snappy 0.3.0 → 0.5.0

data/vendor/snappy/snappy-stubs-internal.h

@@ -31,21 +31,23 @@
 #ifndef THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_
 #define THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_
 
-#ifdef HAVE_CONFIG_H
+#if HAVE_CONFIG_H
 #include "config.h"
 #endif
 
-#include <string>
+#include <stdint.h>
 
-#include <assert.h>
-#include <stdlib.h>
-#include <string.h>
+#include <cassert>
+#include <cstdlib>
+#include <cstring>
+#include <limits>
+#include <string>
 
-#ifdef HAVE_SYS_MMAN_H
+#if HAVE_SYS_MMAN_H
 #include <sys/mman.h>
 #endif
 
-#ifdef HAVE_UNISTD_H
+#if HAVE_UNISTD_H
 #include <unistd.h>
 #endif
 
@@ -67,19 +69,11 @@
 
 #include "snappy-stubs-public.h"
 
-#if defined(__x86_64__)
-
-// Enable 64-bit optimized versions of some routines.
-#define ARCH_K8 1
-
-#elif defined(__ppc64__)
-
+// Used to enable 64-bit optimized versions of some routines.
+#if defined(__PPC64__) || defined(__powerpc64__)
 #define ARCH_PPC 1
-
-#elif defined(__aarch64__)
-
+#elif defined(__aarch64__) || defined(_M_ARM64)
 #define ARCH_ARM 1
-
 #endif
 
 // Needed by OS X, among others.
@@ -93,223 +87,83 @@
 #ifdef ARRAYSIZE
 #undef ARRAYSIZE
 #endif
-#define ARRAYSIZE(a) (sizeof(a) / sizeof(*(a)))
+#define ARRAYSIZE(a) int{sizeof(a) / sizeof(*(a))}
 
 // Static prediction hints.
-#ifdef HAVE_BUILTIN_EXPECT
+#if HAVE_BUILTIN_EXPECT
 #define SNAPPY_PREDICT_FALSE(x) (__builtin_expect(x, 0))
 #define SNAPPY_PREDICT_TRUE(x) (__builtin_expect(!!(x), 1))
 #else
 #define SNAPPY_PREDICT_FALSE(x) x
 #define SNAPPY_PREDICT_TRUE(x) x
-#endif
-
-// This is only used for recomputing the tag byte table used during
-// decompression; for simplicity we just remove it from the open-source
-// version (anyone who wants to regenerate it can just do the call
-// themselves within main()).
-#define DEFINE_bool(flag_name, default_value, description) \
-  bool FLAGS_ ## flag_name = default_value
-#define DECLARE_bool(flag_name) \
-  extern bool FLAGS_ ## flag_name
-
-namespace snappy {
-
-static const uint32 kuint32max = static_cast<uint32>(0xFFFFFFFF);
-static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL);
+#endif  // HAVE_BUILTIN_EXPECT
 
-// Potentially unaligned loads and stores.
-
-// x86, PowerPC, and ARM64 can simply do these loads and stores native.
-
-#if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || \
-    defined(__aarch64__)
-
-#define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p))
-#define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p))
-#define UNALIGNED_LOAD64(_p) (*reinterpret_cast<const uint64 *>(_p))
-
-#define UNALIGNED_STORE16(_p, _val) (*reinterpret_cast<uint16 *>(_p) = (_val))
-#define UNALIGNED_STORE32(_p, _val) (*reinterpret_cast<uint32 *>(_p) = (_val))
-#define UNALIGNED_STORE64(_p, _val) (*reinterpret_cast<uint64 *>(_p) = (_val))
+// Inlining hints.
+#if HAVE_ATTRIBUTE_ALWAYS_INLINE
+#define SNAPPY_ATTRIBUTE_ALWAYS_INLINE __attribute__((always_inline))
+#else
+#define SNAPPY_ATTRIBUTE_ALWAYS_INLINE
+#endif  // HAVE_ATTRIBUTE_ALWAYS_INLINE
 
-// ARMv7 and newer support native unaligned accesses, but only of 16-bit
-// and 32-bit values (not 64-bit); older versions either raise a fatal signal,
-// do an unaligned read and rotate the words around a bit, or do the reads very
-// slowly (trip through kernel mode). There's no simple #define that says just
-// “ARMv7 or higher”, so we have to filter away all ARMv5 and ARMv6
-// sub-architectures.
-//
-// This is a mess, but there's not much we can do about it.
-//
-// To further complicate matters, only LDR instructions (single reads) are
-// allowed to be unaligned, not LDRD (two reads) or LDM (many reads). Unless we
-// explicitly tell the compiler that these accesses can be unaligned, it can and
-// will combine accesses. On armcc, the way to signal this is done by accessing
-// through the type (uint32 __packed *), but GCC has no such attribute
-// (it ignores __attribute__((packed)) on individual variables). However,
-// we can tell it that a _struct_ is unaligned, which has the same effect,
-// so we do that.
-
-#elif defined(__arm__) && \
-      !defined(__ARM_ARCH_4__) && \
-      !defined(__ARM_ARCH_4T__) && \
-      !defined(__ARM_ARCH_5__) && \
-      !defined(__ARM_ARCH_5T__) && \
-      !defined(__ARM_ARCH_5TE__) && \
-      !defined(__ARM_ARCH_5TEJ__) && \
-      !defined(__ARM_ARCH_6__) && \
-      !defined(__ARM_ARCH_6J__) && \
-      !defined(__ARM_ARCH_6K__) && \
-      !defined(__ARM_ARCH_6Z__) && \
-      !defined(__ARM_ARCH_6ZK__) && \
-      !defined(__ARM_ARCH_6T2__)
-
-#if __GNUC__
-#define ATTRIBUTE_PACKED __attribute__((__packed__))
+#if HAVE_BUILTIN_PREFETCH
+#define SNAPPY_PREFETCH(ptr) __builtin_prefetch(ptr, 0, 3)
 #else
-#define ATTRIBUTE_PACKED
+#define SNAPPY_PREFETCH(ptr) (void)(ptr)
 #endif
 
-namespace base {
-namespace internal {
-
-struct Unaligned16Struct {
-  uint16 value;
-  uint8 dummy;  // To make the size non-power-of-two.
-} ATTRIBUTE_PACKED;
-
-struct Unaligned32Struct {
-  uint32 value;
-  uint8 dummy;  // To make the size non-power-of-two.
-} ATTRIBUTE_PACKED;
-
-}  // namespace internal
-}  // namespace base
-
-#define UNALIGNED_LOAD16(_p) \
-    ((reinterpret_cast<const ::snappy::base::internal::Unaligned16Struct *>(_p))->value)
-#define UNALIGNED_LOAD32(_p) \
-    ((reinterpret_cast<const ::snappy::base::internal::Unaligned32Struct *>(_p))->value)
-
-#define UNALIGNED_STORE16(_p, _val) \
-    ((reinterpret_cast< ::snappy::base::internal::Unaligned16Struct *>(_p))->value = \
-         (_val))
-#define UNALIGNED_STORE32(_p, _val) \
-    ((reinterpret_cast< ::snappy::base::internal::Unaligned32Struct *>(_p))->value = \
-         (_val))
-
-// TODO: NEON supports unaligned 64-bit loads and stores.
-// See if that would be more efficient on platforms supporting it,
-// at least for copies.
-
-inline uint64 UNALIGNED_LOAD64(const void *p) {
-  uint64 t;
-  memcpy(&t, p, sizeof t);
-  return t;
-}
-
-inline void UNALIGNED_STORE64(void *p, uint64 v) {
-  memcpy(p, &v, sizeof v);
-}
-
-#else
+// Stubbed version of ABSL_FLAG.
+//
+// In the open source version, flags can only be changed at compile time.
+#define SNAPPY_FLAG(flag_type, flag_name, default_value, help) \
+  flag_type FLAGS_ ## flag_name = default_value
 
-// These functions are provided for architectures that don't support
-// unaligned loads and stores.
+namespace snappy {
 
-inline uint16 UNALIGNED_LOAD16(const void *p) {
-  uint16 t;
-  memcpy(&t, p, sizeof t);
-  return t;
-}
+// Stubbed version of absl::GetFlag().
+template <typename T>
+inline T GetFlag(T flag) { return flag; }
 
-inline uint32 UNALIGNED_LOAD32(const void *p) {
-  uint32 t;
-  memcpy(&t, p, sizeof t);
-  return t;
-}
+static const uint32_t kuint32max = std::numeric_limits<uint32_t>::max();
+static const int64_t kint64max = std::numeric_limits<int64_t>::max();
 
-inline uint64 UNALIGNED_LOAD64(const void *p) {
-  uint64 t;
-  memcpy(&t, p, sizeof t);
-  return t;
-}
+// Potentially unaligned loads and stores.
 
-inline void UNALIGNED_STORE16(void *p, uint16 v) {
-  memcpy(p, &v, sizeof v);
+inline uint16_t UNALIGNED_LOAD16(const void *p) {
+  // Compiles to a single movzx/ldrh on clang/gcc/msvc.
+  uint16_t v;
+  std::memcpy(&v, p, sizeof(v));
+  return v;
 }
 
-inline void UNALIGNED_STORE32(void *p, uint32 v) {
-  memcpy(p, &v, sizeof v);
+inline uint32_t UNALIGNED_LOAD32(const void *p) {
+  // Compiles to a single mov/ldr on clang/gcc/msvc.
+  uint32_t v;
+  std::memcpy(&v, p, sizeof(v));
+  return v;
 }
 
-inline void UNALIGNED_STORE64(void *p, uint64 v) {
-  memcpy(p, &v, sizeof v);
+inline uint64_t UNALIGNED_LOAD64(const void *p) {
+  // Compiles to a single mov/ldr on clang/gcc/msvc.
+  uint64_t v;
+  std::memcpy(&v, p, sizeof(v));
+  return v;
 }
 
-#endif
-
-// The following guarantees declaration of the byte swap functions.
-#if defined(SNAPPY_IS_BIG_ENDIAN)
-
-#ifdef HAVE_SYS_BYTEORDER_H
-#include <sys/byteorder.h>
-#endif
-
-#ifdef HAVE_SYS_ENDIAN_H
-#include <sys/endian.h>
-#endif
-
-#ifdef _MSC_VER
-#include <stdlib.h>
-#define bswap_16(x) _byteswap_ushort(x)
-#define bswap_32(x) _byteswap_ulong(x)
-#define bswap_64(x) _byteswap_uint64(x)
-
-#elif defined(__APPLE__)
-// Mac OS X / Darwin features
-#include <libkern/OSByteOrder.h>
-#define bswap_16(x) OSSwapInt16(x)
-#define bswap_32(x) OSSwapInt32(x)
-#define bswap_64(x) OSSwapInt64(x)
-
-#elif defined(HAVE_BYTESWAP_H)
-#include <byteswap.h>
-
-#elif defined(bswap32)
-// FreeBSD defines bswap{16,32,64} in <sys/endian.h> (already #included).
-#define bswap_16(x) bswap16(x)
-#define bswap_32(x) bswap32(x)
-#define bswap_64(x) bswap64(x)
-
-#elif defined(BSWAP_64)
-// Solaris 10 defines BSWAP_{16,32,64} in <sys/byteorder.h> (already #included).
-#define bswap_16(x) BSWAP_16(x)
-#define bswap_32(x) BSWAP_32(x)
-#define bswap_64(x) BSWAP_64(x)
-
-#else
-
-inline uint16 bswap_16(uint16 x) {
-  return (x << 8) | (x >> 8);
+inline void UNALIGNED_STORE16(void *p, uint16_t v) {
+  // Compiles to a single mov/strh on clang/gcc/msvc.
+  std::memcpy(p, &v, sizeof(v));
 }
 
-inline uint32 bswap_32(uint32 x) {
-  x = ((x & 0xff00ff00UL) >> 8) | ((x & 0x00ff00ffUL) << 8);
-  return (x >> 16) | (x << 16);
+inline void UNALIGNED_STORE32(void *p, uint32_t v) {
+  // Compiles to a single mov/str on clang/gcc/msvc.
+  std::memcpy(p, &v, sizeof(v));
 }
 
-inline uint64 bswap_64(uint64 x) {
-  x = ((x & 0xff00ff00ff00ff00ULL) >> 8) | ((x & 0x00ff00ff00ff00ffULL) << 8);
-  x = ((x & 0xffff0000ffff0000ULL) >> 16) | ((x & 0x0000ffff0000ffffULL) << 16);
-  return (x >> 32) | (x << 32);
+inline void UNALIGNED_STORE64(void *p, uint64_t v) {
+  // Compiles to a single mov/str on clang/gcc/msvc.
+  std::memcpy(p, &v, sizeof(v));
 }
 
-#endif
-
-#endif  // defined(SNAPPY_IS_BIG_ENDIAN)
-
 // Convert to little-endian storage, opposite of network format.
 // Convert x from host to little endian: x = LittleEndian.FromHost(x);
 // convert x from little endian to host: x = LittleEndian.ToHost(x);
@@ -321,44 +175,110 @@ inline uint64 bswap_64(uint64 x) {
 //    x = LittleEndian.Load16(p);
 class LittleEndian {
  public:
-  // Conversion functions.
-#if defined(SNAPPY_IS_BIG_ENDIAN)
-
-  static uint16 FromHost16(uint16 x) { return bswap_16(x); }
-  static uint16 ToHost16(uint16 x) { return bswap_16(x); }
-
-  static uint32 FromHost32(uint32 x) { return bswap_32(x); }
-  static uint32 ToHost32(uint32 x) { return bswap_32(x); }
-
-  static bool IsLittleEndian() { return false; }
-
-#else  // !defined(SNAPPY_IS_BIG_ENDIAN)
-
-  static uint16 FromHost16(uint16 x) { return x; }
-  static uint16 ToHost16(uint16 x) { return x; }
-
-  static uint32 FromHost32(uint32 x) { return x; }
-  static uint32 ToHost32(uint32 x) { return x; }
+  // Functions to do unaligned loads and stores in little-endian order.
+  static inline uint16_t Load16(const void *ptr) {
+    // Compiles to a single mov/str on recent clang and gcc.
+#if SNAPPY_IS_BIG_ENDIAN
+    const uint8_t* const buffer = reinterpret_cast<const uint8_t*>(ptr);
+    return (static_cast<uint16_t>(buffer[0])) |
+            (static_cast<uint16_t>(buffer[1]) << 8);
+#else
+    // memcpy() turns into a single instruction early in the optimization
+    // pipeline (relatively to a series of byte accesses). So, using memcpy
+    // instead of byte accesses may lead to better decisions in more stages of
+    // the optimization pipeline.
+    uint16_t value;
+    std::memcpy(&value, ptr, 2);
+    return value;
+#endif
+  }
 
-  static bool IsLittleEndian() { return true; }
+  static inline uint32_t Load32(const void *ptr) {
+    // Compiles to a single mov/str on recent clang and gcc.
+#if SNAPPY_IS_BIG_ENDIAN
+    const uint8_t* const buffer = reinterpret_cast<const uint8_t*>(ptr);
+    return (static_cast<uint32_t>(buffer[0])) |
+            (static_cast<uint32_t>(buffer[1]) << 8) |
+            (static_cast<uint32_t>(buffer[2]) << 16) |
+            (static_cast<uint32_t>(buffer[3]) << 24);
+#else
+    // See Load16() for the rationale of using memcpy().
+    uint32_t value;
+    std::memcpy(&value, ptr, 4);
+    return value;
+#endif
+  }
 
-#endif  // !defined(SNAPPY_IS_BIG_ENDIAN)
+  static inline uint64_t Load64(const void *ptr) {
+    // Compiles to a single mov/str on recent clang and gcc.
+#if SNAPPY_IS_BIG_ENDIAN
+    const uint8_t* const buffer = reinterpret_cast<const uint8_t*>(ptr);
+    return (static_cast<uint64_t>(buffer[0])) |
+            (static_cast<uint64_t>(buffer[1]) << 8) |
+            (static_cast<uint64_t>(buffer[2]) << 16) |
+            (static_cast<uint64_t>(buffer[3]) << 24) |
+            (static_cast<uint64_t>(buffer[4]) << 32) |
+            (static_cast<uint64_t>(buffer[5]) << 40) |
+            (static_cast<uint64_t>(buffer[6]) << 48) |
+            (static_cast<uint64_t>(buffer[7]) << 56);
+#else
+    // See Load16() for the rationale of using memcpy().
+    uint64_t value;
+    std::memcpy(&value, ptr, 8);
+    return value;
+#endif
+  }
 
-  // Functions to do unaligned loads and stores in little-endian order.
-  static uint16 Load16(const void *p) {
-    return ToHost16(UNALIGNED_LOAD16(p));
+  static inline void Store16(void *dst, uint16_t value) {
+    // Compiles to a single mov/str on recent clang and gcc.
+#if SNAPPY_IS_BIG_ENDIAN
+    uint8_t* const buffer = reinterpret_cast<uint8_t*>(dst);
+    buffer[0] = static_cast<uint8_t>(value);
+    buffer[1] = static_cast<uint8_t>(value >> 8);
+#else
+    // See Load16() for the rationale of using memcpy().
+    std::memcpy(dst, &value, 2);
+#endif
   }
 
-  static void Store16(void *p, uint16 v) {
-    UNALIGNED_STORE16(p, FromHost16(v));
+  static void Store32(void *dst, uint32_t value) {
+    // Compiles to a single mov/str on recent clang and gcc.
+#if SNAPPY_IS_BIG_ENDIAN
+    uint8_t* const buffer = reinterpret_cast<uint8_t*>(dst);
+    buffer[0] = static_cast<uint8_t>(value);
+    buffer[1] = static_cast<uint8_t>(value >> 8);
+    buffer[2] = static_cast<uint8_t>(value >> 16);
+    buffer[3] = static_cast<uint8_t>(value >> 24);
+#else
+    // See Load16() for the rationale of using memcpy().
+    std::memcpy(dst, &value, 4);
+#endif
   }
 
-  static uint32 Load32(const void *p) {
-    return ToHost32(UNALIGNED_LOAD32(p));
+  static void Store64(void* dst, uint64_t value) {
+    // Compiles to a single mov/str on recent clang and gcc.
+#if SNAPPY_IS_BIG_ENDIAN
+    uint8_t* const buffer = reinterpret_cast<uint8_t*>(dst);
+    buffer[0] = static_cast<uint8_t>(value);
+    buffer[1] = static_cast<uint8_t>(value >> 8);
+    buffer[2] = static_cast<uint8_t>(value >> 16);
+    buffer[3] = static_cast<uint8_t>(value >> 24);
+    buffer[4] = static_cast<uint8_t>(value >> 32);
+    buffer[5] = static_cast<uint8_t>(value >> 40);
+    buffer[6] = static_cast<uint8_t>(value >> 48);
+    buffer[7] = static_cast<uint8_t>(value >> 56);
+#else
+    // See Load16() for the rationale of using memcpy().
+    std::memcpy(dst, &value, 8);
+#endif
   }
 
-  static void Store32(void *p, uint32 v) {
-    UNALIGNED_STORE32(p, FromHost32(v));
+  static inline constexpr bool IsLittleEndian() {
+#if SNAPPY_IS_BIG_ENDIAN
+    return false;
+#else
+    return true;
+#endif  // SNAPPY_IS_BIG_ENDIAN
   }
 };
 
@@ -366,19 +286,17 @@ class LittleEndian {
 class Bits {
  public:
   // Return floor(log2(n)) for positive integer n.
-  static int Log2FloorNonZero(uint32 n);
+  static int Log2FloorNonZero(uint32_t n);
 
   // Return floor(log2(n)) for positive integer n.  Returns -1 iff n == 0.
-  static int Log2Floor(uint32 n);
+  static int Log2Floor(uint32_t n);
 
   // Return the first set least / most significant bit, 0-indexed.  Returns an
   // undefined value if n == 0.  FindLSBSetNonZero() is similar to ffs() except
   // that it's 0-indexed.
-  static int FindLSBSetNonZero(uint32 n);
+  static int FindLSBSetNonZero(uint32_t n);
 
-#if defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
-  static int FindLSBSetNonZero64(uint64 n);
-#endif  // defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
+  static int FindLSBSetNonZero64(uint64_t n);
 
  private:
   // No copying
@@ -386,9 +304,9 @@ class Bits {
   void operator=(const Bits&);
 };
 
-#ifdef HAVE_BUILTIN_CTZ
+#if HAVE_BUILTIN_CTZ
 
-inline int Bits::Log2FloorNonZero(uint32 n) {
+inline int Bits::Log2FloorNonZero(uint32_t n) {
   assert(n != 0);
   // (31 ^ x) is equivalent to (31 - x) for x in [0, 31]. An easy proof
   // represents subtraction in base 2 and observes that there's no carry.
@@ -399,66 +317,52 @@ inline int Bits::Log2FloorNonZero(uint32 n) {
   return 31 ^ __builtin_clz(n);
 }
 
-inline int Bits::Log2Floor(uint32 n) {
+inline int Bits::Log2Floor(uint32_t n) {
   return (n == 0) ? -1 : Bits::Log2FloorNonZero(n);
 }
 
-inline int Bits::FindLSBSetNonZero(uint32 n) {
+inline int Bits::FindLSBSetNonZero(uint32_t n) {
   assert(n != 0);
   return __builtin_ctz(n);
 }
 
-#if defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
-inline int Bits::FindLSBSetNonZero64(uint64 n) {
-  assert(n != 0);
-  return __builtin_ctzll(n);
-}
-#endif  // defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
-
 #elif defined(_MSC_VER)
 
-inline int Bits::Log2FloorNonZero(uint32 n) {
+inline int Bits::Log2FloorNonZero(uint32_t n) {
   assert(n != 0);
+  // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long.
   unsigned long where;
   _BitScanReverse(&where, n);
   return static_cast<int>(where);
 }
 
-inline int Bits::Log2Floor(uint32 n) {
+inline int Bits::Log2Floor(uint32_t n) {
+  // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long.
   unsigned long where;
   if (_BitScanReverse(&where, n))
     return static_cast<int>(where);
   return -1;
 }
 
-inline int Bits::FindLSBSetNonZero(uint32 n) {
+inline int Bits::FindLSBSetNonZero(uint32_t n) {
   assert(n != 0);
+  // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long.
   unsigned long where;
   if (_BitScanForward(&where, n))
     return static_cast<int>(where);
   return 32;
 }
 
-#if defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
-inline int Bits::FindLSBSetNonZero64(uint64 n) {
-  assert(n != 0);
-  unsigned long where;
-  if (_BitScanForward64(&where, n))
-    return static_cast<int>(where);
-  return 64;
-}
-#endif  // defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
-
 #else  // Portable versions.
 
-inline int Bits::Log2FloorNonZero(uint32 n) {
+inline int Bits::Log2FloorNonZero(uint32_t n) {
   assert(n != 0);
 
   int log = 0;
-  uint32 value = n;
+  uint32_t value = n;
   for (int i = 4; i >= 0; --i) {
     int shift = (1 << i);
-    uint32 x = value >> shift;
+    uint32_t x = value >> shift;
     if (x != 0) {
       value = x;
       log += shift;
@@ -468,16 +372,16 @@ inline int Bits::Log2FloorNonZero(uint32 n) {
   return log;
 }
 
-inline int Bits::Log2Floor(uint32 n) {
+inline int Bits::Log2Floor(uint32_t n) {
   return (n == 0) ? -1 : Bits::Log2FloorNonZero(n);
 }
 
-inline int Bits::FindLSBSetNonZero(uint32 n) {
+inline int Bits::FindLSBSetNonZero(uint32_t n) {
   assert(n != 0);
 
   int rc = 31;
   for (int i = 4, shift = 1 << 4; i >= 0; --i) {
-    const uint32 x = n << shift;
+    const uint32_t x = n << shift;
     if (x != 0) {
       n = x;
       rc -= shift;
@@ -487,27 +391,48 @@ inline int Bits::FindLSBSetNonZero(uint32 n) {
   return rc;
 }
 
-#if defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
+#endif  // End portable versions.
+
+#if HAVE_BUILTIN_CTZ
+
+inline int Bits::FindLSBSetNonZero64(uint64_t n) {
+  assert(n != 0);
+  return __builtin_ctzll(n);
+}
+
+#elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_ARM64))
+// _BitScanForward64() is only available on x64 and ARM64.
+
+inline int Bits::FindLSBSetNonZero64(uint64_t n) {
+  assert(n != 0);
+  // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long.
+  unsigned long where;
+  if (_BitScanForward64(&where, n))
+    return static_cast<int>(where);
+  return 64;
+}
+
+#else  // Portable version.
+
 // FindLSBSetNonZero64() is defined in terms of FindLSBSetNonZero().
-inline int Bits::FindLSBSetNonZero64(uint64 n) {
+inline int Bits::FindLSBSetNonZero64(uint64_t n) {
   assert(n != 0);
 
-  const uint32 bottombits = static_cast<uint32>(n);
+  const uint32_t bottombits = static_cast<uint32_t>(n);
   if (bottombits == 0) {
-    // Bottom bits are zero, so scan in top bits
-    return 32 + FindLSBSetNonZero(static_cast<uint32>(n >> 32));
+    // Bottom bits are zero, so scan the top bits.
+    return 32 + FindLSBSetNonZero(static_cast<uint32_t>(n >> 32));
   } else {
     return FindLSBSetNonZero(bottombits);
   }
 }
-#endif  // defined(ARCH_K8) || defined(ARCH_PPC) || defined(ARCH_ARM)
 
-#endif  // End portable versions.
+#endif  // HAVE_BUILTIN_CTZ
 
 // Variable-length integer encoding.
 class Varint {
  public:
-  // Maximum lengths of varint encoding of uint32.
+  // Maximum lengths of varint encoding of uint32_t.
   static const int kMax32 = 5;
 
   // Attempts to parse a varint32 from a prefix of the bytes in [ptr,limit-1].
@@ -516,23 +441,23 @@ class Varint {
   // past the last byte of the varint32. Else returns NULL.  On success,
   // "result <= limit".
   static const char* Parse32WithLimit(const char* ptr, const char* limit,
-                                      uint32* OUTPUT);
+                                      uint32_t* OUTPUT);
 
   // REQUIRES   "ptr" points to a buffer of length sufficient to hold "v".
   // EFFECTS    Encodes "v" into "ptr" and returns a pointer to the
   //            byte just past the last encoded byte.
-  static char* Encode32(char* ptr, uint32 v);
+  static char* Encode32(char* ptr, uint32_t v);
 
   // EFFECTS    Appends the varint representation of "value" to "*s".
-  static void Append32(std::string* s, uint32 value);
+  static void Append32(std::string* s, uint32_t value);
 };
 
 inline const char* Varint::Parse32WithLimit(const char* p,
                                             const char* l,
-                                            uint32* OUTPUT) {
+                                            uint32_t* OUTPUT) {
   const unsigned char* ptr = reinterpret_cast<const unsigned char*>(p);
   const unsigned char* limit = reinterpret_cast<const unsigned char*>(l);
-  uint32 b, result;
+  uint32_t b, result;
   if (ptr >= limit) return NULL;
   b = *(ptr++); result = b & 127;          if (b < 128) goto done;
   if (ptr >= limit) return NULL;
@@ -549,30 +474,30 @@ inline const char* Varint::Parse32WithLimit(const char* p,
   return reinterpret_cast<const char*>(ptr);
 }
 
-inline char* Varint::Encode32(char* sptr, uint32 v) {
+inline char* Varint::Encode32(char* sptr, uint32_t v) {
   // Operate on characters as unsigneds
-  unsigned char* ptr = reinterpret_cast<unsigned char*>(sptr);
-  static const int B = 128;
-  if (v < (1<<7)) {
-    *(ptr++) = v;
-  } else if (v < (1<<14)) {
-    *(ptr++) = v | B;
-    *(ptr++) = v>>7;
-  } else if (v < (1<<21)) {
-    *(ptr++) = v | B;
-    *(ptr++) = (v>>7) | B;
-    *(ptr++) = v>>14;
-  } else if (v < (1<<28)) {
-    *(ptr++) = v | B;
-    *(ptr++) = (v>>7) | B;
-    *(ptr++) = (v>>14) | B;
-    *(ptr++) = v>>21;
+  uint8_t* ptr = reinterpret_cast<uint8_t*>(sptr);
+  static const uint8_t B = 128;
+  if (v < (1 << 7)) {
+    *(ptr++) = static_cast<uint8_t>(v);
+  } else if (v < (1 << 14)) {
+    *(ptr++) = static_cast<uint8_t>(v | B);
+    *(ptr++) = static_cast<uint8_t>(v >> 7);
+  } else if (v < (1 << 21)) {
+    *(ptr++) = static_cast<uint8_t>(v | B);
+    *(ptr++) = static_cast<uint8_t>((v >> 7) | B);
+    *(ptr++) = static_cast<uint8_t>(v >> 14);
+  } else if (v < (1 << 28)) {
+    *(ptr++) = static_cast<uint8_t>(v | B);
+    *(ptr++) = static_cast<uint8_t>((v >> 7) | B);
+    *(ptr++) = static_cast<uint8_t>((v >> 14) | B);
+    *(ptr++) = static_cast<uint8_t>(v >> 21);
   } else {
-    *(ptr++) = v | B;
-    *(ptr++) = (v>>7) | B;
-    *(ptr++) = (v>>14) | B;
-    *(ptr++) = (v>>21) | B;
-    *(ptr++) = v>>28;
+    *(ptr++) = static_cast<uint8_t>(v | B);
+    *(ptr++) = static_cast<uint8_t>((v>>7) | B);
+    *(ptr++) = static_cast<uint8_t>((v>>14) | B);
+    *(ptr++) = static_cast<uint8_t>((v>>21) | B);
+    *(ptr++) = static_cast<uint8_t>(v >> 28);
   }
   return reinterpret_cast<char*>(ptr);
 }