[ELF] Read the call graph profile from object files.

[Bigcheese]

Summary

This uses the call graph profile embedded in the object files to construct the call graph.

This is read from a SHT_LLVM_CALL_GRAPH_PROFILE (0x6fff4c02) section as (uint32_t, uint32_t, uint64_t) tuples as (from symbol index, to symbol index, weight).

This depends on https://reviews.llvm.org/D44965 for llvm.

[• espindola]

ELF/InputFiles.cpp

219

Why do you need to store this in Config? The code that is walking CallGraphProfileSymbols could just as easily walk CGProfile, no?

[grimar]

This should be very useful. Needs rebasing though.

I think Rafael is right and eliminating Config->CallGraphProfileSymbols would be better.
You could create a helper function to call from Driver.cpp that would scan over
CGProfile and add graph edges probably.

[Bigcheese]

Address review comments.

[Bigcheese]

ping

[grimar]

ELF/Driver.cpp

676

Can you add a test to check this line?

685

CGProfile is used only once so I think you can just inline it:

for (const Elf_CGProfile_Impl<ELFT> &CGPE : Obj->CGProfile)

[Bigcheese]

• Added test for unorderable symbols from object files
• Removed unneeded temporary

[ruiu]

LGTM

ELF/Driver.cpp

691

I believe you can use {} instead of std::make_pair().

ELF/Driver.cpp

   }
 }
 
+template <class ELFT> static void readCallGraphsFromObjectFiles() {
+  auto FindSection = [&](const Symbol *Sym) -> const InputSectionBase * {
+    warnUnorderableSymbol(Sym);
+    if (const auto *SymD = dyn_cast<Defined>(Sym))
+      return dyn_cast_or_null<InputSectionBase>(SymD->Section);
+    return nullptr;
+  };
+
+  for (auto File : ObjectFiles) {
+    auto *Obj = cast<ObjFile<ELFT>>(File);
+    ArrayRef<Elf_CGProfile_Impl<ELFT>> CGProfile = Obj->CGProfile;
+    for (const Elf_CGProfile_Impl<ELFT> &CGPE : CGProfile) {
+      const InputSectionBase *FromSB =
+          FindSection(&Obj->getSymbol(CGPE.cgp_from));
+      const InputSectionBase *ToSB = FindSection(&Obj->getSymbol(CGPE.cgp_to));
+      if (!FromSB || !ToSB)
+        continue;
+      Config->CallGraphProfile[std::make_pair(FromSB, ToSB)] += CGPE.cgp_weight;
+    }
+  }
+}
+
 static bool getCompressDebugSections(opt::InputArgList &Args) {
   StringRef S = Args.getLastArgValue(OPT_compress_debug_sections, "none");
   if (S == "none")

   if (auto *Arg = Args.getLastArg(OPT_call_graph_ordering_file))
     if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
       readCallGraph(*Buffer);
+  readCallGraphsFromObjectFiles<ELFT>();
 
   // Write the result to the file.
   writeResult<ELFT>();

ELF/InputFiles.h

   typedef typename ELFT::Sym Elf_Sym;
   typedef typename ELFT::Shdr Elf_Shdr;
   typedef typename ELFT::Word Elf_Word;
+  typedef typename ELFT::CGProfile Elf_CGProfile;
 
   StringRef getShtGroupSignature(ArrayRef<Elf_Shdr> Sections,
                                  const Elf_Shdr &Sec);

   // Pointer to this input file's .llvm_addrsig section, if it has one.
   const Elf_Shdr *AddrsigSec = nullptr;
 
+  // SHT_LLVM_CALL_GRAPH_PROFILE table
+  ArrayRef<Elf_CGProfile> CGProfile;
+
 private:
   void
   initializeSections(llvm::DenseSet<llvm::CachedHashStringRef> &ComdatGroups);

ELF/InputFiles.cpp

       continue;
     const Elf_Shdr &Sec = ObjSections[I];
 
+    if (Sec.sh_type == ELF::SHT_LLVM_CALL_GRAPH_PROFILE)
+      CGProfile = check(
+          this->getObj().template getSectionContentsAsArray<Elf_CGProfile>(
+              &Sec));
+
     // SHF_EXCLUDE'ed sections are discarded by the linker. However,
     // if -r is given, we'll let the final link discard such sections.
     // This is compatible with GNU.

test/ELF/cgprofile-obj.s

+# REQUIRES: x86
+
+# RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t
+# RUN: ld.lld -e A %t -o %t2
+# RUN: llvm-readobj -symbols %t2 | FileCheck %s
+
+    .section    .text.D,"ax",@progbits
+D:
+    retq
+
+    .section    .text.C,"ax",@progbits
+    .globl  C
+C:
+    retq
+
+    .section    .text.B,"ax",@progbits
+    .globl  B
+B:
+    retq
+
+    .section    .text.A,"ax",@progbits
+    .globl  A
+A:
+Aa:
+    retq
+
+    .cg_profile A, B, 10
+    .cg_profile A, B, 10
+    .cg_profile Aa, B, 80
+    .cg_profile A, C, 40
+    .cg_profile B, C, 30
+    .cg_profile C, D, 90
+
+# CHECK:          Name: D
+# CHECK-NEXT:     Value: 0x201003
+# CHECK:          Name: A
+# CHECK-NEXT:     Value: 0x201000
+# CHECK:          Name: B
+# CHECK-NEXT:     Value: 0x201001
+# CHECK:          Name: C
+# CHECK-NEXT:     Value: 0x201002