diff --git a/file/filename.cc b/file/filename.cc
index 968adbaa7..0ac4e6ef8 100644
--- a/file/filename.cc
+++ b/file/filename.cc
@@ -461,8 +461,8 @@ Status GetInfoLogFiles(Env* env, const std::string& db_log_dir,
 std::string NormalizePath(const std::string& path) {
   std::string dst;
   for (auto c : path) {
-    if (!dst.empty() && c == kFilePathSeparator &&
-        dst.back() == kFilePathSeparator) {
+    if (!dst.empty() && (c == kFilePathSeparator || c == '/') &&
+        (dst.back() == kFilePathSeparator || dst.back() == '/')) {
       continue;
     }
     dst.push_back(c);
diff --git a/port/win/port_win.cc b/port/win/port_win.cc
index 2d99a7a9b..34f6b28de 100644
--- a/port/win/port_win.cc
+++ b/port/win/port_win.cc
@@ -100,6 +100,14 @@ bool CondVar::TimedWait(uint64_t abs_time_us) {
 
   // Caller must ensure that mutex is held prior to calling this method
   std::unique_lock<std::mutex> lk(mu_->getLock(), std::adopt_lock);
+
+  // Work around https://github.com/microsoft/STL/issues/369
+#if defined(_MSC_VER) && (!defined(_MSVC_STL_UPDATE) || _MSVC_STL_UPDATE < 202008L)
+  if (relTimeUs == microseconds::zero()) {
+    lk.unlock();
+    lk.lock();
+  }
+#endif
 #ifndef NDEBUG
   mu_->locked_ = false;
 #endif
diff --git a/port/win/port_win.h b/port/win/port_win.h
index a3ffd559c..2c5b8ff05 100644
--- a/port/win/port_win.h
+++ b/port/win/port_win.h
@@ -283,7 +283,7 @@ extern const size_t kPageSize;
 #endif
 
 static inline void AsmVolatilePause() {
-#if defined(_M_IX86) || defined(_M_X64)
+#if defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) || defined(_M_ARM)
   YieldProcessor();
 #endif
   // it would be nice to get "wfe" on ARM here
diff --git a/third-party/folly/folly/chrono/Hardware.h b/third-party/folly/folly/chrono/Hardware.h
index ec7be82e8..6635b8717 100644
--- a/third-party/folly/folly/chrono/Hardware.h
+++ b/third-party/folly/folly/chrono/Hardware.h
@@ -10,7 +10,7 @@
 #include <chrono>
 #include <cstdint>
 
-#if _MSC_VER
+#if _MSC_VER && (defined(_M_IX86) || defined(_M_X64))
 extern "C" std::uint64_t __rdtsc();
 #pragma intrinsic(__rdtsc)
 #endif
@@ -18,7 +18,7 @@ extern "C" std::uint64_t __rdtsc();
 namespace folly {
 
 inline std::uint64_t hardware_timestamp() {
-#if _MSC_VER
+#if _MSC_VER && (defined(_M_IX86) || defined(_M_X64))
   return __rdtsc();
 #elif __GNUC__ && (__i386__ || FOLLY_X64)
   return __builtin_ia32_rdtsc();
diff --git a/util/math.h b/util/math.h
index 64cdb2f44..514943164 100644
--- a/util/math.h
+++ b/util/math.h
@@ -13,24 +13,75 @@
 
 namespace ROCKSDB_NAMESPACE {
 
+#if defined(_MSC_VER) && !defined(_M_X64)
+namespace detail {
+template <typename T>
+int BitsSetToOneFallback(T v) {
+  const int kBits = static_cast<int>(sizeof(T)) * 8;
+  static_assert((kBits & (kBits - 1)) == 0, "must be power of two bits");
+  // we static_cast these bit patterns in order to truncate them to the correct size
+  v = static_cast<T>(v - ((v >> 1) & static_cast<T>(0x5555555555555555ull)));
+  v = static_cast<T>((v & static_cast<T>(0x3333333333333333ull)) +
+                     ((v >> 2) & static_cast<T>(0x3333333333333333ull)));
+  v = static_cast<T>((v + (v >> 4)) & static_cast<T>(0x0F0F0F0F0F0F0F0Full));
+  for (int shift_bits = 8; shift_bits < kBits; shift_bits <<= 1) {
+    v += static_cast<T>(v >> shift_bits);
+  }
+  // we want the bottom "slot" that's big enough to store kDigits
+  // (and including) kBits.
+  return static_cast<int>(v & static_cast<T>(kBits | (kBits - 1)));
+}
+
+} // namespace detail
+#endif
+
+// Number of bits set to 1. Also known as "population count".
 template <typename T>
 inline int BitsSetToOne(T v) {
   static_assert(std::is_integral<T>::value, "non-integral type");
 #ifdef _MSC_VER
   static_assert(sizeof(T) <= sizeof(uint64_t), "type too big");
-  if (sizeof(T) > sizeof(uint32_t)) {
-    return static_cast<int>(__popcnt64(static_cast<uint64_t>(v)));
-  } else {
+  if (sizeof(T) < sizeof(uint32_t)) {
+    // This bit mask is to avoid a compiler warning on unused path
+    constexpr auto mm = 8 * sizeof(uint32_t) - 1;
+    // The bit mask is to neutralize sign extension on small signed types
+    constexpr uint32_t m = (uint32_t{1} << ((8 * sizeof(T)) & mm)) - 1;
+#if defined(_M_X64) || defined(_M_IX86)
+    return static_cast<int>(__popcnt(static_cast<uint32_t>(v) & m));
+#else
+    return static_cast<int>(detail::BitsSetToOneFallback(v) & m);
+#endif
+  } else if (sizeof(T) == sizeof(uint32_t)) {
+#if defined(_M_X64) || defined(_M_IX86)
     return static_cast<int>(__popcnt(static_cast<uint32_t>(v)));
+#else
+    return detail::BitsSetToOneFallback(static_cast<uint32_t>(v));
+#endif
+  } else {
+#ifdef _M_X64
+    return static_cast<int>(__popcnt64(static_cast<uint64_t>(v)));
+#elif defined(_M_IX86)
+    return static_cast<int>(
+        __popcnt(static_cast<uint32_t>(static_cast<uint64_t>(v) >> 32) +
+                 __popcnt(static_cast<uint32_t>(v))));
+#else
+    return detail::BitsSetToOneFallback(static_cast<uint64_t>(v));
+#endif
   }
 #else
   static_assert(sizeof(T) <= sizeof(unsigned long long), "type too big");
-  if (sizeof(T) > sizeof(unsigned long)) {
-    return __builtin_popcountll(static_cast<unsigned long long>(v));
-  } else if (sizeof(T) > sizeof(unsigned int)) {
+  if (sizeof(T) < sizeof(unsigned int)) {
+    // This bit mask is to avoid a compiler warning on unused path
+    constexpr auto mm = 8 * sizeof(unsigned int) - 1;
+    // This bit mask is to neutralize sign extension on small signed types
+    constexpr unsigned int m = (1U << ((8 * sizeof(T)) & mm)) - 1;
+    return __builtin_popcount(static_cast<unsigned int>(v) & m);
+  } else if (sizeof(T) == sizeof(unsigned int)) {
+    return __builtin_popcount(static_cast<unsigned int>(v));
+  } else if (sizeof(T) <= sizeof(unsigned long)) {
     return __builtin_popcountl(static_cast<unsigned long>(v));
   } else {
-    return __builtin_popcount(static_cast<unsigned int>(v));
+    return __builtin_popcountll(static_cast<unsigned long long>(v));
   }
 #endif
 }
diff --git a/util/xxh3p.h b/util/xxh3p.h
index d1fc2bba2..c63f0e36d 100644
--- a/util/xxh3p.h
+++ b/util/xxh3p.h
@@ -253,7 +253,7 @@ XXH_FORCE_INLINE U64x2 XXH_vec_mule(U32x4 a, U32x4 b) {
 #if defined(XXH_NO_PREFETCH)
 #  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
 #else
-#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() is not defined outside of x86/x64 */
 #    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
 #    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
 #  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )