From c139abc5098bfe355f897d9b8bb4d1026cd4d3bd Mon Sep 17 00:00:00 2001
From: Alex Guteniev <gutenev@gmail.com>
Date: Tue, 4 Aug 2020 22:11:37 +0300
Subject: [PATCH] Use _mm_pause, load loop, and SRLWOCK

Resolves #370 , resolves #680
---
 stl/inc/atomic     |  6 ++--
 stl/inc/xatomic.h  |  6 ++++
 stl/src/atomic.cpp | 90 ++++++++++++++++++++++++++++++++++++----------
 3 files changed, 82 insertions(+), 20 deletions(-)

diff --git a/stl/inc/atomic b/stl/inc/atomic
index 633280f5d5..1a57fe4416 100644
--- a/stl/inc/atomic
+++ b/stl/inc/atomic
@@ -392,8 +392,10 @@ void _Atomic_wait_direct(
 
 #if 1 // TRANSITION, ABI
 inline void _Atomic_lock_spinlock(long& _Spinlock) noexcept {
-    while (_InterlockedExchange(&_Spinlock, 1)) {
-        _YIELD_PROCESSOR();
+    while (_InterlockedExchange(&_Spinlock, 1) != 0) {
+        while (__iso_volatile_load32(&reinterpret_cast<int&>(_Spinlock)) != 0) {
+            _YIELD_PROCESSOR();
+        }
     }
 }
 
diff --git a/stl/inc/xatomic.h b/stl/inc/xatomic.h
index 8498f18c1a..a607b545c0 100644
--- a/stl/inc/xatomic.h
+++ b/stl/inc/xatomic.h
@@ -28,7 +28,13 @@ _STL_DISABLE_CLANG_WARNINGS
 #define _INTRIN_ACQUIRE(x) x
 #define _INTRIN_RELEASE(x) x
 #define _INTRIN_ACQ_REL(x) x
+#ifdef _M_CEE_PURE
 #define _YIELD_PROCESSOR()
+#else //  ^^^ _M_CEE_PURE / !_M_CEE_PURE vvv
+extern "C" __MACHINEX86_X64(void _mm_pause(void))
+#define _YIELD_PROCESSOR() _mm_pause()
+#endif // ^^^ !_M_CEE_PURE ^^^
+
 
 #elif defined(_M_ARM) || defined(_M_ARM64)
 #define _INTRIN_RELAXED(x) _CONCAT(x, _nf)
diff --git a/stl/src/atomic.cpp b/stl/src/atomic.cpp
index b2c0eeb00b..a9c54cb8a4 100644
--- a/stl/src/atomic.cpp
+++ b/stl/src/atomic.cpp
@@ -4,33 +4,87 @@
 // implement shared_ptr spin lock
 
 #include <yvals.h>
-
+#include <atomic>
 #include <intrin.h>
-#pragma warning(disable : 4793)
+#include <libloaderapi.h>
+#include <synchapi.h>
 
-_EXTERN_C
+#pragma warning(disable : 4793) // '%s' : function is compiled as native code
 
-// SPIN LOCK FOR shared_ptr ATOMIC OPERATIONS
-volatile long _Shared_ptr_flag;
+namespace {
+    // MUTEX FOR shared_ptr ATOMIC OPERATIONS
+    SRWLOCK _Shared_ptr_lock = SRWLOCK_INIT;
 
-_CRTIMP2_PURE void __cdecl _Lock_shared_ptr_spin_lock() { // spin until _Shared_ptr_flag successfully set
-#ifdef _M_ARM
-    while (_InterlockedExchange_acq(&_Shared_ptr_flag, 1)) {
-        __yield();
+#if _STL_WIN32_WINNT < _STL_WIN32_WINNT_VISTA
+    // SPIN LOCK FOR shared_ptr ATOMIC OPERATIONS
+    long _Shared_ptr_flag;
+
+
+    struct _Srw_functions_table {
+        _STD atomic<decltype(&::AcquireSRWLockExclusive)> _Pfn_AcquireSRWLockExclusive{nullptr};
+        _STD atomic<decltype(&::ReleaseSRWLockExclusive)> _Pfn_ReleaseSRWLockExclusive{nullptr};
+    };
+
+    _Srw_functions_table _Table;
+
+    bool _Acuire_srw_functions() {
+        enum class _Shared_ptr_api_level : char {
+            __has_nothing = false,
+            __has_srwlock = true,
+            __not_set,
+        };
+        static _STD atomic<_Shared_ptr_api_level> _Api_level{_Shared_ptr_api_level::__not_set};
+
+        _Shared_ptr_api_level _Current = _Api_level.load(_STD memory_order_acquire);
+        if (_Current == _Shared_ptr_api_level::__not_set) {
+            _Current          = _Shared_ptr_api_level::__has_nothing;
+            HMODULE _Kernel32 = GetModuleHandleW(L"Kernel32.dll");
+            if (_Kernel32 != nullptr) {
+                auto _Pfn_AcquireSRWLockExclusive = reinterpret_cast<decltype(&::AcquireSRWLockExclusive)>(
+                    GetProcAddress(_Kernel32, "AcquireSRWLockExclusive"));
+                auto _Pfn_ReleaseSRWLockExclusive = reinterpret_cast<decltype(&::ReleaseSRWLockExclusive)>(
+                    GetProcAddress(_Kernel32, "ReleaseSRWLockExclusive"));
+                if (_Pfn_AcquireSRWLockExclusive != nullptr && _Pfn_ReleaseSRWLockExclusive != nullptr) {
+                    _Table._Pfn_AcquireSRWLockExclusive.store(_Pfn_AcquireSRWLockExclusive, _STD memory_order_relaxed);
+                    _Table._Pfn_ReleaseSRWLockExclusive.store(_Pfn_ReleaseSRWLockExclusive, _STD memory_order_relaxed);
+                    _Current = _Shared_ptr_api_level::__has_srwlock;
+                }
+            }
+            _Api_level.store(_Current, _STD memory_order_release);
+        }
+        return reinterpret_cast<bool&>(_Current);
     }
-#else // _M_ARM
-    while (_interlockedbittestandset(&_Shared_ptr_flag, 0)) { // set bit 0
+#endif // ^^^ _STL_WIN32_WINNT < _STL_WIN32_WINNT_VISTA ^^^
+} // unnamed namespace
+
+_EXTERN_C
+
+_CRTIMP2_PURE void __cdecl _Lock_shared_ptr_spin_lock() { // spin until _Shared_ptr_flag successfully set
+#if _STL_WIN32_WINNT >= _STL_WIN32_WINNT_VISTA
+    AcquireSRWLockExclusive(_Shared_ptr_lock);
+#else // ^^^ _STL_WIN32_WINNT >= _STL_WIN32_WINNT_VISTA / _STL_WIN32_WINNT < _STL_WIN32_WINNT_VISTA vvv
+    if (_Acuire_srw_functions()) {
+        _Table._Pfn_AcquireSRWLockExclusive.load(_STD memory_order_relaxed)(&_Shared_ptr_lock);
+    } else {
+        while (_interlockedbittestandset(&_Shared_ptr_flag, 0) != 0) { // set bit 0
+            while (__iso_volatile_load32(reinterpret_cast<int*>(&_Shared_ptr_flag)) != 0) {
+                YieldProcessor();
+            }
+        }
     }
-#endif // _M_ARM
+#endif // ^^^ _STL_WIN32_WINNT < _STL_WIN32_WINNT_VISTA ^^^
 }
 
 _CRTIMP2_PURE void __cdecl _Unlock_shared_ptr_spin_lock() { // release previously obtained lock
-#ifdef _M_ARM
-    __dmb(_ARM_BARRIER_ISH);
-    __iso_volatile_store32(reinterpret_cast<volatile int*>(&_Shared_ptr_flag), 0);
-#else // _M_ARM
-    _interlockedbittestandreset(&_Shared_ptr_flag, 0); // reset bit 0
-#endif // _M_ARM
+#if _STL_WIN32_WINNT >= _STL_WIN32_WINNT_VISTA
+    ReleaseSRWLockExclusive(&_Shared_ptr_lock);
+#else // ^^^ _STL_WIN32_WINNT >= _STL_WIN32_WINNT_VISTA / _STL_WIN32_WINNT < _STL_WIN32_WINNT_VISTA vvv
+    if (_Acuire_srw_functions()) {
+        _Table._Pfn_ReleaseSRWLockExclusive.load(_STD memory_order_relaxed)(&_Shared_ptr_lock);
+    } else {
+        _interlockedbittestandreset(&_Shared_ptr_flag, 0); // reset bit 0
+    }
+#endif // ^^^ _STL_WIN32_WINNT < _STL_WIN32_WINNT_VISTA ^^^
 }
 
 _END_EXTERN_C