[lld] [lld][ELF] Extend profile guided function ordering to ELF binaries (PR #117514)

[Ellis Hoag via llvm-commits]

================
@@ -14,13 +14,138 @@
 #ifndef LLD_MACHO_BPSECTION_ORDERER_H
 #define LLD_MACHO_BPSECTION_ORDERER_H
 
+#include "InputSection.h"
+#include "Relocations.h"
+#include "Symbols.h"
+#include "lld/Common/BPSectionOrdererBase.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/TinyPtrVector.h"
 
 namespace lld::macho {
 
 class InputSection;
 
+class BPSymbolMacho : public BPSymbol {
+  const Symbol *sym;
+
+public:
+  explicit BPSymbolMacho(const Symbol *s) : sym(s) {}
+
+  llvm::StringRef getName() const override { return sym->getName(); }
+
+  const Defined *asDefined() const {
+    return llvm::dyn_cast_or_null<Defined>(sym);
+  }
+
+  std::optional<uint64_t> getValue() const override {
+    if (auto *d = asDefined())
+      return d->value;
+    return {};
+  }
+
+  std::optional<uint64_t> getSize() const override {
+    if (auto *d = asDefined())
+      return d->size;
+    return {};
+  }
+
+  const Symbol *getSymbol() const { return sym; }
+};
+
+class BPSectionMacho : public BPSectionBase {
+  const InputSection *isec;
+  uint64_t sectionIdx;
+
+public:
+  explicit BPSectionMacho(const InputSection *sec, uint64_t sectionIdx)
+      : isec(sec), sectionIdx(sectionIdx) {}
+
+  const InputSection *getSection() const { return isec; }
+
+  uint64_t getSize() const override { return isec->getSize(); }
+
+  uint64_t getSectionIdx() const { return sectionIdx; }
+
+  bool isCodeSection() const override { return macho::isCodeSection(isec); }
+
+  bool hasValidData() const override {
+    return isec && !isec->data.empty() && isec->data.data();
+  }
+
+  SmallVector<std::unique_ptr<BPSymbol>> getSymbols() const override {
+    SmallVector<std::unique_ptr<BPSymbol>> symbols;
+    for (auto *sym : isec->symbols)
+      if (auto *d = llvm::dyn_cast_or_null<Defined>(sym))
+        symbols.emplace_back(std::make_unique<BPSymbolMacho>(d));
+    return symbols;
+  }
+
+  // Linkage names can be prefixed with "_" or "l_" on Mach-O. See
+  // Mangler::getNameWithPrefix() for details.
+  std::optional<StringRef>
+  getResolvedLinkageName(llvm::StringRef name) const override {
+    if (name.consume_front("_") || name.consume_front("l_"))
+      return name;
+    return {};
+  }
+
+  void
+  getSectionHashes(llvm::SmallVectorImpl<uint64_t> &hashes) const override {
+    constexpr unsigned windowSize = 4;
+
+    // Calculate content hashes
+    size_t dataSize = isec->data.size();
+    for (size_t i = 0; i < dataSize; i++) {
+      auto window = isec->data.drop_front(i).take_front(windowSize);
+      hashes.push_back(xxHash64(window));
+    }
+
+    // Calculate relocation hashes
+    for (const auto &r : isec->relocs) {
+      if (r.length == 0 || r.referent.isNull() || r.offset >= isec->data.size())
+        continue;
+
+      uint64_t relocHash = getRelocHash(r, this);
+      uint32_t start = (r.offset < windowSize) ? 0 : r.offset - windowSize + 1;
+      for (uint32_t i = start; i < r.offset + r.length; i++) {
+        auto window = isec->data.drop_front(i).take_front(windowSize);
+        hashes.push_back(xxHash64(window) + relocHash);
+      }
+    }
+
+    llvm::sort(hashes);
+    hashes.erase(std::unique(hashes.begin(), hashes.end()), hashes.end());
+  }
+
+  const InputSection *getInputSection() const { return isec; }
+
+  static bool classof(const BPSectionBase *s) { return true; }
+
+private:
+  static uint64_t getRelocHash(const Reloc &reloc,
+                               const BPSectionMacho *section) {
+    auto *isec = reloc.getReferentInputSection();
+    std::optional<uint64_t> sectionIdx;
+    if (isec && isec == section->getSection())
+      sectionIdx = section->getSectionIdx();
----------------
ellishg wrote:

I see the problem now. In the original code we need the section index map to get the index of the referent section (where the relocation points to), which could be different from the section that the relocation is in. My apologies, we may need to keep the map around unless you can think of a different solution.

https://github.com/llvm/llvm-project/blob/3c464d23682b0f9e6f70965e8f8f3861c9ba5417/lld/MachO/BPSectionOrderer.cpp#L46-L50

https://github.com/llvm/llvm-project/pull/117514