A follow-up to the patch for bug #56405 "Deadlock in the MDL deadlock

detector". This patch addresses performance regression in OLTP_RO/MyISAM
test on Windows introduced by the fix for bug #56405. Thus it makes
original patch acceptable as a solution for bug #56585 "Slowdown of
readonly sysbench benchmarks (e.g point_select) on Windows 5.5".

With this patch, MySQL will use native Windows condition variables and 
reader-writer locks  if  they are supported by the OS.

This speeds up MyISAM and the effect comes mostly from using native
rwlocks. Native conditions improve scalability with higher number of 
concurrent users in other situations, e.g for prlocks.

Benchmark numbers for this patch as measured on Win2008R2 quad
core machine are attached to the bug report.
( direct link http://bugs.mysql.com/file.php?id=15883 )

Note that currently we require at least Windows7/WS2008R2 for 
reader-writer locks, even though native rwlock is available also on Vista.
Reason is that "trylock" APIs are missing on Vista, and trylock is used in
the server (in a single place in query cache).

While this patch could have been written differently, to enable the native
rwlock optimization also on Vista/WS2008 (e.g using native locks everywhere
but portable implementation in query cache), this would come at the 
expense of the code clarity, as it would introduce a new  "try-able" rwlock
type, to handle Vista case.

Another way to improve performance for the special case 
(OLTP_RO/MYISAM/Vista) would be to eliminate "trylock" usage from server,
 but this is outside of the scope here.


Native conditions variables are used beginning with Vista though the effect
of using condition variables alone is not measurable in this benchmark.
But when used together with native rwlocks on Win7, native conditions improve 
performance in high-concurrency OLTP_RO/MyISAM (128 and more sysbench 
users).

1 files changed, 190 insertions, 42 deletions

diff --git a/mysys/my_wincond.c b/mysys/my_wincond.c
index ad1636011db..58c09e332d6 100644
--- a/mysys/my_wincond.c
+++ b/mysys/my_wincond.c
@@ -24,7 +24,108 @@
 #include <process.h>
 #include <sys/timeb.h>
 
-int pthread_cond_init(pthread_cond_t *cond, const pthread_condattr_t *attr)
+
+/*
+  Windows native condition variables. We use runtime loading / function 
+  pointers, because they are not available on XP 
+*/
+
+/* Prototypes and function pointers for condition variable functions */
+typedef VOID (WINAPI * InitializeConditionVariableProc) 
+  (PCONDITION_VARIABLE ConditionVariable);
+
+typedef BOOL (WINAPI * SleepConditionVariableCSProc)
+  (PCONDITION_VARIABLE ConditionVariable,
+  PCRITICAL_SECTION CriticalSection, 
+  DWORD dwMilliseconds);
+
+typedef VOID (WINAPI * WakeAllConditionVariableProc)
+ (PCONDITION_VARIABLE ConditionVariable);
+
+typedef VOID (WINAPI * WakeConditionVariableProc)
+  (PCONDITION_VARIABLE ConditionVariable);
+
+static InitializeConditionVariableProc my_InitializeConditionVariable;
+static SleepConditionVariableCSProc my_SleepConditionVariableCS;
+static WakeAllConditionVariableProc my_WakeAllConditionVariable;
+static WakeConditionVariableProc my_WakeConditionVariable;
+
+
+/**
+ Indicates if we have native condition variables,
+ initialized first time pthread_cond_init is called.
+*/
+
+static BOOL have_native_conditions= FALSE;
+
+
+/**
+  Check if native conditions can be used, load function pointers 
+*/
+
+static void check_native_cond_availability(void)
+{
+  HMODULE module= GetModuleHandle("kernel32");
+
+  my_InitializeConditionVariable= (InitializeConditionVariableProc)
+    GetProcAddress(module, "InitializeConditionVariable");
+  my_SleepConditionVariableCS= (SleepConditionVariableCSProc)
+    GetProcAddress(module, "SleepConditionVariableCS");
+  my_WakeAllConditionVariable= (WakeAllConditionVariableProc)
+    GetProcAddress(module, "WakeAllConditionVariable");
+  my_WakeConditionVariable= (WakeConditionVariableProc)
+    GetProcAddress(module, "WakeConditionVariable");
+
+  if (my_InitializeConditionVariable)
+    have_native_conditions= TRUE;
+}
+
+
+
+/**
+  Convert abstime to milliseconds
+*/
+
+static DWORD get_milliseconds(const struct timespec *abstime)
+{
+  long long millis; 
+  union ft64 now;
+
+  if (abstime == NULL)
+   return INFINITE;
+
+  GetSystemTimeAsFileTime(&now.ft);
+
+  /*
+    Calculate time left to abstime
+    - subtract start time from current time(values are in 100ns units)
+    - convert to millisec by dividing with 10000
+  */
+  millis= (abstime->tv.i64 - now.i64) / 10000;
+  
+  /* Don't allow the timeout to be negative */
+  if (millis < 0)
+    return 0;
+
+  /*
+    Make sure the calculated timeout does not exceed original timeout
+    value which could cause "wait for ever" if system time changes
+  */
+  if (millis > abstime->max_timeout_msec)
+    millis= abstime->max_timeout_msec;
+  
+  if (millis > UINT_MAX)
+    millis= UINT_MAX;
+
+  return (DWORD)millis;
+}
+
+
+/*
+  Old (pre-vista) implementation using events
+*/
+
+static int legacy_cond_init(pthread_cond_t *cond, const pthread_condattr_t *attr)
 {
   cond->waiting= 0;
   InitializeCriticalSection(&cond->lock_waiting);
@@ -53,7 +154,8 @@ int pthread_cond_init(pthread_cond_t *cond, const pthread_condattr_t *attr)
   return 0;
 }
 
-int pthread_cond_destroy(pthread_cond_t *cond)
+
+static int legacy_cond_destroy(pthread_cond_t *cond)
 {
   DeleteCriticalSection(&cond->lock_waiting);
 
@@ -65,48 +167,13 @@ int pthread_cond_destroy(pthread_cond_t *cond)
 }
 
 
-int pthread_cond_wait(pthread_cond_t *cond, pthread_mutex_t *mutex)
-{
-  return pthread_cond_timedwait(cond,mutex,NULL);
-}
-
-
-int pthread_cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex,
+static int legacy_cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex,
                            struct timespec *abstime)
 {
   int result;
-  long timeout; 
-  union ft64 now;
-
-  if( abstime != NULL )
-  {
-    GetSystemTimeAsFileTime(&now.ft);
-
-    /*
-      Calculate time left to abstime
-      - subtract start time from current time(values are in 100ns units)
-      - convert to millisec by dividing with 10000
-    */
-    timeout= (long)((abstime->tv.i64 - now.i64) / 10000);
-    
-    /* Don't allow the timeout to be negative */
-    if (timeout < 0)
-      timeout= 0L;
-
-    /*
-      Make sure the calucated timeout does not exceed original timeout
-      value which could cause "wait for ever" if system time changes
-    */
-    if (timeout > abstime->max_timeout_msec)
-      timeout= abstime->max_timeout_msec;
-
-  }
-  else
-  {
-    /* No time specified; don't expire */
-    timeout= INFINITE;
-  }
+  DWORD timeout; 
 
+  timeout= get_milliseconds(abstime);
   /* 
     Block access if previous broadcast hasn't finished.
     This is just for safety and should normally not
@@ -142,7 +209,7 @@ int pthread_cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex,
   return result == WAIT_TIMEOUT ? ETIMEDOUT : 0;
 }
 
-int pthread_cond_signal(pthread_cond_t *cond)
+static int legacy_cond_signal(pthread_cond_t *cond)
 {
   EnterCriticalSection(&cond->lock_waiting);
   
@@ -155,7 +222,7 @@ int pthread_cond_signal(pthread_cond_t *cond)
 }
 
 
-int pthread_cond_broadcast(pthread_cond_t *cond)
+static int legacy_cond_broadcast(pthread_cond_t *cond)
 {
   EnterCriticalSection(&cond->lock_waiting);
   /*
@@ -177,6 +244,87 @@ int pthread_cond_broadcast(pthread_cond_t *cond)
 }
 
 
+/* 
+ Posix API functions. Just choose between native and legacy implementation.
+*/
+
+int pthread_cond_init(pthread_cond_t *cond, const pthread_condattr_t *attr)
+{
+  /*
+    Once initialization is used here rather than in my_init(), to 
+    1) avoid  my_init() pitfalls- undefined order in which initialization should
+    run
+    2) be potentially useful C++ (in static constructors that run before main())
+    3) just to simplify the API.
+    Also, the overhead of my_pthread_once is very small.
+  */
+  static my_pthread_once_t once_control= MY_PTHREAD_ONCE_INIT;
+  my_pthread_once(&once_control, check_native_cond_availability);
+
+  if (have_native_conditions)
+  {
+    my_InitializeConditionVariable(&cond->native_cond);
+    return 0;
+  }
+  else 
+    return legacy_cond_init(cond, attr);
+}
+
+
+int pthread_cond_destroy(pthread_cond_t *cond)
+{
+  if (have_native_conditions)
+    return 0; /* no destroy function */
+  else
+    return legacy_cond_destroy(cond);
+}
+
+
+int pthread_cond_broadcast(pthread_cond_t *cond)
+{
+  if (have_native_conditions)
+  {
+    my_WakeAllConditionVariable(&cond->native_cond);
+    return 0;
+  }
+  else
+    return legacy_cond_broadcast(cond);
+}
+
+
+int pthread_cond_signal(pthread_cond_t *cond)
+{
+  if (have_native_conditions)
+  {
+    my_WakeConditionVariable(&cond->native_cond);
+    return 0;
+  }
+  else
+    return legacy_cond_signal(cond);
+}
+
+
+int pthread_cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex,
+  struct timespec *abstime)
+{
+  if (have_native_conditions)
+  {
+    DWORD timeout= get_milliseconds(abstime);
+    if (!my_SleepConditionVariableCS(&cond->native_cond, mutex, timeout))
+      return ETIMEDOUT;
+    return 0;  
+  }
+  else
+    return legacy_cond_timedwait(cond, mutex, abstime);
+}
+
+
+int pthread_cond_wait(pthread_cond_t *cond, pthread_mutex_t *mutex)
+{
+  return pthread_cond_timedwait(cond, mutex, NULL);
+}
+
+
 int pthread_attr_init(pthread_attr_t *connect_att)
 {
   connect_att->dwStackSize	= 0;