Optimizations for Windows

- Set opt_lg_chunk based on run-time OS setting
- Verify LG_PAGE is compatible with run-time OS setting
- When targeting Windows Vista or newer, use SRWLOCK instead of CRITICAL_SECTION
- When targeting Windows Vista or newer, statically initialize init_lock

include/jemalloc/internal/mutex.h

@@ -26,7 +26,11 @@ typedef struct malloc_mutex_s malloc_mutex_t;
 
 struct malloc_mutex_s {
 #ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+	SRWLOCK         	lock;
+#  else
 	CRITICAL_SECTION	lock;
+#  endif
 #elif (defined(JEMALLOC_OSSPIN))
 	OSSpinLock		lock;
 #elif (defined(JEMALLOC_MUTEX_INIT_CB))
@@ -70,7 +74,11 @@ malloc_mutex_lock(malloc_mutex_t *mutex)
 
 	if (isthreaded) {
 #ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+		AcquireSRWLockExclusive(&mutex->lock);
+#  else
 		EnterCriticalSection(&mutex->lock);
+#  endif
 #elif (defined(JEMALLOC_OSSPIN))
 		OSSpinLockLock(&mutex->lock);
 #else
@@ -85,7 +93,11 @@ malloc_mutex_unlock(malloc_mutex_t *mutex)
 
 	if (isthreaded) {
 #ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+		ReleaseSRWLockExclusive(&mutex->lock);
+#  else
 		LeaveCriticalSection(&mutex->lock);
+#  endif
 #elif (defined(JEMALLOC_OSSPIN))
 		OSSpinLockUnlock(&mutex->lock);
 #else

src/chunk.c

@@ -5,7 +5,7 @@
 /* Data. */
 
 const char	*opt_dss = DSS_DEFAULT;
-size_t		opt_lg_chunk = LG_CHUNK_DEFAULT;
+size_t		opt_lg_chunk = 0;
 
 /* Used exclusively for gdump triggering. */
 static size_t	curchunks;
@@ -535,6 +535,21 @@ chunks_rtree_node_alloc(size_t nelms)
 bool
 chunk_boot(void)
 {
+#ifdef _WIN32
+	SYSTEM_INFO info;
+	GetSystemInfo(&info);
+
+	/* Verify actual page size is equal to or an integral multiple of configured page size */
+	if (info.dwPageSize & ((1U << LG_PAGE) - 1))
+		return (true);
+
+	/* Configure chunksize (if not set) to match granularity (usually 64K), so pages_map will always take fast path */
+	if (!opt_lg_chunk)
+		opt_lg_chunk = ffs((int)info.dwAllocationGranularity) - 1;
+#else
+	if (!opt_lg_chunk)
+		opt_lg_chunk = LG_CHUNK_DEFAULT;
+#endif
 
 	/* Set variables according to the value of opt_lg_chunk. */
 	chunksize = (ZU(1) << opt_lg_chunk);

src/jemalloc.c

@@ -175,6 +175,9 @@ static bool			malloc_initializer = NO_INITIALIZER;
 
 /* Used to avoid initialization races. */
 #ifdef _WIN32
+#if _WIN32_WINNT >= 0x0600
+static malloc_mutex_t	init_lock = SRWLOCK_INIT;
+#else
 static malloc_mutex_t	init_lock;
 
 JEMALLOC_ATTR(constructor)
@@ -190,7 +193,7 @@ _init_init_lock(void)
 JEMALLOC_SECTION(".CRT$XCU") JEMALLOC_ATTR(used)
 static const void (WINAPI *init_init_lock)(void) = _init_init_lock;
 #endif
-
+#endif
 #else
 static malloc_mutex_t	init_lock = MALLOC_MUTEX_INITIALIZER;
 #endif

src/mutex.c

@@ -73,9 +73,13 @@ malloc_mutex_init(malloc_mutex_t *mutex)
 {
 
 #ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+	InitializeSRWLock(&mutex->lock);
+#  else
 	if (!InitializeCriticalSectionAndSpinCount(&mutex->lock,
 	    _CRT_SPINCOUNT))
 		return (true);
+#  endif
 #elif (defined(JEMALLOC_OSSPIN))
 	mutex->lock = 0;
 #elif (defined(JEMALLOC_MUTEX_INIT_CB))