X-Git-Url: http://plrg.eecs.uci.edu/git/?p=folly.git;a=blobdiff_plain;f=folly%2FRWSpinLock.h;h=2f177ed001bdc71142fe4b76218ab9819892f1f4;hp=a5b6053467ab8c6f8e686c04f1ca0312fbae62e8;hb=eba5e7f2236a0950957a8a247a0e83385149d13c;hpb=fe33916f8b77db93ca0737ad9d00cd48d216c6e4

diff --git a/folly/RWSpinLock.h b/folly/RWSpinLock.h
index a5b60534..2f177ed0 100644
--- a/folly/RWSpinLock.h
+++ b/folly/RWSpinLock.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2016 Facebook, Inc.
+ * Copyright 2017 Facebook, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -81,8 +81,7 @@
  * @author Xin Liu <xliux@fb.com>
  */
 
-#ifndef FOLLY_RWSPINLOCK_H_
-#define FOLLY_RWSPINLOCK_H_
+#pragma once
 
 /*
 ========================================================================
@@ -139,29 +138,29 @@ pthread_rwlock_t Read        728698     24us       101ns     7.28ms     194us
 #include <folly/Portability.h>
 #include <folly/portability/Asm.h>
 
-#if defined(__GNUC__) && \
-  (defined(__i386) || FOLLY_X64 || \
-   defined(ARCH_K8))
-# define RW_SPINLOCK_USE_X86_INTRINSIC_
-# include <x86intrin.h>
+#if defined(__GNUC__) && (defined(__i386) || FOLLY_X64 || defined(ARCH_K8))
+#define RW_SPINLOCK_USE_X86_INTRINSIC_
+#include <x86intrin.h>
 #elif defined(_MSC_VER) && defined(FOLLY_X64)
-# define RW_SPINLOCK_USE_X86_INTRINSIC_
+#define RW_SPINLOCK_USE_X86_INTRINSIC_
+#elif FOLLY_AARCH64
+#define RW_SPINLOCK_USE_X86_INTRINSIC_
 #else
-# undef RW_SPINLOCK_USE_X86_INTRINSIC_
+#undef RW_SPINLOCK_USE_X86_INTRINSIC_
 #endif
 
 // iOS doesn't define _mm_cvtsi64_si128 and friends
-#if (FOLLY_SSE >= 2) && !FOLLY_MOBILE
+#if (FOLLY_SSE >= 2) && !FOLLY_MOBILE && FOLLY_X64
 #define RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
 #else
 #undef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
 #endif
 
+#include <algorithm>
 #include <atomic>
 #include <string>
-#include <algorithm>
+#include <thread>
 
-#include <sched.h>
 #include <glog/logging.h>
 
 #include <folly/Likely.h>
@@ -193,9 +192,11 @@ class RWSpinLock {
 
   // Lockable Concept
   void lock() {
-    int count = 0;
+    uint_fast32_t count = 0;
     while (!LIKELY(try_lock())) {
-      if (++count > 1000) sched_yield();
+      if (++count > 1000) {
+        std::this_thread::yield();
+      }
     }
   }
 
@@ -207,9 +208,11 @@ class RWSpinLock {
 
   // SharedLockable Concept
   void lock_shared() {
-    int count = 0;
+    uint_fast32_t count = 0;
     while (!LIKELY(try_lock_shared())) {
-      if (++count > 1000) sched_yield();
+      if (++count > 1000) {
+        std::this_thread::yield();
+      }
     }
   }
 
@@ -225,9 +228,11 @@ class RWSpinLock {
 
   // UpgradeLockable Concept
   void lock_upgrade() {
-    int count = 0;
+    uint_fast32_t count = 0;
     while (!try_lock_upgrade()) {
-      if (++count > 1000) sched_yield();
+      if (++count > 1000) {
+        std::this_thread::yield();
+      }
     }
   }
 
@@ -239,7 +244,9 @@ class RWSpinLock {
   void unlock_upgrade_and_lock() {
     int64_t count = 0;
     while (!try_unlock_upgrade_and_lock()) {
-      if (++count > 1000) sched_yield();
+      if (++count > 1000) {
+        std::this_thread::yield();
+      }
     }
   }
 
@@ -308,8 +315,10 @@ class RWSpinLock {
 
   class ReadHolder {
    public:
-    explicit ReadHolder(RWSpinLock* lock = nullptr) : lock_(lock) {
-      if (lock_) lock_->lock_shared();
+    explicit ReadHolder(RWSpinLock* lock) : lock_(lock) {
+      if (lock_) {
+        lock_->lock_shared();
+      }
     }
 
     explicit ReadHolder(RWSpinLock& lock) : lock_(&lock) {
@@ -323,12 +332,16 @@ class RWSpinLock {
     // down-grade
     explicit ReadHolder(UpgradedHolder&& upgraded) : lock_(upgraded.lock_) {
       upgraded.lock_ = nullptr;
-      if (lock_) lock_->unlock_upgrade_and_lock_shared();
+      if (lock_) {
+        lock_->unlock_upgrade_and_lock_shared();
+      }
     }
 
     explicit ReadHolder(WriteHolder&& writer) : lock_(writer.lock_) {
       writer.lock_ = nullptr;
-      if (lock_) lock_->unlock_and_lock_shared();
+      if (lock_) {
+        lock_->unlock_and_lock_shared();
+      }
     }
 
     ReadHolder& operator=(ReadHolder&& other) {
@@ -340,13 +353,23 @@ class RWSpinLock {
     ReadHolder(const ReadHolder& other) = delete;
     ReadHolder& operator=(const ReadHolder& other) = delete;
 
-    ~ReadHolder() { if (lock_) lock_->unlock_shared(); }
+    ~ReadHolder() {
+      if (lock_) {
+        lock_->unlock_shared();
+      }
+    }
 
     void reset(RWSpinLock* lock = nullptr) {
-      if (lock == lock_) return;
-      if (lock_) lock_->unlock_shared();
+      if (lock == lock_) {
+        return;
+      }
+      if (lock_) {
+        lock_->unlock_shared();
+      }
       lock_ = lock;
-      if (lock_) lock_->lock_shared();
+      if (lock_) {
+        lock_->lock_shared();
+      }
     }
 
     void swap(ReadHolder* other) {
@@ -361,8 +384,10 @@ class RWSpinLock {
 
   class UpgradedHolder {
    public:
-    explicit UpgradedHolder(RWSpinLock* lock = nullptr) : lock_(lock) {
-      if (lock_) lock_->lock_upgrade();
+    explicit UpgradedHolder(RWSpinLock* lock) : lock_(lock) {
+      if (lock_) {
+        lock_->lock_upgrade();
+      }
     }
 
     explicit UpgradedHolder(RWSpinLock& lock) : lock_(&lock) {
@@ -372,7 +397,9 @@ class RWSpinLock {
     explicit UpgradedHolder(WriteHolder&& writer) {
       lock_ = writer.lock_;
       writer.lock_ = nullptr;
-      if (lock_) lock_->unlock_and_lock_upgrade();
+      if (lock_) {
+        lock_->unlock_and_lock_upgrade();
+      }
     }
 
     UpgradedHolder(UpgradedHolder&& other) noexcept : lock_(other.lock_) {
@@ -388,13 +415,23 @@ class RWSpinLock {
     UpgradedHolder(const UpgradedHolder& other) = delete;
     UpgradedHolder& operator =(const UpgradedHolder& other) = delete;
 
-    ~UpgradedHolder() { if (lock_) lock_->unlock_upgrade(); }
+    ~UpgradedHolder() {
+      if (lock_) {
+        lock_->unlock_upgrade();
+      }
+    }
 
     void reset(RWSpinLock* lock = nullptr) {
-      if (lock == lock_) return;
-      if (lock_) lock_->unlock_upgrade();
+      if (lock == lock_) {
+        return;
+      }
+      if (lock_) {
+        lock_->unlock_upgrade();
+      }
       lock_ = lock;
-      if (lock_) lock_->lock_upgrade();
+      if (lock_) {
+        lock_->lock_upgrade();
+      }
     }
 
     void swap(UpgradedHolder* other) {
@@ -410,8 +447,10 @@ class RWSpinLock {
 
   class WriteHolder {
    public:
-    explicit WriteHolder(RWSpinLock* lock = nullptr) : lock_(lock) {
-      if (lock_) lock_->lock();
+    explicit WriteHolder(RWSpinLock* lock) : lock_(lock) {
+      if (lock_) {
+        lock_->lock();
+      }
     }
 
     explicit WriteHolder(RWSpinLock& lock) : lock_(&lock) {
@@ -422,7 +461,9 @@ class RWSpinLock {
     explicit WriteHolder(UpgradedHolder&& upgraded) {
       lock_ = upgraded.lock_;
       upgraded.lock_ = nullptr;
-      if (lock_) lock_->unlock_upgrade_and_lock();
+      if (lock_) {
+        lock_->unlock_upgrade_and_lock();
+      }
     }
 
     WriteHolder(WriteHolder&& other) noexcept : lock_(other.lock_) {
@@ -438,13 +479,23 @@ class RWSpinLock {
     WriteHolder(const WriteHolder& other) = delete;
     WriteHolder& operator =(const WriteHolder& other) = delete;
 
-    ~WriteHolder () { if (lock_) lock_->unlock(); }
+    ~WriteHolder() {
+      if (lock_) {
+        lock_->unlock();
+      }
+    }
 
     void reset(RWSpinLock* lock = nullptr) {
-      if (lock == lock_) return;
-      if (lock_) lock_->unlock();
+      if (lock == lock_) {
+        return;
+      }
+      if (lock_) {
+        lock_->unlock();
+      }
       lock_ = lock;
-      if (lock_) lock_->lock();
+      if (lock_) {
+        lock_->lock();
+      }
     }
 
     void swap(WriteHolder* other) {
@@ -458,12 +509,6 @@ class RWSpinLock {
     RWSpinLock* lock_;
   };
 
-  // Synchronized<> adaptors
-  friend void acquireRead(RWSpinLock& l) { return l.lock_shared(); }
-  friend void acquireReadWrite(RWSpinLock& l) { return l.lock(); }
-  friend void releaseRead(RWSpinLock& l) { return l.unlock_shared(); }
-  friend void releaseReadWrite(RWSpinLock& l) { return l.unlock(); }
-
  private:
   std::atomic<int32_t> bits_;
 };
@@ -486,13 +531,14 @@ struct RWTicketIntTrait<64> {
 
 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
   static __m128i make128(const uint16_t v[4]) {
-    return _mm_set_epi16(0, 0, 0, 0, v[3], v[2], v[1], v[0]);
+    return _mm_set_epi16(0, 0, 0, 0,
+        short(v[3]), short(v[2]), short(v[1]), short(v[0]));
   }
   static inline __m128i fromInteger(uint64_t from) {
-    return _mm_cvtsi64_si128(from);
+    return _mm_cvtsi64_si128(int64_t(from));
   }
   static inline uint64_t toInteger(__m128i in) {
-    return _mm_cvtsi128_si64(in);
+    return uint64_t(_mm_cvtsi128_si64(in));
   }
   static inline uint64_t addParallel(__m128i in, __m128i kDelta) {
     return toInteger(_mm_add_epi16(in, kDelta));
@@ -508,24 +554,26 @@ struct RWTicketIntTrait<32> {
 
 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
   static __m128i make128(const uint8_t v[4]) {
-    return _mm_set_epi8(0, 0, 0, 0, 0, 0, 0, 0,
-        0, 0, 0, 0, v[3], v[2], v[1], v[0]);
+    return _mm_set_epi8(
+        0, 0, 0, 0,
+        0, 0, 0, 0,
+        0, 0, 0, 0,
+        char(v[3]), char(v[2]), char(v[1]), char(v[0]));
   }
   static inline __m128i fromInteger(uint32_t from) {
-    return _mm_cvtsi32_si128(from);
+    return _mm_cvtsi32_si128(int32_t(from));
   }
   static inline uint32_t toInteger(__m128i in) {
-    return _mm_cvtsi128_si32(in);
+    return uint32_t(_mm_cvtsi128_si32(in));
   }
   static inline uint32_t addParallel(__m128i in, __m128i kDelta) {
     return toInteger(_mm_add_epi8(in, kDelta));
   }
 #endif
 };
-}  // detail
-
+} // namespace detail
 
-template<size_t kBitWidth, bool kFavorWriter=false>
+template <size_t kBitWidth, bool kFavorWriter = false>
 class RWTicketSpinLockT {
   typedef detail::RWTicketIntTrait<kBitWidth> IntTraitType;
   typedef typename detail::RWTicketIntTrait<kBitWidth>::FullInt FullInt;
@@ -545,13 +593,13 @@ class RWTicketSpinLockT {
   } ticket;
 
  private: // Some x64-specific utilities for atomic access to ticket.
-  template<class T> static T load_acquire(T* addr) {
+  template <class T> static T load_acquire(T* addr) {
     T t = *addr; // acquire barrier
     asm_volatile_memory();
     return t;
   }
 
-  template<class T>
+  template <class T>
   static void store_release(T* addr, T v) {
     asm_volatile_memory();
     *addr = v; // release barrier
@@ -592,7 +640,9 @@ class RWTicketSpinLockT {
   bool try_lock() {
     RWTicket t;
     FullInt old = t.whole = load_acquire(&ticket.whole);
-    if (t.users != t.write) return false;
+    if (t.users != t.write) {
+      return false;
+    }
     ++t.users;
     return __sync_bool_compare_and_swap(&ticket.whole, old, t.whole);
   }
@@ -604,16 +654,18 @@ class RWTicketSpinLockT {
    * turns.
    */
   void writeLockAggressive() {
-    // sched_yield() is needed here to avoid a pathology if the number
+    // std::this_thread::yield() is needed here to avoid a pathology if the number
     // of threads attempting concurrent writes is >= the number of real
     // cores allocated to this process. This is less likely than the
     // corresponding situation in lock_shared(), but we still want to
     // avoid it
-    int count = 0;
+    uint_fast32_t count = 0;
     QuarterInt val = __sync_fetch_and_add(&ticket.users, 1);
     while (val != load_acquire(&ticket.write)) {
       asm_volatile_pause();
-      if (UNLIKELY(++count > 1000)) sched_yield();
+      if (UNLIKELY(++count > 1000)) {
+        std::this_thread::yield();
+      }
     }
   }
 
@@ -626,7 +678,7 @@ class RWTicketSpinLockT {
     // there are a lot of competing readers.  The aggressive spinning
     // can help to avoid starving writers.
     //
-    // We don't worry about sched_yield() here because the caller
+    // We don't worry about std::this_thread::yield() here because the caller
     // has already explicitly abandoned fairness.
     while (!try_lock()) {}
   }
@@ -640,9 +692,9 @@ class RWTicketSpinLockT {
   void unlock() {
     RWTicket t;
     t.whole = load_acquire(&ticket.whole);
-    FullInt old = t.whole;
 
 #ifdef RW_SPINLOCK_USE_SSE_INSTRUCTIONS_
+    FullInt old = t.whole;
     // SSE2 can reduce the lock and unlock overhead by 10%
     static const QuarterInt kDeltaBuf[4] = { 1, 1, 0, 0 };   // write/read/user
     static const __m128i kDelta = IntTraitType::make128(kDeltaBuf);
@@ -656,13 +708,15 @@ class RWTicketSpinLockT {
   }
 
   void lock_shared() {
-    // sched_yield() is important here because we can't grab the
+    // std::this_thread::yield() is important here because we can't grab the
     // shared lock if there is a pending writeLockAggressive, so we
     // need to let threads that already have a shared lock complete
-    int count = 0;
+    uint_fast32_t count = 0;
     while (!LIKELY(try_lock_shared())) {
       asm_volatile_pause();
-      if (UNLIKELY((++count & 1023) == 0)) sched_yield();
+      if (UNLIKELY((++count & 1023) == 0)) {
+        std::this_thread::yield();
+      }
     }
   }
 
@@ -684,7 +738,7 @@ class RWTicketSpinLockT {
   }
 
   void unlock_shared() {
-    QuarterInt val = __sync_fetch_and_add(&ticket.write, 1);
+    __sync_fetch_and_add(&ticket.write, 1);
   }
 
   class WriteHolder;
@@ -695,13 +749,16 @@ class RWTicketSpinLockT {
     ReadHolder(ReadHolder const&) = delete;
     ReadHolder& operator=(ReadHolder const&) = delete;
 
-    explicit ReadHolder(RWSpinLock *lock = nullptr) :
-      lock_(lock) {
-      if (lock_) lock_->lock_shared();
+    explicit ReadHolder(RWSpinLock* lock) : lock_(lock) {
+      if (lock_) {
+        lock_->lock_shared();
+      }
     }
 
     explicit ReadHolder(RWSpinLock &lock) : lock_ (&lock) {
-      if (lock_) lock_->lock_shared();
+      if (lock_) {
+        lock_->lock_shared();
+      }
     }
 
     // atomically unlock the write-lock from writer and acquire the read-lock
@@ -713,13 +770,19 @@ class RWTicketSpinLockT {
     }
 
     ~ReadHolder() {
-      if (lock_) lock_->unlock_shared();
+      if (lock_) {
+        lock_->unlock_shared();
+      }
     }
 
     void reset(RWSpinLock *lock = nullptr) {
-      if (lock_) lock_->unlock_shared();
+      if (lock_) {
+        lock_->unlock_shared();
+      }
       lock_ = lock;
-      if (lock_) lock_->lock_shared();
+      if (lock_) {
+        lock_->lock_shared();
+      }
     }
 
     void swap(ReadHolder *other) {
@@ -735,22 +798,34 @@ class RWTicketSpinLockT {
     WriteHolder(WriteHolder const&) = delete;
     WriteHolder& operator=(WriteHolder const&) = delete;
 
-    explicit WriteHolder(RWSpinLock *lock = nullptr) : lock_(lock) {
-      if (lock_) lock_->lock();
+    explicit WriteHolder(RWSpinLock* lock) : lock_(lock) {
+      if (lock_) {
+        lock_->lock();
+      }
     }
     explicit WriteHolder(RWSpinLock &lock) : lock_ (&lock) {
-      if (lock_) lock_->lock();
+      if (lock_) {
+        lock_->lock();
+      }
     }
 
     ~WriteHolder() {
-      if (lock_) lock_->unlock();
+      if (lock_) {
+        lock_->unlock();
+      }
     }
 
     void reset(RWSpinLock *lock = nullptr) {
-      if (lock == lock_) return;
-      if (lock_) lock_->unlock();
+      if (lock == lock_) {
+        return;
+      }
+      if (lock_) {
+        lock_->unlock();
+      }
       lock_ = lock;
-      if (lock_) lock_->lock();
+      if (lock_) {
+        lock_->lock();
+      }
     }
 
     void swap(WriteHolder *other) {
@@ -761,20 +836,6 @@ class RWTicketSpinLockT {
     friend class ReadHolder;
     RWSpinLock *lock_;
   };
-
-  // Synchronized<> adaptors.
-  friend void acquireRead(RWTicketSpinLockT& mutex) {
-    mutex.lock_shared();
-  }
-  friend void acquireReadWrite(RWTicketSpinLockT& mutex) {
-    mutex.lock();
-  }
-  friend void releaseRead(RWTicketSpinLockT& mutex) {
-    mutex.unlock_shared();
-  }
-  friend void releaseReadWrite(RWTicketSpinLockT& mutex) {
-    mutex.unlock();
-  }
 };
 
 typedef RWTicketSpinLockT<32> RWTicketSpinLock32;
@@ -782,10 +843,8 @@ typedef RWTicketSpinLockT<64> RWTicketSpinLock64;
 
 #endif  // RW_SPINLOCK_USE_X86_INTRINSIC_
 
-}  // namespace folly
+} // namespace folly
 
 #ifdef RW_SPINLOCK_USE_X86_INTRINSIC_
 #undef RW_SPINLOCK_USE_X86_INTRINSIC_
 #endif
-
-#endif  // FOLLY_RWSPINLOCK_H_