Commit | Line | Data |
---|---|---|
05780f4b MM |
1 | /* |
2 | * Matt McCutchen's Big Integer Library | |
05780f4b MM |
3 | */ |
4 | ||
5 | #include "BigUnsigned.hh" | |
6 | ||
b3fe29df | 7 | // The "management" routines that used to be here are now in NumberlikeArray.hh. |
05780f4b MM |
8 | |
9 | /* | |
10 | * The steps for construction of a BigUnsigned | |
11 | * from an integral value x are as follows: | |
12 | * 1. If x is zero, create an empty BigUnsigned and stop. | |
13 | * 2. If x is negative, throw an exception. | |
14 | * 3. Allocate a one-block number array. | |
15 | * 4. If x is of a signed type, convert x to the unsigned | |
16 | * type of the same length. | |
17 | * 5. Expand x to a Blk, and store it in the number array. | |
b3fe29df MM |
18 | * |
19 | * Since 2005.01.06, NumberlikeArray uses `NULL' rather | |
20 | * than a real array if one of zero length is needed. | |
21 | * These constructors implicitly call NumberlikeArray's | |
a8b42b68 | 22 | * default constructor, which sets `blk = NULL, cap = len = 0'. |
b3fe29df MM |
23 | * So if the input number is zero, they can just return. |
24 | * See remarks in `NumberlikeArray.hh'. | |
05780f4b MM |
25 | */ |
26 | ||
27 | BigUnsigned::BigUnsigned(unsigned long x) { | |
b3fe29df MM |
28 | if (x == 0) |
29 | ; // NumberlikeArray already did all the work | |
30 | else { | |
05780f4b | 31 | cap = 1; |
a8b42b68 | 32 | blk = new Blk[1]; |
05780f4b MM |
33 | len = 1; |
34 | blk[0] = Blk(x); | |
35 | } | |
36 | } | |
37 | ||
38 | BigUnsigned::BigUnsigned(long x) { | |
b3fe29df MM |
39 | if (x == 0) |
40 | ; | |
41 | else if (x > 0) { | |
05780f4b | 42 | cap = 1; |
a8b42b68 | 43 | blk = new Blk[1]; |
05780f4b MM |
44 | len = 1; |
45 | blk[0] = Blk(x); | |
46 | } else | |
b3fe29df | 47 | throw "BigUnsigned::BigUnsigned(long): Cannot construct a BigUnsigned from a negative number"; |
05780f4b MM |
48 | } |
49 | ||
50 | BigUnsigned::BigUnsigned(unsigned int x) { | |
b3fe29df MM |
51 | if (x == 0) |
52 | ; | |
53 | else { | |
05780f4b | 54 | cap = 1; |
a8b42b68 | 55 | blk = new Blk[1]; |
05780f4b MM |
56 | len = 1; |
57 | blk[0] = Blk(x); | |
58 | } | |
59 | } | |
60 | ||
61 | BigUnsigned::BigUnsigned(int x) { | |
b3fe29df MM |
62 | if (x == 0) |
63 | ; | |
64 | else if (x > 0) { | |
05780f4b | 65 | cap = 1; |
a8b42b68 | 66 | blk = new Blk[1]; |
05780f4b MM |
67 | len = 1; |
68 | blk[0] = Blk(x); | |
69 | } else | |
b3fe29df | 70 | throw "BigUnsigned::BigUnsigned(int): Cannot construct a BigUnsigned from a negative number"; |
05780f4b MM |
71 | } |
72 | ||
73 | BigUnsigned::BigUnsigned(unsigned short x) { | |
b3fe29df MM |
74 | if (x == 0) |
75 | ; | |
76 | else { | |
05780f4b | 77 | cap = 1; |
a8b42b68 | 78 | blk = new Blk[1]; |
05780f4b MM |
79 | len = 1; |
80 | blk[0] = Blk(x); | |
81 | } | |
82 | } | |
83 | ||
84 | BigUnsigned::BigUnsigned(short x) { | |
b3fe29df MM |
85 | if (x == 0) |
86 | ; | |
87 | else if (x > 0) { | |
05780f4b | 88 | cap = 1; |
a8b42b68 | 89 | blk = new Blk[1]; |
05780f4b MM |
90 | len = 1; |
91 | blk[0] = Blk(x); | |
92 | } else | |
b3fe29df | 93 | throw "BigUnsigned::BigUnsigned(short): Cannot construct a BigUnsigned from a negative number"; |
05780f4b MM |
94 | } |
95 | ||
96 | // CONVERTERS | |
97 | /* | |
98 | * The steps for conversion of a BigUnsigned to an | |
99 | * integral type are as follows: | |
100 | * 1. If the BigUnsigned is zero, return zero. | |
101 | * 2. If it is more than one block long or its lowest | |
102 | * block has bits set out of the range of the target | |
103 | * type, throw an exception. | |
104 | * 3. Otherwise, convert the lowest block to the | |
105 | * target type and return it. | |
106 | */ | |
107 | ||
108 | namespace { | |
109 | // These masks are used to test whether a Blk has bits | |
110 | // set out of the range of a smaller integral type. Note | |
111 | // that this range is not considered to include the sign bit. | |
112 | const BigUnsigned::Blk lMask = ~0 >> 1; | |
113 | const BigUnsigned::Blk uiMask = (unsigned int)(~0); | |
114 | const BigUnsigned::Blk iMask = uiMask >> 1; | |
115 | const BigUnsigned::Blk usMask = (unsigned short)(~0); | |
116 | const BigUnsigned::Blk sMask = usMask >> 1; | |
117 | } | |
118 | ||
119 | BigUnsigned::operator unsigned long() const { | |
120 | if (len == 0) | |
121 | return 0; | |
122 | else if (len == 1) | |
123 | return (unsigned long) blk[0]; | |
124 | else | |
125 | throw "BigUnsigned::operator unsigned long: Value is too big for an unsigned long"; | |
126 | } | |
127 | ||
128 | BigUnsigned::operator long() const { | |
129 | if (len == 0) | |
130 | return 0; | |
131 | else if (len == 1 && (blk[0] & lMask) == blk[0]) | |
132 | return (long) blk[0]; | |
133 | else | |
134 | throw "BigUnsigned::operator long: Value is too big for a long"; | |
135 | } | |
136 | ||
137 | BigUnsigned::operator unsigned int() const { | |
138 | if (len == 0) | |
139 | return 0; | |
140 | else if (len == 1 && (blk[0] & uiMask) == blk[0]) | |
141 | return (unsigned int) blk[0]; | |
142 | else | |
143 | throw "BigUnsigned::operator unsigned int: Value is too big for an unsigned int"; | |
144 | } | |
145 | ||
146 | BigUnsigned::operator int() const { | |
147 | if (len == 0) | |
148 | return 0; | |
149 | else if (len == 1 && (blk[0] & iMask) == blk[0]) | |
150 | return (int) blk[0]; | |
151 | else | |
152 | throw "BigUnsigned::operator int: Value is too big for an int"; | |
153 | } | |
154 | ||
155 | BigUnsigned::operator unsigned short() const { | |
156 | if (len == 0) | |
157 | return 0; | |
158 | else if (len == 1 && (blk[0] & usMask) == blk[0]) | |
159 | return (unsigned short) blk[0]; | |
160 | else | |
161 | throw "BigUnsigned::operator unsigned short: Value is too big for an unsigned short"; | |
162 | } | |
163 | ||
164 | BigUnsigned::operator short() const { | |
165 | if (len == 0) | |
166 | return 0; | |
167 | else if (len == 1 && (blk[0] & sMask) == blk[0]) | |
168 | return (short) blk[0]; | |
169 | else | |
170 | throw "BigUnsigned::operator short: Value is too big for a short"; | |
171 | } | |
172 | ||
173 | // COMPARISON | |
174 | BigUnsigned::CmpRes BigUnsigned::compareTo(const BigUnsigned &x) const { | |
175 | // A bigger length implies a bigger number. | |
176 | if (len < x.len) | |
177 | return less; | |
178 | else if (len > x.len) | |
179 | return greater; | |
180 | else { | |
181 | // Compare blocks one by one from left to right. | |
182 | Index i = len; | |
183 | while (i > 0) { | |
184 | i--; | |
185 | if (blk[i] == x.blk[i]) | |
186 | continue; | |
187 | else if (blk[i] > x.blk[i]) | |
188 | return greater; | |
189 | else | |
190 | return less; | |
191 | } | |
192 | // If no blocks differed, the numbers are equal. | |
193 | return equal; | |
194 | } | |
195 | } | |
196 | ||
197 | // PUT-HERE OPERATIONS | |
198 | ||
4efbb076 MM |
199 | /* |
200 | * Below are implementations of the four basic arithmetic operations | |
201 | * for `BigUnsigned's. Their purpose is to use a mechanism that can | |
202 | * calculate the sum, difference, product, and quotient/remainder of | |
203 | * two individual blocks in order to calculate the sum, difference, | |
204 | * product, and quotient/remainder of two multi-block BigUnsigned | |
205 | * numbers. | |
206 | * | |
207 | * As alluded to in the comment before class `BigUnsigned', | |
208 | * these algorithms bear a remarkable similarity (in purpose, if | |
209 | * not in implementation) to the way humans operate on big numbers. | |
210 | * The built-in `+', `-', `*', `/' and `%' operators are analogous | |
211 | * to elementary-school ``math facts'' and ``times tables''; the | |
212 | * four routines below are analogous to ``long division'' and its | |
213 | * relatives. (Only a computer can ``memorize'' a times table with | |
214 | * 18446744073709551616 entries! (For 32-bit blocks.)) | |
215 | * | |
216 | * The discovery of these four algorithms, called the ``classical | |
217 | * algorithms'', marked the beginning of the study of computer science. | |
218 | * See Section 4.3.1 of Knuth's ``The Art of Computer Programming''. | |
219 | */ | |
220 | ||
8c16728a MM |
221 | /* |
222 | * On most calls to put-here operations, it's safe to read the inputs little by | |
223 | * little and write the outputs little by little. However, if one of the | |
224 | * inputs is coming from the same variable into which the output is to be | |
225 | * stored (an "aliased" call), we risk overwriting the input before we read it. | |
226 | * In this case, we first compute the result into a temporary BigUnsigned | |
227 | * variable and then copy it into the requested output variable *this. | |
228 | * Each put-here operation uses the DOTR_ALIASED macro (Do The Right Thing on | |
229 | * aliased calls) to generate code for this check. | |
230 | * | |
231 | * I adopted this approach on 2007.02.13 (see Assignment Operators in | |
232 | * BigUnsigned.hh). Before then, put-here operations rejected aliased calls | |
233 | * with an exception. I think doing the right thing is better. | |
234 | * | |
235 | * Some of the put-here operations can probably handle aliased calls safely | |
236 | * without the extra copy because (for example) they process blocks strictly | |
237 | * right-to-left. At some point I might determine which ones don't need the | |
238 | * copy, but my reasoning would need to be verified very carefully. For now | |
239 | * I'll leave in the copy. | |
240 | */ | |
241 | #define DOTR_ALIASED(cond, op) \ | |
242 | if (cond) { \ | |
243 | BigUnsigned tmpThis; \ | |
244 | tmpThis.op; \ | |
245 | *this = tmpThis; \ | |
246 | return; \ | |
247 | } | |
248 | ||
05780f4b MM |
249 | // Addition |
250 | void BigUnsigned::add(const BigUnsigned &a, const BigUnsigned &b) { | |
8c16728a | 251 | DOTR_ALIASED(this == &a || this == &b, add(a, b)); |
05780f4b MM |
252 | // If one argument is zero, copy the other. |
253 | if (a.len == 0) { | |
254 | operator =(b); | |
255 | return; | |
256 | } else if (b.len == 0) { | |
257 | operator =(a); | |
258 | return; | |
259 | } | |
4efbb076 | 260 | // Some variables... |
05780f4b MM |
261 | // Carries in and out of an addition stage |
262 | bool carryIn, carryOut; | |
263 | Blk temp; | |
264 | Index i; | |
265 | // a2 points to the longer input, b2 points to the shorter | |
266 | const BigUnsigned *a2, *b2; | |
267 | if (a.len >= b.len) { | |
268 | a2 = &a; | |
269 | b2 = &b; | |
270 | } else { | |
271 | a2 = &b; | |
272 | b2 = &a; | |
273 | } | |
274 | // Set prelimiary length and make room in this BigUnsigned | |
275 | len = a2->len + 1; | |
276 | allocate(len); | |
277 | // For each block index that is present in both inputs... | |
278 | for (i = 0, carryIn = false; i < b2->len; i++) { | |
279 | // Add input blocks | |
280 | temp = a2->blk[i] + b2->blk[i]; | |
281 | // If a rollover occurred, the result is less than either input. | |
282 | // This test is used many times in the BigUnsigned code. | |
283 | carryOut = (temp < a2->blk[i]); | |
284 | // If a carry was input, handle it | |
285 | if (carryIn) { | |
286 | temp++; | |
287 | carryOut |= (temp == 0); | |
288 | } | |
289 | blk[i] = temp; // Save the addition result | |
290 | carryIn = carryOut; // Pass the carry along | |
291 | } | |
292 | // If there is a carry left over, increase blocks until | |
293 | // one does not roll over. | |
294 | for (; i < a2->len && carryIn; i++) { | |
295 | temp = a2->blk[i] + 1; | |
296 | carryIn = (temp == 0); | |
297 | blk[i] = temp; | |
298 | } | |
299 | // If the carry was resolved but the larger number | |
300 | // still has blocks, copy them over. | |
301 | for (; i < a2->len; i++) | |
302 | blk[i] = a2->blk[i]; | |
303 | // Set the extra block if there's still a carry, decrease length otherwise | |
304 | if (carryIn) | |
305 | blk[i] = 1; | |
306 | else | |
307 | len--; | |
308 | } | |
309 | ||
310 | // Subtraction | |
311 | void BigUnsigned::subtract(const BigUnsigned &a, const BigUnsigned &b) { | |
8c16728a | 312 | DOTR_ALIASED(this == &a || this == &b, subtract(a, b)); |
05780f4b MM |
313 | // If b is zero, copy a. If a is shorter than b, the result is negative. |
314 | if (b.len == 0) { | |
315 | operator =(a); | |
316 | return; | |
317 | } else if (a.len < b.len) | |
318 | throw "BigUnsigned::subtract: Negative result in unsigned calculation"; | |
4efbb076 | 319 | // Some variables... |
05780f4b MM |
320 | bool borrowIn, borrowOut; |
321 | Blk temp; | |
322 | Index i; | |
323 | // Set preliminary length and make room | |
324 | len = a.len; | |
325 | allocate(len); | |
326 | // For each block index that is present in both inputs... | |
327 | for (i = 0, borrowIn = false; i < b.len; i++) { | |
328 | temp = a.blk[i] - b.blk[i]; | |
329 | // If a reverse rollover occurred, the result is greater than the block from a. | |
330 | borrowOut = (temp > a.blk[i]); | |
331 | // Handle an incoming borrow | |
332 | if (borrowIn) { | |
333 | borrowOut |= (temp == 0); | |
334 | temp--; | |
335 | } | |
336 | blk[i] = temp; // Save the subtraction result | |
337 | borrowIn = borrowOut; // Pass the borrow along | |
338 | } | |
339 | // If there is a borrow left over, decrease blocks until | |
340 | // one does not reverse rollover. | |
341 | for (; i < a.len && borrowIn; i++) { | |
342 | borrowIn = (a.blk[i] == 0); | |
343 | blk[i] = a.blk[i] - 1; | |
344 | } | |
345 | // If there's still a borrow, the result is negative. | |
346 | // Throw an exception, but zero out this object first just in case. | |
347 | if (borrowIn) { | |
348 | len = 0; | |
349 | throw "BigUnsigned::subtract: Negative result in unsigned calculation"; | |
350 | } else // Copy over the rest of the blocks | |
351 | for (; i < a.len; i++) | |
352 | blk[i] = a.blk[i]; | |
353 | // Zap leading zeros | |
354 | zapLeadingZeros(); | |
355 | } | |
356 | ||
4efbb076 MM |
357 | /* |
358 | * About the multiplication and division algorithms: | |
359 | * | |
360 | * I searched unsucessfully for fast built-in operations like the `b_0' | |
361 | * and `c_0' Knuth describes in Section 4.3.1 of ``The Art of Computer | |
362 | * Programming'' (replace `place' by `Blk'): | |
363 | * | |
364 | * ``b_0[:] multiplication of a one-place integer by another one-place | |
365 | * integer, giving a two-place answer; | |
366 | * | |
367 | * ``c_0[:] division of a two-place integer by a one-place integer, | |
368 | * provided that the quotient is a one-place integer, and yielding | |
369 | * also a one-place remainder.'' | |
370 | * | |
371 | * I also missed his note that ``[b]y adjusting the word size, if | |
372 | * necessary, nearly all computers will have these three operations | |
373 | * available'', so I gave up on trying to use algorithms similar to his. | |
374 | * A future version of the library might include such algorithms; I | |
375 | * would welcome contributions from others for this. | |
376 | * | |
377 | * I eventually decided to use bit-shifting algorithms. To multiply `a' | |
378 | * and `b', we zero out the result. Then, for each `1' bit in `a', we | |
379 | * shift `b' left the appropriate amount and add it to the result. | |
380 | * Similarly, to divide `a' by `b', we shift `b' left varying amounts, | |
381 | * repeatedly trying to subtract it from `a'. When we succeed, we note | |
382 | * the fact by setting a bit in the quotient. While these algorithms | |
383 | * have the same O(n^2) time complexity as Knuth's, the ``constant factor'' | |
384 | * is likely to be larger. | |
385 | * | |
386 | * Because I used these algorithms, which require single-block addition | |
387 | * and subtraction rather than single-block multiplication and division, | |
388 | * the innermost loops of all four routines are very similar. Study one | |
389 | * of them and all will become clear. | |
390 | */ | |
391 | ||
392 | /* | |
393 | * This is a little inline function used by both the multiplication | |
394 | * routine and the division routine. | |
395 | * | |
396 | * `getShiftedBlock' returns the `x'th block of `num << y'. | |
397 | * `y' may be anything from 0 to N - 1, and `x' may be anything from | |
398 | * 0 to `num.len'. | |
399 | * | |
400 | * Two things contribute to this block: | |
401 | * | |
402 | * (1) The `N - y' low bits of `num.blk[x]', shifted `y' bits left. | |
403 | * | |
404 | * (2) The `y' high bits of `num.blk[x-1]', shifted `N - y' bits right. | |
405 | * | |
406 | * But we must be careful if `x == 0' or `x == num.len', in | |
407 | * which case we should use 0 instead of (2) or (1), respectively. | |
408 | * | |
409 | * If `y == 0', then (2) contributes 0, as it should. However, | |
410 | * in some computer environments, for a reason I cannot understand, | |
411 | * `a >> b' means `a >> (b % N)'. This means `num.blk[x-1] >> (N - y)' | |
412 | * will return `num.blk[x-1]' instead of the desired 0 when `y == 0'; | |
413 | * the test `y == 0' handles this case specially. | |
414 | */ | |
415 | inline BigUnsigned::Blk getShiftedBlock(const BigUnsigned &num, | |
416 | BigUnsigned::Index x, unsigned int y) { | |
417 | BigUnsigned::Blk part1 = (x == 0 || y == 0) ? 0 : (num.blk[x - 1] >> (BigUnsigned::N - y)); | |
418 | BigUnsigned::Blk part2 = (x == num.len) ? 0 : (num.blk[x] << y); | |
419 | return part1 | part2; | |
420 | } | |
421 | ||
05780f4b MM |
422 | // Multiplication |
423 | void BigUnsigned::multiply(const BigUnsigned &a, const BigUnsigned &b) { | |
8c16728a | 424 | DOTR_ALIASED(this == &a || this == &b, multiply(a, b)); |
05780f4b MM |
425 | // If either a or b is zero, set to zero. |
426 | if (a.len == 0 || b.len == 0) { | |
427 | len = 0; | |
428 | return; | |
429 | } | |
4efbb076 MM |
430 | /* |
431 | * Overall method: | |
432 | * | |
433 | * Set this = 0. | |
434 | * For each 1-bit of `a' (say the `i2'th bit of block `i'): | |
435 | * Add `b << (i blocks and i2 bits)' to *this. | |
436 | */ | |
05780f4b MM |
437 | // Variables for the calculation |
438 | Index i, j, k; | |
439 | unsigned int i2; | |
4efbb076 | 440 | Blk temp; |
05780f4b MM |
441 | bool carryIn, carryOut; |
442 | // Set preliminary length and make room | |
443 | len = a.len + b.len; | |
444 | allocate(len); | |
445 | // Zero out this object | |
446 | for (i = 0; i < len; i++) | |
447 | blk[i] = 0; | |
448 | // For each block of the first number... | |
449 | for (i = 0; i < a.len; i++) { | |
450 | // For each 1-bit of that block... | |
4efbb076 | 451 | for (i2 = 0; i2 < N; i2++) { |
26a5f52b | 452 | if ((a.blk[i] & (Blk(1) << i2)) == 0) |
05780f4b | 453 | continue; |
4efbb076 MM |
454 | /* |
455 | * Add b to this, shifted left i blocks and i2 bits. | |
05780f4b | 456 | * j is the index in b, and k = i + j is the index in this. |
4efbb076 MM |
457 | * |
458 | * `getShiftedBlock', a short inline function defined above, | |
459 | * is now used for the bit handling. It replaces the more | |
460 | * complex `bHigh' code, in which each run of the loop dealt | |
461 | * immediately with the low bits and saved the high bits to | |
462 | * be picked up next time. The last run of the loop used to | |
463 | * leave leftover high bits, which were handled separately. | |
464 | * Instead, this loop runs an additional time with j == b.len. | |
465 | * These changes were made on 2005.01.11. | |
466 | */ | |
467 | for (j = 0, k = i, carryIn = false; j <= b.len; j++, k++) { | |
468 | /* | |
469 | * The body of this loop is very similar to the body of the first loop | |
470 | * in `add', except that this loop does a `+=' instead of a `+'. | |
471 | */ | |
472 | temp = blk[k] + getShiftedBlock(b, j, i2); | |
05780f4b MM |
473 | carryOut = (temp < blk[k]); |
474 | if (carryIn) { | |
475 | temp++; | |
476 | carryOut |= (temp == 0); | |
477 | } | |
478 | blk[k] = temp; | |
479 | carryIn = carryOut; | |
05780f4b | 480 | } |
4efbb076 MM |
481 | // No more extra iteration to deal with `bHigh'. |
482 | // Roll-over a carry as necessary. | |
05780f4b MM |
483 | for (; carryIn; k++) { |
484 | blk[k]++; | |
485 | carryIn = (blk[k] == 0); | |
486 | } | |
487 | } | |
488 | } | |
489 | // Zap possible leading zero | |
490 | if (blk[len - 1] == 0) | |
491 | len--; | |
492 | } | |
493 | ||
494 | /* | |
495 | * DIVISION WITH REMAINDER | |
496 | * The functionality of divide, modulo, and %= is included in this one monstrous call, | |
497 | * which deserves some explanation. | |
498 | * | |
499 | * The division *this / b is performed. | |
500 | * Afterwards, q has the quotient, and *this has the remainder. | |
501 | * Thus, a call is like q = *this / b, *this %= b. | |
502 | * | |
503 | * This seemingly bizarre pattern of inputs and outputs has a justification. The | |
504 | * ``put-here operations'' are supposed to be fast. Therefore, they accept inputs | |
505 | * and provide outputs in the most convenient places so that no value ever needs | |
506 | * to be copied in its entirety. That way, the client can perform exactly the | |
507 | * copying it needs depending on where the inputs are and where it wants the output. | |
8c16728a MM |
508 | * A better name for this function might be "modWithQuotient", but I would rather |
509 | * not change the name now. | |
05780f4b MM |
510 | */ |
511 | void BigUnsigned::divideWithRemainder(const BigUnsigned &b, BigUnsigned &q) { | |
8c16728a MM |
512 | /* |
513 | * Defending against aliased calls is a bit tricky because we are | |
514 | * writing to both *this and q. | |
515 | * | |
516 | * It would be silly to try to write quotient and remainder to the | |
517 | * same variable. Rule that out right away. | |
518 | */ | |
519 | if (this == &q) | |
520 | throw "BigUnsigned::divideWithRemainder: Cannot write quotient and remainder into the same variable"; | |
521 | /* | |
522 | * Now *this and q are separate, so the only concern is that b might be | |
523 | * aliased to one of them. If so, use a temporary copy of b. | |
524 | */ | |
525 | if (this == &b || &q == &b) { | |
526 | BigUnsigned tmpB(b); | |
527 | divideWithRemainder(tmpB, q); | |
528 | return; | |
529 | } | |
05780f4b MM |
530 | |
531 | /* | |
532 | * Note that the mathematical definition of mod (I'm trusting Knuth) is somewhat | |
533 | * different from the way the normal C++ % operator behaves in the case of division by 0. | |
534 | * This function does it Knuth's way. | |
535 | * | |
536 | * We let a / 0 == 0 (it doesn't matter) and a % 0 == a, no exceptions thrown. | |
537 | * This allows us to preserve both Knuth's demand that a mod 0 == a | |
538 | * and the useful property that (a / b) * b + (a % b) == a. | |
539 | */ | |
540 | if (b.len == 0) { | |
541 | q.len = 0; | |
542 | return; | |
543 | } | |
544 | ||
545 | /* | |
546 | * If *this.len < b.len, then *this < b, and we can be sure that b doesn't go into | |
547 | * *this at all. The quotient is 0 and *this is already the remainder (so leave it alone). | |
548 | */ | |
549 | if (len < b.len) { | |
550 | q.len = 0; | |
551 | return; | |
552 | } | |
553 | ||
554 | /* | |
555 | * At this point we know *this > b > 0. (Whew!) | |
556 | */ | |
557 | ||
05780f4b | 558 | /* |
4efbb076 MM |
559 | * Overall method: |
560 | * | |
561 | * For each appropriate i and i2, decreasing: | |
562 | * Try to subtract (b << (i blocks and i2 bits)) from *this. | |
563 | * (`work2' holds the result of this subtraction.) | |
564 | * If the result is nonnegative: | |
565 | * Turn on bit i2 of block i of the quotient q. | |
566 | * Save the result of the subtraction back into *this. | |
567 | * Otherwise: | |
568 | * Bit i2 of block i remains off, and *this is unchanged. | |
569 | * | |
570 | * Eventually q will contain the entire quotient, and *this will | |
571 | * be left with the remainder. | |
05780f4b MM |
572 | * |
573 | * We use work2 to temporarily store the result of a subtraction. | |
4efbb076 MM |
574 | * work2[x] corresponds to blk[x], not blk[x+i], since 2005.01.11. |
575 | * If the subtraction is successful, we copy work2 back to blk. | |
576 | * (There's no `work1'. In a previous version, when division was | |
577 | * coded for a read-only dividend, `work1' played the role of | |
578 | * the here-modifiable `*this' and got the remainder.) | |
579 | * | |
580 | * We never touch the i lowest blocks of either blk or work2 because | |
581 | * they are unaffected by the subtraction: we are subtracting | |
582 | * (b << (i blocks and i2 bits)), which ends in at least `i' zero blocks. | |
583 | */ | |
05780f4b MM |
584 | // Variables for the calculation |
585 | Index i, j, k; | |
586 | unsigned int i2; | |
4efbb076 | 587 | Blk temp; |
05780f4b MM |
588 | bool borrowIn, borrowOut; |
589 | ||
2f145f11 MM |
590 | /* |
591 | * Make sure we have an extra zero block just past the value. | |
2f145f11 | 592 | * |
4efbb076 MM |
593 | * When we attempt a subtraction, we might shift `b' so |
594 | * its first block begins a few bits left of the dividend, | |
595 | * and then we'll try to compare these extra bits with | |
596 | * a nonexistent block to the left of the dividend. The | |
597 | * extra zero block ensures sensible behavior; we need | |
598 | * an extra block in `work2' for exactly the same reason. | |
599 | * | |
600 | * See below `divideWithRemainder' for the interesting and | |
601 | * amusing story of this section of code. | |
2f145f11 | 602 | */ |
4efbb076 | 603 | Index origLen = len; // Save real length. |
be1bdfe2 MM |
604 | // 2006.05.03: Copy the number and then change the length! |
605 | allocateAndCopy(len + 1); // Get the space. | |
4efbb076 | 606 | len++; // Increase the length. |
4efbb076 | 607 | blk[origLen] = 0; // Zero the extra block. |
05780f4b | 608 | |
4efbb076 MM |
609 | // work2 holds part of the result of a subtraction; see above. |
610 | Blk *work2 = new Blk[len]; | |
05780f4b MM |
611 | |
612 | // Set preliminary length for quotient and make room | |
2f145f11 | 613 | q.len = origLen - b.len + 1; |
05780f4b MM |
614 | q.allocate(q.len); |
615 | // Zero out the quotient | |
616 | for (i = 0; i < q.len; i++) | |
617 | q.blk[i] = 0; | |
618 | ||
619 | // For each possible left-shift of b in blocks... | |
620 | i = q.len; | |
621 | while (i > 0) { | |
622 | i--; | |
623 | // For each possible left-shift of b in bits... | |
4efbb076 | 624 | // (Remember, N is the number of bits in a Blk.) |
05780f4b | 625 | q.blk[i] = 0; |
4efbb076 | 626 | i2 = N; |
05780f4b MM |
627 | while (i2 > 0) { |
628 | i2--; | |
629 | /* | |
4efbb076 MM |
630 | * Subtract b, shifted left i blocks and i2 bits, from *this, |
631 | * and store the answer in work2. In the for loop, `k == i + j'. | |
05780f4b MM |
632 | * |
633 | * Compare this to the middle section of `multiply'. They | |
4efbb076 MM |
634 | * are in many ways analogous. See especially the discussion |
635 | * of `getShiftedBlock'. | |
05780f4b | 636 | */ |
4efbb076 MM |
637 | for (j = 0, k = i, borrowIn = false; j <= b.len; j++, k++) { |
638 | temp = blk[k] - getShiftedBlock(b, j, i2); | |
05780f4b MM |
639 | borrowOut = (temp > blk[k]); |
640 | if (borrowIn) { | |
641 | borrowOut |= (temp == 0); | |
642 | temp--; | |
643 | } | |
4efbb076 MM |
644 | // Since 2005.01.11, indices of `work2' directly match those of `blk', so use `k'. |
645 | work2[k] = temp; | |
05780f4b | 646 | borrowIn = borrowOut; |
05780f4b | 647 | } |
4efbb076 MM |
648 | // No more extra iteration to deal with `bHigh'. |
649 | // Roll-over a borrow as necessary. | |
650 | for (; k < origLen && borrowIn; k++) { | |
05780f4b | 651 | borrowIn = (blk[k] == 0); |
4efbb076 | 652 | work2[k] = blk[k] - 1; |
05780f4b | 653 | } |
4efbb076 MM |
654 | /* |
655 | * If the subtraction was performed successfully (!borrowIn), | |
656 | * set bit i2 in block i of the quotient. | |
657 | * | |
658 | * Then, copy the portion of work2 filled by the subtraction | |
659 | * back to *this. This portion starts with block i and ends-- | |
660 | * where? Not necessarily at block `i + b.len'! Well, we | |
661 | * increased k every time we saved a block into work2, so | |
662 | * the region of work2 we copy is just [i, k). | |
663 | */ | |
05780f4b | 664 | if (!borrowIn) { |
26a5f52b | 665 | q.blk[i] |= (Blk(1) << i2); |
4efbb076 | 666 | while (k > i) { |
05780f4b | 667 | k--; |
4efbb076 | 668 | blk[k] = work2[k]; |
05780f4b MM |
669 | } |
670 | } | |
671 | } | |
672 | } | |
673 | // Zap possible leading zero in quotient | |
674 | if (q.blk[q.len - 1] == 0) | |
675 | q.len--; | |
676 | // Zap any/all leading zeros in remainder | |
677 | zapLeadingZeros(); | |
678 | // Deallocate temporary array. | |
679 | // (Thanks to Brad Spencer for noticing my accidental omission of this!) | |
680 | delete [] work2; | |
681 | ||
05780f4b | 682 | } |
4efbb076 MM |
683 | /* |
684 | * The out-of-bounds accesses story: | |
685 | * | |
686 | * On 2005.01.06 or 2005.01.07 (depending on your time zone), | |
687 | * Milan Tomic reported out-of-bounds memory accesses in | |
688 | * the Big Integer Library. To investigate the problem, I | |
689 | * added code to bounds-check every access to the `blk' array | |
690 | * of a `NumberlikeArray'. | |
691 | * | |
692 | * This gave me warnings that fell into two categories of false | |
693 | * positives. The bounds checker was based on length, not | |
694 | * capacity, and in two places I had accessed memory that I knew | |
695 | * was inside the capacity but that wasn't inside the length: | |
696 | * | |
697 | * (1) The extra zero block at the left of `*this'. Earlier | |
698 | * versions said `allocateAndCopy(len + 1); blk[len] = 0;' | |
699 | * but did not increment `len'. | |
700 | * | |
701 | * (2) The entire digit array in the conversion constructor | |
702 | * ``BigUnsignedInABase(BigUnsigned)''. It was allocated with | |
703 | * a conservatively high capacity, but the length wasn't set | |
704 | * until the end of the constructor. | |
705 | * | |
706 | * To simplify matters, I changed both sections of code so that | |
707 | * all accesses occurred within the length. The messages went | |
708 | * away, and I told Milan that I couldn't reproduce the problem, | |
709 | * sending a development snapshot of the bounds-checked code. | |
710 | * | |
711 | * Then, on 2005.01.09-10, he told me his debugger still found | |
712 | * problems, specifically at the line `delete [] work2'. | |
713 | * It was `work2', not `blk', that was causing the problems; | |
714 | * this possibility had not occurred to me at all. In fact, | |
715 | * the problem was that `work2' needed an extra block just | |
716 | * like `*this'. Go ahead and laugh at me for finding (1) | |
717 | * without seeing what was actually causing the trouble. :-) | |
718 | * | |
719 | * The 2005.01.11 version fixes this problem. I hope this is | |
720 | * the last of my memory-related bloopers. So this is what | |
721 | * starts happening to your C++ code if you use Java too much! | |
722 | */ | |
05780f4b MM |
723 | |
724 | // Bitwise and | |
725 | void BigUnsigned::bitAnd(const BigUnsigned &a, const BigUnsigned &b) { | |
8c16728a | 726 | DOTR_ALIASED(this == &a || this == &b, bitAnd(a, b)); |
05780f4b MM |
727 | len = (a.len >= b.len) ? b.len : a.len; |
728 | allocate(len); | |
729 | Index i; | |
730 | for (i = 0; i < len; i++) | |
731 | blk[i] = a.blk[i] & b.blk[i]; | |
732 | zapLeadingZeros(); | |
733 | } | |
734 | ||
735 | // Bitwise or | |
736 | void BigUnsigned::bitOr(const BigUnsigned &a, const BigUnsigned &b) { | |
8c16728a | 737 | DOTR_ALIASED(this == &a || this == &b, bitOr(a, b)); |
05780f4b MM |
738 | Index i; |
739 | const BigUnsigned *a2, *b2; | |
740 | if (a.len >= b.len) { | |
741 | a2 = &a; | |
742 | b2 = &b; | |
743 | } else { | |
744 | a2 = &b; | |
745 | b2 = &a; | |
746 | } | |
747 | allocate(a2->len); | |
748 | for (i = 0; i < b2->len; i++) | |
749 | blk[i] = a2->blk[i] | b2->blk[i]; | |
750 | for (; i < a2->len; i++) | |
751 | blk[i] = a2->blk[i]; | |
752 | len = a2->len; | |
753 | } | |
754 | ||
755 | // Bitwise xor | |
756 | void BigUnsigned::bitXor(const BigUnsigned &a, const BigUnsigned &b) { | |
8c16728a | 757 | DOTR_ALIASED(this == &a || this == &b, bitXor(a, b)); |
05780f4b MM |
758 | Index i; |
759 | const BigUnsigned *a2, *b2; | |
760 | if (a.len >= b.len) { | |
761 | a2 = &a; | |
762 | b2 = &b; | |
763 | } else { | |
764 | a2 = &b; | |
765 | b2 = &a; | |
766 | } | |
3aaa5ce6 | 767 | allocate(a2->len); |
05780f4b MM |
768 | for (i = 0; i < b2->len; i++) |
769 | blk[i] = a2->blk[i] ^ b2->blk[i]; | |
770 | for (; i < a2->len; i++) | |
771 | blk[i] = a2->blk[i]; | |
772 | len = a2->len; | |
773 | zapLeadingZeros(); | |
774 | } | |
775 | ||
776 | // INCREMENT/DECREMENT OPERATORS | |
777 | ||
778 | // Prefix increment | |
779 | void BigUnsigned::operator ++() { | |
780 | Index i; | |
781 | bool carry = true; | |
782 | for (i = 0; i < len && carry; i++) { | |
783 | blk[i]++; | |
784 | carry = (blk[i] == 0); | |
785 | } | |
786 | if (carry) { | |
787 | // Matt fixed a bug 2004.12.24: next 2 lines used to say allocateAndCopy(len + 1) | |
918d66f2 MM |
788 | // Matt fixed another bug 2006.04.24: |
789 | // old number only has len blocks, so copy before increasing length | |
790 | allocateAndCopy(len + 1); | |
05780f4b | 791 | len++; |
05780f4b MM |
792 | blk[i] = 1; |
793 | } | |
794 | } | |
795 | ||
796 | // Postfix increment: same as prefix | |
797 | void BigUnsigned::operator ++(int) { | |
798 | operator ++(); | |
799 | } | |
800 | ||
801 | // Prefix decrement | |
802 | void BigUnsigned::operator --() { | |
803 | if (len == 0) | |
804 | throw "BigUnsigned::operator --(): Cannot decrement an unsigned zero"; | |
805 | Index i; | |
806 | bool borrow = true; | |
807 | for (i = 0; borrow; i++) { | |
808 | borrow = (blk[i] == 0); | |
809 | blk[i]--; | |
810 | } | |
811 | // Zap possible leading zero (there can only be one) | |
812 | if (blk[len - 1] == 0) | |
813 | len--; | |
814 | } | |
815 | ||
816 | // Postfix decrement: same as prefix | |
817 | void BigUnsigned::operator --(int) { | |
818 | operator --(); | |
819 | } |