[lld-macho] Deduplicate fixed-width literals

[int3]

Summary

Conceptually, the implementation is pretty straightforward: we put each
literal value into a hashtable, and then write out the keys of that
hashtable at the end.

In contrast with ELF, the Mach-O format does not support variable-length
literals that aren't strings. Its literals are either 4, 8, or 16 bytes
in length. LLD-ELF dedups its literals via sorting + uniq'ing, but since
we don't need to worry about overly-long values, we should be able to do
a faster job by just hashing.

That said, the implementation right now is far from optimal, because we
add to those hashtables serially. To parallelize this, we'll need a
basic concurrent hashtable (only needs to support concurrent writes w/o
interleave reads), which shouldn't be to hard to implement, but I'd like
to punt on it for now.

Numbers for linking chromium_framework on my 3.2 GHz 16-Core Intel Xeon W:

    N           Min           Max        Median           Avg        Stddev
x  20          4.27          4.39         4.315        4.3225   0.033225703
+  20          4.36          4.82          4.44        4.4845    0.13152846
Difference at 95.0% confidence
        0.162 +/- 0.0613971
        3.74783% +/- 1.42041%
        (Student's t, pooled s = 0.0959262)

This corresponds to binary size savings of 2MB out of 335MB, or 0.6%.
It's not a great tradeoff as-is, but as mentioned our implementation can
be signficantly optimized, and literal dedup will unlock more
opportunities for ICF to identify identical structures that reference
the same literals.

[int3]

remove redundant test change

[int3]

lld/test/MachO/mattrs.ll

14

the LLVM IR in this test generated literals, which got moved to different addresses in this diff. But this test doesn't actually care about the locations of the literals, so I've changed it accordingly.

[tschuett]

This does not compile:

std::vector<String<4>> strings4;
strings4.reserve(number_of_4_strings*sizeof(String<4>));
for (size_t i = 0, e = isec->data.size() / 4; i < e; ++i) {
  strings4.emplace_back(String<4>(xxx));
}
strings4.sort();
auto bla = std::unique(strings4.begin(), strings4.end());

The hash map will have a lot of mallocs. I would expect sort + unique to be faster. Maybe even parallel sort. String<4> could be like a StringRef, but with fixed size (4).

The for-loop could be llvm::parallel_for_each_n(...)

[tschuett]

In D103113#2782094, @tschuett wrote:

This does not compile:

std::vector<String<4>> strings4;
strings4.reserve(number_of_4_strings*sizeof(String<4>));
for (size_t i = 0, e = isec->data.size() / 4; i < e; ++i) {
  strings4.emplace_back(String<4>(xxx));
}
strings4.sort();
auto bla = std::unique(strings4.begin(), strings4.end());

The hash map will have a lot of mallocs. I would expect sort + unique to be faster. Maybe even parallel sort.

The emplace loop could even be a memcpy?!? Or do everything in-place?

[int3]

The hash map will have a lot of mallocs. I would expect sort + unique to be faster.

I guess I could reserve() memory for the hashmap before inserting...

But as the commit message notes, I'm aware this is far from an optimal implementation, and I think we can revisit this later after building out our other features. I'm mostly interested in this to see how much of a win it'll unlock when coupled with ICF.

[tschuett]

In D103113#2783198, @int3 wrote:

The hash map will have a lot of mallocs. I would expect sort + unique to be faster.

I guess I could reserve() memory for the hashmap before inserting...

But as the commit message notes, I'm aware this is far from an optimal implementation, and I think we can revisit this later after building out our other features. I'm mostly interested in this to see how much of a win it'll unlock when coupled with ICF.

No worries.

[int3]

update

[gkm]

This was very pleasant to read & review!

lld/MachO/SyntheticSections.cpp

1160–1166

What is the purpose of the braces around these case bodies?

[int3]

lld/MachO/SyntheticSections.cpp

1160–1166

Declaring variables whose scope can leak out of the switch-case raises the error "cannot jump from switch statement to this case label... jump bypasses variable initialization". In this case it's not actually needed since i and value are scoped within the for loop, but nonetheless I like putting braces around non-trivial case blocks so I don't have to worry about this issue.

[int3]

Thanks!

lld/MachO/InputFiles.cpp

 }
 
 template <class Section>
 void ObjFile::parseSections(ArrayRef<Section> sections) {
   subsections.reserve(sections.size());
   auto *buf = reinterpret_cast<const uint8_t *>(mb.getBufferStart());
 
   for (const Section &sec : sections) {
-    if (config->dedupLiterals && sectionType(sec.flags) == S_CSTRING_LITERALS) {
+    if (config->dedupLiterals &&
+        (sectionType(sec.flags) == S_CSTRING_LITERALS ||
+         isWordLiteralSection(sec.flags))) {
       if (sec.nreloc)
         fatal(toString(this) + " contains relocations in " + sec.segname + "," +
               sec.sectname +
               ", so LLD cannot deduplicate literals. Try re-running without "
               "--deduplicate-literals.");
 
-      auto *isec = make<CStringInputSection>();
+      InputSection *isec;
+      if (sectionType(sec.flags) == S_CSTRING_LITERALS) {
+        isec = make<CStringInputSection>();
         parseSection(this, buf, sec, isec);
-      isec->splitIntoPieces(); // FIXME: parallelize this?
+        // FIXME: parallelize this?
+        cast<CStringInputSection>(isec)->splitIntoPieces();
+      } else {
+        isec = make<WordLiteralInputSection>();
+        parseSection(this, buf, sec, isec);
+      }
       subsections.push_back({{0, isec}});
     } else {
       auto *isec = make<ConcatInputSection>();
       parseSection(this, buf, sec, isec);
       if (!(isDebugSection(isec->flags) &&
             isec->segname == segment_names::dwarf)) {
         subsections.push_back({{0, isec}});
       } else {

lld/MachO/InputSection.h

 class InputFile;
 class OutputSection;
 
 class InputSection {
 public:
   enum Kind {
     ConcatKind,
     CStringLiteralKind,
+    WordLiteralKind,
   };
 
   Kind kind() const { return sectionKind; }
   virtual ~InputSection() = default;
   virtual uint64_t getSize() const { return data.size(); }
   uint64_t getFileSize() const;
   // Translates \p off -- an offset relative to this InputSection -- into an
   // offset from the beginning of its parent OutputSection.
public:
 
   static bool classof(const InputSection *isec) {
     return isec->kind() == CStringLiteralKind;
   }
 
   std::vector<StringPiece> pieces;
 };
 
+class WordLiteralInputSection : public InputSection {
+public:
+  WordLiteralInputSection() : InputSection(WordLiteralKind) {}
+  uint64_t getFileOffset(uint64_t off) const override;
+  uint64_t getOffset(uint64_t off) const override;
+
+  static bool classof(const InputSection *isec) {
+    return isec->kind() == WordLiteralKind;
+  }
+};
+
 inline uint8_t sectionType(uint32_t flags) {
   return flags & llvm::MachO::SECTION_TYPE;
 }
 
 inline bool isZeroFill(uint32_t flags) {
   return llvm::MachO::isVirtualSection(sectionType(flags));
 }
 
 inline bool isThreadLocalVariables(uint32_t flags) {
   return sectionType(flags) == llvm::MachO::S_THREAD_LOCAL_VARIABLES;
 }
 
 // These sections contain the data for initializing thread-local variables.
 inline bool isThreadLocalData(uint32_t flags) {
   return sectionType(flags) == llvm::MachO::S_THREAD_LOCAL_REGULAR ||
          sectionType(flags) == llvm::MachO::S_THREAD_LOCAL_ZEROFILL;
 }
 
 inline bool isDebugSection(uint32_t flags) {
   return (flags & llvm::MachO::SECTION_ATTRIBUTES_USR) ==
          llvm::MachO::S_ATTR_DEBUG;
 }
 
+inline bool isWordLiteralSection(uint32_t flags) {
+  return sectionType(flags) == llvm::MachO::S_4BYTE_LITERALS ||
+         sectionType(flags) == llvm::MachO::S_8BYTE_LITERALS ||
+         sectionType(flags) == llvm::MachO::S_16BYTE_LITERALS;
+}
+
 bool isCodeSection(const InputSection *);
 
 extern std::vector<InputSection *> inputSections;
 
 namespace section_names {
 
 constexpr const char authGot[] = "__auth_got";
 constexpr const char authPtr[] = "__auth_ptr";

 constexpr const char export_[] = "__export";
 constexpr const char functionStarts[] = "__func_starts";
 constexpr const char got[] = "__got";
 constexpr const char header[] = "__mach_header";
 constexpr const char indirectSymbolTable[] = "__ind_sym_tab";
 constexpr const char const_[] = "__const";
 constexpr const char lazySymbolPtr[] = "__la_symbol_ptr";
 constexpr const char lazyBinding[] = "__lazy_binding";
+constexpr const char literals[] = "__literals";
 constexpr const char moduleInitFunc[] = "__mod_init_func";
 constexpr const char moduleTermFunc[] = "__mod_term_func";
 constexpr const char nonLazySymbolPtr[] = "__nl_symbol_ptr";
 constexpr const char objcCatList[] = "__objc_catlist";
 constexpr const char objcClassList[] = "__objc_classlist";
 constexpr const char objcConst[] = "__objc_const";
 constexpr const char objcImageInfo[] = "__objc_imageinfo";
 constexpr const char objcNonLazyCatList[] = "__objc_nlcatlist";

lld/MachO/InputSection.cpp

 }
 
 uint64_t CStringInputSection::getOffset(uint64_t off) const {
   const StringPiece &piece = getStringPiece(off);
   uint64_t addend = off - piece.inSecOff;
   return piece.outSecOff + addend;
 }
 
+uint64_t WordLiteralInputSection::getFileOffset(uint64_t off) const {
+  return parent->fileOff + getOffset(off);
+}
+
+uint64_t WordLiteralInputSection::getOffset(uint64_t off) const {
+  auto *osec = cast<WordLiteralSection>(parent);
+  const uint8_t *buf = data.data();
+  switch (sectionType(flags)) {
+  case S_4BYTE_LITERALS:
+    return osec->getLiteral4Offset(buf + off);
+  case S_8BYTE_LITERALS:
+    return osec->getLiteral8Offset(buf + off);
+  case S_16BYTE_LITERALS:
+    return osec->getLiteral16Offset(buf + off);
+  default:
+    llvm_unreachable("invalid literal section type");
+  }
+}
+
 bool macho::isCodeSection(const InputSection *isec) {
   uint32_t type = sectionType(isec->flags);
   if (type != S_REGULAR && type != S_COALESCED)
     return false;
 
   uint32_t attr = isec->flags & SECTION_ATTRIBUTES_USR;
   if (attr == S_ATTR_PURE_INSTRUCTIONS)
     return true;

lld/MachO/SyntheticSections.h

 #include "Target.h"
 
 #include "llvm/ADT/Hashing.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/MC/StringTableBuilder.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 
+#include <unordered_map>
+
 namespace llvm {
 class DWARFUnit;
 } // namespace llvm
 
 namespace lld {
 namespace macho {
 
 class Defined;
public:
   bool isNeeded() const override { return !entries.empty(); }
   void finalize() override;
   void writeTo(uint8_t *buf) const override;
   const llvm::SetVector<Symbol *> &getEntries() const { return entries; }
   // Returns whether the symbol was added. Note that every stubs entry will
   // have a corresponding entry in the LazyPointerSection.
   bool addEntry(Symbol *);
   uint64_t getVA(uint32_t stubsIndex) const {
+    assert(isFinal || target->usesThunks());
     // ConcatOutputSection::finalize() can seek the address of a
     // stub before its address is assigned. Before __stubs is
     // finalized, return a contrived out-of-range address.
     return isFinal ? addr + stubsIndex * target->stubSize
                    : TargetInfo::outOfRangeVA;
   }
 
   bool isFinal = false; // is address assigned?
public:
   void writeTo(uint8_t *buf) const override { builder.write(buf); }
 
   std::vector<CStringInputSection *> inputs;
 
 private:
   llvm::StringTableBuilder builder;
 };
 
+/*
+ * This section contains deduplicated literal values. The 16-byte values are
+ * laid out first, followed by the 8- and then the 4-byte ones.
+ */
+class WordLiteralSection : public SyntheticSection {
+public:
+  using UInt128 = std::pair<uint64_t, uint64_t>;
+  // I don't think the standard guarantees the size of a pair, so let's make
+  // sure it's exact -- that way we can construct it via `mmap`.
+  static_assert(sizeof(UInt128) == 16, "");
+
+  WordLiteralSection();
+  void addInput(WordLiteralInputSection *);
+  void writeTo(uint8_t *buf) const override;
+
+  uint64_t getSize() const override {
+    return literal16Map.size() * 16 + literal8Map.size() * 8 +
+           literal4Map.size() * 4;
+  }
+
+  bool isNeeded() const override {
+    return !literal16Map.empty() || !literal4Map.empty() ||
+           !literal8Map.empty();
+  }
+
+  uint64_t getLiteral16Offset(const uint8_t *buf) const {
+    return literal16Map.at(*reinterpret_cast<const UInt128 *>(buf)) * 16;
+  }
+
+  uint64_t getLiteral8Offset(const uint8_t *buf) const {
+    return literal16Map.size() * 16 +
+           literal8Map.at(*reinterpret_cast<const uint64_t *>(buf)) * 8;
+  }
+
+  uint64_t getLiteral4Offset(const uint8_t *buf) const {
+    return literal16Map.size() * 16 + literal8Map.size() * 8 +
+           literal4Map.at(*reinterpret_cast<const uint32_t *>(buf)) * 4;
+  }
+
+private:
+  template <class T> struct Hasher {
+    llvm::hash_code operator()(T v) const { return llvm::hash_value(v); }
+  };
+  // We're using unordered_map instead of DenseMap here because we need to
+  // support all possible integer values -- there are no suitable tombstone
+  // values for DenseMap.
+  std::unordered_map<UInt128, uint64_t, Hasher<UInt128>> literal16Map;
+  std::unordered_map<uint64_t, uint64_t> literal8Map;
+  std::unordered_map<uint32_t, uint64_t> literal4Map;
+};
+
 struct InStruct {
   MachHeaderSection *header = nullptr;
   CStringSection *cStringSection = nullptr;
+  WordLiteralSection *wordLiteralSection = nullptr;
   RebaseSection *rebase = nullptr;
   BindingSection *binding = nullptr;
   WeakBindingSection *weakBinding = nullptr;
   LazyBindingSection *lazyBinding = nullptr;
   ExportSection *exports = nullptr;
   GotSection *got = nullptr;
   TlvPointerSection *tlvPointers = nullptr;
   LazyPointerSection *lazyPointers = nullptr;

lld/MachO/SyntheticSections.cpp

for (CStringInputSection *isec : inputs) {
     for (size_t i = 0, e = isec->pieces.size(); i != e; ++i) {
       isec->pieces[i].outSecOff =
           builder.getOffset(isec->getCachedHashStringRef(i));
       isec->isFinal = true;
     }
   }
 }
 
+// This section is actually emitted as __TEXT,__const by ld64, but clang may
+// emit input sections of that name, and LLD doesn't currently support mixing
+// synthetic and concat-type OutputSections. To work around this, I've given
+// our merged-literals section a different name.
+WordLiteralSection::WordLiteralSection()
+    : SyntheticSection(segment_names::text, section_names::literals) {
+  align = 16;
+}
+
+void WordLiteralSection::addInput(WordLiteralInputSection *isec) {
+  isec->parent = this;
+  // We do all processing of the InputSection here, so it will be effectively
+  // finalized.
+  isec->isFinal = true;
+  const uint8_t *buf = isec->data.data();
+  switch (sectionType(isec->flags)) {
+  case S_4BYTE_LITERALS: {
+    for (size_t i = 0, e = isec->data.size() / 4; i < e; ++i) {
+      uint32_t value = *reinterpret_cast<const uint32_t *>(buf + i * 4);
+      literal4Map.emplace(value, literal4Map.size());
+    }
+    break;
+  }
+  case S_8BYTE_LITERALS: {
+    for (size_t i = 0, e = isec->data.size() / 8; i < e; ++i) {
+      uint64_t value = *reinterpret_cast<const uint64_t *>(buf + i * 8);
+      literal8Map.emplace(value, literal8Map.size());
+    }
+    break;
+  }
+  case S_16BYTE_LITERALS: {
+    for (size_t i = 0, e = isec->data.size() / 16; i < e; ++i) {
+      UInt128 value = *reinterpret_cast<const UInt128 *>(buf + i * 16);
+      literal16Map.emplace(value, literal16Map.size());
+    }
+    break;
+  }
+  default:
+    llvm_unreachable("invalid literal section type");
+  }
+}
+
+void WordLiteralSection::writeTo(uint8_t *buf) const {
+  // Note that we don't attempt to do any endianness conversion in addInput(),
+  // so we don't do it here either -- just write out the original value,
+  // byte-for-byte.
+  for (const auto &p : literal16Map)
+    memcpy(buf + p.second * 16, &p.first, 16);
+  buf += literal16Map.size() * 16;
+
+  for (const auto &p : literal8Map)
+    memcpy(buf + p.second * 8, &p.first, 8);
+  buf += literal8Map.size() * 8;
+
+  for (const auto &p : literal4Map)
+    memcpy(buf + p.second * 4, &p.first, 4);
+}
+
 void macho::createSyntheticSymbols() {
   auto addHeaderSymbol = [](const char *name) {
     symtab->addSynthetic(name, in.header->isec, /*value=*/0,
                          /*privateExtern=*/true, /*includeInSymtab=*/false,
                          /*referencedDynamically=*/false);
   };
 
   switch (config->outputType) {

lld/MachO/Writer.cpp

template <class LP> void Writer::createOutputSections() {
   }
 
   // Then add input sections to output sections.
   DenseMap<NamePair, ConcatOutputSection *> concatOutputSections;
   for (const auto &p : enumerate(inputSections)) {
     InputSection *isec = p.value();
     if (isec->shouldOmitFromOutput())
       continue;
+    OutputSection *osec;
     if (auto *concatIsec = dyn_cast<ConcatInputSection>(isec)) {
       NamePair names = maybeRenameSection({isec->segname, isec->name});
-      ConcatOutputSection *&osec = concatOutputSections[names];
-      if (osec == nullptr) {
-        osec = make<ConcatOutputSection>(names.second);
-        osec->inputOrder = p.index();
-      }
-      osec->addInput(concatIsec);
+      ConcatOutputSection *&concatOsec = concatOutputSections[names];
+      if (concatOsec == nullptr)
+        concatOsec = make<ConcatOutputSection>(names.second);
+      concatOsec->addInput(concatIsec);
+      osec = concatOsec;
     } else if (auto *cStringIsec = dyn_cast<CStringInputSection>(isec)) {
-      if (in.cStringSection->inputs.empty())
-        in.cStringSection->inputOrder = p.index();
       in.cStringSection->addInput(cStringIsec);
+      osec = in.cStringSection;
+    } else if (auto *litIsec = dyn_cast<WordLiteralInputSection>(isec)) {
+      in.wordLiteralSection->addInput(litIsec);
+      osec = in.wordLiteralSection;
+    } else {
+      llvm_unreachable("unhandled InputSection type");
     }
+    osec->inputOrder = std::min(osec->inputOrder, static_cast<int>(p.index()));
   }
 
   // Once all the inputs are added, we can finalize the output section
   // properties and create the corresponding output segments.
   for (const auto &it : concatOutputSections) {
     StringRef segname = it.first.first;
     ConcatOutputSection *osec = it.second;
     if (segname == segment_names::ld) {
template <class LP> void Writer::run() {
   writeOutputFile();
 }
 
 template <class LP> void macho::writeResult() { Writer().run<LP>(); }
 
 void macho::createSyntheticSections() {
   in.header = make<MachHeaderSection>();
   in.cStringSection = config->dedupLiterals ? make<CStringSection>() : nullptr;
+  in.wordLiteralSection =
+      config->dedupLiterals ? make<WordLiteralSection>() : nullptr;
   in.rebase = make<RebaseSection>();
   in.binding = make<BindingSection>();
   in.weakBinding = make<WeakBindingSection>();
   in.lazyBinding = make<LazyBindingSection>();
   in.exports = make<ExportSection>();
   in.got = make<GotSection>();
   in.tlvPointers = make<TlvPointerSection>();
   in.lazyPointers = make<LazyPointerSection>();

lld/test/MachO/literal-dedup.s

+# REQUIRES: x86
+# RUN: rm -rf %t; split-file %s %t
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/test.s -o %t/test.o
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/qux.s -o %t/qux.o
+# RUN: %lld -dylib --deduplicate-literals %t/test.o %t/qux.o -o %t/test
+# RUN: llvm-objdump --macho --section="__TEXT,__literals" --section="__DATA,ptrs" --syms %t/test | FileCheck %s
+# RUN: llvm-readobj --section-headers %t/test | FileCheck %s --check-prefix=HEADER
+
+# CHECK:      Contents of (__TEXT,__literals) section
+# CHECK-NEXT: [[#%.16x,DEADBEEF16:]] ef be ad de ef be ad de ef be ad de ef be ad de
+# CHECK-NEXT: [[#%.16x,FEEDFACE16:]] ce fa ed fe ce fa ed fe ce fa ed fe ce fa ed fe
+# CHECK-NEXT: [[#%.16x,DEADBEEF8:]]  ef be ad de ef be ad de ce fa ed fe ce fa ed fe
+# CHECK-NEXT: [[#%.16x,DEADBEEF4:]]  ef be ad de ce fa ed fe
+# CHECK-NEXT: Contents of (__DATA,ptrs) section
+# CHECK-NEXT: 0000000000001000  0x[[#%x,DEADBEEF16]]
+# CHECK-NEXT: 0000000000001008  0x[[#%x,DEADBEEF16]]
+# CHECK-NEXT: 0000000000001010  0x[[#%x,FEEDFACE16]]
+# CHECK-NEXT: 0000000000001018  0x[[#%x,DEADBEEF16]]
+# CHECK-NEXT: 0000000000001020  0x[[#%x,DEADBEEF8]]
+# CHECK-NEXT: 0000000000001028  0x[[#%x,DEADBEEF8]]
+# CHECK-NEXT: 0000000000001030  0x[[#%x,DEADBEEF8 + 8]]
+# CHECK-NEXT: 0000000000001038  0x[[#%x,DEADBEEF8]]
+# CHECK-NEXT: 0000000000001040  0x[[#%x,DEADBEEF4]]
+# CHECK-NEXT: 0000000000001048  0x[[#%x,DEADBEEF4]]
+# CHECK-NEXT: 0000000000001050  0x[[#%x,DEADBEEF4 + 4]]
+# CHECK-NEXT: 0000000000001058  0x[[#%x,DEADBEEF4]]
+
+## Make sure the symbol addresses are correct too.
+# CHECK:     SYMBOL TABLE:
+# CHECK-DAG: [[#DEADBEEF16]] g     O __TEXT,__literals _qux16
+# CHECK-DAG: [[#DEADBEEF8]]  g     O __TEXT,__literals _qux8
+# CHECK-DAG: [[#DEADBEEF4]]  g     O __TEXT,__literals _qux4
+
+## Make sure we set the right alignment and flags.
+# HEADER:        Name: __literals
+# HEADER-NEXT:   Segment: __TEXT
+# HEADER-NEXT:   Address:
+# HEADER-NEXT:   Size:
+# HEADER-NEXT:   Offset:
+# HEADER-NEXT:   Alignment: 4
+# HEADER-NEXT:   RelocationOffset:
+# HEADER-NEXT:   RelocationCount: 0
+# HEADER-NEXT:   Type: Regular
+# HEADER-NEXT:   Attributes [ (0x0)
+# HEADER-NEXT:   ]
+# HEADER-NEXT:   Reserved1: 0x0
+# HEADER-NEXT:   Reserved2: 0x0
+# HEADER-NEXT:   Reserved3: 0x0
+
+#--- test.s
+.literal4
+.p2align 2
+L._foo4:
+  .long 0xdeadbeef
+L._bar4:
+  .long 0xdeadbeef
+L._baz4:
+  .long 0xfeedface
+
+.literal8
+L._foo8:
+  .quad 0xdeadbeefdeadbeef
+L._bar8:
+  .quad 0xdeadbeefdeadbeef
+L._baz8:
+  .quad 0xfeedfacefeedface
+
+.literal16
+L._foo16:
+  .quad 0xdeadbeefdeadbeef
+  .quad 0xdeadbeefdeadbeef
+L._bar16:
+  .quad 0xdeadbeefdeadbeef
+  .quad 0xdeadbeefdeadbeef
+L._baz16:
+  .quad 0xfeedfacefeedface
+  .quad 0xfeedfacefeedface
+
+.section __DATA,ptrs,literal_pointers
+.quad L._foo16
+.quad L._bar16
+.quad L._baz16
+.quad _qux16
+
+.quad L._foo8
+.quad L._bar8
+.quad L._baz8
+.quad _qux8
+
+.quad L._foo4
+.quad L._bar4
+.quad L._baz4
+.quad _qux4
+
+#--- qux.s
+.globl _qux4, _qux8, _qux16
+
+.literal4
+.p2align 2
+_qux4:
+  .long 0xdeadbeef
+
+.literal8
+_qux8:
+  .quad 0xdeadbeefdeadbeef
+
+.literal16
+_qux16:
+  .quad 0xdeadbeefdeadbeef
+  .quad 0xdeadbeefdeadbeef

lld/test/MachO/mattrs.ll

 ; REQUIRES: x86
 ; RUN: llvm-as %s -o %t.o
 
 ;; Verify that LTO behavior can be tweaked using -mattr.
 
 ; RUN: %lld -mcpu haswell -mllvm -mattr=+fma %t.o -o %t.dylib -dylib
 ; RUN: llvm-objdump -d --section="__text" --no-leading-addr --no-show-raw-insn %t.dylib | FileCheck %s --check-prefix=FMA
 
 ; RUN: %lld -mcpu haswell -mllvm -mattr=-fma %t.o -o %t.dylib -dylib
 ; RUN: llvm-objdump -d --section="__text" --no-leading-addr --no-show-raw-insn %t.dylib | FileCheck %s --check-prefix=NO-FMA
 
 ; FMA:      <_foo>:
 ; FMA-NEXT: vrcpss       %xmm0, %xmm0, %xmm1
-; FMA-NEXT: vfmsub213ss  7(%rip), %xmm1, %xmm0
+; FMA-NEXT: vfmsub213ss  [[#]](%rip), %xmm1, %xmm0
 ; FMA-NEXT: vfnmadd132ss %xmm1, %xmm1, %xmm0
 ; FMA-NEXT: retq
 
 ; NO-FMA:      <_foo>:
 ; NO-FMA-NEXT: vrcpss %xmm0, %xmm0, %xmm1
 ; NO-FMA-NEXT: vmulss %xmm1, %xmm0, %xmm0
-; NO-FMA-NEXT: vmovss 16(%rip), %xmm2
+; NO-FMA-NEXT: vmovss [[#]](%rip), %xmm2
 ; NO-FMA-NEXT: vsubss %xmm0, %xmm2, %xmm0
 ; NO-FMA-NEXT: vmulss %xmm0, %xmm1, %xmm0
 ; NO-FMA-NEXT: vaddss %xmm0, %xmm1, %xmm0
 ; NO-FMA-NEXT: retq
 
 target triple = "x86_64-apple-darwin"
 target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 
 define float @foo(float %x) #0 {
   %div = fdiv fast float 1.0, %x
   ret float %div
 }
 
 attributes #0 = { "unsafe-fp-math"="true" "reciprocal-estimates"="divf,vec-divf" }