amd64 atomic ops

lock-free alloc (WL#3229), lock-free hash (WL#3230)
bit functions made inline


include/Makefile.am:
  lf.h added
mysys/Makefile.am:
  lf_hash.c lf_dynarray.c lf_alloc-pin.c
include/atomic/nolock.h:
  amd64 atomic ops
include/atomic/rwlock.h:
  s/rw_lock/mutex/g
include/atomic/x86-gcc.h:
  amd64 atomic ops
  try PAUSE
include/my_global.h:
  STATIC_INLINE
mysys/mf_keycache.c:
  make bit functions inline
mysys/my_atomic.c:
  STATIC_INLINE
mysys/my_bitmap.c:
  make bit functions inline
sql/ha_myisam.cc:
  make bit functions inline
sql/item_func.cc:
  make bit functions inline
include/my_atomic.h:
  STATIC_INLINE
mysys/my_bit.c:
  make bit functions inline
sql/sql_select.cc:
  make bit functions inline
storage/myisam/mi_create.c:
  make bit functions inline
storage/myisam/mi_test2.c:
  make bit functions inline
storage/myisam/myisamchk.c:
  make bit functions inline
mysys/my_init.c:
  thread_size moved to mysys
sql/mysql_priv.h:
  thread_size moved to mysys
sql/set_var.cc:
  thread_size moved to mysys
include/my_sys.h:
  thread_size moved to mysys
sql/mysqld.cc:
  thread_size moved to mysys
sql/sql_parse.cc:
  thread_size moved to mysys
sql/sql_test.cc:
  thread_size moved to mysys
include/lf.h:
  dylf_dynarray refactored to remove 65536 elements limit
mysys/lf_alloc-pin.c:
  dylf_dynarray refactored to remove 65536 elements limit
mysys/lf_dynarray.c:
  dylf_dynarray refactored to remove 65536 elements limit
mysys/lf_hash.c:
  dylf_dynarray refactored to remove 65536 elements limit
unittest/mysys/my_atomic-t.c:
  fix to commit (remove debug code)

1 files changed, 107 insertions, 0 deletions

diff --git a/include/my_bit.h b/include/my_bit.h
new file mode 100644
index 00000000000..71bbe2d4ba3
--- /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(uint 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(uint key);
+extern uint my_count_bits(ulonglong v);
+extern uint my_count_bits_ushort(ushort v);
+#endif