[Clang] Harmonize Split DWARF options with llc

[aaronpuchert]

Summary

With Split DWARF the resulting object file (then called skeleton CU)
contains the file name of another ("DWO") file with the debug info.
This can be a problem for remote compilation, as it will contain the
name of the file on the compilation server, not on the client.

To use Split DWARF with remote compilation, one needs to either

• make sure only relative paths are used, and mirror the build directory structure of the client on the server,
• inject the desired file name on the client directly.

Since llc already supports the latter solution, we're just copying that
over. We allow setting the actual output filename separately from the
value of the DW_AT_[GNU_]dwo_name attribute in the skeleton CU.

Fixes PR40276.

[dblaikie]

Pleasue include mention of the bug (PR40276) in the commit message & clarify that while this is useful for some remote compilation models, it's not strictly necessary/the only way to do it (a remote compilation model that keeps relative paths and uses compilation-dir instead can work without this attribute)

Use llvm-dwarfdump rather than llvm-objdump to dump the contents of the debug_info section and test the dwo_name there (rather than dumping hex) & probably the objdump part of the test isn't needed? (& I guess there's an LLVM patch to add the rest of this functionality?)

[aaronpuchert]

In D59673#1438716, @dblaikie wrote:

Use llvm-dwarfdump rather than llvm-objdump to dump the contents of the debug_info section and test the dwo_name there (rather than dumping hex)

I didn't know about llvm-dwarfdump, I wondered why llvm-objdump wouldn't pretty-print the debug info as objdump does. That makes a lot of sense then.

probably the objdump part of the test isn't needed?

Actually I only need to check that the DWO file is there (under the proper file name). Any ideas?

(& I guess there's an LLVM patch to add the rest of this functionality?)

What do we have to do there? Should we add the option to llc as well?

[aaronpuchert]

Use llvm-dwarfdump to inspect debug info, remove unneeded flags.

[dblaikie]

In D59673#1438888, @aaronpuchert wrote:

In D59673#1438716, @dblaikie wrote:

Use llvm-dwarfdump rather than llvm-objdump to dump the contents of the debug_info section and test the dwo_name there (rather than dumping hex)

I didn't know about llvm-dwarfdump, I wondered why llvm-objdump wouldn't pretty-print the debug info as objdump does. That makes a lot of sense then.

probably the objdump part of the test isn't needed?

Actually I only need to check that the DWO file is there (under the proper file name). Any ideas?

Ah, fair. You could actually test the dwo_name is accurate in the .dwo file (I added the dwo_name to the .dwo file so that multi-level dwp error messages could be more informative)

(& I guess there's an LLVM patch to add the rest of this functionality?)

What do we have to do there? Should we add the option to llc as well?

Yep - pretty much anything in MCOptions should be surfaced through llc for llvm-level testing.

[aaronpuchert]

In D59673#1452269, @dblaikie wrote:

Ah, fair. You could actually test the dwo_name is accurate in the .dwo file (I added the dwo_name to the .dwo file so that multi-level dwp error messages could be more informative)

Ok, I'll just check the dwo_name for both files then.

(& I guess there's an LLVM patch to add the rest of this functionality?)

What do we have to do there? Should we add the option to llc as well?

Yep - pretty much anything in MCOptions should be surfaced through llc for llvm-level testing.

I was about to add this when I discovered that there is such an option already. The command line clang -cc1 -split-dwarf-file a -fsplit-dwarf-dwo-name-attr b corresponds to llc -split-dwarf-output a -split-dwarf-file b. That seems a bit unfortunate, especially the different meanings of -split-dwarf-file. Should/can we strive to emulate the behavior of llc in Clang instead? That would perhaps require changes to the behavior of -split-dwarf-file. Here's how the options behave with llc:

Option	∅	-split-dwarf-file a
∅	No split	Split DI in same object, set DW_AT_GNU_dwo_name = a, no extra file
-split-dwarf-output b	No split	Split DI, set DW_AT_GNU_dwo_name = a, separate file b

Currently, Clang's -enable-split-dwarf[=split] -split-dwarf-file a does the same as -split-dwarf-output a -split-dwarf-file a in llc, and -enable-split-dwarf=single -split-dwarf-file a does the same as -split-dwarf-file a in llc.

[aaronpuchert]

Ok, here is an idea. We introduce -split-dwarf-output in Clang instead of -fsplit-dwarf-dwo-name-attr. If given, it overrides the output file name for the Split DWARF file, which we otherwise take from -split-dwarf-file. The option is obviously incompatible with -enable-split-dwarf=single, so we will disallow that. This should be backwards-compatible, and bring the behavior of llc and clang -cc1 closer together. What do you think?

[dblaikie]

In D59673#1460605, @aaronpuchert wrote:

Ok, here is an idea. We introduce -split-dwarf-output in Clang instead of -fsplit-dwarf-dwo-name-attr. If given, it overrides the output file name for the Split DWARF file, which we otherwise take from -split-dwarf-file. The option is obviously incompatible with -enable-split-dwarf=single, so we will disallow that. This should be backwards-compatible, and bring the behavior of llc and clang -cc1 closer together. What do you think?

Sure, I think the naming's a bit weird (but hard to come up with good names for any of this) - but consistency seems like a good first pass at least, given we're halfway there anyway.

[aaronpuchert]

In D59673#1461983, @dblaikie wrote:

Sure, I think the naming's a bit weird (but hard to come up with good names for any of this)

Agreed, -split-dwarf-output is pretty clear, but -split-dwarf-file could be both the actual filename or the attribute.

There is test/CodeGen/split-debug-filename.c checking that we emit !DICompileUnit({{.*}}, splitDebugFilename: "foo.dwo" into the IR under some circumstances, but I'm not sure why. It seems to be ignored by llc. What's the intended use for this? Your commit message in rC301063 suggests that this is needed for implicit modules. How does this work? I'm asking of course because I'm not sure whether we might need to emit both file names there.

[dblaikie]

In D59673#1465975, @aaronpuchert wrote:

In D59673#1461983, @dblaikie wrote:

Sure, I think the naming's a bit weird (but hard to come up with good names for any of this)

Agreed, -split-dwarf-output is pretty clear, but -split-dwarf-file could be both the actual filename or the attribute.

There is test/CodeGen/split-debug-filename.c checking that we emit !DICompileUnit({{.*}}, splitDebugFilename: "foo.dwo" into the IR under some circumstances, but I'm not sure why. It seems to be ignored by llc.

For implicit modular debug info I think it probably isn't ignored. See DwarfDebug.cpp:~630.

What's the intended use for this? Your commit message in rC301063 suggests that this is needed for implicit modules. How does this work? I'm asking of course because I'm not sure whether we might need to emit both file names there.

So implicit modules can use a sort of pseudo-split DWARF. The .pcm (module file) itself is an object file in this mode - with the split DWARF and a section containing the AST bitcode. Then in the normal source files that use the .pcm, they get the usual non-split DWARF, plus skeleton CUs that reference the .pcm file (& contain some other attributes for regenerating it in case it's been cleaned up from the module cache). This is why the file name can be (& has to be) passed down through the IR - it /can/ be, because this compilation isn't choosing tnhe file name (so there's no issue with the .dwo file name changing between IR generation and object generation (as there is with LTO - where the compilation doesn't know the final object name, only the linker knows that)) and it /must/ be done this way due to multiple skeleton CUs if the main CU references multiple .pcm debug info.

[aaronpuchert]

Introduce -split-dwarf-output instead to mirror the behavior of llc, which already allows to set the attribute separately from the output file name.

[aaronpuchert]

lib/CodeGen/BackendUtil.cpp

1345 ↗

(On Diff #196149)

@pcc Your documentation for DwoPath suggests that this should be the actual output filename. However, the test that you added together with this line in rC333677 doesn't fail whatever garbage I write into that field here. What can I add to that so that it fails when we don't do the right thing here?

[aaronpuchert]

Adapt one more test case.

[aaronpuchert]

Ping. I'm sorry that the change turned out so big, but note that it doesn't change the behavior of any non-cc1 flags.

[dblaikie]

Might be easier as a few patches - renaming the existing option, adding the new one, then removing the single split dwarf flag handling in favor of implying that by the absence of an output file name. (if I'm reading what this patch does)

include/clang/Basic/CodeGenOptions.h

185–188 ↗

(On Diff #196156)

Comments are a bit inconsistent - "filename for the split debug info" versus "name for the split debug info file" - makes it perhaps a bit harder to see what the important difference is between these two values.

It might be helpful to clarify "used in the skeleton CU" as "Used as the dwo_name in the DWARF" versus "the name of the file to write the .dwo sections to" or something like that?

lib/CodeGen/BackendUtil.cpp

847 ↗

(On Diff #196156)

Why did this add a check for EnableSplitDwarf here that wasn't there before?

[aaronpuchert]

In D59673#1521413, @dblaikie wrote:

Might be easier as a few patches - renaming the existing option, adding the new one, then removing the single split dwarf flag handling in favor of implying that by the absence of an output file name.

No problem, makes sense to me. I'll see if it's possible to separate these changes.

if I'm reading what this patch does

Yes, that seems to be how llc behaves, and I want to copy that.

lib/CodeGen/BackendUtil.cpp

847 ↗

(On Diff #196156)

I just changed the order, the check was there as (CodeGenOpts.getSplitDwarfMode() == CodeGenOptions::SplitFileFission) in the second line of the if.

[aaronpuchert]

In D59673#1527487, @aaronpuchert wrote:

In D59673#1521413, @dblaikie wrote:

Might be easier as a few patches - renaming the existing option, adding the new one, then removing the single split dwarf flag handling in favor of implying that by the absence of an output file name.

No problem, makes sense to me. I'll see if it's possible to separate these changes.

I have managed to split this into commits that build and test fine own their own, but there are temporary inconsistencies. If I rename -split-dwarf-file to -split-dwarf-output first, then I have to use that option for -enable-split-dwarf=single, and then I reintroduce -split-dwarf-file and have to change it back. It's not a huge problem, but it might feel a bit weird. On the other hand, the changes are indeed smaller and easier to review. I will push them soon.

[aaronpuchert]

Split from other changes as suggested. A predecessor is in D63130, and a successor will come soon.

[aaronpuchert]

Correct an oversight.

[aaronpuchert]

In D59673#1521413, @dblaikie wrote:

Might be easier as a few patches - renaming the existing option,

D63130

adding the new one,

This change.

then removing the single split dwarf flag handling in favor of implying that by the absence of an output file name. (if I'm reading what this patch does)

D63167

[dblaikie]

Looks good!

[aaronpuchert]

lib/CodeGen/BackendUtil.cpp

1345 ↗

(On Diff #196149)

@pcc Could you (or perhaps @tejohnson) comment on what the intended behavior is here, and how I can change the test so that it fails when I do the wrong thing? Is this the name of the file we write the split debug info to, or is it the value we use for the DW_AT_[GNU_]dwo_name attribute in the skeleton CU?

[tejohnson]

lib/CodeGen/BackendUtil.cpp

1345 ↗

(On Diff #196149)

It is the name of the file the split debug info is written to. If you test by changing the file name given to the -split-dwarf-file option in test/CodeGen/thinlto-split-dwarf.c, make sure you clean up the old one in your test output directory. When I tested it just now, it initially still passed because I had an old one sitting in the output directory from a prior run. Once I removed that it failed with the new name (without changing the corresponding llvm-readobj invocation).

[aaronpuchert]

lib/CodeGen/BackendUtil.cpp

1345 ↗

(On Diff #196149)

Thanks, I didn't consider that. I wasn't even aware that test output is persisted.

It seems DwoPath is used both as output filename and as value for DW_AT_[GNU_]dwo_name in the skeleton CU. llc has separate options for both: -split-dwarf-file for the attribute and -split-dwarf-output for the output filename. I want to do the same for Clang. Then we should probably separate them for LTO, too.

What is the use case for -split-dwarf-file with an individual thin backend? Is that used for distributed/remote builds? Also is there any non-cc1 way to use it? There is --plugin-opt=dwo_dir=... for the LTO linker plugin, but that seems to go a different route.

[tejohnson]

lib/CodeGen/BackendUtil.cpp

1345 ↗

(On Diff #196149)

This is the path taken for distributed ThinLTO, the plugin-opt passed to linker are for in-process ThinLTO. While -split-dwarf-file is a cc1 option, it is normally set automatically from the filename under -gsplit-dwarf (see where it is set in Clang::ConstructJob).

[aaronpuchert]

lib/CodeGen/BackendUtil.cpp

1345 ↗

(On Diff #196149)

Ok, then it makes sense to do the changes for LTO as well.

You're right that -split-dwarf-file is produced for -gsplit-dwarf, but not with LTO, right? When I set -gsplit-dwarf on a vanilla ThinLTO build, then the flag is just ignored and I'm getting the debug info in the final executables/shared objects. (At least with Clang 8.)

I'm asking because this changes the cc1-interface, and while I have adapted Clang::ConstructJob, I'm not sure how distributed ThinLTO works, and whether there have to be changes.

If it works like test/CodeGen/thinlto-split-dwarf.c, using cc1-options, then users might have to adapt after this change.

[aaronpuchert]

Make sure the flags have the same meaning for LTO. Also slightly reworded the HelpText.

[aaronpuchert]

Sorry for making the inline comments disappear, but I had to switch to the big repository.

@tejohnson Could you have a look at the LTO-related changes (BackendUtil.cpp + all files with LTO in the name)?

[tejohnson]

lgtm for the LTO bits. Suggestion below for comment.

llvm/include/llvm/LTO/Config.h

92

Probably needs a comment similar to the one below about being for running individual backends directly. Otherwise (for in process ThinLTO) we use the DwoDir.

[aaronpuchert]

llvm/include/llvm/LTO/Config.h

92

Sure. Any objections to the following wording?

/// The name for the split debug info file used for the DW_AT_[GNU_]dwo_name
/// attribute in the skeleton CU. This should generally only be used when
/// running an individual backend directly via thinBackend(), as otherwise
/// all objects would use the same .dwo file. Not used as output path.

[tejohnson]

llvm/include/llvm/LTO/Config.h

92

Looks good.

[svenvh]

As a followup to r363496, I've added llvm-dwarfdump as a clang test dependency in r364021.

clang/include/clang/Basic/CodeGenOptions.h

public:
   /// The files specified here are linked in to the module before optimizations.
   std::vector<BitcodeFileToLink> LinkBitcodeFiles;
 
   /// The user provided name for the "main file", if non-empty. This is useful
   /// in situations where the input file name does not match the original input
   /// file, for example with -save-temps.
   std::string MainFileName;
 
-  /// The name for the split debug info file that we'll break out. This is used
-  /// in the backend for setting the name in the skeleton cu.
+  /// The name for the split debug info file used for the DW_AT_[GNU_]dwo_name
+  /// attribute in the skeleton CU.
+  std::string SplitDwarfFile;
+
+  /// Output filename for the split debug info, not used in the skeleton CU.
   std::string SplitDwarfOutput;
 
   /// The name of the relocation model to use.
   llvm::Reloc::Model RelocationModel;
 
   /// The thread model to use
   std::string ThreadModel;
 

clang/include/clang/Driver/CC1Options.td

 def fblocks_runtime_optional : Flag<["-"], "fblocks-runtime-optional">,
   HelpText<"Weakly link in the blocks runtime">;
 def fexternc_nounwind : Flag<["-"], "fexternc-nounwind">,
   HelpText<"Assume all functions with C linkage do not unwind">;
 def enable_split_dwarf : Flag<["-"], "enable-split-dwarf">,
   HelpText<"Use DWARF fission in 'split' mode">;
 def enable_split_dwarf_EQ : Joined<["-"], "enable-split-dwarf=">,
   HelpText<"Set DWARF fission mode to either 'split' or 'single'">, Values<"split,single">;
+def split_dwarf_file : Separate<["-"], "split-dwarf-file">,
+  HelpText<"Name of the split dwarf debug info file to encode in the object file">;
 def fno_wchar : Flag<["-"], "fno-wchar">,
   HelpText<"Disable C++ builtin type wchar_t">;
 def fconstant_string_class : Separate<["-"], "fconstant-string-class">,
   MetaVarName<"<class name>">,
   HelpText<"Specify the class to use for constant Objective-C string objects.">;
 def fobjc_arc_cxxlib_EQ : Joined<["-"], "fobjc-arc-cxxlib=">,
   HelpText<"Objective-C++ Automatic Reference Counting standard library kind">, Values<"libc++,libstdc++,none">;
 def fobjc_runtime_has_weak : Flag<["-"], "fobjc-runtime-has-weak">,

clang/lib/CodeGen/BackendUtil.cpp

static void initTargetOptions(llvm::TargetOptions &Options,
   Options.UniqueSectionNames = CodeGenOpts.UniqueSectionNames;
   Options.EmulatedTLS = CodeGenOpts.EmulatedTLS;
   Options.ExplicitEmulatedTLS = CodeGenOpts.ExplicitEmulatedTLS;
   Options.DebuggerTuning = CodeGenOpts.getDebuggerTuning();
   Options.EmitStackSizeSection = CodeGenOpts.StackSizeSection;
   Options.EmitAddrsig = CodeGenOpts.Addrsig;
 
   if (CodeGenOpts.getSplitDwarfMode() != CodeGenOptions::NoFission)
-    Options.MCOptions.SplitDwarfFile = CodeGenOpts.SplitDwarfOutput;
+    Options.MCOptions.SplitDwarfFile = CodeGenOpts.SplitDwarfFile;
   Options.MCOptions.MCRelaxAll = CodeGenOpts.RelaxAll;
   Options.MCOptions.MCSaveTempLabels = CodeGenOpts.SaveTempLabels;
   Options.MCOptions.MCUseDwarfDirectory = !CodeGenOpts.NoDwarfDirectoryAsm;
   Options.MCOptions.MCNoExecStack = CodeGenOpts.NoExecStack;
   Options.MCOptions.MCIncrementalLinkerCompatible =
       CodeGenOpts.IncrementalLinkerCompatible;
   Options.MCOptions.MCPIECopyRelocations = CodeGenOpts.PIECopyRelocations;
   Options.MCOptions.MCFatalWarnings = CodeGenOpts.FatalWarnings;
static void runThinLTOBackend(ModuleSummaryIndex *CombinedIndex, Module *M,
   }
 
   Conf.ProfileRemapping = std::move(ProfileRemapping);
   Conf.UseNewPM = CGOpts.ExperimentalNewPassManager;
   Conf.DebugPassManager = CGOpts.DebugPassManager;
   Conf.RemarksWithHotness = CGOpts.DiagnosticsWithHotness;
   Conf.RemarksFilename = CGOpts.OptRecordFile;
   Conf.RemarksPasses = CGOpts.OptRecordPasses;
-  Conf.DwoPath = CGOpts.SplitDwarfOutput;
+  Conf.SplitDwarfFile = CGOpts.SplitDwarfFile;
+  Conf.SplitDwarfOutput = CGOpts.SplitDwarfOutput;
   switch (Action) {
   case Backend_EmitNothing:
     Conf.PreCodeGenModuleHook = [](size_t Task, const Module &Mod) {
       return false;
     };
     break;
   case Backend_EmitLL:
     Conf.PreCodeGenModuleHook = [&](size_t Task, const Module &Mod) {

clang/lib/CodeGen/CGDebugInfo.cpp

void CGDebugInfo::CreateCompileUnit() {
 
   // Create new compile unit.
   TheCU = DBuilder.createCompileUnit(
       LangTag, CUFile, CGOpts.EmitVersionIdentMetadata ? Producer : "",
       LO.Optimize || CGOpts.PrepareForLTO || CGOpts.PrepareForThinLTO,
       CGOpts.DwarfDebugFlags, RuntimeVers,
       (CGOpts.getSplitDwarfMode() != CodeGenOptions::NoFission)
           ? ""
-          : CGOpts.SplitDwarfOutput,
+          : CGOpts.SplitDwarfFile,
       EmissionKind, DwoId, CGOpts.SplitDwarfInlining,
       CGOpts.DebugInfoForProfiling,
       CGM.getTarget().getTriple().isNVPTX()
           ? llvm::DICompileUnit::DebugNameTableKind::None
           : static_cast<llvm::DICompileUnit::DebugNameTableKind>(
                 CGOpts.DebugNameTable),
       CGOpts.DebugRangesBaseAddress);
 }

clang/lib/Driver/ToolChains/Clang.cpp

RenderDebugOptions(TC, D, RawTriple, Args, EmitCodeView, IsWindowsMSVC,
                      CmdArgs, DebugInfoKind, DwarfFission);
 
   // Add the split debug info name to the command lines here so we
   // can propagate it to the backend.
   bool SplitDWARF = (DwarfFission != DwarfFissionKind::None) &&
                     TC.getTriple().isOSBinFormatELF() &&
                     (isa<AssembleJobAction>(JA) || isa<CompileJobAction>(JA) ||
                      isa<BackendJobAction>(JA));
-  const char *SplitDWARFOut;
   if (SplitDWARF) {
+    const char *SplitDWARFOut = SplitDebugName(Args, Input, Output);
+    CmdArgs.push_back("-split-dwarf-file");
+    CmdArgs.push_back(SplitDWARFOut);
+    if (DwarfFission == DwarfFissionKind::Split) {
       CmdArgs.push_back("-split-dwarf-output");
-    SplitDWARFOut = SplitDebugName(Args, Input, Output);
       CmdArgs.push_back(SplitDWARFOut);
     }
+  }
 
   // Pass the linker version in use.
   if (Arg *A = Args.getLastArg(options::OPT_mlinker_version_EQ)) {
     CmdArgs.push_back("-target-linker-version");
     CmdArgs.push_back(A->getValue());
   }
 
   if (!shouldUseLeafFramePointer(Args, RawTriple))

clang/lib/Frontend/CompilerInvocation.cpp

static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
   }
   Opts.DwarfVersion = getLastArgIntValue(Args, OPT_dwarf_version_EQ, 0, Diags);
   Opts.DebugColumnInfo = Args.hasArg(OPT_dwarf_column_info);
   Opts.EmitCodeView = Args.hasArg(OPT_gcodeview);
   Opts.CodeViewGHash = Args.hasArg(OPT_gcodeview_ghash);
   Opts.MacroDebugInfo = Args.hasArg(OPT_debug_info_macro);
   Opts.WholeProgramVTables = Args.hasArg(OPT_fwhole_program_vtables);
   Opts.LTOVisibilityPublicStd = Args.hasArg(OPT_flto_visibility_public_std);
+  Opts.SplitDwarfFile = Args.getLastArgValue(OPT_split_dwarf_file);
   Opts.SplitDwarfOutput = Args.getLastArgValue(OPT_split_dwarf_output);
   Opts.SplitDwarfInlining = !Args.hasArg(OPT_fno_split_dwarf_inlining);
 
   if (Arg *A =
           Args.getLastArg(OPT_enable_split_dwarf, OPT_enable_split_dwarf_EQ)) {
     if (A->getOption().matches(options::OPT_enable_split_dwarf)) {
       Opts.setSplitDwarfMode(CodeGenOptions::SplitFileFission);
     } else {

clang/test/CodeGen/split-debug-filename.c

 // REQUIRES: x86-registered-target
-// RUN: %clang_cc1 -debug-info-kind=limited -split-dwarf-output foo.dwo -S -emit-llvm -o - %s | FileCheck %s
-// RUN: %clang_cc1 -debug-info-kind=limited -enable-split-dwarf -split-dwarf-output foo.dwo -S -emit-llvm -o - %s | FileCheck --check-prefix=VANILLA %s
-// RUN: %clang_cc1 -triple x86_64-unknown-linux -debug-info-kind=limited -enable-split-dwarf -split-dwarf-output %t.dwo -emit-obj -o - %s | llvm-readobj -S - | FileCheck --check-prefix=O %s
+// RUN: %clang_cc1 -debug-info-kind=limited -split-dwarf-file foo.dwo -S -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -debug-info-kind=limited -enable-split-dwarf -split-dwarf-file foo.dwo -S -emit-llvm -o - %s | FileCheck --check-prefix=VANILLA %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux -debug-info-kind=limited -enable-split-dwarf -split-dwarf-file %t.dwo -split-dwarf-output %t.dwo -emit-obj -o - %s | llvm-readobj -S - | FileCheck --check-prefix=O %s
 // RUN: llvm-readobj -S %t.dwo | FileCheck --check-prefix=DWO %s
 
 int main (void) {
   return 0;
 }
 
 // Testing to ensure that the dwo name gets output into the compile unit.
 // CHECK: !DICompileUnit({{.*}}, splitDebugFilename: "foo.dwo"
 
 // Testing to ensure that the dwo name is not output into the compile unit if
 // it's for vanilla split-dwarf rather than split-dwarf for implicit modules.
 // VANILLA-NOT: splitDebugFilename
 
 // O-NOT: .dwo
 // DWO: .dwo

clang/test/CodeGen/split-debug-output.c

+// REQUIRES: x86-registered-target
+// RUN: %clang_cc1 -triple x86_64-unknown-linux -debug-info-kind=limited -enable-split-dwarf -split-dwarf-file foo.dwo -split-dwarf-output %t -emit-obj -o - %s | llvm-dwarfdump -debug-info - | FileCheck %s
+// RUN: llvm-dwarfdump -debug-info %t | FileCheck %s
+
+int f() { return 0; }
+
+// CHECK: DW_AT_GNU_dwo_name ("foo.dwo")

clang/test/CodeGen/split-debug-single-file.c

 // REQUIRES: x86-registered-target
 
 // Testing to ensure -enable-split-dwarf=single allows to place .dwo sections into regular output object.
 //  RUN: %clang_cc1 -debug-info-kind=limited -triple x86_64-unknown-linux \
-//  RUN:   -enable-split-dwarf=single -split-dwarf-output %t.o -emit-obj -o %t.o %s
+//  RUN:   -enable-split-dwarf=single -split-dwarf-file %t.o -emit-obj -o %t.o %s
 //  RUN: llvm-readobj -S %t.o | FileCheck --check-prefix=MODE-SINGLE %s
 //  MODE-SINGLE: .dwo
 
 // Testing to ensure -enable-split-dwarf=split does not place .dwo sections into regular output object.
 //  RUN: %clang_cc1 -debug-info-kind=limited -triple x86_64-unknown-linux \
-//  RUN:   -enable-split-dwarf=split -split-dwarf-output %t.dwo -emit-obj -o %t.o %s
+//  RUN:   -enable-split-dwarf=split -split-dwarf-file %t.dwo -split-dwarf-output %t.dwo -emit-obj -o %t.o %s
 //  RUN: llvm-readobj -S %t.o | FileCheck --check-prefix=MODE-SPLIT %s
 //  MODE-SPLIT-NOT: .dwo
 
 int main (void) {
   return 0;
 }

clang/test/CodeGen/thinlto-split-dwarf.c

 // REQUIRES: x86-registered-target
 
 // RUN: %clang_cc1 -debug-info-kind=limited -triple x86_64-unknown-linux-gnu \
 // RUN:   -flto=thin -emit-llvm-bc \
 // RUN:   -o %t.o %s
 
 // RUN: llvm-lto2 run -thinlto-distributed-indexes %t.o \
 // RUN:   -o %t2.index \
 // RUN:   -r=%t.o,main,px
 
 // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu \
 // RUN:   -emit-obj -fthinlto-index=%t.o.thinlto.bc \
-// RUN:   -o %t.native.o -split-dwarf-output %t.native.dwo -x ir %t.o
+// RUN:   -o %t.native.o -split-dwarf-file %t.file.dwo \
+// RUN:   -split-dwarf-output %t.output.dwo -x ir %t.o
 
-// RUN: llvm-readobj -S %t.native.o | FileCheck --check-prefix=O %s
-// RUN: llvm-readobj -S %t.native.dwo | FileCheck --check-prefix=DWO %s
+// RUN: llvm-dwarfdump %t.native.o | FileCheck --check-prefix=O %s
+// RUN: llvm-dwarfdump %t.output.dwo | FileCheck --check-prefix=DWO %s
 
-// O-NOT: .dwo
-// DWO: .dwo
+// O: DW_AT_GNU_dwo_name ("{{.*}}.file.dwo")
+// O-NOT: DW_TAG_subprogram
+// DWO: DW_TAG_subprogram
 
 int main() {}

clang/test/Driver/split-debug.c

 // Check that we split debug output properly
 //
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-ACTIONS < %t %s
 //
-// CHECK-ACTIONS: "-split-dwarf-output" "split-debug.dwo"
+// CHECK-ACTIONS: "-split-dwarf-file" "split-debug.dwo" "-split-dwarf-output" "split-debug.dwo"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-ACTIONS < %t %s
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf=split -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-ACTIONS < %t %s
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf=single -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-ACTIONS-SINGLE-SPLIT < %t %s
 //
 // CHECK-ACTIONS-SINGLE-SPLIT: "-enable-split-dwarf=single"
-// CHECK-ACTIONS-SINGLE-SPLIT: "-split-dwarf-output" "split-debug.o"
+// CHECK-ACTIONS-SINGLE-SPLIT: "-split-dwarf-file" "split-debug.o"
+// CHECK-ACTIONS-SINGLE-SPLIT-NOT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf=single -c -### -o %tfoo.o %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-SINGLE-SPLIT-FILENAME < %t %s
 //
-// CHECK-SINGLE-SPLIT-FILENAME: "-split-dwarf-output" "{{.*}}foo.o"
+// CHECK-SINGLE-SPLIT-FILENAME: "-split-dwarf-file" "{{.*}}foo.o"
+// CHECK-SINGLE-SPLIT-FILENAME-NOT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-macosx -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-NO-ACTIONS < %t %s
 //
 // CHECK-NO-ACTIONS-NOT: -split-dwarf
 
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -o Bad.x -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-BAD < %t %s
 //
 // CHECK-BAD-NOT: "Bad.dwo"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-OPTION < %t %s
 //
 // RUN: %clang -target x86_64-pc-freebsd12 -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-OPTION < %t %s
 //
 // RUN: %clang -target amdgcn-amd-amdhsa -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-OPTION < %t %s
 //
-// CHECK-OPTION: "-split-dwarf-output" "split-debug.dwo"
+// CHECK-OPTION: "-split-dwarf-file" "split-debug.dwo" "-split-dwarf-output" "split-debug.dwo"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-ASM < %t %s
 //
 // CHECK-ASM-NOT: objcopy
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -no-integrated-as -gsplit-dwarf -c -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-IAS < %t %s
 //
 // CHECK-IAS: objcopy
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -gmlt -fno-split-dwarf-inlining -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-GMLT-WITH-SPLIT < %t %s
 //
 // CHECK-GMLT-WITH-SPLIT: "-enable-split-dwarf"
 // CHECK-GMLT-WITH-SPLIT: "-debug-info-kind=line-tables-only"
+// CHECK-GMLT-WITH-SPLIT: "-split-dwarf-file"
 // CHECK-GMLT-WITH-SPLIT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -g -fno-split-dwarf-inlining -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-NOINLINE-WITHOUT-SPLIT < %t %s
 //
 // CHECK-NOINLINE-WITHOUT-SPLIT-NOT: "-enable-split-dwarf"
 // CHECK-NOINLINE-WITHOUT-SPLIT: "-fno-split-dwarf-inlining"
 // CHECK-NOINLINE-WITHOUT-SPLIT: "-debug-info-kind=limited"

 // CHECK-SPLIT-WITH-NOINL: "-debug-info-kind=limited"
 // CHECK-SPLIT-WITH-NOINL: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -gmlt -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-GMLT-OVER-SPLIT < %t %s
 //
 // CHECK-GMLT-OVER-SPLIT-NOT: "-enable-split-dwarf"
 // CHECK-GMLT-OVER-SPLIT: "-debug-info-kind=line-tables-only"
+// CHECK-GMLT-OVER-SPLIT-NOT: "-split-dwarf-file"
 // CHECK-GMLT-OVER-SPLIT-NOT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gmlt -gsplit-dwarf -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-SPLIT-OVER-GMLT < %t %s
 //
 // CHECK-SPLIT-OVER-GMLT: "-enable-split-dwarf" "-debug-info-kind=limited"
+// CHECK-SPLIT-OVER-GMLT: "-split-dwarf-file"
 // CHECK-SPLIT-OVER-GMLT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -g0 -fno-split-dwarf-inlining -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-G0-OVER-SPLIT < %t %s
 //
 // CHECK-G0-OVER-SPLIT-NOT: "-enable-split-dwarf"
 // CHECK-G0-OVER-SPLIT-NOT: "-debug-info-kind
+// CHECK-G0-OVER-SPLIT-NOT: "-split-dwarf-file"
 // CHECK-G0-OVER-SPLIT-NOT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf -g0 -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-G0-OVER-SPLIT < %t %s
 // RUN: %clang -target x86_64-unknown-linux-gnu -gsplit-dwarf=split -g0 -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-G0-OVER-SPLIT < %t %s
 //
 // CHECK-G0-OVER-SPLIT-NOT: "-enable-split-dwarf"
 // CHECK-G0-OVER-SPLIT-NOT: "-debug-info-kind
+// CHECK-G0-OVER-SPLIT-NOT: "-split-dwarf-file"
 // CHECK-G0-OVER-SPLIT-NOT: "-split-dwarf-output"
 
 // RUN: %clang -target x86_64-unknown-linux-gnu -g0 -gsplit-dwarf -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-SPLIT-OVER-G0 < %t %s
 // RUN: %clang -target x86_64-unknown-linux-gnu -g0 -gsplit-dwarf=split -S -### %s 2> %t
 // RUN: FileCheck -check-prefix=CHECK-SPLIT-OVER-G0 < %t %s
 //
 // CHECK-SPLIT-OVER-G0: "-enable-split-dwarf" "-debug-info-kind=limited"
+// CHECK-SPLIT-OVER-G0: "-split-dwarf-file"
 // CHECK-SPLIT-OVER-G0: "-split-dwarf-output"

llvm/include/llvm/LTO/Config.h

struct Config {
   std::string SampleProfile;
 
   /// Name remapping file for profile data.
   std::string ProfileRemapping;
 
   /// The directory to store .dwo files.
   std::string DwoDir;
 
+  /// The name for the split debug info file used for the DW_AT_[GNU_]dwo_name
+  /// attribute in the skeleton CU.
+  std::string SplitDwarfFile;
+
   /// The path to write a .dwo file to. This should generally only be used when
   /// running an individual backend directly via thinBackend(), as otherwise
-  /// all .dwo files will be written to the same path.
-  std::string DwoPath;
+  /// all .dwo files will be written to the same path. Not used in skeleton CU.
+  std::string SplitDwarfOutput;
 
   /// Optimization remarks file path.
   std::string RemarksFilename = "";
 
   /// Optimization remarks pass filter.
   std::string RemarksPasses = "";
 
   /// Whether to emit optimization remarks with hotness informations.

llvm/lib/LTO/LTOBackend.cpp

 }
 
 void codegen(Config &Conf, TargetMachine *TM, AddStreamFn AddStream,
              unsigned Task, Module &Mod) {
   if (Conf.PreCodeGenModuleHook && !Conf.PreCodeGenModuleHook(Task, Mod))
     return;
 
   std::unique_ptr<ToolOutputFile> DwoOut;
-  SmallString<1024> DwoFile(Conf.DwoPath);
+  SmallString<1024> DwoFile(Conf.SplitDwarfOutput);
   if (!Conf.DwoDir.empty()) {
     std::error_code EC;
     if (auto EC = llvm::sys::fs::create_directories(Conf.DwoDir))
       report_fatal_error("Failed to create directory " + Conf.DwoDir + ": " +
                          EC.message());
 
     DwoFile = Conf.DwoDir;
     sys::path::append(DwoFile, std::to_string(Task) + ".dwo");
-  }
+    TM->Options.MCOptions.SplitDwarfFile = DwoFile.str().str();
+  } else
+    TM->Options.MCOptions.SplitDwarfFile = Conf.SplitDwarfFile;
 
   if (!DwoFile.empty()) {
     std::error_code EC;
-    TM->Options.MCOptions.SplitDwarfFile = DwoFile.str().str();
     DwoOut = llvm::make_unique<ToolOutputFile>(DwoFile, EC, sys::fs::F_None);
     if (EC)
       report_fatal_error("Failed to open " + DwoFile + ": " + EC.message());
   }
 
   auto Stream = AddStream(Task);
   legacy::PassManager CodeGenPasses;
   if (TM->addPassesToEmitFile(CodeGenPasses, *Stream->OS,