: 11, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !8, retainedNodes: !14)
+!5 = !DISubroutineType(types: !6)
+!6 = !{!7}
+!7 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!8 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !3, producer: "clang version 14.0.0 ([e-mail]:llvm/llvm-project.git e1d09ac2d118825452bfc26e44565f7f4122fd6d)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, globals: !9, splitDebugInlining: false, nameTableKind: None)
+!9 = !{!0, !10, !15}
+!10 = !DIGlobalVariableExpression(var: !11, expr: !DIExpression())
+!11 = distinct !DIGlobalVariable(name: "C", scope: !12, file: !3, line: 20, type: !7, isLocal: true, isDefinition: true)
+!12 = distinct !DILexicalBlock(scope: !13, file: !3, line: 19, column: 3)
+!13 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !3, file: !3, line: 18, type: !5, scopeLine: 18, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !8, retainedNodes: !14)
+!14 = !{}
+!15 = !DIGlobalVariableExpression(var: !16, expr: !DIExpression())
+!16 = distinct !DIGlobalVariable(name: "A", scope: !17, file: !3, line: 5, type: !7, isLocal: false, isDefinition: true)
+!17 = distinct !DILexicalBlock(scope: !18, file: !3, line: 4, column: 3)
+!18 = distinct !DISubprogram(name: "removed", linkageName: "_Z7removedv", scope: !3, file: !3, line: 3, type: !5, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !8, retainedNodes: !14)
+!19 = !{i32 7, !"Dwarf Version", i32 4}
+!20 = !{i32 2, !"Debug Info Version", i32 3}
+!21 = !{i32 1, !"wchar_size", i32 4}
+!22 = !{i32 7, !"uwtable", i32 1}
+!23 = !{i32 7, !"frame-pointer", i32 2}
+!24 = !{!"clang version 14.0.0 ([e-mail]:llvm/llvm-project.git e1d09ac2d118825452bfc26e44565f7f4122fd6d)"}
+!25 = !DILocation(line: 14, column: 13, scope: !2)
+!26 = !DILocation(line: 14, column: 5, scope: !2)
+!27 = !DILocation(line: 21, column: 12, scope: !12)
+!28 = !DILocation(line: 6, column: 13, scope: !17, inlinedAt: !29)
+!29 = distinct !DILocation(line: 21, column: 18, scope: !12)
+!30 = !DILocation(line: 21, column: 16, scope: !12)
+!31 = !DILocation(line: 14, column: 13, scope: !2, inlinedAt: !32)
+!32 = distinct !DILocation(line: 21, column: 30, scope: !12)
+!33 = !DILocation(line: 21, column: 28, scope: !12)
+!34 = !DILocation(line: 21, column: 5, scope: !12)

llvm/test/DebugInfo/Generic/lexical-block-types.ll

+; RUN: %llc_dwarf -O0 -filetype=obj < %s | llvm-dwarfdump -debug-info - | FileCheck --implicit-check-not "{{DW_TAG|NULL}}" %s
+; REQUIRES: object-emission
+
+; inline __attribute__((always_inline))
+; void removed() {
+;   struct A1 { int i; };
+;   typedef int Int1;
+;   {
+;     struct I1 { Int1 j; };
+;     struct C1 { typedef char Char1; Char1 c; };
+;     A1 a1; a1.i++;
+;     {
+;       I1 i1; i1.j++;
+;       C1 c1; c1.c++;
+;     }
+;   }
+; }
+;
+; __attribute__((always_inline))
+; void not_removed() {
+;   struct A2 { int i; };
+;   typedef int Int2;
+;   {
+;     struct I2 { Int2 j; };
+;     struct C2 { typedef char Char2; Char2 c; };
+;     A2 a2; a2.i++;
+;     {
+;       I2 i2; i2.j++;
+;       C2 c2; c2.c++;
+;     }
+;   }
+; }
+;
+; void foo() {
+;   struct A3 { int i; };
+;   typedef int Int3;
+;   {
+;     struct I3 { Int3 j; };
+;     {
+;       struct C3 { typedef char Char3; Char3 c; };
+;       A3 a3; a3.i++;
+;       {
+;         I3 i3; i3.j++;
+;         C3 c3; c3.c++;
+;       }
+;     }
+;   }
+;   removed();
+;   not_removed();
+; }
+;
+; CHECK: DW_TAG_compile_unit
+
+; Out-of-line definition of `not_removed()` shouldn't contain any debug info for types.
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_abstract_origin  {{.*}} "_Z11not_removedv"
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_abstract_origin  {{.*}} "a2"
+; CHECK:       DW_TAG_lexical_block
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_abstract_origin  {{.*}} "i2"
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_abstract_origin  {{.*}} "c2"
+; CHECK:         NULL
+; CHECK:       NULL
+; CHECK:     NULL
+
+; Abstract definition of `removed()`.
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("removed")
+; CHECK:     DW_AT_inline (DW_INL_inlined)
+
+; I1 and C1 defined in the first lexical block, typedef Char1 is a child of C1.
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_name ("a1")
+; CHECK:         DW_AT_type {{.*}} "A1"
+; CHECK:       DW_TAG_lexical_block
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_type {{.*}} "I1"
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_type {{.*}} "C1"
+; CHECK:         NULL
+; CHECK:       DW_TAG_structure_type
+; CHECK:         DW_AT_name ("I1")
+; CHECK:         DW_TAG_member
+; CHECK:           DW_AT_type {{.*}} "Int1"
+; CHECK:         NULL
+; CHECK:       DW_TAG_structure_type
+; CHECK:         DW_AT_name ("C1")
+; CHECK:         DW_TAG_member
+; CHECK:           DW_AT_type {{.*}} "C1::Char1"
+; CHECK:         DW_TAG_typedef
+; CHECK:           DW_AT_name ("Char1")
+; CHECK:         NULL
+; CHECK:       NULL
+
+; A1 and typedef Int1 defined in subprogram scope.
+; CHECK:     DW_TAG_structure_type
+; CHECK:       DW_AT_name ("A1")
+; CHECK:       DW_TAG_member
+; CHECK:       NULL
+; CHECK:     DW_TAG_typedef
+; CHECK:       DW_AT_name ("Int1")
+; CHECK:     NULL
+
+; CHECK:   DW_TAG_base_type
+; CHECK:   DW_TAG_base_type
+
+; Abstract definition of `not_removed()`.
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("not_removed")
+; CHECK:     DW_AT_inline (DW_INL_inlined)
+
+; I2 and C2 defined in the first lexical block, typedef Char2 is a child of C2.
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_name ("a2")
+; CHECK:         DW_AT_type {{.*}} "A2"
+; CHECK:       DW_TAG_lexical_block
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_name ("i2")
+; CHECK:           DW_AT_type {{.*}} "I2"
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_name ("c2")
+; CHECK:           DW_AT_type {{.*}} "C2"
+; CHECK:         NULL
+; CHECK:       DW_TAG_structure_type
+; CHECK:         DW_AT_name ("I2")
+; CHECK:         DW_TAG_member
+; CHECK:           DW_AT_type {{.*}} "Int2"
+; CHECK:         NULL
+; CHECK:       DW_TAG_structure_type
+; CHECK:         DW_AT_name ("C2")
+; CHECK:         DW_TAG_member
+; CHECK:           DW_AT_type {{.*}} "C2::Char2"
+; CHECK:         DW_TAG_typedef
+; CHECK:           DW_AT_name ("Char2")
+; CHECK:         NULL
+; CHECK:       NULL
+
+; A2 and typedef Int2 defined in subprogram scope.
+; CHECK:     DW_TAG_structure_type
+; CHECK:       DW_AT_name ("A2")
+; CHECK:       DW_TAG_member
+; CHECK:       NULL
+; CHECK:     DW_TAG_typedef
+; CHECK:       DW_AT_name ("Int2")
+; CHECK:     NULL
+
+; Definition of `foo()`.
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_AT_name ("foo")
+
+; CHECK:     DW_TAG_lexical_block
+; CHECK:       DW_TAG_variable
+; CHECK:         DW_AT_name ("a3")
+; CHECK:         DW_AT_type {{.*}} "A3"
+; CHECK:       DW_TAG_lexical_block
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_name ("i3")
+; CHECK:           DW_AT_type {{.*}} "I3"
+; CHECK:         DW_TAG_variable
+; CHECK:           DW_AT_name ("c3")
+; CHECK:           DW_AT_type {{.*}} "C3"
+; CHECK:         NULL
+
+; C3 has the inner lexical block scope, typedef Char3 is a child of C3.
+; CHECK:       DW_TAG_structure_type
+; CHECK:         DW_AT_name ("C3")
+; CHECK:         DW_TAG_member
+; CHECK:           DW_AT_type {{.*}} "C3::Char3"
+; CHECK:         DW_TAG_typedef
+; CHECK:           DW_AT_name ("Char3")
+; CHECK:         NULL
+; CHECK:       NULL
+
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:       DW_AT_abstract_origin  {{.*}} "_Z7removedv"
+; CHECK:       DW_TAG_lexical_block
+; CHECK:         DW_TAG_variable
+; CHECK:         DW_TAG_lexical_block
+; CHECK:           DW_TAG_variable
+; CHECK:           DW_TAG_variable
+; CHECK:           NULL
+; CHECK:         NULL
+; CHECK:       NULL
+
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:       DW_AT_abstract_origin  {{.*}} "_Z11not_removedv"
+; CHECK:       DW_TAG_lexical_block
+; CHECK:         DW_TAG_variable
+; CHECK:         DW_TAG_lexical_block
+; CHECK:           DW_TAG_variable
+; CHECK:           DW_TAG_variable
+; CHECK:           NULL
+; CHECK:         NULL
+; CHECK:       NULL
+
+; A3 and typedef Int3 defined in the subprogram scope.
+; FIXME: I3 has subprogram scope here, but should be in the outer lexical block
+; (which is ommitted). It wasn't placed correctly, because it's the only non-scope
+; entity in the block and it isn't listed in retainedTypes; we simply wasn't aware
+; about it while deciding whether to create a lexical block or not.
+; CHECK:     DW_TAG_structure_type
+; CHECK:       DW_AT_name ("A3")
+; CHECK:       DW_TAG_member
+; CHECK:       NULL
+; CHECK:     DW_TAG_structure_type
+; CHECK:       DW_AT_name ("I3")
+; CHECK:       DW_TAG_member
+; CHECK:         DW_AT_type {{.*}} "Int3"
+; CHECK:       NULL
+; CHECK:     DW_TAG_typedef
+; CHECK:       DW_AT_name ("Int3")
+; CHECK:     NULL
+; CHECK:   NULL
+
+%struct.A2 = type { i32 }
+%struct.I2 = type { i32 }
+%struct.C2 = type { i8 }
+%struct.A1 = type { i32 }
+%struct.I1 = type { i32 }
+%struct.C1 = type { i8 }
+%struct.A3 = type { i32 }
+%struct.I3 = type { i32 }
+%struct.C3 = type { i8 }
+
+define dso_local void @_Z11not_removedv() !dbg !8 {
+entry:
+  %a2 = alloca %struct.A2, align 4
+  %i2 = alloca %struct.I2, align 4
+  %c2 = alloca %struct.C2, align 1
+  call void @llvm.dbg.declare(metadata %struct.A2* %a2, metadata !12, metadata !DIExpression()), !dbg !18
+  %i = getelementptr inbounds %struct.A2, %struct.A2* %a2, i32 0, i32 0, !dbg !19
+  %0 = load i32, i32* %i, align 4, !dbg !20
+  %inc = add nsw i32 %0, 1, !dbg !20
+  store i32 %inc, i32* %i, align 4, !dbg !20
+  call void @llvm.dbg.declare(metadata %struct.I2* %i2, metadata !21, metadata !DIExpression()), !dbg !27
+  %j = getelementptr inbounds %struct.I2, %struct.I2* %i2, i32 0, i32 0, !dbg !28
+  %1 = load i32, i32* %j, align 4, !dbg !29
+  %inc1 = add nsw i32 %1, 1, !dbg !29
+  store i32 %inc1, i32* %j, align 4, !dbg !29
+  call void @llvm.dbg.declare(metadata %struct.C2* %c2, metadata !30, metadata !DIExpression()), !dbg !36
+  %c = getelementptr inbounds %struct.C2, %struct.C2* %c2, i32 0, i32 0, !dbg !37
+  %2 = load i8, i8* %c, align 1, !dbg !38
+  %inc2 = add i8 %2, 1, !dbg !38
+  store i8 %inc2, i8* %c, align 1, !dbg !38
+  ret void, !dbg !39
+}
+
+declare void @llvm.dbg.declare(metadata, metadata, metadata)
+
+define dso_local void @_Z3foov() !dbg !40 {
+entry:
+  %a1.i = alloca %struct.A1, align 4
+  %i1.i = alloca %struct.I1, align 4
+  %c1.i = alloca %struct.C1, align 1
+  %a2.i = alloca %struct.A2, align 4
+  %i2.i = alloca %struct.I2, align 4
+  %c2.i = alloca %struct.C2, align 1
+  %a3 = alloca %struct.A3, align 4
+  %i3 = alloca %struct.I3, align 4
+  %c3 = alloca %struct.C3, align 1
+  call void @llvm.dbg.declare(metadata %struct.A3* %a3, metadata !41, metadata !DIExpression()), !dbg !47
+  %i = getelementptr inbounds %struct.A3, %struct.A3* %a3, i32 0, i32 0, !dbg !48
+  %0 = load i32, i32* %i, align 4, !dbg !49
+  %inc = add nsw i32 %0, 1, !dbg !49
+  store i32 %inc, i32* %i, align 4, !dbg !49
+  call void @llvm.dbg.declare(metadata %struct.I3* %i3, metadata !50, metadata !DIExpression()), !dbg !56
+  %j = getelementptr inbounds %struct.I3, %struct.I3* %i3, i32 0, i32 0, !dbg !57
+  %1 = load i32, i32* %j, align 4, !dbg !58
+  %inc1 = add nsw i32 %1, 1, !dbg !58
+  store i32 %inc1, i32* %j, align 4, !dbg !58
+  call void @llvm.dbg.declare(metadata %struct.C3* %c3, metadata !59, metadata !DIExpression()), !dbg !64
+  %c = getelementptr inbounds %struct.C3, %struct.C3* %c3, i32 0, i32 0, !dbg !65
+  %2 = load i8, i8* %c, align 1, !dbg !66
+  %inc2 = add i8 %2, 1, !dbg !66
+  store i8 %inc2, i8* %c, align 1, !dbg !66
+  call void @llvm.dbg.declare(metadata %struct.A1* %a1.i, metadata !67, metadata !DIExpression()), !dbg !73
+  %i.i3 = getelementptr inbounds %struct.A1, %struct.A1* %a1.i, i32 0, i32 0, !dbg !75
+  %3 = load i32, i32* %i.i3, align 4, !dbg !76
+  %inc.i4 = add nsw i32 %3, 1, !dbg !76
+  store i32 %inc.i4, i32* %i.i3, align 4, !dbg !76
+  call void @llvm.dbg.declare(metadata %struct.I1* %i1.i, metadata !77, metadata !DIExpression()), !dbg !83
+  %j.i5 = getelementptr inbounds %struct.I1, %struct.I1* %i1.i, i32 0, i32 0, !dbg !84
+  %4 = load i32, i32* %j.i5, align 4, !dbg !85
+  %inc1.i6 = add nsw i32 %4, 1, !dbg !85
+  store i32 %inc1.i6, i32* %j.i5, align 4, !dbg !85
+  call void @llvm.dbg.declare(metadata %struct.C1* %c1.i, metadata !86, metadata !DIExpression()), !dbg !91
+  %c.i7 = getelementptr inbounds %struct.C1, %struct.C1* %c1.i, i32 0, i32 0, !dbg !92
+  %5 = load i8, i8* %c.i7, align 1, !dbg !93
+  %inc2.i8 = add i8 %5, 1, !dbg !93
+  store i8 %inc2.i8, i8* %c.i7, align 1, !dbg !93
+  call void @llvm.dbg.declare(metadata %struct.A2* %a2.i, metadata !12, metadata !DIExpression()), !dbg !94
+  %i.i = getelementptr inbounds %struct.A2, %struct.A2* %a2.i, i32 0, i32 0, !dbg !96
+  %6 = load i32, i32* %i.i, align 4, !dbg !97
+  %inc.i = add nsw i32 %6, 1, !dbg !97
+  store i32 %inc.i, i32* %i.i, align 4, !dbg !97
+  call void @llvm.dbg.declare(metadata %struct.I2* %i2.i, metadata !21, metadata !DIExpression()), !dbg !98
+  %j.i = getelementptr inbounds %struct.I2, %struct.I2* %i2.i, i32 0, i32 0, !dbg !99
+  %7 = load i32, i32* %j.i, align 4, !dbg !100
+  %inc1.i = add nsw i32 %7, 1, !dbg !100
+  store i32 %inc1.i, i32* %j.i, align 4, !dbg !100
+  call void @llvm.dbg.declare(metadata %struct.C2* %c2.i, metadata !30, metadata !DIExpression()), !dbg !101
+  %c.i = getelementptr inbounds %struct.C2, %struct.C2* %c2.i, i32 0, i32 0, !dbg !102
+  %8 = load i8, i8* %c.i, align 1, !dbg !103
+  %inc2.i = add i8 %8, 1, !dbg !103
+  store i8 %inc2.i, i8* %c.i, align 1, !dbg !103
+  ret void, !dbg !104
+}
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3, !4, !5, !6}
+!llvm.ident = !{!7}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 14.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "test.cpp", directory: "/")
+!2 = !{i32 7, !"Dwarf Version", i32 4}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!4 = !{i32 1, !"wchar_size", i32 4}
+!5 = !{i32 7, !"uwtable", i32 1}
+!6 = !{i32 7, !"frame-pointer", i32 2}
+!7 = !{!"clang version 14.0.0"}
+!8 = distinct !DISubprogram(name: "not_removed", linkageName: "_Z11not_removedv", scope: !1, file: !1, line: 17, type: !9, scopeLine: 17, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !11)
+!9 = !DISubroutineType(types: !10)
+!10 = !{null}
+!11 = !{}
+!12 = !DILocalVariable(name: "a2", scope: !13, file: !1, line: 23, type: !14)
+!13 = distinct !DILexicalBlock(scope: !8, file: !1, line: 20, column: 3)
+!14 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "A2", scope: !8, file: !1, line: 18, size: 32, flags: DIFlagTypePassByValue, elements: !15)
+!15 = !{!16}
+!16 = !DIDerivedType(tag: DW_TAG_member, name: "i", scope: !14, file: !1, line: 18, baseType: !17, size: 32)
+!17 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!18 = !DILocation(line: 23, column: 8, scope: !13)
+!19 = !DILocation(line: 23, column: 15, scope: !13)
+!20 = !DILocation(line: 23, column: 16, scope: !13)
+!21 = !DILocalVariable(name: "i2", scope: !22, file: !1, line: 25, type: !23)
+!22 = distinct !DILexicalBlock(scope: !13, file: !1, line: 24, column: 5)
+!23 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "I2", scope: !13, file: !1, line: 21, size: 32, flags: DIFlagTypePassByValue, elements: !24)
+!24 = !{!25}
+!25 = !DIDerivedType(tag: DW_TAG_member, name: "j", scope: !23, file: !1, line: 21, baseType: !26, size: 32)
+!26 = !DIDerivedType(tag: DW_TAG_typedef, name: "Int2", scope: !8, file: !1, line: 19, baseType: !17)
+!27 = !DILocation(line: 25, column: 10, scope: !22)
+!28 = !DILocation(line: 25, column: 17, scope: !22)
+!29 = !DILocation(line: 25, column: 18, scope: !22)
+!30 = !DILocalVariable(name: "c2", scope: !22, file: !1, line: 26, type: !31)
+!31 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "C2", scope: !13, file: !1, line: 22, size: 8, flags: DIFlagTypePassByValue, elements: !32)
+!32 = !{!33}
+!33 = !DIDerivedType(tag: DW_TAG_member, name: "c", scope: !31, file: !1, line: 22, baseType: !34, size: 8)
+!34 = !DIDerivedType(tag: DW_TAG_typedef, name: "Char2", scope: !31, file: !1, line: 22, baseType: !35)
+!35 = !DIBasicType(name: "char", size: 8, encoding: DW_ATE_signed_char)
+!36 = !DILocation(line: 26, column: 10, scope: !22)
+!37 = !DILocation(line: 26, column: 17, scope: !22)
+!38 = !DILocation(line: 26, column: 18, scope: !22)
+!39 = !DILocation(line: 29, column: 1, scope: !8)
+!40 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !1, file: !1, line: 31, type: !9, scopeLine: 31, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !11)
+!41 = !DILocalVariable(name: "a3", scope: !42, file: !1, line: 38, type: !44)
+!42 = distinct !DILexicalBlock(scope: !43, file: !1, line: 36, column: 5)
+!43 = distinct !DILexicalBlock(scope: !40, file: !1, line: 34, column: 3)
+!44 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "A3", scope: !40, file: !1, line: 32, size: 32, flags: DIFlagTypePassByValue, elements: !45)
+!45 = !{!46}
+!46 = !DIDerivedType(tag: DW_TAG_member, name: "i", scope: !44, file: !1, line: 32, baseType: !17, size: 32)
+!47 = !DILocation(line: 38, column: 10, scope: !42)
+!48 = !DILocation(line: 38, column: 17, scope: !42)
+!49 = !DILocation(line: 38, column: 18, scope: !42)
+!50 = !DILocalVariable(name: "i3", scope: !51, file: !1, line: 40, type: !52)
+!51 = distinct !DILexicalBlock(scope: !42, file: !1, line: 39, column: 7)
+!52 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "I3", scope: !43, file: !1, line: 35, size: 32, flags: DIFlagTypePassByValue, elements: !53)
+!53 = !{!54}
+!54 = !DIDerivedType(tag: DW_TAG_member, name: "j", scope: !52, file: !1, line: 35, baseType: !55, size: 32)
+!55 = !DIDerivedType(tag: DW_TAG_typedef, name: "Int3", scope: !40, file: !1, line: 33, baseType: !17)
+!56 = !DILocation(line: 40, column: 12, scope: !51)
+!57 = !DILocation(line: 40, column: 19, scope: !51)
+!58 = !DILocation(line: 40, column: 20, scope: !51)
+!59 = !DILocalVariable(name: "c3", scope: !51, file: !1, line: 41, type: !60)
+!60 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "C3", scope: !42, file: !1, line: 37, size: 8, flags: DIFlagTypePassByValue, elements: !61)
+!61 = !{!62}
+!62 = !DIDerivedType(tag: DW_TAG_member, name: "c", scope: !60, file: !1, line: 37, baseType: !63, size: 8)
+!63 = !DIDerivedType(tag: DW_TAG_typedef, name: "Char3", scope: !60, file: !1, line: 37, baseType: !35)
+!64 = !DILocation(line: 41, column: 12, scope: !51)
+!65 = !DILocation(line: 41, column: 19, scope: !51)
+!66 = !DILocation(line: 41, column: 20, scope: !51)
+!67 = !DILocalVariable(name: "a1", scope: !68, file: !1, line: 8, type: !70)
+!68 = distinct !DILexicalBlock(scope: !69, file: !1, line: 5, column: 3)
+!69 = distinct !DISubprogram(name: "removed", linkageName: "_Z7removedv", scope: !1, file: !1, line: 2, type: !9, scopeLine: 2, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !11)
+!70 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "A1", scope: !69, file: !1, line: 3, size: 32, flags: DIFlagTypePassByValue, elements: !71, identifier: "_ZTSZ7removedvE2A1")
+!71 = !{!72}
+!72 = !DIDerivedType(tag: DW_TAG_member, name: "i", scope: !70, file: !1, line: 3, baseType: !17, size: 32)
+!73 = !DILocation(line: 8, column: 8, scope: !68, inlinedAt: !74)
+!74 = distinct !DILocation(line: 45, column: 3, scope: !40)
+!75 = !DILocation(line: 8, column: 15, scope: !68, inlinedAt: !74)
+!76 = !DILocation(line: 8, column: 16, scope: !68, inlinedAt: !74)
+!77 = !DILocalVariable(name: "i1", scope: !78, file: !1, line: 10, type: !79)
+!78 = distinct !DILexicalBlock(scope: !68, file: !1, line: 9, column: 5)
+!79 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "I1", scope: !68, file: !1, line: 6, size: 32, flags: DIFlagTypePassByValue, elements: !80, identifier: "_ZTSZ7removedvE2I1")
+!80 = !{!81}
+!81 = !DIDerivedType(tag: DW_TAG_member, name: "j", scope: !79, file: !1, line: 6, baseType: !82, size: 32)
+!82 = !DIDerivedType(tag: DW_TAG_typedef, name: "Int1", scope: !69, file: !1, line: 4, baseType: !17)
+!83 = !DILocation(line: 10, column: 10, scope: !78, inlinedAt: !74)
+!84 = !DILocation(line: 10, column: 17, scope: !78, inlinedAt: !74)
+!85 = !DILocation(line: 10, column: 18, scope: !78, inlinedAt: !74)
+!86 = !DILocalVariable(name: "c1", scope: !78, file: !1, line: 11, type: !87)
+!87 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "C1", scope: !68, file: !1, line: 7, size: 8, flags: DIFlagTypePassByValue, elements: !88, identifier: "_ZTSZ7removedvE2C1")
+!88 = !{!89}
+!89 = !DIDerivedType(tag: DW_TAG_member, name: "c", scope: !87, file: !1, line: 7, baseType: !90, size: 8)
+!90 = !DIDerivedType(tag: DW_TAG_typedef, name: "Char1", scope: !87, file: !1, line: 7, baseType: !35)
+!91 = !DILocation(line: 11, column: 10, scope: !78, inlinedAt: !74)
+!92 = !DILocation(line: 11, column: 17, scope: !78, inlinedAt: !74)
+!93 = !DILocation(line: 11, column: 18, scope: !78, inlinedAt: !74)
+!94 = !DILocation(line: 23, column: 8, scope: !13, inlinedAt: !95)
+!95 = distinct !DILocation(line: 46, column: 3, scope: !40)
+!96 = !DILocation(line: 23, column: 15, scope: !13, inlinedAt: !95)
+!97 = !DILocation(line: 23, column: 16, scope: !13, inlinedAt: !95)
+!98 = !DILocation(line: 25, column: 10, scope: !22, inlinedAt: !95)
+!99 = !DILocation(line: 25, column: 17, scope: !22, inlinedAt: !95)
+!100 = !DILocation(line: 25, column: 18, scope: !22, inlinedAt: !95)
+!101 = !DILocation(line: 26, column: 10, scope: !22, inlinedAt: !95)
+!102 = !DILocation(line: 26, column: 17, scope: !22, inlinedAt: !95)
+!103 = !DILocation(line: 26, column: 18, scope: !22, inlinedAt: !95)
+!104 = !DILocation(line: 47, column: 1, scope: !40)

llvm/test/DebugInfo/Generic/namespace.ll

 ; CHECK:   DW_AT_MIPS_linkage_name
 ; CHECK:   DW_AT_name ("f1")
 ; CHECK: [[FUNC1:.*]]: DW_TAG_subprogram
 ; CHECK:   DW_AT_MIPS_linkage_name
 ; CHECK:   DW_AT_name ("f1")
 ; CHECK:   DW_TAG_formal_parameter
 ; CHECK:   NULL
 
+; CHECK: [[FUNC_FWD:0x[0-9a-f]*]]:{{.*}}DW_TAG_subprogram
+; CHECK:   DW_AT_name ("func_fwd")
+; CHECK-NOT: DW_AT_declaration
+
+; CHECK: [[I:0x[0-9a-f]*]]:{{ *}}DW_TAG_variable
+; CHECK:   DW_AT_name ("i")
+; CHECK: [[VAR_FWD:0x[0-9a-f]*]]:{{ *}}DW_TAG_variable
+; CHECK:   DW_AT_name ("var_fwd")
+
 ; CHECK: [[FOO:0x[0-9a-f]*]]:{{ *}}DW_TAG_structure_type
 ; CHECK:   DW_AT_name ("foo")
 ; CHECK:   DW_AT_declaration
 ; CHECK: [[BAR:0x[0-9a-f]*]]:{{ *}}DW_TAG_structure_type
 ; CHECK:   DW_AT_name ("bar")
 
-; CHECK: [[I:0x[0-9a-f]*]]:{{ *}}DW_TAG_variable
-; CHECK:   DW_AT_name ("i")
-
 ; CHECK: [[BAZ:0x[0-9a-f]*]]:{{.*}}DW_TAG_typedef
 ; CHECK:   DW_AT_name ("baz")
 
 ; CHECK: [[VAR_DECL:0x[0-9a-f]*]]:{{.*}}DW_TAG_variable
 ; CHECK:   DW_AT_name ("var_decl")
 ; CHECK:   DW_AT_declaration
 
 ; CHECK: [[FUNC_DECL:0x[0-9a-f]*]]:{{.*}}DW_TAG_subprogram
 ; CHECK:   DW_AT_name ("func_decl")
 ; CHECK:   DW_AT_declaration
-
-; CHECK: [[VAR_FWD:0x[0-9a-f]*]]:{{ *}}DW_TAG_variable
-; CHECK:   DW_AT_name ("var_fwd")
-
-; CHECK: [[FUNC_FWD:0x[0-9a-f]*]]:{{.*}}DW_TAG_subprogram
-; CHECK:   DW_AT_name ("func_fwd")
-; CHECK-NOT: DW_AT_declaration
 ; CHECK: NULL
 
 ; CHECK: DW_TAG_imported_module
 ; CHECK:   DW_AT_decl_file ([[F2:.*]])
 ; CHECK:   DW_AT_decl_line (15)
 ; CHECK:   DW_AT_import ([[NS2]])
 ; CHECK: DW_TAG_imported_declaration
 ; CHECK: NULL
 
 ; CHECK: DW_TAG_subprogram
 ; CHECK: DW_TAG_subprogram
 ; CHECK:   DW_AT_MIPS_linkage_name
 ; CHECK:   DW_AT_name ("func")
 ; CHECK:   DW_TAG_formal_parameter
+; CHECK:   DW_TAG_lexical_block
+; CHECK:     DW_TAG_imported_module
+; CHECK:       DW_AT_decl_file ([[F2]])
+; CHECK:       DW_AT_decl_line (23)
+; CHECK:       DW_AT_import {{.*}}
+; CHECK:     NULL
 ; CHECK:   DW_TAG_imported_module
 ; CHECK:     DW_AT_decl_file ([[F2:.*]])
 ; CHECK:     DW_AT_decl_line (26)
 ; CHECK:     DW_AT_import ([[NS1]])
 ; CHECK:   DW_TAG_imported_declaration
 ; CHECK:     DW_AT_decl_file ([[F2]])
 ; CHECK:     DW_AT_decl_line (27)
 ; CHECK:     DW_AT_import ([[FOO]])

 ; CHECK:   DW_TAG_imported_declaration
 ; CHECK:     DW_AT_decl_file ([[F2]])
 ; CHECK:     DW_AT_decl_line (36)
 ; CHECK:     DW_AT_import ([[VAR_FWD]])
 ; CHECK:   DW_TAG_imported_declaration
 ; CHECK:     DW_AT_decl_file ([[F2]])
 ; CHECK:     DW_AT_decl_line (37)
 ; CHECK:     DW_AT_import ([[FUNC_FWD]])
-; CHECK:   DW_TAG_lexical_block
-; CHECK:     DW_TAG_imported_module
-; CHECK:       DW_AT_decl_file ([[F2]])
-; CHECK:       DW_AT_decl_line (23)
-; CHECK:       DW_AT_import {{.*}}
-; CHECK:     NULL
 ; CHECK:   NULL
 
-; CHECK: DW_TAG_base_type
 ; CHECK: DW_TAG_subprogram
+; CHECK: DW_TAG_base_type
 ; CHECK: DW_TAG_imported_module
 ; CHECK:   DW_AT_decl_file ([[F2:.*]])
 ; CHECK:   DW_AT_decl_line (18)
 ; CHECK:   DW_AT_import ([[NS1]])
 ; CHECK: DW_TAG_imported_declaration
 ; CHECK: DW_TAG_base_type
 ; CHECK: NULL
 

llvm/test/DebugInfo/Generic/split-dwarf-local-import.ll

+; RUN: %llc_dwarf -O1 -filetype=obj -split-dwarf-file=%t.dwo < %s | llvm-dwarfdump -debug-info - | FileCheck %s --implicit-check-not "{{DW_TAG|NULL}}"
+; REQUIRES: object-emission
+
+; CHECK-LABEL: debug_info contents
+; CHECK: DW_TAG_compile_unit
+; CHECK:   DW_AT_GNU_dwo_name
+; CHECK:   DW_AT_GNU_dwo_id
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:     NULL
+; CHECK:   NULL
+
+; CHECK: DW_TAG_compile_unit
+; CHECK-NOT: DW_AT_GNU_dwo_name
+; CHECK-NOT: DW_AT_GNU_dwo_id
+; CHECK:   DW_TAG_subprogram
+; CHECK:   NULL
+
+; CHECK-LABEL: debug_info.dwo contents
+
+; CHECK: DW_TAG_compile_unit
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_TAG_imported_declaration
+; CHECK:     NULL
+; CHECK:   DW_TAG_subprogram
+; CHECK:     DW_TAG_inlined_subroutine
+; CHECK:     NULL
+; CHECK:   DW_TAG_namespace
+; CHECK:     DW_TAG_structure_type
+; CHECK:     NULL
+; CHECK:   DW_TAG_base_type
+; CHECK:   NULL
+
+%"struct.ns::t1" = type { i8 }
+
+declare void @llvm.dbg.declare(metadata, metadata, metadata)
+
+declare dso_local void @_Z3pinv() local_unnamed_addr
+
+define dso_local i32 @main() local_unnamed_addr !dbg !18 {
+entry:
+  call void @llvm.dbg.declare(metadata %"struct.ns::t1"* undef, metadata !22, metadata !DIExpression()), !dbg !23
+  call void @_Z3pinv(), !dbg !25
+  ret i32 0, !dbg !26
+}
+
+!llvm.dbg.cu = !{!0, !10}
+!llvm.ident = !{!12, !12}
+!llvm.module.flags = !{!13, !14, !15, !16, !17}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 14.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, imports: !2, splitDebugInlining: true, nameTableKind: GNU)
+!1 = !DIFile(filename: "a.cpp", directory: "/")
+!2 = !{!3}
+!3 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !4, entity: !8, file: !1, line: 3)
+!4 = distinct !DISubprogram(name: "f1", linkageName: "_Z2f1v", scope: !1, file: !1, line: 3, type: !5, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !7)
+!5 = !DISubroutineType(types: !6)
+!6 = !{null}
+!7 = !{}
+!8 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "t1", scope: !9, file: !1, line: 1, size: 8, flags: DIFlagTypePassByValue, elements: !7, identifier: "_ZTSN2ns2t1E")
+!9 = !DINamespace(name: "ns", scope: null)
+!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, producer: "clang version 14.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: true, nameTableKind: GNU)
+!11 = !DIFile(filename: "b.cpp", directory: "/")
+!12 = !{!"clang version 14.0.0"}
+!13 = !{i32 7, !"Dwarf Version", i32 4}
+!14 = !{i32 2, !"Debug Info Version", i32 3}
+!15 = !{i32 1, !"wchar_size", i32 4}
+!16 = !{i32 7, !"uwtable", i32 1}
+!17 = !{i32 7, !"frame-pointer", i32 2}
+!18 = distinct !DISubprogram(name: "main", scope: !11, file: !11, line: 2, type: !19, scopeLine: 2, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !10, retainedNodes: !7)
+!19 = !DISubroutineType(types: !20)
+!20 = !{!21}
+!21 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!22 = !DILocalVariable(name: "v1", scope: !4, file: !1, line: 3, type: !8)
+!23 = !DILocation(line: 3, column: 37, scope: !4, inlinedAt: !24)
+!24 = distinct !DILocation(line: 3, column: 3, scope: !18)
+!25 = !DILocation(line: 3, column: 41, scope: !4, inlinedAt: !24)
+!26 = !DILocation(line: 4, column: 1, scope: !18)

llvm/test/DebugInfo/Generic/split-dwarf-local-import2.ll

+; RUN: %llc_dwarf -split-dwarf-file=%t.dwo < %s | FileCheck %s
+
+; Ensure function-local DW_TAG_imported_declaration get skipped if parent subprogram was not emitted.
+; CHECK-NOT: DW_TAG_imported_declaration
+
+define void @f1() !dbg !13 {
+lbl:
+  ret void, !dbg !16
+}
+
+define void @f2() !dbg !22 {
+lbl:
+  ret void, !dbg !23
+}
+
+!llvm.dbg.cu = !{!0, !2, !10}
+!llvm.module.flags = !{!12}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, emissionKind: FullDebug)
+!1 = !DIFile(filename: "a.cc", directory: "")
+!2 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !3, emissionKind: FullDebug, imports: !4)
+!3 = !DIFile(filename: "b.cc", directory: "")
+!4 = !{!5}
+!5 = !DIImportedEntity(tag: DW_TAG_imported_declaration, scope: !6, entity: !7)
+!6 = !DISubprogram(scope: null, spFlags: DISPFlagOptimized)
+!7 = !DINamespace(scope: !2)
+!8 = !DISubroutineType(types: !9)
+!9 = !{}
+!10 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !11, emissionKind: FullDebug)
+!11 = !DIFile(filename: "c.cc", directory: "")
+!12 = !{i32 2, !"Debug Info Version", i32 3}
+!13 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
+!16 = !DILocation(line: 0, scope: !17, inlinedAt: !18)
+!17 = distinct !DISubprogram(scope: null, unit: !10)
+!18 = !DILocation(line: 0, scope: !21)
+!21 = !DILexicalBlockFile(scope: !13, discriminator: 0)
+!22 = distinct !DISubprogram(scope: null, type: !8, spFlags: DISPFlagDefinition, unit: !0)
+!23 = !DILocation(line: 0, scope: !24, inlinedAt: !25)
+!24 = distinct !DISubprogram(scope: null, unit: !2)
+!25 = !DILocation(line: 0, scope: !22)

llvm/test/Instrumentation/InstrProfiling/debug-info-correlate.ll

 ; CHECK:      ![[SCOPE]] = {{.*}} !DISubprogram(name: "foo"
 ; CHECK:      ![[ANNOTATIONS]] = !{![[NAME:[0-9]+]], ![[HASH:[0-9]+]], ![[COUNTERS:[0-9]+]]}
 ; CHECK:      ![[NAME]] = !{!"Function Name", !"foo"}
 ; CHECK:      ![[HASH]] = !{!"CFG Hash", i64 12345678}
 ; CHECK:      ![[COUNTERS]] = !{!"Num Counters", i32 2}
 
 define void @_Z3foov() !dbg !12 {
   call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__profn_foo, i32 0, i32 0), i64 12345678, i32 2, i32 0)
-  ret void
+  ret void, !dbg !17
 }
 
 declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
 
 !llvm.dbg.cu = !{!0}
 !llvm.module.flags = !{!2, !3, !4, !5, !6, !7, !8, !9, !10}
 !llvm.ident = !{!11}
 

 !9 = !{i32 7, !"uwtable", i32 1}
 !10 = !{i32 7, !"frame-pointer", i32 1}
 !11 = !{!"clang version 14.0.0"}
 !12 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !13, file: !13, line: 1, type: !14, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !16)
 !13 = !DIFile(filename: "debug-info-correlate.cpp", directory: "")
 !14 = !DISubroutineType(types: !15)
 !15 = !{null}
 !16 = !{}
+!17 = !DILocation(line: 1, scope: !12)
 
 ; CHECK-DWARF: DW_TAG_compile_unit
 ; CHECK-DWARF:   DW_TAG_subprogram
 ; CHECK-DWARF:     DW_AT_name ("foo")
 ; CHECK-DWARF:     DW_TAG_variable
 ; CHECK-DWARF:       DW_AT_name ("__profc_foo")
 ; CHECK-DWARF:       DW_AT_type ({{.*}} "Profile Data Type")
 ; CHECK-DWARF:       DW_TAG_LLVM_annotation