Merge bk-internal.mysql.com:/home/bk/mysql-maria

into  janus.mylan:/usr/home/serg/Abk/mysql-maria


configure.in:
  Auto merged
include/my_global.h:
  Auto merged
include/my_sys.h:
  Auto merged
mysys/Makefile.am:
  Auto merged
sql/item_func.cc:
  Auto merged
sql/mysql_priv.h:
  Auto merged
sql/mysqld.cc:
  Auto merged
sql/set_var.cc:
  Auto merged
sql/sql_parse.cc:
  Auto merged
sql/sql_select.cc:
  Auto merged
sql/sql_test.cc:
  Auto merged
storage/maria/Makefile.am:
  Auto merged
storage/maria/ha_maria.cc:
  Auto merged
storage/myisam/ha_myisam.cc:
  Auto merged
unittest/Makefile.am:
  Auto merged
unittest/mysys/my_atomic-t.c:
  merged

10 files changed, 416 insertions, 49 deletions

diff --git a/include/Makefile.am b/include/Makefile.am
index bf7dcd265be..d5e75d31593 100644
--- a/include/Makefile.am
+++ b/include/Makefile.am
@@ -23,7 +23,7 @@ pkginclude_HEADERS =	my_dbug.h m_string.h my_sys.h my_list.h my_xml.h \
 			my_getopt.h sslopt-longopts.h my_dir.h typelib.h \
 			sslopt-vars.h sslopt-case.h sql_common.h keycache.h \
 			mysql_time.h mysql/plugin.h $(BUILT_SOURCES)
-noinst_HEADERS =	config-win.h config-netware.h \
+noinst_HEADERS =	config-win.h config-netware.h lf.h \
 			heap.h maria.h myisamchk.h my_bitmap.h my_uctype.h \
 			myisam.h myisampack.h myisammrg.h ft_global.h\
 			mysys_err.h my_base.h help_start.h help_end.h \
diff --git a/include/atomic/nolock.h b/include/atomic/nolock.h
index 1151a334b06..a696e008f03 100644
--- a/include/atomic/nolock.h
+++ b/include/atomic/nolock.h
@@ -14,19 +14,20 @@
    along with this program; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
 
-#if defined(__i386__) || defined(_M_IX86)
-
-#ifdef MY_ATOMIC_MODE_DUMMY
-#  define LOCK ""
-#else
-#  define LOCK "lock"
-#endif
-
-#ifdef __GNUC__
-#include "x86-gcc.h"
-#elif defined(_MSC_VER)
-#include "x86-msvc.h"
-#endif
+#if defined(__i386__) || defined(_M_IX86) || defined(__x86_64__)
+
+#  ifdef MY_ATOMIC_MODE_DUMMY
+#    define LOCK_prefix ""
+#  else
+#    define LOCK_prefix "lock"
+#  endif
+
+#  ifdef __GNUC__
+#    include "x86-gcc.h"
+#  elif defined(_MSC_VER)
+#    error Broken!
+#    include "x86-msvc.h"
+#  endif
 #endif
 
 #ifdef make_atomic_cas_body
diff --git a/include/atomic/rwlock.h b/include/atomic/rwlock.h
index 12d0dd3f069..3d8edb9e27e 100644
--- a/include/atomic/rwlock.h
+++ b/include/atomic/rwlock.h
@@ -14,7 +14,7 @@
    along with this program; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
 
-typedef struct {pthread_rwlock_t rw;} my_atomic_rwlock_t;
+typedef struct {pthread_mutex_t rw;} my_atomic_rwlock_t;
 
 #ifdef MY_ATOMIC_MODE_DUMMY
 /*
@@ -32,13 +32,18 @@ typedef struct {pthread_rwlock_t rw;} my_atomic_rwlock_t;
 #define my_atomic_rwlock_wrunlock(name)
 #define MY_ATOMIC_MODE "dummy (non-atomic)"
 #else
-#define my_atomic_rwlock_destroy(name)     pthread_rwlock_destroy(& (name)->rw)
-#define my_atomic_rwlock_init(name)        pthread_rwlock_init(& (name)->rw, 0)
-#define my_atomic_rwlock_rdlock(name)      pthread_rwlock_rdlock(& (name)->rw)
-#define my_atomic_rwlock_wrlock(name)      pthread_rwlock_wrlock(& (name)->rw)
-#define my_atomic_rwlock_rdunlock(name)    pthread_rwlock_unlock(& (name)->rw)
-#define my_atomic_rwlock_wrunlock(name)    pthread_rwlock_unlock(& (name)->rw)
-#define MY_ATOMIC_MODE "rwlocks"
+/*
+  we're using read-write lock macros but map them to mutex locks, and they're
+  faster. Still, having semantically rich API we can change the
+  underlying implementation, if necessary.
+*/
+#define my_atomic_rwlock_destroy(name)     pthread_mutex_destroy(& (name)->rw)
+#define my_atomic_rwlock_init(name)        pthread_mutex_init(& (name)->rw, 0)
+#define my_atomic_rwlock_rdlock(name)      pthread_mutex_lock(& (name)->rw)
+#define my_atomic_rwlock_wrlock(name)      pthread_mutex_lock(& (name)->rw)
+#define my_atomic_rwlock_rdunlock(name)    pthread_mutex_unlock(& (name)->rw)
+#define my_atomic_rwlock_wrunlock(name)    pthread_mutex_unlock(& (name)->rw)
+#define MY_ATOMIC_MODE "mutex"
 #endif
 
 #define make_atomic_add_body(S)     int ## S sav; sav= *a; *a+= v; v=sav;
diff --git a/include/atomic/x86-gcc.h b/include/atomic/x86-gcc.h
index 5a72f318a61..0be8fdf9244 100644
--- a/include/atomic/x86-gcc.h
+++ b/include/atomic/x86-gcc.h
@@ -20,10 +20,18 @@
   architectures support double-word (128-bit) cas.
 */
 
-#ifdef MY_ATOMIC_NO_XADD
-#define MY_ATOMIC_MODE "gcc-x86" LOCK "-no-xadd"
+#ifdef __x86_64__
+#  ifdef MY_ATOMIC_NO_XADD
+#    define MY_ATOMIC_MODE "gcc-amd64" LOCK_prefix "-no-xadd"
+#  else
+#    define MY_ATOMIC_MODE "gcc-amd64" LOCK_prefix
+#  endif
 #else
-#define MY_ATOMIC_MODE "gcc-x86" LOCK
+#  ifdef MY_ATOMIC_NO_XADD
+#    define MY_ATOMIC_MODE "gcc-x86" LOCK_prefix "-no-xadd"
+#  else
+#    define MY_ATOMIC_MODE "gcc-x86" LOCK_prefix
+#  endif
 #endif
 
 /* fix -ansi errors while maintaining readability */
@@ -33,12 +41,12 @@
 
 #ifndef MY_ATOMIC_NO_XADD
 #define make_atomic_add_body(S)					\
-  asm volatile (LOCK "; xadd %0, %1;" : "+r" (v) , "+m" (*a))
+  asm volatile (LOCK_prefix "; xadd %0, %1;" : "+r" (v) , "+m" (*a))
 #endif
 #define make_atomic_swap_body(S)				\
-  asm volatile ("; xchg %0, %1;" : "+r" (v) , "+m" (*a))
+  asm volatile ("xchg %0, %1;" : "+r" (v) , "+m" (*a))
 #define make_atomic_cas_body(S)					\
-  asm volatile (LOCK "; cmpxchg %3, %0; setz %2;"		\
+  asm volatile (LOCK_prefix "; cmpxchg %3, %0; setz %2;"	\
                : "+m" (*a), "+a" (*cmp), "=q" (ret): "r" (set))
 
 #ifdef MY_ATOMIC_MODE_DUMMY
@@ -47,13 +55,16 @@
 #else
 /*
   Actually 32-bit reads/writes are always atomic on x86
-  But we add LOCK here anyway to force memory barriers
+  But we add LOCK_prefix here anyway to force memory barriers
 */
 #define make_atomic_load_body(S)				\
   ret=0;							\
-  asm volatile (LOCK "; cmpxchg %2, %0"				\
+  asm volatile (LOCK_prefix "; cmpxchg %2, %0"			\
                : "+m" (*a), "+a" (ret): "r" (ret))
 #define make_atomic_store_body(S)				\
-  asm volatile ("; xchg %0, %1;" : "+m" (*a) : "r" (v))
+  asm volatile ("; xchg %0, %1;" : "+m" (*a), "+r" (v))
 #endif
 
+/* TODO test on intel whether the below helps. on AMD it makes no difference */
+//#define LF_BACKOFF ({asm volatile ("rep; nop"); 1; })
+
diff --git a/include/atomic/x86-msvc.h b/include/atomic/x86-msvc.h
index d4024a854fb..8f3e55aaed7 100644
--- a/include/atomic/x86-msvc.h
+++ b/include/atomic/x86-msvc.h
@@ -26,19 +26,19 @@
 #ifndef _atomic_h_cleanup_
 #define _atomic_h_cleanup_ "atomic/x86-msvc.h"
 
-#define MY_ATOMIC_MODE "msvc-x86" LOCK
+#define MY_ATOMIC_MODE "msvc-x86" LOCK_prefix
 
 #define make_atomic_add_body(S)				\
   _asm {						\
     _asm mov   reg_ ## S, v				\
-    _asm LOCK  xadd *a, reg_ ## S			\
+    _asm LOCK_prefix  xadd *a, reg_ ## S		\
     _asm movzx v, reg_ ## S				\
   }
 #define make_atomic_cas_body(S)				\
   _asm {						\
     _asm mov    areg_ ## S, *cmp			\
     _asm mov    reg2_ ## S, set				\
-    _asm LOCK cmpxchg *a, reg2_ ## S			\
+    _asm LOCK_prefix cmpxchg *a, reg2_ ## S		\
     _asm mov    *cmp, areg_ ## S			\
     _asm setz   al					\
     _asm movzx  ret, al					\
@@ -56,13 +56,13 @@
 #else
 /*
   Actually 32-bit reads/writes are always atomic on x86
-  But we add LOCK here anyway to force memory barriers
+  But we add LOCK_prefix here anyway to force memory barriers
 */
 #define make_atomic_load_body(S)			\
   _asm {						\
     _asm mov    areg_ ## S, 0				\
     _asm mov    reg2_ ## S, areg_ ## S			\
-    _asm LOCK cmpxchg *a, reg2_ ## S			\
+    _asm LOCK_prefix cmpxchg *a, reg2_ ## S		\
     _asm mov    ret, areg_ ## S				\
   }
 #define make_atomic_store_body(S)			\
diff --git a/include/lf.h b/include/lf.h
new file mode 100644
index 00000000000..4c6765b2d40
--- /dev/null
+++ b/include/lf.h
@@ -0,0 +1,240 @@
+
+/*
+  TODO
+    1. copyright
+    6. reduce the number of memory barriers
+*/
+
+#ifndef _lf_h
+#define _lf_h
+
+#include <my_atomic.h>
+
+/*
+  Generic helpers
+*/
+
+#define lock_wrap(f,t,proto_args, args, lock)   \
+t _ ## f proto_args;                            \
+static inline t f  proto_args                   \
+{                                               \
+  t ret;                                        \
+  my_atomic_rwlock_wrlock(lock);                \
+  ret= _ ## f args;                             \
+  my_atomic_rwlock_wrunlock(lock);              \
+  return ret;                                   \
+}
+
+#define lock_wrap_void(f,proto_args, args, lock) \
+void _ ## f proto_args;                         \
+static inline void f proto_args                 \
+{                                               \
+  my_atomic_rwlock_wrlock(lock);                \
+  _ ## f args;                                  \
+  my_atomic_rwlock_wrunlock(lock);              \
+}
+
+#define nolock_wrap(f,t,proto_args, args)       \
+t _ ## f proto_args;                            \
+static inline t f  proto_args                   \
+{                                               \
+  return _ ## f args;                           \
+}
+
+#define nolock_wrap_void(f,proto_args, args)    \
+void _ ## f proto_args;                         \
+static inline void f proto_args                 \
+{                                               \
+  _ ## f args;                                  \
+}
+
+/*
+  dynamic array
+
+  4 levels of 256 elements each mean 4311810304 elements in an array - it
+  should be enough for a while
+*/
+#define LF_DYNARRAY_LEVEL_LENGTH 256
+#define LF_DYNARRAY_LEVELS       4
+
+typedef struct {
+  void * volatile level[LF_DYNARRAY_LEVELS];
+  uint size_of_element;
+  my_atomic_rwlock_t lock;
+} LF_DYNARRAY;
+
+typedef int (*lf_dynarray_func)(void *, void *);
+
+void lf_dynarray_init(LF_DYNARRAY *array, uint element_size);
+void lf_dynarray_destroy(LF_DYNARRAY *array);
+
+nolock_wrap(lf_dynarray_nr, int,
+            (LF_DYNARRAY *array, void *el),
+            (array,el));
+
+nolock_wrap(lf_dynarray_value, void *,
+            (LF_DYNARRAY *array, uint idx),
+            (array,idx));
+
+lock_wrap(lf_dynarray_lvalue, void *,
+          (LF_DYNARRAY *array, uint idx),
+          (array,idx),
+          &array->lock);
+nolock_wrap(lf_dynarray_iterate, int,
+            (LF_DYNARRAY *array, lf_dynarray_func func, void *arg),
+            (array,func,arg));
+
+/*
+  pin manager for memory allocator
+*/
+
+#define LF_PINBOX_PINS 4
+#define LF_PURGATORY_SIZE 10
+
+typedef void lf_pinbox_free_func(void *, void *);
+
+typedef struct {
+  LF_DYNARRAY pinstack;
+  lf_pinbox_free_func *free_func;
+  void * free_func_arg;
+  uint32 volatile pinstack_top_ver;         /* this is a versioned pointer */
+  uint32 volatile pins_in_stack;            /* number of elements in array */
+} LF_PINBOX;
+
+/* we want sizeof(LF_PINS) to be close to 128 to avoid false sharing */
+typedef struct {
+  void * volatile pin[LF_PINBOX_PINS];
+  void * purgatory[LF_PURGATORY_SIZE];
+  LF_PINBOX *pinbox;
+  uint32 purgatory_count;
+  uint32 volatile link;
+  char pad[128-sizeof(uint32)*2
+              -sizeof(void *)*(LF_PINBOX_PINS+LF_PURGATORY_SIZE+1)];
+} LF_PINS;
+
+#define lf_rwlock_by_pins(PINS)   \
+  my_atomic_rwlock_wrlock(&(PINS)->pinbox->pinstack.lock)
+#define lf_rwunlock_by_pins(PINS) \
+  my_atomic_rwlock_wrunlock(&(PINS)->pinbox->pinstack.lock)
+
+/*
+  compile-time assert, to require "no less than N" pins
+  it's enough if it'll fail on at least one compiler, so
+  we'll enable it on GCC only, which supports zero-length arrays.
+*/
+#if defined(__GNUC__) && defined(MY_LF_EXTRA_DEBUG)
+#define LF_REQUIRE_PINS(N)                                      \
+  static const char require_pins[LF_PINBOX_PINS-N];             \
+  static const int LF_NUM_PINS_IN_THIS_FILE=N;
+#define _lf_pin(PINS, PIN, ADDR)                                \
+  (                                                             \
+    my_atomic_storeptr(&(PINS)->pin[PIN], (ADDR)),              \
+    assert(PIN < LF_NUM_PINS_IN_THIS_FILE)                      \
+  )
+#else
+#define LF_REQUIRE_PINS(N)
+#define _lf_pin(PINS, PIN, ADDR)  my_atomic_storeptr(&(PINS)->pin[PIN], (ADDR))
+#endif
+
+#define _lf_unpin(PINS, PIN)      _lf_pin(PINS, PIN, NULL)
+#define lf_pin(PINS, PIN, ADDR)   \
+  do {                            \
+    lf_rwlock_by_pins(PINS);      \
+    _lf_pin(PINS, PIN, ADDR);     \
+    lf_rwunlock_by_pins(PINS);    \
+  } while (0)
+#define lf_unpin(PINS, PIN)  lf_pin(PINS, PIN, NULL)
+#define _lf_assert_pin(PINS, PIN) assert((PINS)->pin[PIN] != 0)
+#define _lf_assert_unpin(PINS, PIN) assert((PINS)->pin[PIN]==0)
+
+void lf_pinbox_init(LF_PINBOX *pinbox, lf_pinbox_free_func *free_func,
+                    void * free_func_arg);
+void lf_pinbox_destroy(LF_PINBOX *pinbox);
+
+lock_wrap(lf_pinbox_get_pins, LF_PINS *,
+          (LF_PINBOX *pinbox),
+          (pinbox),
+          &pinbox->pinstack.lock);
+lock_wrap_void(lf_pinbox_put_pins,
+               (LF_PINS *pins),
+               (pins),
+               &pins->pinbox->pinstack.lock);
+#if 0
+lock_wrap_void(lf_pinbox_real_free,
+               (LF_PINS *pins),
+               (pins),
+               &pins->pinbox->pinstack.lock);
+#endif
+lock_wrap_void(lf_pinbox_free,
+               (LF_PINS *pins, void *addr),
+               (pins,addr),
+               &pins->pinbox->pinstack.lock);
+
+/*
+  memory allocator
+*/
+
+typedef struct st_lf_allocator {
+  LF_PINBOX pinbox;
+  void * volatile top;
+  uint element_size;
+  uint32 volatile mallocs;
+} LF_ALLOCATOR;
+
+void lf_alloc_init(LF_ALLOCATOR *allocator, uint size);
+void lf_alloc_destroy(LF_ALLOCATOR *allocator);
+uint lf_alloc_in_pool(LF_ALLOCATOR *allocator);
+#define _lf_alloc_free(PINS, PTR)   _lf_pinbox_free((PINS), (PTR))
+#define lf_alloc_free(PINS, PTR)    lf_pinbox_free((PINS), (PTR))
+#define _lf_alloc_get_pins(ALLOC)   _lf_pinbox_get_pins(&(ALLOC)->pinbox)
+#define lf_alloc_get_pins(ALLOC)    lf_pinbox_get_pins(&(ALLOC)->pinbox)
+#define _lf_alloc_put_pins(PINS)    _lf_pinbox_put_pins(PINS)
+#define lf_alloc_put_pins(PINS)     lf_pinbox_put_pins(PINS)
+#define lf_alloc_real_free(ALLOC,ADDR) my_free((gptr)(ADDR), MYF(0))
+
+lock_wrap(lf_alloc_new, void *,
+          (LF_PINS *pins),
+          (pins),
+          &pins->pinbox->pinstack.lock);
+
+/*
+  extendible hash
+*/
+#include <hash.h>
+
+#define LF_HASH_UNIQUE 1
+
+typedef struct {
+  LF_DYNARRAY array;                    /* hash itself */
+  LF_ALLOCATOR alloc;                   /* allocator for elements */
+  hash_get_key get_key;                 /* see HASH */
+  CHARSET_INFO *charset;                /* see HASH */
+  uint key_offset, key_length;          /* see HASH */
+  uint element_size, flags;             /* LF_HASH_UNIQUE, etc */
+  int32 volatile size;                  /* size of array */
+  int32 volatile count;                 /* number of elements in the hash */
+} LF_HASH;
+
+void lf_hash_init(LF_HASH *hash, uint element_size, uint flags,
+                  uint key_offset, uint key_length, hash_get_key get_key,
+                  CHARSET_INFO *charset);
+void lf_hash_destroy(LF_HASH *hash);
+int lf_hash_insert(LF_HASH *hash, LF_PINS *pins, const void *data);
+void *lf_hash_search(LF_HASH *hash, LF_PINS *pins, const void *key, uint keylen);
+int lf_hash_delete(LF_HASH *hash, LF_PINS *pins, const void *key, uint keylen);
+#define _lf_hash_get_pins(HASH)   _lf_alloc_get_pins(&(HASH)->alloc)
+#define lf_hash_get_pins(HASH)    lf_alloc_get_pins(&(HASH)->alloc)
+#define _lf_hash_put_pins(PINS)   _lf_pinbox_put_pins(PINS)
+#define lf_hash_put_pins(PINS)    lf_pinbox_put_pins(PINS)
+
+/*
+  cleanup
+*/
+
+#undef lock_wrap_void
+#undef lock_wrap
+#undef nolock_wrap_void
+#undef nolock_wrap
+
+#endif
+
diff --git a/include/my_atomic.h b/include/my_atomic.h
index 9a319f84451..921b55e68a2 100644
--- a/include/my_atomic.h
+++ b/include/my_atomic.h
@@ -36,7 +36,7 @@
 #ifdef HAVE_INLINE
 
 #define make_atomic_add(S)					\
-static inline int ## S my_atomic_add ## S(			\
+STATIC_INLINE int ## S my_atomic_add ## S(			\
                         int ## S volatile *a, int ## S v)	\
 {								\
   make_atomic_add_body(S);					\
@@ -44,7 +44,7 @@ static inline int ## S my_atomic_add ## S(			\
 }
 
 #define make_atomic_swap(S)					\
-static inline int ## S my_atomic_swap ## S(			\
+STATIC_INLINE int ## S my_atomic_swap ## S(			\
                          int ## S volatile *a, int ## S v)	\
 {								\
   make_atomic_swap_body(S);					\
@@ -52,7 +52,7 @@ static inline int ## S my_atomic_swap ## S(			\
 }
 
 #define make_atomic_cas(S)					\
-static inline int my_atomic_cas ## S(int ## S volatile *a,	\
+STATIC_INLINE int my_atomic_cas ## S(int ## S volatile *a,	\
                             int ## S *cmp, int ## S set)	\
 {								\
   int8 ret;							\
@@ -61,7 +61,7 @@ static inline int my_atomic_cas ## S(int ## S volatile *a,	\
 }
 
 #define make_atomic_load(S)					\
-static inline int ## S my_atomic_load ## S(int ## S volatile *a) \
+STATIC_INLINE int ## S my_atomic_load ## S(int ## S volatile *a) \
 {								\
   int ## S ret;						\
   make_atomic_load_body(S);					\
@@ -69,7 +69,7 @@ static inline int ## S my_atomic_load ## S(int ## S volatile *a) \
 }
 
 #define make_atomic_store(S)					\
-static inline void my_atomic_store ## S(			\
+STATIC_INLINE void my_atomic_store ## S(			\
                      int ## S volatile *a, int ## S v)	\
 {								\
   make_atomic_store_body(S);					\
@@ -118,6 +118,11 @@ make_atomic_swap(16)
 make_atomic_swap(32)
 make_atomic_swap(ptr)
 
+#ifdef _atomic_h_cleanup_
+#include _atomic_h_cleanup_
+#undef _atomic_h_cleanup_
+#endif
+
 #undef make_atomic_add
 #undef make_atomic_cas
 #undef make_atomic_load
@@ -130,9 +135,8 @@ make_atomic_swap(ptr)
 #undef make_atomic_swap_body
 #undef intptr
 
-#ifdef _atomic_h_cleanup_
-#include _atomic_h_cleanup_
-#undef _atomic_h_cleanup_
+#ifndef LF_BACKOFF
+#define LF_BACKOFF (1)
 #endif
 
 #if SIZEOF_CHARP == SIZEOF_INT
diff --git a/include/my_bit.h b/include/my_bit.h
new file mode 100644
index 00000000000..58e8bb39683
--- /dev/null
+++ b/include/my_bit.h
@@ -0,0 +1,107 @@
+/*
+  Some useful bit functions
+*/
+
+#ifdef HAVE_INLINE
+
+extern const char _my_bits_nbits[256];
+extern const uchar _my_bits_reverse_table[256];
+
+/*
+  Find smallest X in 2^X >= value
+  This can be used to divide a number with value by doing a shift instead
+*/
+
+STATIC_INLINE uint my_bit_log2(ulong value)
+{
+  uint bit;
+  for (bit=0 ; value > 1 ; value>>=1, bit++) ;
+  return bit;
+}
+
+STATIC_INLINE uint my_count_bits(ulonglong v)
+{
+#if SIZEOF_LONG_LONG > 4
+  /* The following code is a bit faster on 16 bit machines than if we would
+     only shift v */
+  ulong v2=(ulong) (v >> 32);
+  return (uint) (uchar) (_my_bits_nbits[(uchar)  v] +
+                         _my_bits_nbits[(uchar) (v >> 8)] +
+                         _my_bits_nbits[(uchar) (v >> 16)] +
+                         _my_bits_nbits[(uchar) (v >> 24)] +
+                         _my_bits_nbits[(uchar) (v2)] +
+                         _my_bits_nbits[(uchar) (v2 >> 8)] +
+                         _my_bits_nbits[(uchar) (v2 >> 16)] +
+                         _my_bits_nbits[(uchar) (v2 >> 24)]);
+#else
+  return (uint) (uchar) (_my_bits_nbits[(uchar)  v] +
+                         _my_bits_nbits[(uchar) (v >> 8)] +
+                         _my_bits_nbits[(uchar) (v >> 16)] +
+                         _my_bits_nbits[(uchar) (v >> 24)]);
+#endif
+}
+
+STATIC_INLINE uint my_count_bits_ushort(ushort v)
+{
+  return _my_bits_nbits[v];
+}
+
+
+/*
+  Next highest power of two
+
+  SYNOPSIS
+    my_round_up_to_next_power()
+    v		Value to check
+
+  RETURN
+    Next or equal power of 2
+    Note: 0 will return 0
+
+  NOTES
+    Algorithm by Sean Anderson, according to:
+    http://graphics.stanford.edu/~seander/bithacks.html
+    (Orignal code public domain)
+
+    Comments shows how this works with 01100000000000000000000000001011
+*/
+
+STATIC_INLINE uint32 my_round_up_to_next_power(uint32 v)
+{
+  v--;			/* 01100000000000000000000000001010 */
+  v|= v >> 1;		/* 01110000000000000000000000001111 */
+  v|= v >> 2;		/* 01111100000000000000000000001111 */
+  v|= v >> 4;		/* 01111111110000000000000000001111 */
+  v|= v >> 8;		/* 01111111111111111100000000001111 */
+  v|= v >> 16;		/* 01111111111111111111111111111111 */
+  return v+1;		/* 10000000000000000000000000000000 */
+}
+
+STATIC_INLINE uint32 my_clear_highest_bit(uint32 v)
+{
+  uint32 w=v >> 1;
+  w|= w >> 1;
+  w|= w >> 2;
+  w|= w >> 4;
+  w|= w >> 8;
+  w|= w >> 16;
+  return v & w;
+}
+
+STATIC_INLINE uint32 my_reverse_bits(uint32 key)
+{
+  return
+    (_my_bits_reverse_table[ key      & 255] << 24) |
+    (_my_bits_reverse_table[(key>> 8) & 255] << 16) |
+    (_my_bits_reverse_table[(key>>16) & 255] <<  8) |
+     _my_bits_reverse_table[(key>>24)      ];
+}
+
+#else
+extern uint my_bit_log2(ulong value);
+extern uint32 my_round_up_to_next_power(uint32 v);
+uint32 my_clear_highest_bit(uint32 v);
+uint32 my_reverse_bits(uint32 key);
+extern uint my_count_bits(ulonglong v);
+extern uint my_count_bits_ushort(ushort v);
+#endif
diff --git a/include/my_global.h b/include/my_global.h
index 579379e7506..7f57b43f89f 100644
--- a/include/my_global.h
+++ b/include/my_global.h
@@ -192,6 +192,8 @@
 #endif
 #undef inline_test_2
 #undef inline_test_1
+/* helper macro for "instantiating" inline functions */
+#define STATIC_INLINE static inline
 
 /*
   The following macros are used to control inlining a bit more than
diff --git a/include/my_sys.h b/include/my_sys.h
index 115183bf9c2..fc83c583201 100644
--- a/include/my_sys.h
+++ b/include/my_sys.h
@@ -216,6 +216,7 @@ extern int (*error_handler_hook)(uint my_err, const char *str,myf MyFlags);
 extern int (*fatal_error_handler_hook)(uint my_err, const char *str,
 				       myf MyFlags);
 extern uint my_file_limit;
+extern ulong my_thread_stack_size;
 
 #ifdef HAVE_LARGE_PAGES
 extern my_bool my_use_large_pages;
@@ -821,10 +822,6 @@ extern int packfrm(const void *, uint, const void **, uint *);
 extern int unpackfrm(const void **, uint *, const void *);
 
 extern ha_checksum my_checksum(ha_checksum crc, const byte *mem, uint count);
-extern uint my_bit_log2(ulong value);
-extern uint32 my_round_up_to_next_power(uint32 v);
-extern uint my_count_bits(ulonglong v);
-extern uint my_count_bits_ushort(ushort v);
 extern void my_sleep(ulong m_seconds);
 extern ulong crc32(ulong crc, const uchar *buf, uint len);
 extern uint my_set_max_open_files(uint files);
@@ -840,7 +837,7 @@ extern int my_getncpus();
 #ifndef MAP_NOSYNC
 #define MAP_NOSYNC      0
 #endif
-#ifndef MAP_NORESERVE   
+#ifndef MAP_NORESERVE
 #define MAP_NORESERVE 0         /* For irix and AIX */
 #endif