[llvm] r317579 - Reland "Correct dwarf unwind information in function epilogue for X86"

Wed Nov 8 10:43:17 PST 2017

I confirmed this caused a regression for AArch64 and I am reducing a test
case.

On Wed, Nov 8, 2017 at 9:34 AM, Reid Kleckner <rnk at google.com> wrote:

> I suspect this change introduced a bug in compact unwind info on AArch64:
> https://logs.chromium.org/v/?s=chromium%2Fbb%2Fchromium.clan
> g%2FToTiOS%2F250%2F%2B%2Frecipes%2Fsteps%2Fcompile%2F0%2Fstdout
>
> FAILED: ios_clang_arm64/obj/third_party/opus/opus/pitch.o
> ...
> Assertion failed: (StackSize == 0 && "We already have the CFA offset!"),
> function generateCompactUnwindEncoding, file /b/c/b/ToTiOS/src/third_party/
> llvm/lib/Target/AArch64/MCTargetDesc/AArch64AsmBackend.cpp, line 487.
> 0  clang                    0x00000001035d7fb8
> llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 40
> 1  clang                    0x00000001035d86b6
> llvm::sys::PrintStackTraceOnErrorSignal(llvm::StringRef, bool) + 838
> 2  libsystem_platform.dylib 0x00007fffeb995bba _sigtramp + 26
> 3  clang                    0x0000000103294e79
> llvm::MCCVLineEntry::Make(llvm::MCObjectStreamer*) + 25
> 4  libsystem_c.dylib        0x00007fffeb81c420 abort + 129
> 5  libsystem_c.dylib        0x00007fffeb7e3893 basename_r + 0
> 6  clang                    0x0000000101e22067
> llvm::createAArch64beAsmBackend(llvm::Target const&, llvm::MCRegisterInfo
> const&, llvm::Triple const&, llvm::StringRef, llvm::MCTargetOptions const&)
> + 8631
> 7  clang                    0x00000001032bf4ef
> llvm::MCStreamer::generateCompactUnwindEncodings(llvm::MCAsmBackend*) + 79
> 8  clang                    0x00000001032a1196
> llvm::MCDwarfFrameEmitter::Emit(llvm::MCObjectStreamer&,
> llvm::MCAsmBackend*, bool) + 38
> 9  clang                    0x00000001032b806e
> llvm::MCObjectStreamer::EmitFrames(llvm::MCAsmBackend*) + 62
> 10 clang                    0x00000001032b12f5
> llvm::createMachOStreamer(llvm::MCContext&, std::__1::unique_ptr<llvm::MCAsmBackend,
> std::__1::default_delete<llvm::MCAsmBackend> >&&,
> llvm::raw_pwrite_stream&, std::__1::unique_ptr<llvm::MCCodeEmitter,
> std::__1::default_delete<llvm::MCCodeEmitter> >&&, bool, bool, bool) +
> 5749
> 11 clang                    0x00000001032c37f2 llvm::MCStreamer::Finish()
> + 306
> 12 clang                    0x0000000103d6e271
> llvm::AsmPrinter::doFinalization(llvm::Module&) + 3489
> 13 clang                    0x0000000102ffa740
> llvm::FPPassManager::doFinalization(llvm::Module&) + 80
> 14 clang                    0x0000000102ffac8d
> llvm::legacy::PassManagerImpl::run(llvm::Module&) + 1277
> 15 clang                    0x000000010380cd33
> clang::EmitBackendOutput(clang::DiagnosticsEngine&,
> clang::HeaderSearchOptions const&, clang::CodeGenOptions const&,
> clang::TargetOptions const&, clang::LangOptions const&, llvm::DataLayout
> const&, llvm::Module*, clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream,
> std::__1::default_delete<llvm::raw_pwrite_stream> >) + 15571
> 16 clang                    0x0000000103a44ed5
> clang::EmitObjAction::EmitObjAction(llvm::LLVMContext*) + 2341
> 17 clang                    0x0000000103d10ccc
> clang::MultiplexConsumer::HandleTranslationUnit(clang::ASTContext&) + 44
> 18 clang                    0x00000001044cdfb5
> clang::ParseAST(clang::Sema&, bool, bool) + 469
> 19 clang                    0x0000000103ccdfdc
> clang::FrontendAction::Execute() + 76
> 20 clang                    0x0000000103c86e41
> clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) + 1217
> 21 clang                    0x0000000103d340fa
> clang::ExecuteCompilerInvocation(clang::CompilerInstance*) + 4970
> 22 clang                    0x0000000101c86a57
> cc1_main(llvm::ArrayRef<char const*>, char const*, void*) + 1399
> 23 clang                    0x0000000101c850df main + 11951
> 24 libdyld.dylib            0x00007fffeb788255 start + 1
>
> I'll try to get a reproducer.
>
> On Tue, Nov 7, 2017 at 6:40 AM, Petar Jovanovic via llvm-commits <
> llvm-commits at lists.llvm.org> wrote:
>
>> Author: petarj
>> Date: Tue Nov  7 06:40:27 2017
>> New Revision: 317579
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=317579&view=rev
>> Log:
>> Reland "Correct dwarf unwind information in function epilogue for X86"
>>
>> Reland r317100 with minor fix regarding ComputeCommonTailLength function
>> in
>> BranchFolding.cpp. Skipping top CFI instructions block needs to executed
>> on
>> several more return points in ComputeCommonTailLength().
>>
>> Original r317100 message:
>>
>> "Correct dwarf unwind information in function epilogue for X86"
>>
>> This patch aims to provide correct dwarf unwind information in function
>> epilogue for X86.
>>
>> It consists of two parts. The first part inserts CFI instructions that set
>> appropriate cfa offset and cfa register in emitEpilogue() in
>> X86FrameLowering. This part is X86 specific.
>>
>> The second part is platform independent and ensures that:
>>
>> - CFI instructions do not affect code generation
>> - Unwind information remains correct when a function is modified by
>>   different passes. This is done in a late pass by analyzing information
>>   about cfa offset and cfa register in BBs and inserting additional CFI
>>   directives where necessary.
>>
>> Changed CFI instructions so that they:
>>
>> - are duplicable
>> - are not counted as instructions when tail duplicating or tail merging
>> - can be compared as equal
>>
>> Added CFIInstrInserter pass:
>>
>> - analyzes each basic block to determine cfa offset and register valid at
>>   its entry and exit
>> - verifies that outgoing cfa offset and register of predecessor blocks
>> match
>>   incoming values of their successors
>> - inserts additional CFI directives at basic block beginning to correct
>> the
>>   rule for calculating CFA
>>
>> Having CFI instructions in function epilogue can cause incorrect CFA
>> calculation rule for some basic blocks. This can happen if, due to basic
>> block reordering, or the existence of multiple epilogue blocks, some of
>> the
>> blocks have wrong cfa offset and register values set by the epilogue block
>> above them.
>>
>> CFIInstrInserter is currently run only on X86, but can be used by any
>> target
>> that implements support for adding CFI instructions in epilogue.
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171108/7c125264/attachment.html>