js/src/jsbit.h

   1 /* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
   2 /* ***** BEGIN LICENSE BLOCK *****
   3  * Version: MPL 1.1/GPL 2.0/LGPL 2.1
   4  *
   5  * The contents of this file are subject to the Mozilla Public License Version
   6  * 1.1 (the "License"); you may not use this file except in compliance with
   7  * the License. You may obtain a copy of the License at
   8  * http://www.mozilla.org/MPL/
   9  *
  10  * Software distributed under the License is distributed on an "AS IS" basis,
  11  * WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
  12  * for the specific language governing rights and limitations under the
  13  * License.
  14  *
  15  * The Original Code is Mozilla Communicator client code, released
  16  * March 31, 1998.
  17  *
  18  * The Initial Developer of the Original Code is
  19  * Netscape Communications Corporation.
  20  * Portions created by the Initial Developer are Copyright (C) 1998
  21  * the Initial Developer. All Rights Reserved.
  22  *
  23  * Contributor(s):
  24  *
  25  * Alternatively, the contents of this file may be used under the terms of
  26  * either of the GNU General Public License Version 2 or later (the "GPL"),
  27  * or the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
  28  * in which case the provisions of the GPL or the LGPL are applicable instead
  29  * of those above. If you wish to allow use of your version of this file only
  30  * under the terms of either the GPL or the LGPL, and not to allow others to
  31  * use your version of this file under the terms of the MPL, indicate your
  32  * decision by deleting the provisions above and replace them with the notice
  33  * and other provisions required by the GPL or the LGPL. If you do not delete
  34  * the provisions above, a recipient may use your version of this file under
  35  * the terms of any one of the MPL, the GPL or the LGPL.
  36  *
  37  * ***** END LICENSE BLOCK ***** */
  38
  39 #ifndef jsbit_h___
  40 #define jsbit_h___
  41
  42 #include "jstypes.h"
  43 #include "jscompat.h"
  44 #include "jsutil.h"
  45
  46 JS_BEGIN_EXTERN_C
  47
  48 /*
  49 ** A jsbitmap_t is a long integer that can be used for bitmaps
  50 */
  51 typedef jsuword     jsbitmap_t;     /* NSPR name, a la Unix system types */
  52 typedef jsbitmap_t  jsbitmap;       /* JS-style scalar typedef name */
  53
  54 #define JS_BITMAP_SIZE(bits)    (JS_HOWMANY(bits, JS_BITS_PER_WORD) *         \
  55                                  sizeof(jsbitmap))
  56
  57 #define JS_TEST_BIT(_map,_bit)  ((_map)[(_bit)>>JS_BITS_PER_WORD_LOG2] &      \
  58                                  ((jsbitmap)1<<((_bit)&(JS_BITS_PER_WORD-1))))
  59 #define JS_SET_BIT(_map,_bit)   ((_map)[(_bit)>>JS_BITS_PER_WORD_LOG2] |=     \
  60                                  ((jsbitmap)1<<((_bit)&(JS_BITS_PER_WORD-1))))
  61 #define JS_CLEAR_BIT(_map,_bit) ((_map)[(_bit)>>JS_BITS_PER_WORD_LOG2] &=     \
  62                                  ~((jsbitmap)1<<((_bit)&(JS_BITS_PER_WORD-1))))
  63
  64 /*
  65 ** Compute the log of the least power of 2 greater than or equal to n
  66 */
  67 extern JS_PUBLIC_API(JSIntn) JS_CeilingLog2(JSUint32 i);
  68
  69 /*
  70 ** Compute the log of the greatest power of 2 less than or equal to n
  71 */
  72 extern JS_PUBLIC_API(JSIntn) JS_FloorLog2(JSUint32 i);
  73
  74 /*
  75  * Replace bit-scanning code sequences with CPU-specific instructions to
  76  * speedup calculations of ceiling/floor log2.
  77  *
  78  * With GCC 3.4 or later we can use __builtin_clz for that, see bug 327129.
  79  *
  80  * SWS: Added MSVC intrinsic bitscan support.  See bugs 349364 and 356856.
  81  */
  82 #if defined(_WIN32) && (_MSC_VER >= 1300) && (defined(_M_IX86) || defined(_M_AMD64) || defined(_M_X64))
  83
  84 unsigned char _BitScanForward(unsigned long * Index, unsigned long Mask);
  85 unsigned char _BitScanReverse(unsigned long * Index, unsigned long Mask);
  86 # pragma intrinsic(_BitScanForward,_BitScanReverse)
  87
  88 __forceinline static int
  89 __BitScanForward32(unsigned int val)
  90 {
  91     unsigned long idx;
  92
  93     _BitScanForward(&idx, (unsigned long)val);
  94     return (int)idx;
  95 }
  96 __forceinline static int
  97 __BitScanReverse32(unsigned int val)
  98 {
  99     unsigned long idx;
 100
 101     _BitScanReverse(&idx, (unsigned long)val);
 102     return (int)(31-idx);
 103 }
 104 # define js_bitscan_ctz32(val)  __BitScanForward32(val)
 105 # define js_bitscan_clz32(val)  __BitScanReverse32(val)
 106 # define JS_HAS_BUILTIN_BITSCAN32
 107
 108 #if defined(_M_AMD64) || defined(_M_X64)
 109 unsigned char _BitScanForward64(unsigned long * Index, unsigned __int64 Mask);
 110 unsigned char _BitScanReverse64(unsigned long * Index, unsigned __int64 Mask);
 111 # pragma intrinsic(_BitScanForward64,_BitScanReverse64)
 112
 113 __forceinline static int
 114 __BitScanForward64(unsigned __int64 val)
 115 {
 116     unsigned long idx;
 117
 118     _BitScanForward64(&idx, val);
 119     return (int)idx;
 120 }
 121 __forceinline static int
 122 __BitScanReverse64(unsigned __int64 val)
 123 {
 124     unsigned long idx;
 125
 126     _BitScanReverse64(&idx, val);
 127     return (int)(63-idx);
 128 }
 129 # define js_bitscan_ctz64(val)  __BitScanForward64(val)
 130 # define js_bitscan_clz64(val)  __BitScanReverse64(val)
 131 # define JS_HAS_BUILTIN_BITSCAN64
 132 #endif
 133 #elif (__GNUC__ >= 4) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 4)
 134
 135 # define js_bitscan_ctz32(val)  __builtin_ctz(val)
 136 # define js_bitscan_clz32(val)  __builtin_clz(val)
 137 # define JS_HAS_BUILTIN_BITSCAN32
 138 # if (JS_BYTES_PER_WORD == 8)
 139 #  define js_bitscan_ctz64(val)  __builtin_ctzll(val)
 140 #  define js_bitscan_clz64(val)  __builtin_clzll(val)
 141 #  define JS_HAS_BUILTIN_BITSCAN64
 142 # endif
 143
 144 #endif
 145
 146 /*
 147 ** Macro version of JS_CeilingLog2: Compute the log of the least power of
 148 ** 2 greater than or equal to _n. The result is returned in _log2.
 149 */
 150 #ifdef JS_HAS_BUILTIN_BITSCAN32
 151 /*
 152  * Use intrinsic function or count-leading-zeros to calculate ceil(log2(_n)).
 153  * The macro checks for "n <= 1" and not "n != 0" as js_bitscan_clz32(0) is
 154  * undefined.
 155  */
 156 # define JS_CEILING_LOG2(_log2,_n)                                            \
 157     JS_BEGIN_MACRO                                                            \
 158         unsigned int j_ = (unsigned int)(_n);                                 \
 159         (_log2) = (j_ <= 1 ? 0 : 32 - js_bitscan_clz32(j_ - 1));              \
 160     JS_END_MACRO
 161 #else
 162 # define JS_CEILING_LOG2(_log2,_n)                                            \
 163     JS_BEGIN_MACRO                                                            \
 164         JSUint32 j_ = (JSUint32)(_n);                                         \
 165         (_log2) = 0;                                                          \
 166         if ((j_) & ((j_)-1))                                                  \
 167             (_log2) += 1;                                                     \
 168         if ((j_) >> 16)                                                       \
 169             (_log2) += 16, (j_) >>= 16;                                       \
 170         if ((j_) >> 8)                                                        \
 171             (_log2) += 8, (j_) >>= 8;                                         \
 172         if ((j_) >> 4)                                                        \
 173             (_log2) += 4, (j_) >>= 4;                                         \
 174         if ((j_) >> 2)                                                        \
 175             (_log2) += 2, (j_) >>= 2;                                         \
 176         if ((j_) >> 1)                                                        \
 177             (_log2) += 1;                                                     \
 178     JS_END_MACRO
 179 #endif
 180
 181 /*
 182 ** Macro version of JS_FloorLog2: Compute the log of the greatest power of
 183 ** 2 less than or equal to _n. The result is returned in _log2.
 184 **
 185 ** This is equivalent to finding the highest set bit in the word.
 186 */
 187 #ifdef JS_HAS_BUILTIN_BITSCAN32
 188 /*
 189  * Use js_bitscan_clz32 or count-leading-zeros to calculate floor(log2(_n)).
 190  * Since js_bitscan_clz32(0) is undefined, the macro set the loweset bit to 1
 191  * to ensure 0 result when _n == 0.
 192  */
 193 # define JS_FLOOR_LOG2(_log2,_n)                                              \
 194     JS_BEGIN_MACRO                                                            \
 195         (_log2) = 31 - js_bitscan_clz32(((unsigned int)(_n)) | 1);            \
 196     JS_END_MACRO
 197 #else
 198 # define JS_FLOOR_LOG2(_log2,_n)                                              \
 199     JS_BEGIN_MACRO                                                            \
 200         JSUint32 j_ = (JSUint32)(_n);                                         \
 201         (_log2) = 0;                                                          \
 202         if ((j_) >> 16)                                                       \
 203             (_log2) += 16, (j_) >>= 16;                                       \
 204         if ((j_) >> 8)                                                        \
 205             (_log2) += 8, (j_) >>= 8;                                         \
 206         if ((j_) >> 4)                                                        \
 207             (_log2) += 4, (j_) >>= 4;                                         \
 208         if ((j_) >> 2)                                                        \
 209             (_log2) += 2, (j_) >>= 2;                                         \
 210         if ((j_) >> 1)                                                        \
 211             (_log2) += 1;                                                     \
 212     JS_END_MACRO
 213 #endif
 214
 215 /*
 216  * Internal function.
 217  * Compute the log of the least power of 2 greater than or equal to n. This is
 218  * a version of JS_CeilingLog2 that operates on unsigned integers with
 219  * CPU-dependant size.
 220  */
 221 #define JS_CEILING_LOG2W(n) ((n) <= 1 ? 0 : 1 + JS_FLOOR_LOG2W((n) - 1))
 222
 223 /*
 224  * Internal function.
 225  * Compute the log of the greatest power of 2 less than or equal to n.
 226  * This is a version of JS_FloorLog2 that operates on unsigned integers with
 227  * CPU-dependant size and requires that n != 0.
 228  */
 229 #define JS_FLOOR_LOG2W(n) (JS_ASSERT((n) != 0), js_FloorLog2wImpl(n))
 230
 231 #if JS_BYTES_PER_WORD == 4
 232
 233 # ifdef JS_HAS_BUILTIN_BITSCAN32
 234 #  define js_FloorLog2wImpl(n)                                                \
 235     ((size_t)(JS_BITS_PER_WORD - 1 - js_bitscan_clz32(n)))
 236 # else
 237 #  define js_FloorLog2wImpl(n) ((size_t)JS_FloorLog2(n))
 238 #endif
 239
 240 #elif JS_BYTES_PER_WORD == 8
 241
 242 # ifdef JS_HAS_BUILTIN_BITSCAN64
 243 #  define js_FloorLog2wImpl(n)                                                \
 244     ((size_t)(JS_BITS_PER_WORD - 1 - js_bitscan_clz64(n)))
 245 # else
 246 extern size_t js_FloorLog2wImpl(size_t n);
 247 # endif
 248
 249 #else
 250
 251 # error "NOT SUPPORTED"
 252
 253 #endif
 254
 255 namespace js {
 256
 257 inline size_t
 258 CountTrailingZeros(size_t n)
 259 {
 260     JS_ASSERT(n != 0);
 261 #if JS_BYTES_PER_WORD != 4 && JS_BYTES_PER_WORD != 8
 262 # error "NOT SUPPORTED"
 263 #endif
 264
 265 #if JS_BYTES_PER_WORD == 4 && defined JS_HAS_BUILTIN_BITSCAN32
 266     return js_bitscan_ctz32(n);
 267 #elif JS_BYTES_PER_WORD == 8 && defined JS_HAS_BUILTIN_BITSCAN64
 268     return js_bitscan_ctz64(n);
 269 #else
 270     size_t count = 0;
 271 # if JS_BYTES_PER_WORD == 8
 272     if (!(n & size_t(0xFFFFFFFFU))) { count += 32; n >>= 32; }
 273 # endif
 274     if (!(n & 0xFFFF)) { count += 16; n >>= 16; }
 275     if (!(n & 0xFF))   { count += 8;  n >>= 8; }
 276     if (!(n & 0xF))    { count += 4;  n >>= 4; }
 277     if (!(n & 0x3))    { count += 2;  n >>= 2; }
 278     if (!(n & 0x1))    { count += 1;  n >>= 1; }
 279     return count + 1 - (n & 0x1);
 280 #endif
 281 }
 282
 283 }
 284
 285 /*
 286  * Macros for rotate left. There is no rotate operation in the C Language so
 287  * the construct (a << 4) | (a >> 28) is used instead. Most compilers convert
 288  * this to a rotate instruction but some versions of MSVC don't without a
 289  * little help.  To get MSVC to generate a rotate instruction, we have to use
 290  * the _rotl intrinsic and use a pragma to make _rotl inline.
 291  *
 292  * MSVC in VS2005 will do an inline rotate instruction on the above construct.
 293  */
 294
 295 #if defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_AMD64) || \
 296     defined(_M_X64))
 297 #include <stdlib.h>
 298 #pragma intrinsic(_rotl)
 299 #define JS_ROTATE_LEFT32(a, bits) _rotl(a, bits)
 300 #else
 301 #define JS_ROTATE_LEFT32(a, bits) (((a) << (bits)) | ((a) >> (32 - (bits))))
 302 #endif
 303
 304 JS_END_EXTERN_C
 305 #endif /* jsbit_h___ */