<html dir="ltr">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style type="text/css" id="owaParaStyle"></style>
</head>
<body fpstyle="1" ocsi="0">
<div style="direction: ltr;font-family: Tahoma;color: #000000;font-size: 10pt;">There is one approach how to deal with this problem at:
<div><br>
</div>
<div><a href="https://reviews.llvm.org/D40979" target="_blank" style="font-family: "Segoe UI", Helvetica, Arial, sans-serif; font-size: 13.3333px;">https://reviews.llvm.org/D40979</a></div>
<div>
<div style="font-family: Times New Roman; color: #000000; font-size: 16px"><br>
</div>
<div style="font-family: Times New Roman; color: #000000; font-size: 16px">
<hr tabindex="-1">
<div id="divRpF361564" style="direction: ltr;"><font face="Tahoma" size="2" color="#000000"><b>From:</b> Reid Kleckner [rnk@google.com]<br>
<b>Sent:</b> Wednesday, November 8, 2017 10:21 PM<br>
<b>To:</b> Petar Jovanovic<br>
<b>Cc:</b> llvm-commits<br>
<b>Subject:</b> Re: [llvm] r317579 - Reland "Correct dwarf unwind information in function epilogue for X86"<br>
</font><br>
</div>
<div></div>
<div>
<div dir="ltr">There is a design issue with making CFI_INSTRUCTION duplicable. Darwin compact unwind info emission can't handle multiple prologue setups. I will revert this and put the reduced test case in the commit message.</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">On Wed, Nov 8, 2017 at 10:43 AM, Reid Kleckner <span dir="ltr">
<<a href="mailto:rnk@google.com" target="_blank">rnk@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex; border-left:1px #ccc solid; padding-left:1ex">
<div dir="ltr">I confirmed this caused a regression for AArch64 and I am reducing a test case.
<div>
<div class="h5"><br>
<div class="gmail_extra"><br>
<div class="gmail_quote">On Wed, Nov 8, 2017 at 9:34 AM, Reid Kleckner <span dir="ltr">
<<a href="mailto:rnk@google.com" target="_blank">rnk@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex; border-left:1px #ccc solid; padding-left:1ex">
<div dir="ltr">I suspect this change introduced a bug in compact unwind info on AArch64:
<div><a href="https://logs.chromium.org/v/?s=chromium%2Fbb%2Fchromium.clang%2FToTiOS%2F250%2F%2B%2Frecipes%2Fsteps%2Fcompile%2F0%2Fstdout" target="_blank">https://logs.chromium.org/v/?s<wbr>=chromium%2Fbb%2Fchromium.clan<wbr>g%2FToTiOS%2F250%2F%2B%2Frecip<wbr>es%2Fsteps%2Fcompile%2F0%2Fstd<wbr>out</a><br>
</div>
<div><br>
</div>
<div>
<div>FAILED: ios_clang_arm64/obj/third_part<wbr>y/opus/opus/pitch.o </div>
<div>...</div>
<div>Assertion failed: (StackSize == 0 && "We already have the CFA offset!"), function generateCompactUnwindEncoding, file /b/c/b/ToTiOS/src/third_party/<wbr>llvm/lib/Target/AArch64/MCTarg<wbr>etDesc/AArch64AsmBackend.cpp, line 487.</div>
<div>0 clang 0x00000001035d7fb8 llvm::sys::PrintStackTrace(llv<wbr>m::raw_ostream&) + 40</div>
<div>1 clang 0x00000001035d86b6 llvm::sys::PrintStackTraceOnEr<wbr>rorSignal(llvm::StringRef, bool) + 838</div>
<div>2 libsystem_platform.dylib 0x00007fffeb995bba _sigtramp + 26</div>
<div>3 clang 0x0000000103294e79 llvm::MCCVLineEntry::Make(llvm<wbr>::MCObjectStreamer*) + 25</div>
<div>4 libsystem_c.dylib 0x00007fffeb81c420 abort + 129</div>
<div>5 libsystem_c.dylib 0x00007fffeb7e3893 basename_r + 0</div>
<div>6 clang 0x0000000101e22067 llvm::createAArch64beAsmBacken<wbr>d(llvm::Target const&, llvm::MCRegisterInfo const&, llvm::Triple const&, llvm::StringRef, llvm::MCTargetOptions const&) + 8631</div>
<div>7 clang 0x00000001032bf4ef llvm::MCStreamer::generateComp<wbr>actUnwindEncodings(llvm::MCAsm<wbr>Backend*) + 79</div>
<div>8 clang 0x00000001032a1196 llvm::MCDwarfFrameEmitter::Emi<wbr>t(llvm::MCObjectStreamer&, llvm::MCAsmBackend*, bool) + 38</div>
<div>9 clang 0x00000001032b806e llvm::MCObjectStreamer::EmitFr<wbr>ames(llvm::MCAsmBackend*) + 62</div>
<div>10 clang 0x00000001032b12f5 llvm::createMachOStreamer(llvm<wbr>::MCContext&, std::__1::unique_ptr<llvm::MCA<wbr>smBackend, std::__1::default_delete<llvm:<wbr>:MCAsmBackend> >&&, llvm::raw_pwrite_stream&, std::__1::unique_ptr<llvm::MCC<wbr>odeEmitter,
std::__1::default_delete<llvm:<wbr>:MCCodeEmitter> >&&, bool, bool, bool) + 5749</div>
<div>11 clang 0x00000001032c37f2 llvm::MCStreamer::Finish() + 306</div>
<div>12 clang 0x0000000103d6e271 llvm::AsmPrinter::doFinalizati<wbr>on(llvm::Module&) + 3489</div>
<div>13 clang 0x0000000102ffa740 llvm::FPPassManager::doFinaliz<wbr>ation(llvm::Module&) + 80</div>
<div>14 clang 0x0000000102ffac8d llvm::legacy::PassManagerImpl:<wbr>:run(llvm::Module&) + 1277</div>
<div>15 clang 0x000000010380cd33 clang::EmitBackendOutput(clang<wbr>::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<wbr>_pwrite_stream, std::__1::default_delete<llvm:<wbr>:raw_pwrite_stream> >) + 15571</div>
<div>16 clang 0x0000000103a44ed5 clang::EmitObjAction::EmitObjA<wbr>ction(llvm::LLVMContext*) + 2341</div>
<div>17 clang 0x0000000103d10ccc clang::MultiplexConsumer::Hand<wbr>leTranslationUnit(clang::ASTCo<wbr>ntext&) + 44</div>
<div>18 clang 0x00000001044cdfb5 clang::ParseAST(clang::Sema&, bool, bool) + 469</div>
<div>19 clang 0x0000000103ccdfdc clang::FrontendAction::Execute<wbr>() + 76</div>
<div>20 clang 0x0000000103c86e41 clang::CompilerInstance::Execu<wbr>teAction(clang::FrontendAction<wbr>&) + 1217</div>
<div>21 clang 0x0000000103d340fa clang::ExecuteCompilerInvocati<wbr>on(clang::CompilerInstance*) + 4970</div>
<div>22 clang 0x0000000101c86a57 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) + 1399</div>
<div>23 clang 0x0000000101c850df main + 11951</div>
<div>24 libdyld.dylib 0x00007fffeb788255 start + 1</div>
<div><br>
</div>
</div>
<div>I'll try to get a reproducer.</div>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">On Tue, Nov 7, 2017 at 6:40 AM, Petar Jovanovic via llvm-commits
<span dir="ltr"><<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex; border-left:1px #ccc solid; padding-left:1ex">
Author: petarj<br>
Date: Tue Nov 7 06:40:27 2017<br>
New Revision: 317579<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=317579&view=rev" rel="noreferrer" target="_blank">
http://llvm.org/viewvc/llvm-pr<wbr>oject?rev=317579&view=rev</a><br>
Log:<br>
Reland "Correct dwarf unwind information in function epilogue for X86"<br>
<br>
Reland r317100 with minor fix regarding ComputeCommonTailLength function in<br>
BranchFolding.cpp. Skipping top CFI instructions block needs to executed on<br>
several more return points in ComputeCommonTailLength().<br>
<br>
Original r317100 message:<br>
<br>
"Correct dwarf unwind information in function epilogue for X86"<br>
<br>
This patch aims to provide correct dwarf unwind information in function<br>
epilogue for X86.<br>
<br>
It consists of two parts. The first part inserts CFI instructions that set<br>
appropriate cfa offset and cfa register in emitEpilogue() in<br>
X86FrameLowering. This part is X86 specific.<br>
<br>
The second part is platform independent and ensures that:<br>
<br>
- CFI instructions do not affect code generation<br>
- Unwind information remains correct when a function is modified by<br>
different passes. This is done in a late pass by analyzing information<br>
about cfa offset and cfa register in BBs and inserting additional CFI<br>
directives where necessary.<br>
<br>
Changed CFI instructions so that they:<br>
<br>
- are duplicable<br>
- are not counted as instructions when tail duplicating or tail merging<br>
- can be compared as equal<br>
<br>
Added CFIInstrInserter pass:<br>
<br>
- analyzes each basic block to determine cfa offset and register valid at<br>
its entry and exit<br>
- verifies that outgoing cfa offset and register of predecessor blocks match<br>
incoming values of their successors<br>
- inserts additional CFI directives at basic block beginning to correct the<br>
rule for calculating CFA<br>
<br>
Having CFI instructions in function epilogue can cause incorrect CFA<br>
calculation rule for some basic blocks. This can happen if, due to basic<br>
block reordering, or the existence of multiple epilogue blocks, some of the<br>
blocks have wrong cfa offset and register values set by the epilogue block<br>
above them.<br>
<br>
CFIInstrInserter is currently run only on X86, but can be used by any target<br>
that implements support for adding CFI instructions in epilogue.<br>
<br>
</blockquote>
</div>
</div>
</blockquote>
</div>
</div>
</div>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</div>
</div>
</div>
</div>
</body>
</html>