| 1 |
/*=========================================================================== |
| 2 |
* |
| 3 |
* PUBLIC DOMAIN NOTICE |
| 4 |
* National Center for Biotechnology Information |
| 5 |
* |
| 6 |
* This software/database is a "United States Government Work" under the |
| 7 |
* terms of the United States Copyright Act. It was written as part of |
| 8 |
* the author's official duties as a United States Government employee and |
| 9 |
* thus cannot be copyrighted. This software/database is freely available |
| 10 |
* to the public for use. The National Library of Medicine and the U.S. |
| 11 |
* Government have not placed any restriction on its use or reproduction. |
| 12 |
* |
| 13 |
* Although all reasonable efforts have been taken to ensure the accuracy |
| 14 |
* and reliability of the software and data, the NLM and the U.S. |
| 15 |
* Government do not and cannot warrant the performance or results that |
| 16 |
* may be obtained by using this software or data. The NLM and the U.S. |
| 17 |
* Government disclaim all warranties, express or implied, including |
| 18 |
* warranties of performance, merchantability or fitness for any particular |
| 19 |
* purpose. |
| 20 |
* |
| 21 |
* Please cite the author in any work or product based on this material. |
| 22 |
* |
| 23 |
* =========================================================================== |
| 24 |
* |
| 25 |
*/ |
| 26 |
|
| 27 |
#ifndef _h_noarch_bitstr_ |
| 28 |
#define _h_noarch_bitstr_ |
| 29 |
|
| 30 |
#ifndef _h_bitstr_ |
| 31 |
#error "don't include <noarch/bitstr.h> directly - use <bitstr.h>" |
| 32 |
#endif |
| 33 |
|
| 34 |
#ifndef _h_klib_defs_ |
| 35 |
#include <klib/defs.h> |
| 36 |
#endif |
| 37 |
|
| 38 |
#include <byteswap.h> |
| 39 |
|
| 40 |
#ifdef __cplusplus |
| 41 |
extern "C" { |
| 42 |
#endif |
| 43 |
|
| 44 |
/* bitcpy |
| 45 |
* copy a string of bits from source to dest |
| 46 |
* |
| 47 |
* both source and dest may have non-byte aligned pointers |
| 48 |
* the number of bits to copy need not be byte aligned |
| 49 |
* |
| 50 |
* depending upon architecture and OS conventions, the word |
| 51 |
* size may be adjusted to 1, 2, or 4 bytes, where the base |
| 52 |
* pointers are always word aligned. |
| 53 |
* |
| 54 |
* bits in memory are always treated as big-endian, meaning |
| 55 |
* that on multi-byte fetches and stores, we perform byte-swapping |
| 56 |
* if there are shifts or masks |
| 57 |
*/ |
| 58 |
static __inline__ |
| 59 |
void bitcpy ( void *dbase, bitsz_t doff, const void *sbase, bitsz_t soff, bitsz_t sz ) |
| 60 |
{ |
| 61 |
/* noop if sz == 0 */ |
| 62 |
if ( sz != 0 ) |
| 63 |
{ |
| 64 |
/* loop counter and destination word count */ |
| 65 |
size_t i, dcountz; |
| 66 |
|
| 67 |
/* left & right masks and working register */ |
| 68 |
WRD lmask, rmask, reg; |
| 69 |
|
| 70 |
/* produce word-aligned pointers */ |
| 71 |
#if WRDSIZE == 8 |
| 72 |
/* 1-4. all at once */ |
| 73 |
WRD *dst = ( WRD* ) dbase + ( doff >> WRDSHIFT ); |
| 74 |
const WRD *src = ( const WRD* ) sbase + ( soff >> WRDSHIFT ); |
| 75 |
#else |
| 76 |
/* 1. capture word alignment adjustment */ |
| 77 |
size_t dadjust = ( size_t ) dbase & ( WRDSIZE / 8 - 1 ); |
| 78 |
size_t sadjust = ( size_t ) sbase & ( WRDSIZE / 8 - 1 ); |
| 79 |
|
| 80 |
/* 2. create word-aligned pointers */ |
| 81 |
WRD *dst = ( WRD* ) ( ( size_t ) dbase - dadjust ); |
| 82 |
const WRD *src = ( const WRD* ) ( ( size_t ) sbase - sadjust ); |
| 83 |
|
| 84 |
/* 3. incorporate alignment adjustment into offset bits */ |
| 85 |
doff += dadjust << 3; |
| 86 |
soff += sadjust << 3; |
| 87 |
|
| 88 |
/* 4. readjust pointers based upon offset */ |
| 89 |
dst += doff >> WRDSHIFT; |
| 90 |
src += soff >> WRDSHIFT; |
| 91 |
#endif |
| 92 |
/* 5. restate offsets */ |
| 93 |
doff &= ( WRDSIZE - 1 ); |
| 94 |
soff &= ( WRDSIZE - 1 ); |
| 95 |
|
| 96 |
/* calculate number of words - 1 in dst */ |
| 97 |
dcountz = ( doff + sz + ( WRDSIZE - 1 ) - WRDSIZE ) >> WRDSHIFT; |
| 98 |
|
| 99 |
/* calculate masks */ |
| 100 |
lmask = rmask = ~ 0; |
| 101 |
lmask >>= doff; |
| 102 |
rmask >>= ( doff + sz ) & ( WRDSIZE - 1 ); |
| 103 |
if ( ( WRD ) ( rmask + 1 ) == 0 ) |
| 104 |
rmask = 0; |
| 105 |
|
| 106 |
/* prime register with masked dst [ 0 ] */ |
| 107 |
reg = BSWAP ( dst [ 0 ] ) & ~ lmask; |
| 108 |
|
| 109 |
/* if source and destination are aligned */ |
| 110 |
if ( doff == soff ) |
| 111 |
{ |
| 112 |
/* merge src [ 0 ] into reg through mask */ |
| 113 |
reg |= BSWAP ( src [ 0 ] ) & lmask; |
| 114 |
|
| 115 |
#if WRDSIZE > 8 |
| 116 |
/* straight copies don't need byteswap |
| 117 |
other than on first and last words |
| 118 |
put first word back into little-endian |
| 119 |
for remainder of loop */ |
| 120 |
if ( dcountz > 0 ) |
| 121 |
{ |
| 122 |
reg = BSWAP ( reg ); |
| 123 |
#endif |
| 124 |
/* aligned buffers have n:n word ratio */ |
| 125 |
for ( i = 0; i < dcountz; ) |
| 126 |
{ |
| 127 |
dst [ i ] = reg; |
| 128 |
reg = src [ ++ i ]; |
| 129 |
} |
| 130 |
|
| 131 |
#if WRDSIZE > 8 |
| 132 |
/* revert to big-endian */ |
| 133 |
reg = BSWAP ( reg ); |
| 134 |
} |
| 135 |
#endif |
| 136 |
} |
| 137 |
|
| 138 |
/* shifting alignment */ |
| 139 |
else |
| 140 |
{ |
| 141 |
/* source count may differ from dest count */ |
| 142 |
size_t scountz = ( soff + sz + ( WRDSIZE - 1 ) - WRDSIZE ) >> WRDSHIFT; |
| 143 |
|
| 144 |
/* use double-word accumulator */ |
| 145 |
ACC acc = BSWAP ( src [ 0 ] ); |
| 146 |
|
| 147 |
/* shift amount */ |
| 148 |
int shift = ( int ) doff - ( int ) soff; |
| 149 |
if ( shift > 0 ) |
| 150 |
{ |
| 151 |
/* take only valid bits in shifted initial src */ |
| 152 |
reg |= ( WRD ) ( acc >> shift ) & lmask; |
| 153 |
|
| 154 |
/* because "shift" > 0, we know "dcountz" >= "scountz" */ |
| 155 |
for ( acc <<= WRDSIZE, i = 0; i < scountz; acc <<= WRDSIZE ) |
| 156 |
{ |
| 157 |
dst [ i ] = BSWAP ( reg ); |
| 158 |
++ i; |
| 159 |
acc |= BSWAP ( src [ i ] ); |
| 160 |
reg = ( WRD ) ( acc >> shift ); |
| 161 |
} |
| 162 |
|
| 163 |
/* if "dcountz" > "scountz" */ |
| 164 |
if ( i < dcountz ) |
| 165 |
{ |
| 166 |
dst [ i ] = BSWAP ( reg ); |
| 167 |
reg = ( WRD ) ( acc >> shift ); |
| 168 |
} |
| 169 |
} |
| 170 |
|
| 171 |
else |
| 172 |
{ |
| 173 |
/* need single word read-ahead and right-shift */ |
| 174 |
shift += WRDSIZE; |
| 175 |
|
| 176 |
/* because "shift" was < 0, we know "dcountz" <= "scountz" */ |
| 177 |
for ( acc <<= WRDSIZE, i = 0; i < dcountz; acc <<= WRDSIZE ) |
| 178 |
{ |
| 179 |
acc |= BSWAP ( src [ i + 1 ] ); |
| 180 |
reg |= ( WRD ) ( acc >> shift ) & lmask; |
| 181 |
dst [ i ++ ] = BSWAP ( reg ); |
| 182 |
lmask = ~ 0; |
| 183 |
reg = 0; |
| 184 |
} |
| 185 |
|
| 186 |
/* if "dcountz" < "scountz" */ |
| 187 |
if ( i < scountz ) |
| 188 |
acc |= BSWAP ( src [ scountz ] ); |
| 189 |
|
| 190 |
reg |= ( WRD ) ( acc >> shift ) & lmask; |
| 191 |
} |
| 192 |
} |
| 193 |
|
| 194 |
/* mask off unused bytes from src */ |
| 195 |
reg &= ~ rmask; |
| 196 |
|
| 197 |
/* bring in saved bits from dst */ |
| 198 |
reg |= BSWAP ( dst [ dcountz ] ) & rmask; |
| 199 |
|
| 200 |
/* write out last word */ |
| 201 |
dst [ dcountz ] = BSWAP ( reg ); |
| 202 |
} |
| 203 |
} |
| 204 |
|
| 205 |
/* bitcmp |
| 206 |
* performs bitwise a - b, returning result as int |
| 207 |
* result value has no meaning, only sign |
| 208 |
* where < 0 means a < b, > 0 means a > b, and 0 means a == b |
| 209 |
* |
| 210 |
* since the comparison produces a tri-state indicator of |
| 211 |
* relative magnitude, the order of "a" and "b" is important. |
| 212 |
* furthermore, the difference operator must be evaluated |
| 213 |
* left to right, because the result indicates more than |
| 214 |
* equality. |
| 215 |
* |
| 216 |
* see bitcpy for general word alignment information |
| 217 |
*/ |
| 218 |
static __inline__ |
| 219 |
int bitcmp ( const void *abase, bitsz_t aoff, const void *bbase, bitsz_t boff, bitsz_t sz ) |
| 220 |
{ |
| 221 |
int diff = 0; |
| 222 |
|
| 223 |
if ( sz != 0 ) |
| 224 |
{ |
| 225 |
/* loop counter and left word count */ |
| 226 |
size_t i, lcountz; |
| 227 |
|
| 228 |
/* left & right masks and working registers */ |
| 229 |
WRD lmask, rmask, lreg, rreg; |
| 230 |
|
| 231 |
/* produce word-aligned pointers */ |
| 232 |
#if WRDSIZE == 8 |
| 233 |
/* 1-4. all at once */ |
| 234 |
const WRD *left = ( const WRD* ) abase + ( aoff >> WRDSHIFT ); |
| 235 |
const WRD *right = ( const WRD* ) bbase + ( boff >> WRDSHIFT ); |
| 236 |
#else |
| 237 |
/* 1. capture word alignment adjustment */ |
| 238 |
size_t aadjust = ( size_t ) abase & ( WRDSIZE / 8 - 1 ); |
| 239 |
size_t badjust = ( size_t ) bbase & ( WRDSIZE / 8 - 1 ); |
| 240 |
|
| 241 |
/* 2. create word-aligned pointers */ |
| 242 |
const WRD *left = ( const WRD* ) ( ( size_t ) abase - aadjust ); |
| 243 |
const WRD *right = ( const WRD* ) ( ( size_t ) bbase - badjust ); |
| 244 |
|
| 245 |
/* 3. incorporate alignment adjustment into offset bits */ |
| 246 |
aoff += aadjust << 3; |
| 247 |
boff += badjust << 3; |
| 248 |
|
| 249 |
/* 4. readjust pointers based upon offset */ |
| 250 |
left += aoff >> WRDSHIFT; |
| 251 |
right += boff >> WRDSHIFT; |
| 252 |
#endif |
| 253 |
/* 5. restate offsets */ |
| 254 |
aoff &= ( WRDSIZE - 1 ); |
| 255 |
boff &= ( WRDSIZE - 1 ); |
| 256 |
|
| 257 |
/* calculate number of words - 1 in left |
| 258 |
since we know a-priori that "sz" > 0, we |
| 259 |
know that the left and right counts must be |
| 260 |
at least 1. our loops treat the last word |
| 261 |
specially, so calculate a loop counter that |
| 262 |
excludes the last word */ |
| 263 |
lcountz = ( aoff + sz + ( WRDSIZE - 1 ) - WRDSIZE ) >> WRDSHIFT; |
| 264 |
|
| 265 |
/* calculate masks */ |
| 266 |
lmask = rmask = ~ 0; |
| 267 |
lmask >>= aoff; |
| 268 |
rmask >>= ( aoff + sz ) & ( WRDSIZE - 1 ); |
| 269 |
if ( ( WRD ) ( rmask + 1 ) == 0 ) |
| 270 |
rmask = 0; |
| 271 |
|
| 272 |
/* significant bits from left [ 0 ] */ |
| 273 |
lreg = BSWAP ( left [ 0 ] ) & lmask; |
| 274 |
|
| 275 |
/* if source and destination are aligned */ |
| 276 |
if ( aoff == boff ) |
| 277 |
{ |
| 278 |
/* test against right bits through mask */ |
| 279 |
rreg = BSWAP ( right [ 0 ] ) & lmask; |
| 280 |
|
| 281 |
/* produce a difference of all but the last |
| 282 |
aligned word, where initial word has been |
| 283 |
left-masked. the last word is tested below. */ |
| 284 |
for ( i = 1; i <= lcountz; ++ i ) |
| 285 |
{ |
| 286 |
diff = ( int ) lreg - ( int ) rreg; |
| 287 |
if ( diff != 0 ) |
| 288 |
return diff; |
| 289 |
|
| 290 |
/* byte-swapping occurs on little-endian architectures */ |
| 291 |
lreg = BSWAP ( left [ i ] ); |
| 292 |
rreg = BSWAP ( right [ i ] ); |
| 293 |
} |
| 294 |
|
| 295 |
/* fall out to end for masked comparison of last word */ |
| 296 |
} |
| 297 |
|
| 298 |
/* shifting alignment */ |
| 299 |
else |
| 300 |
{ |
| 301 |
/* right count may differ from left count |
| 302 |
since alignments differ, the span of "sz" |
| 303 |
bits may hit a different number of words in |
| 304 |
the left array than in the right. */ |
| 305 |
size_t rcountz = ( boff + sz + ( WRDSIZE - 1 ) - WRDSIZE ) >> WRDSHIFT; |
| 306 |
|
| 307 |
/* use double-word accumulator |
| 308 |
note that the extra bits get ignored */ |
| 309 |
ACC acc = BSWAP ( right [ 0 ] ); |
| 310 |
|
| 311 |
/* shift amount: positive if "b" needs to be right shifted. |
| 312 |
NOTE - since the comparison must be successively performed |
| 313 |
from left to right ( see above ), shifting is ALWAYS toward |
| 314 |
right, making for special handling when "shift" < 0 ( see below ) */ |
| 315 |
int shift = ( int ) aoff - ( int ) boff; |
| 316 |
if ( shift > 0 ) |
| 317 |
{ |
| 318 |
/* initial word from right operand, aligned with left */ |
| 319 |
rreg = ( WRD ) ( acc >> shift ) & lmask; |
| 320 |
|
| 321 |
/* "shift" > 0 means "lcountz" >= "rcountz" */ |
| 322 |
for ( acc <<= WRDSIZE, i = 1; i <= rcountz; acc <<= WRDSIZE, ++ i ) |
| 323 |
{ |
| 324 |
/* compare words at i-1 */ |
| 325 |
diff = ( int ) lreg - ( int ) rreg; |
| 326 |
if ( diff != 0 ) |
| 327 |
return diff; |
| 328 |
|
| 329 |
/* accumulate next word from right operand */ |
| 330 |
acc |= BSWAP ( right [ i ] ); |
| 331 |
|
| 332 |
/* bring in next word from left operand */ |
| 333 |
lreg = BSWAP ( left [ i ] ); |
| 334 |
|
| 335 |
/* produce aligned word from right operand */ |
| 336 |
rreg = ( WRD ) ( acc >> shift ); |
| 337 |
} |
| 338 |
|
| 339 |
/* if there is one more word in left */ |
| 340 |
if ( lcountz > rcountz ) |
| 341 |
{ |
| 342 |
/* compare penultimate */ |
| 343 |
diff = ( int ) lreg - ( int ) rreg; |
| 344 |
if ( diff != 0 ) |
| 345 |
return diff; |
| 346 |
|
| 347 |
/* get last word in left */ |
| 348 |
lreg = BSWAP ( left [ lcountz ] ); |
| 349 |
|
| 350 |
/* last word from right is already in "acc" */ |
| 351 |
rreg = ( WRD ) ( acc >> shift ); |
| 352 |
} |
| 353 |
|
| 354 |
/* fall out to end for masked comparison of last word */ |
| 355 |
} |
| 356 |
|
| 357 |
else |
| 358 |
{ |
| 359 |
/* since all shifts must be toward right ( due to left to right |
| 360 |
comparison ), this alignment will require a pre-fetch from |
| 361 |
right operand into accumulator, and adjusting the negative |
| 362 |
shift amount to a positive right-shift. */ |
| 363 |
shift += WRDSIZE; |
| 364 |
|
| 365 |
/* since "shift" was negative, we know "lcountz" <= "rcountz", |
| 366 |
so use "lcountz" as loop limit. pre-shift "acc" as loop init */ |
| 367 |
for ( acc <<= WRDSIZE, i = 1; i <= lcountz; acc <<= WRDSIZE, ++ i ) |
| 368 |
{ |
| 369 |
/* accumulate next word from right operand */ |
| 370 |
acc |= BSWAP ( right [ i ] ); |
| 371 |
|
| 372 |
/* produce aligned word from right operand */ |
| 373 |
rreg = ( WRD ) ( acc >> shift ) & lmask; |
| 374 |
|
| 375 |
/* now test against left */ |
| 376 |
diff = ( int ) lreg - ( int ) rreg; |
| 377 |
if ( diff != 0 ) |
| 378 |
return diff; |
| 379 |
|
| 380 |
/* bring in next word from left operand */ |
| 381 |
lreg = BSWAP ( left [ i ] ); |
| 382 |
|
| 383 |
/* no more left mask */ |
| 384 |
lmask = ~ 0; |
| 385 |
} |
| 386 |
|
| 387 |
/* if there is one more word in right */ |
| 388 |
if ( lcountz < rcountz ) |
| 389 |
acc |= BSWAP ( right [ rcountz ] ); |
| 390 |
|
| 391 |
/* produce "rreg" from "acc" */ |
| 392 |
rreg = ( WRD ) ( acc >> shift ) & lmask; |
| 393 |
|
| 394 |
/* fall out to end for masked comparison of last word */ |
| 395 |
} |
| 396 |
} |
| 397 |
|
| 398 |
/* mask off unused bytes from right */ |
| 399 |
lreg &= ~ rmask; |
| 400 |
rreg &= ~ rmask; |
| 401 |
|
| 402 |
/* perform final comparison */ |
| 403 |
diff = ( int ) lreg - ( int ) rreg; |
| 404 |
} |
| 405 |
|
| 406 |
return diff; |
| 407 |
} |
| 408 |
|
| 409 |
#ifdef __cplusplus |
| 410 |
} |
| 411 |
#endif |
| 412 |
|
| 413 |
#endif /* _h_noarch_bitstr_ */ |