1 //===-- APInt.cpp - Implement APInt class ---------------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements a class to represent arbitrary precision integer
11 // constant values and provide a variety of arithmetic operations on them.
12 //
13 //===----------------------------------------------------------------------===//
14
15 #include "llvm/ADT/APInt.h"
16 #include "llvm/ADT/ArrayRef.h"
17 #include "llvm/ADT/FoldingSet.h"
18 #include "llvm/ADT/Hashing.h"
19 #include "llvm/ADT/SmallString.h"
20 #include "llvm/ADT/StringRef.h"
21 #include "llvm/Config/llvm-config.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/ErrorHandling.h"
24 #include "llvm/Support/MathExtras.h"
25 #include "llvm/Support/raw_ostream.h"
26 #include <climits>
27 #include <cmath>
28 #include <cstdlib>
29 #include <cstring>
30 using namespace llvm;
31
32 #define DEBUG_TYPE "apint"
33
34 /// A utility function for allocating memory, checking for allocation failures,
35 /// and ensuring the contents are zeroed.
getClearedMemory(unsigned numWords)36 inline static uint64_t* getClearedMemory(unsigned numWords) {
37 uint64_t *result = new uint64_t[numWords];
38 memset(result, 0, numWords * sizeof(uint64_t));
39 return result;
40 }
41
42 /// A utility function for allocating memory and checking for allocation
43 /// failure. The content is not zeroed.
getMemory(unsigned numWords)44 inline static uint64_t* getMemory(unsigned numWords) {
45 return new uint64_t[numWords];
46 }
47
48 /// A utility function that converts a character to a digit.
getDigit(char cdigit,uint8_t radix)49 inline static unsigned getDigit(char cdigit, uint8_t radix) {
50 unsigned r;
51
52 if (radix == 16 || radix == 36) {
53 r = cdigit - '0';
54 if (r <= 9)
55 return r;
56
57 r = cdigit - 'A';
58 if (r <= radix - 11U)
59 return r + 10;
60
61 r = cdigit - 'a';
62 if (r <= radix - 11U)
63 return r + 10;
64
65 radix = 10;
66 }
67
68 r = cdigit - '0';
69 if (r < radix)
70 return r;
71
72 return -1U;
73 }
74
75
initSlowCase(uint64_t val,bool isSigned)76 void APInt::initSlowCase(uint64_t val, bool isSigned) {
77 U.pVal = getClearedMemory(getNumWords());
78 U.pVal[0] = val;
79 if (isSigned && int64_t(val) < 0)
80 for (unsigned i = 1; i < getNumWords(); ++i)
81 U.pVal[i] = WORD_MAX;
82 clearUnusedBits();
83 }
84
initSlowCase(const APInt & that)85 void APInt::initSlowCase(const APInt& that) {
86 U.pVal = getMemory(getNumWords());
87 memcpy(U.pVal, that.U.pVal, getNumWords() * APINT_WORD_SIZE);
88 }
89
initFromArray(ArrayRef<uint64_t> bigVal)90 void APInt::initFromArray(ArrayRef<uint64_t> bigVal) {
91 assert(BitWidth && "Bitwidth too small");
92 assert(bigVal.data() && "Null pointer detected!");
93 if (isSingleWord())
94 U.VAL = bigVal[0];
95 else {
96 // Get memory, cleared to 0
97 U.pVal = getClearedMemory(getNumWords());
98 // Calculate the number of words to copy
99 unsigned words = std::min<unsigned>(bigVal.size(), getNumWords());
100 // Copy the words from bigVal to pVal
101 memcpy(U.pVal, bigVal.data(), words * APINT_WORD_SIZE);
102 }
103 // Make sure unused high bits are cleared
104 clearUnusedBits();
105 }
106
APInt(unsigned numBits,ArrayRef<uint64_t> bigVal)107 APInt::APInt(unsigned numBits, ArrayRef<uint64_t> bigVal)
108 : BitWidth(numBits) {
109 initFromArray(bigVal);
110 }
111
APInt(unsigned numBits,unsigned numWords,const uint64_t bigVal[])112 APInt::APInt(unsigned numBits, unsigned numWords, const uint64_t bigVal[])
113 : BitWidth(numBits) {
114 initFromArray(makeArrayRef(bigVal, numWords));
115 }
116
APInt(unsigned numbits,StringRef Str,uint8_t radix)117 APInt::APInt(unsigned numbits, StringRef Str, uint8_t radix)
118 : BitWidth(numbits) {
119 assert(BitWidth && "Bitwidth too small");
120 fromString(numbits, Str, radix);
121 }
122
reallocate(unsigned NewBitWidth)123 void APInt::reallocate(unsigned NewBitWidth) {
124 // If the number of words is the same we can just change the width and stop.
125 if (getNumWords() == getNumWords(NewBitWidth)) {
126 BitWidth = NewBitWidth;
127 return;
128 }
129
130 // If we have an allocation, delete it.
131 if (!isSingleWord())
132 delete [] U.pVal;
133
134 // Update BitWidth.
135 BitWidth = NewBitWidth;
136
137 // If we are supposed to have an allocation, create it.
138 if (!isSingleWord())
139 U.pVal = getMemory(getNumWords());
140 }
141
AssignSlowCase(const APInt & RHS)142 void APInt::AssignSlowCase(const APInt& RHS) {
143 // Don't do anything for X = X
144 if (this == &RHS)
145 return;
146
147 // Adjust the bit width and handle allocations as necessary.
148 reallocate(RHS.getBitWidth());
149
150 // Copy the data.
151 if (isSingleWord())
152 U.VAL = RHS.U.VAL;
153 else
154 memcpy(U.pVal, RHS.U.pVal, getNumWords() * APINT_WORD_SIZE);
155 }
156
157 /// This method 'profiles' an APInt for use with FoldingSet.
Profile(FoldingSetNodeID & ID) const158 void APInt::Profile(FoldingSetNodeID& ID) const {
159 ID.AddInteger(BitWidth);
160
161 if (isSingleWord()) {
162 ID.AddInteger(U.VAL);
163 return;
164 }
165
166 unsigned NumWords = getNumWords();
167 for (unsigned i = 0; i < NumWords; ++i)
168 ID.AddInteger(U.pVal[i]);
169 }
170
171 /// Prefix increment operator. Increments the APInt by one.
operator ++()172 APInt& APInt::operator++() {
173 if (isSingleWord())
174 ++U.VAL;
175 else
176 tcIncrement(U.pVal, getNumWords());
177 return clearUnusedBits();
178 }
179
180 /// Prefix decrement operator. Decrements the APInt by one.
operator --()181 APInt& APInt::operator--() {
182 if (isSingleWord())
183 --U.VAL;
184 else
185 tcDecrement(U.pVal, getNumWords());
186 return clearUnusedBits();
187 }
188
189 /// Adds the RHS APint to this APInt.
190 /// @returns this, after addition of RHS.
191 /// Addition assignment operator.
operator +=(const APInt & RHS)192 APInt& APInt::operator+=(const APInt& RHS) {
193 assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
194 if (isSingleWord())
195 U.VAL += RHS.U.VAL;
196 else
197 tcAdd(U.pVal, RHS.U.pVal, 0, getNumWords());
198 return clearUnusedBits();
199 }
200
operator +=(uint64_t RHS)201 APInt& APInt::operator+=(uint64_t RHS) {
202 if (isSingleWord())
203 U.VAL += RHS;
204 else
205 tcAddPart(U.pVal, RHS, getNumWords());
206 return clearUnusedBits();
207 }
208
209 /// Subtracts the RHS APInt from this APInt
210 /// @returns this, after subtraction
211 /// Subtraction assignment operator.
operator -=(const APInt & RHS)212 APInt& APInt::operator-=(const APInt& RHS) {
213 assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
214 if (isSingleWord())
215 U.VAL -= RHS.U.VAL;
216 else
217 tcSubtract(U.pVal, RHS.U.pVal, 0, getNumWords());
218 return clearUnusedBits();
219 }
220
operator -=(uint64_t RHS)221 APInt& APInt::operator-=(uint64_t RHS) {
222 if (isSingleWord())
223 U.VAL -= RHS;
224 else
225 tcSubtractPart(U.pVal, RHS, getNumWords());
226 return clearUnusedBits();
227 }
228
operator *(const APInt & RHS) const229 APInt APInt::operator*(const APInt& RHS) const {
230 assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
231 if (isSingleWord())
232 return APInt(BitWidth, U.VAL * RHS.U.VAL);
233
234 APInt Result(getMemory(getNumWords()), getBitWidth());
235
236 tcMultiply(Result.U.pVal, U.pVal, RHS.U.pVal, getNumWords());
237
238 Result.clearUnusedBits();
239 return Result;
240 }
241
AndAssignSlowCase(const APInt & RHS)242 void APInt::AndAssignSlowCase(const APInt& RHS) {
243 tcAnd(U.pVal, RHS.U.pVal, getNumWords());
244 }
245
OrAssignSlowCase(const APInt & RHS)246 void APInt::OrAssignSlowCase(const APInt& RHS) {
247 tcOr(U.pVal, RHS.U.pVal, getNumWords());
248 }
249
XorAssignSlowCase(const APInt & RHS)250 void APInt::XorAssignSlowCase(const APInt& RHS) {
251 tcXor(U.pVal, RHS.U.pVal, getNumWords());
252 }
253
operator *=(const APInt & RHS)254 APInt& APInt::operator*=(const APInt& RHS) {
255 assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
256 *this = *this * RHS;
257 return *this;
258 }
259
operator *=(uint64_t RHS)260 APInt& APInt::operator*=(uint64_t RHS) {
261 if (isSingleWord()) {
262 U.VAL *= RHS;
263 } else {
264 unsigned NumWords = getNumWords();
265 tcMultiplyPart(U.pVal, U.pVal, RHS, 0, NumWords, NumWords, false);
266 }
267 return clearUnusedBits();
268 }
269
EqualSlowCase(const APInt & RHS) const270 bool APInt::EqualSlowCase(const APInt& RHS) const {
271 return std::equal(U.pVal, U.pVal + getNumWords(), RHS.U.pVal);
272 }
273
compare(const APInt & RHS) const274 int APInt::compare(const APInt& RHS) const {
275 assert(BitWidth == RHS.BitWidth && "Bit widths must be same for comparison");
276 if (isSingleWord())
277 return U.VAL < RHS.U.VAL ? -1 : U.VAL > RHS.U.VAL;
278
279 return tcCompare(U.pVal, RHS.U.pVal, getNumWords());
280 }
281
compareSigned(const APInt & RHS) const282 int APInt::compareSigned(const APInt& RHS) const {
283 assert(BitWidth == RHS.BitWidth && "Bit widths must be same for comparison");
284 if (isSingleWord()) {
285 int64_t lhsSext = SignExtend64(U.VAL, BitWidth);
286 int64_t rhsSext = SignExtend64(RHS.U.VAL, BitWidth);
287 return lhsSext < rhsSext ? -1 : lhsSext > rhsSext;
288 }
289
290 bool lhsNeg = isNegative();
291 bool rhsNeg = RHS.isNegative();
292
293 // If the sign bits don't match, then (LHS < RHS) if LHS is negative
294 if (lhsNeg != rhsNeg)
295 return lhsNeg ? -1 : 1;
296
297 // Otherwise we can just use an unsigned comparison, because even negative
298 // numbers compare correctly this way if both have the same signed-ness.
299 return tcCompare(U.pVal, RHS.U.pVal, getNumWords());
300 }
301
setBitsSlowCase(unsigned loBit,unsigned hiBit)302 void APInt::setBitsSlowCase(unsigned loBit, unsigned hiBit) {
303 unsigned loWord = whichWord(loBit);
304 unsigned hiWord = whichWord(hiBit);
305
306 // Create an initial mask for the low word with zeros below loBit.
307 uint64_t loMask = WORD_MAX << whichBit(loBit);
308
309 // If hiBit is not aligned, we need a high mask.
310 unsigned hiShiftAmt = whichBit(hiBit);
311 if (hiShiftAmt != 0) {
312 // Create a high mask with zeros above hiBit.
313 uint64_t hiMask = WORD_MAX >> (APINT_BITS_PER_WORD - hiShiftAmt);
314 // If loWord and hiWord are equal, then we combine the masks. Otherwise,
315 // set the bits in hiWord.
316 if (hiWord == loWord)
317 loMask &= hiMask;
318 else
319 U.pVal[hiWord] |= hiMask;
320 }
321 // Apply the mask to the low word.
322 U.pVal[loWord] |= loMask;
323
324 // Fill any words between loWord and hiWord with all ones.
325 for (unsigned word = loWord + 1; word < hiWord; ++word)
326 U.pVal[word] = WORD_MAX;
327 }
328
329 /// Toggle every bit to its opposite value.
flipAllBitsSlowCase()330 void APInt::flipAllBitsSlowCase() {
331 tcComplement(U.pVal, getNumWords());
332 clearUnusedBits();
333 }
334
335 /// Toggle a given bit to its opposite value whose position is given
336 /// as "bitPosition".
337 /// Toggles a given bit to its opposite value.
flipBit(unsigned bitPosition)338 void APInt::flipBit(unsigned bitPosition) {
339 assert(bitPosition < BitWidth && "Out of the bit-width range!");
340 if ((*this)[bitPosition]) clearBit(bitPosition);
341 else setBit(bitPosition);
342 }
343
insertBits(const APInt & subBits,unsigned bitPosition)344 void APInt::insertBits(const APInt &subBits, unsigned bitPosition) {
345 unsigned subBitWidth = subBits.getBitWidth();
346 assert(0 < subBitWidth && (subBitWidth + bitPosition) <= BitWidth &&
347 "Illegal bit insertion");
348
349 // Insertion is a direct copy.
350 if (subBitWidth == BitWidth) {
351 *this = subBits;
352 return;
353 }
354
355 // Single word result can be done as a direct bitmask.
356 if (isSingleWord()) {
357 uint64_t mask = WORD_MAX >> (APINT_BITS_PER_WORD - subBitWidth);
358 U.VAL &= ~(mask << bitPosition);
359 U.VAL |= (subBits.U.VAL << bitPosition);
360 return;
361 }
362
363 unsigned loBit = whichBit(bitPosition);
364 unsigned loWord = whichWord(bitPosition);
365 unsigned hi1Word = whichWord(bitPosition + subBitWidth - 1);
366
367 // Insertion within a single word can be done as a direct bitmask.
368 if (loWord == hi1Word) {
369 uint64_t mask = WORD_MAX >> (APINT_BITS_PER_WORD - subBitWidth);
370 U.pVal[loWord] &= ~(mask << loBit);
371 U.pVal[loWord] |= (subBits.U.VAL << loBit);
372 return;
373 }
374
375 // Insert on word boundaries.
376 if (loBit == 0) {
377 // Direct copy whole words.
378 unsigned numWholeSubWords = subBitWidth / APINT_BITS_PER_WORD;
379 memcpy(U.pVal + loWord, subBits.getRawData(),
380 numWholeSubWords * APINT_WORD_SIZE);
381
382 // Mask+insert remaining bits.
383 unsigned remainingBits = subBitWidth % APINT_BITS_PER_WORD;
384 if (remainingBits != 0) {
385 uint64_t mask = WORD_MAX >> (APINT_BITS_PER_WORD - remainingBits);
386 U.pVal[hi1Word] &= ~mask;
387 U.pVal[hi1Word] |= subBits.getWord(subBitWidth - 1);
388 }
389 return;
390 }
391
392 // General case - set/clear individual bits in dst based on src.
393 // TODO - there is scope for optimization here, but at the moment this code
394 // path is barely used so prefer readability over performance.
395 for (unsigned i = 0; i != subBitWidth; ++i) {
396 if (subBits[i])
397 setBit(bitPosition + i);
398 else
399 clearBit(bitPosition + i);
400 }
401 }
402
extractBits(unsigned numBits,unsigned bitPosition) const403 APInt APInt::extractBits(unsigned numBits, unsigned bitPosition) const {
404 assert(numBits > 0 && "Can't extract zero bits");
405 assert(bitPosition < BitWidth && (numBits + bitPosition) <= BitWidth &&
406 "Illegal bit extraction");
407
408 if (isSingleWord())
409 return APInt(numBits, U.VAL >> bitPosition);
410
411 unsigned loBit = whichBit(bitPosition);
412 unsigned loWord = whichWord(bitPosition);
413 unsigned hiWord = whichWord(bitPosition + numBits - 1);
414
415 // Single word result extracting bits from a single word source.
416 if (loWord == hiWord)
417 return APInt(numBits, U.pVal[loWord] >> loBit);
418
419 // Extracting bits that start on a source word boundary can be done
420 // as a fast memory copy.
421 if (loBit == 0)
422 return APInt(numBits, makeArrayRef(U.pVal + loWord, 1 + hiWord - loWord));
423
424 // General case - shift + copy source words directly into place.
425 APInt Result(numBits, 0);
426 unsigned NumSrcWords = getNumWords();
427 unsigned NumDstWords = Result.getNumWords();
428
429 uint64_t *DestPtr = Result.isSingleWord() ? &Result.U.VAL : Result.U.pVal;
430 for (unsigned word = 0; word < NumDstWords; ++word) {
431 uint64_t w0 = U.pVal[loWord + word];
432 uint64_t w1 =
433 (loWord + word + 1) < NumSrcWords ? U.pVal[loWord + word + 1] : 0;
434 DestPtr[word] = (w0 >> loBit) | (w1 << (APINT_BITS_PER_WORD - loBit));
435 }
436
437 return Result.clearUnusedBits();
438 }
439
getBitsNeeded(StringRef str,uint8_t radix)440 unsigned APInt::getBitsNeeded(StringRef str, uint8_t radix) {
441 assert(!str.empty() && "Invalid string length");
442 assert((radix == 10 || radix == 8 || radix == 16 || radix == 2 ||
443 radix == 36) &&
444 "Radix should be 2, 8, 10, 16, or 36!");
445
446 size_t slen = str.size();
447
448 // Each computation below needs to know if it's negative.
449 StringRef::iterator p = str.begin();
450 unsigned isNegative = *p == '-';
451 if (*p == '-' || *p == '+') {
452 p++;
453 slen--;
454 assert(slen && "String is only a sign, needs a value.");
455 }
456
457 // For radixes of power-of-two values, the bits required is accurately and
458 // easily computed
459 if (radix == 2)
460 return slen + isNegative;
461 if (radix == 8)
462 return slen * 3 + isNegative;
463 if (radix == 16)
464 return slen * 4 + isNegative;
465
466 // FIXME: base 36
467
468 // This is grossly inefficient but accurate. We could probably do something
469 // with a computation of roughly slen*64/20 and then adjust by the value of
470 // the first few digits. But, I'm not sure how accurate that could be.
471
472 // Compute a sufficient number of bits that is always large enough but might
473 // be too large. This avoids the assertion in the constructor. This
474 // calculation doesn't work appropriately for the numbers 0-9, so just use 4
475 // bits in that case.
476 unsigned sufficient
477 = radix == 10? (slen == 1 ? 4 : slen * 64/18)
478 : (slen == 1 ? 7 : slen * 16/3);
479
480 // Convert to the actual binary value.
481 APInt tmp(sufficient, StringRef(p, slen), radix);
482
483 // Compute how many bits are required. If the log is infinite, assume we need
484 // just bit.
485 unsigned log = tmp.logBase2();
486 if (log == (unsigned)-1) {
487 return isNegative + 1;
488 } else {
489 return isNegative + log + 1;
490 }
491 }
492
hash_value(const APInt & Arg)493 hash_code llvm::hash_value(const APInt &Arg) {
494 if (Arg.isSingleWord())
495 return hash_combine(Arg.U.VAL);
496
497 return hash_combine_range(Arg.U.pVal, Arg.U.pVal + Arg.getNumWords());
498 }
499
isSplat(unsigned SplatSizeInBits) const500 bool APInt::isSplat(unsigned SplatSizeInBits) const {
501 assert(getBitWidth() % SplatSizeInBits == 0 &&
502 "SplatSizeInBits must divide width!");
503 // We can check that all parts of an integer are equal by making use of a
504 // little trick: rotate and check if it's still the same value.
505 return *this == rotl(SplatSizeInBits);
506 }
507
508 /// This function returns the high "numBits" bits of this APInt.
getHiBits(unsigned numBits) const509 APInt APInt::getHiBits(unsigned numBits) const {
510 return this->lshr(BitWidth - numBits);
511 }
512
513 /// This function returns the low "numBits" bits of this APInt.
getLoBits(unsigned numBits) const514 APInt APInt::getLoBits(unsigned numBits) const {
515 APInt Result(getLowBitsSet(BitWidth, numBits));
516 Result &= *this;
517 return Result;
518 }
519
520 /// Return a value containing V broadcasted over NewLen bits.
getSplat(unsigned NewLen,const APInt & V)521 APInt APInt::getSplat(unsigned NewLen, const APInt &V) {
522 assert(NewLen >= V.getBitWidth() && "Can't splat to smaller bit width!");
523
524 APInt Val = V.zextOrSelf(NewLen);
525 for (unsigned I = V.getBitWidth(); I < NewLen; I <<= 1)
526 Val |= Val << I;
527
528 return Val;
529 }
530
countLeadingZerosSlowCase() const531 unsigned APInt::countLeadingZerosSlowCase() const {
532 unsigned Count = 0;
533 for (int i = getNumWords()-1; i >= 0; --i) {
534 uint64_t V = U.pVal[i];
535 if (V == 0)
536 Count += APINT_BITS_PER_WORD;
537 else {
538 Count += llvm::countLeadingZeros(V);
539 break;
540 }
541 }
542 // Adjust for unused bits in the most significant word (they are zero).
543 unsigned Mod = BitWidth % APINT_BITS_PER_WORD;
544 Count -= Mod > 0 ? APINT_BITS_PER_WORD - Mod : 0;
545 return Count;
546 }
547
countLeadingOnesSlowCase() const548 unsigned APInt::countLeadingOnesSlowCase() const {
549 unsigned highWordBits = BitWidth % APINT_BITS_PER_WORD;
550 unsigned shift;
551 if (!highWordBits) {
552 highWordBits = APINT_BITS_PER_WORD;
553 shift = 0;
554 } else {
555 shift = APINT_BITS_PER_WORD - highWordBits;
556 }
557 int i = getNumWords() - 1;
558 unsigned Count = llvm::countLeadingOnes(U.pVal[i] << shift);
559 if (Count == highWordBits) {
560 for (i--; i >= 0; --i) {
561 if (U.pVal[i] == WORD_MAX)
562 Count += APINT_BITS_PER_WORD;
563 else {
564 Count += llvm::countLeadingOnes(U.pVal[i]);
565 break;
566 }
567 }
568 }
569 return Count;
570 }
571
countTrailingZerosSlowCase() const572 unsigned APInt::countTrailingZerosSlowCase() const {
573 unsigned Count = 0;
574 unsigned i = 0;
575 for (; i < getNumWords() && U.pVal[i] == 0; ++i)
576 Count += APINT_BITS_PER_WORD;
577 if (i < getNumWords())
578 Count += llvm::countTrailingZeros(U.pVal[i]);
579 return std::min(Count, BitWidth);
580 }
581
countTrailingOnesSlowCase() const582 unsigned APInt::countTrailingOnesSlowCase() const {
583 unsigned Count = 0;
584 unsigned i = 0;
585 for (; i < getNumWords() && U.pVal[i] == WORD_MAX; ++i)
586 Count += APINT_BITS_PER_WORD;
587 if (i < getNumWords())
588 Count += llvm::countTrailingOnes(U.pVal[i]);
589 assert(Count <= BitWidth);
590 return Count;
591 }
592
countPopulationSlowCase() const593 unsigned APInt::countPopulationSlowCase() const {
594 unsigned Count = 0;
595 for (unsigned i = 0; i < getNumWords(); ++i)
596 Count += llvm::countPopulation(U.pVal[i]);
597 return Count;
598 }
599
intersectsSlowCase(const APInt & RHS) const600 bool APInt::intersectsSlowCase(const APInt &RHS) const {
601 for (unsigned i = 0, e = getNumWords(); i != e; ++i)
602 if ((U.pVal[i] & RHS.U.pVal[i]) != 0)
603 return true;
604
605 return false;
606 }
607
isSubsetOfSlowCase(const APInt & RHS) const608 bool APInt::isSubsetOfSlowCase(const APInt &RHS) const {
609 for (unsigned i = 0, e = getNumWords(); i != e; ++i)
610 if ((U.pVal[i] & ~RHS.U.pVal[i]) != 0)
611 return false;
612
613 return true;
614 }
615
byteSwap() const616 APInt APInt::byteSwap() const {
617 assert(BitWidth >= 16 && BitWidth % 16 == 0 && "Cannot byteswap!");
618 if (BitWidth == 16)
619 return APInt(BitWidth, ByteSwap_16(uint16_t(U.VAL)));
620 if (BitWidth == 32)
621 return APInt(BitWidth, ByteSwap_32(unsigned(U.VAL)));
622 if (BitWidth == 48) {
623 unsigned Tmp1 = unsigned(U.VAL >> 16);
624 Tmp1 = ByteSwap_32(Tmp1);
625 uint16_t Tmp2 = uint16_t(U.VAL);
626 Tmp2 = ByteSwap_16(Tmp2);
627 return APInt(BitWidth, (uint64_t(Tmp2) << 32) | Tmp1);
628 }
629 if (BitWidth == 64)
630 return APInt(BitWidth, ByteSwap_64(U.VAL));
631
632 APInt Result(getNumWords() * APINT_BITS_PER_WORD, 0);
633 for (unsigned I = 0, N = getNumWords(); I != N; ++I)
634 Result.U.pVal[I] = ByteSwap_64(U.pVal[N - I - 1]);
635 if (Result.BitWidth != BitWidth) {
636 Result.lshrInPlace(Result.BitWidth - BitWidth);
637 Result.BitWidth = BitWidth;
638 }
639 return Result;
640 }
641
reverseBits() const642 APInt APInt::reverseBits() const {
643 switch (BitWidth) {
644 case 64:
645 return APInt(BitWidth, llvm::reverseBits<uint64_t>(U.VAL));
646 case 32:
647 return APInt(BitWidth, llvm::reverseBits<uint32_t>(U.VAL));
648 case 16:
649 return APInt(BitWidth, llvm::reverseBits<uint16_t>(U.VAL));
650 case 8:
651 return APInt(BitWidth, llvm::reverseBits<uint8_t>(U.VAL));
652 default:
653 break;
654 }
655
656 APInt Val(*this);
657 APInt Reversed(BitWidth, 0);
658 unsigned S = BitWidth;
659
660 for (; Val != 0; Val.lshrInPlace(1)) {
661 Reversed <<= 1;
662 Reversed |= Val[0];
663 --S;
664 }
665
666 Reversed <<= S;
667 return Reversed;
668 }
669
GreatestCommonDivisor(APInt A,APInt B)670 APInt llvm::APIntOps::GreatestCommonDivisor(APInt A, APInt B) {
671 // Fast-path a common case.
672 if (A == B) return A;
673
674 // Corner cases: if either operand is zero, the other is the gcd.
675 if (!A) return B;
676 if (!B) return A;
677
678 // Count common powers of 2 and remove all other powers of 2.
679 unsigned Pow2;
680 {
681 unsigned Pow2_A = A.countTrailingZeros();
682 unsigned Pow2_B = B.countTrailingZeros();
683 if (Pow2_A > Pow2_B) {
684 A.lshrInPlace(Pow2_A - Pow2_B);
685 Pow2 = Pow2_B;
686 } else if (Pow2_B > Pow2_A) {
687 B.lshrInPlace(Pow2_B - Pow2_A);
688 Pow2 = Pow2_A;
689 } else {
690 Pow2 = Pow2_A;
691 }
692 }
693
694 // Both operands are odd multiples of 2^Pow_2:
695 //
696 // gcd(a, b) = gcd(|a - b| / 2^i, min(a, b))
697 //
698 // This is a modified version of Stein's algorithm, taking advantage of
699 // efficient countTrailingZeros().
700 while (A != B) {
701 if (A.ugt(B)) {
702 A -= B;
703 A.lshrInPlace(A.countTrailingZeros() - Pow2);
704 } else {
705 B -= A;
706 B.lshrInPlace(B.countTrailingZeros() - Pow2);
707 }
708 }
709
710 return A;
711 }
712
RoundDoubleToAPInt(double Double,unsigned width)713 APInt llvm::APIntOps::RoundDoubleToAPInt(double Double, unsigned width) {
714 union {
715 double D;
716 uint64_t I;
717 } T;
718 T.D = Double;
719
720 // Get the sign bit from the highest order bit
721 bool isNeg = T.I >> 63;
722
723 // Get the 11-bit exponent and adjust for the 1023 bit bias
724 int64_t exp = ((T.I >> 52) & 0x7ff) - 1023;
725
726 // If the exponent is negative, the value is < 0 so just return 0.
727 if (exp < 0)
728 return APInt(width, 0u);
729
730 // Extract the mantissa by clearing the top 12 bits (sign + exponent).
731 uint64_t mantissa = (T.I & (~0ULL >> 12)) | 1ULL << 52;
732
733 // If the exponent doesn't shift all bits out of the mantissa
734 if (exp < 52)
735 return isNeg ? -APInt(width, mantissa >> (52 - exp)) :
736 APInt(width, mantissa >> (52 - exp));
737
738 // If the client didn't provide enough bits for us to shift the mantissa into
739 // then the result is undefined, just return 0
740 if (width <= exp - 52)
741 return APInt(width, 0);
742
743 // Otherwise, we have to shift the mantissa bits up to the right location
744 APInt Tmp(width, mantissa);
745 Tmp <<= (unsigned)exp - 52;
746 return isNeg ? -Tmp : Tmp;
747 }
748
749 /// This function converts this APInt to a double.
750 /// The layout for double is as following (IEEE Standard 754):
751 /// --------------------------------------
752 /// | Sign Exponent Fraction Bias |
753 /// |-------------------------------------- |
754 /// | 1[63] 11[62-52] 52[51-00] 1023 |
755 /// --------------------------------------
roundToDouble(bool isSigned) const756 double APInt::roundToDouble(bool isSigned) const {
757
758 // Handle the simple case where the value is contained in one uint64_t.
759 // It is wrong to optimize getWord(0) to VAL; there might be more than one word.
760 if (isSingleWord() || getActiveBits() <= APINT_BITS_PER_WORD) {
761 if (isSigned) {
762 int64_t sext = SignExtend64(getWord(0), BitWidth);
763 return double(sext);
764 } else
765 return double(getWord(0));
766 }
767
768 // Determine if the value is negative.
769 bool isNeg = isSigned ? (*this)[BitWidth-1] : false;
770
771 // Construct the absolute value if we're negative.
772 APInt Tmp(isNeg ? -(*this) : (*this));
773
774 // Figure out how many bits we're using.
775 unsigned n = Tmp.getActiveBits();
776
777 // The exponent (without bias normalization) is just the number of bits
778 // we are using. Note that the sign bit is gone since we constructed the
779 // absolute value.
780 uint64_t exp = n;
781
782 // Return infinity for exponent overflow
783 if (exp > 1023) {
784 if (!isSigned || !isNeg)
785 return std::numeric_limits<double>::infinity();
786 else
787 return -std::numeric_limits<double>::infinity();
788 }
789 exp += 1023; // Increment for 1023 bias
790
791 // Number of bits in mantissa is 52. To obtain the mantissa value, we must
792 // extract the high 52 bits from the correct words in pVal.
793 uint64_t mantissa;
794 unsigned hiWord = whichWord(n-1);
795 if (hiWord == 0) {
796 mantissa = Tmp.U.pVal[0];
797 if (n > 52)
798 mantissa >>= n - 52; // shift down, we want the top 52 bits.
799 } else {
800 assert(hiWord > 0 && "huh?");
801 uint64_t hibits = Tmp.U.pVal[hiWord] << (52 - n % APINT_BITS_PER_WORD);
802 uint64_t lobits = Tmp.U.pVal[hiWord-1] >> (11 + n % APINT_BITS_PER_WORD);
803 mantissa = hibits | lobits;
804 }
805
806 // The leading bit of mantissa is implicit, so get rid of it.
807 uint64_t sign = isNeg ? (1ULL << (APINT_BITS_PER_WORD - 1)) : 0;
808 union {
809 double D;
810 uint64_t I;
811 } T;
812 T.I = sign | (exp << 52) | mantissa;
813 return T.D;
814 }
815
816 // Truncate to new width.
trunc(unsigned width) const817 APInt APInt::trunc(unsigned width) const {
818 assert(width < BitWidth && "Invalid APInt Truncate request");
819 assert(width && "Can't truncate to 0 bits");
820
821 if (width <= APINT_BITS_PER_WORD)
822 return APInt(width, getRawData()[0]);
823
824 APInt Result(getMemory(getNumWords(width)), width);
825
826 // Copy full words.
827 unsigned i;
828 for (i = 0; i != width / APINT_BITS_PER_WORD; i++)
829 Result.U.pVal[i] = U.pVal[i];
830
831 // Truncate and copy any partial word.
832 unsigned bits = (0 - width) % APINT_BITS_PER_WORD;
833 if (bits != 0)
834 Result.U.pVal[i] = U.pVal[i] << bits >> bits;
835
836 return Result;
837 }
838
839 // Sign extend to a new width.
sext(unsigned Width) const840 APInt APInt::sext(unsigned Width) const {
841 assert(Width > BitWidth && "Invalid APInt SignExtend request");
842
843 if (Width <= APINT_BITS_PER_WORD)
844 return APInt(Width, SignExtend64(U.VAL, BitWidth));
845
846 APInt Result(getMemory(getNumWords(Width)), Width);
847
848 // Copy words.
849 std::memcpy(Result.U.pVal, getRawData(), getNumWords() * APINT_WORD_SIZE);
850
851 // Sign extend the last word since there may be unused bits in the input.
852 Result.U.pVal[getNumWords() - 1] =
853 SignExtend64(Result.U.pVal[getNumWords() - 1],
854 ((BitWidth - 1) % APINT_BITS_PER_WORD) + 1);
855
856 // Fill with sign bits.
857 std::memset(Result.U.pVal + getNumWords(), isNegative() ? -1 : 0,
858 (Result.getNumWords() - getNumWords()) * APINT_WORD_SIZE);
859 Result.clearUnusedBits();
860 return Result;
861 }
862
863 // Zero extend to a new width.
zext(unsigned width) const864 APInt APInt::zext(unsigned width) const {
865 assert(width > BitWidth && "Invalid APInt ZeroExtend request");
866
867 if (width <= APINT_BITS_PER_WORD)
868 return APInt(width, U.VAL);
869
870 APInt Result(getMemory(getNumWords(width)), width);
871
872 // Copy words.
873 std::memcpy(Result.U.pVal, getRawData(), getNumWords() * APINT_WORD_SIZE);
874
875 // Zero remaining words.
876 std::memset(Result.U.pVal + getNumWords(), 0,
877 (Result.getNumWords() - getNumWords()) * APINT_WORD_SIZE);
878
879 return Result;
880 }
881
zextOrTrunc(unsigned width) const882 APInt APInt::zextOrTrunc(unsigned width) const {
883 if (BitWidth < width)
884 return zext(width);
885 if (BitWidth > width)
886 return trunc(width);
887 return *this;
888 }
889
sextOrTrunc(unsigned width) const890 APInt APInt::sextOrTrunc(unsigned width) const {
891 if (BitWidth < width)
892 return sext(width);
893 if (BitWidth > width)
894 return trunc(width);
895 return *this;
896 }
897
zextOrSelf(unsigned width) const898 APInt APInt::zextOrSelf(unsigned width) const {
899 if (BitWidth < width)
900 return zext(width);
901 return *this;
902 }
903
sextOrSelf(unsigned width) const904 APInt APInt::sextOrSelf(unsigned width) const {
905 if (BitWidth < width)
906 return sext(width);
907 return *this;
908 }
909
910 /// Arithmetic right-shift this APInt by shiftAmt.
911 /// Arithmetic right-shift function.
ashrInPlace(const APInt & shiftAmt)912 void APInt::ashrInPlace(const APInt &shiftAmt) {
913 ashrInPlace((unsigned)shiftAmt.getLimitedValue(BitWidth));
914 }
915
916 /// Arithmetic right-shift this APInt by shiftAmt.
917 /// Arithmetic right-shift function.
ashrSlowCase(unsigned ShiftAmt)918 void APInt::ashrSlowCase(unsigned ShiftAmt) {
919 // Don't bother performing a no-op shift.
920 if (!ShiftAmt)
921 return;
922
923 // Save the original sign bit for later.
924 bool Negative = isNegative();
925
926 // WordShift is the inter-part shift; BitShift is intra-part shift.
927 unsigned WordShift = ShiftAmt / APINT_BITS_PER_WORD;
928 unsigned BitShift = ShiftAmt % APINT_BITS_PER_WORD;
929
930 unsigned WordsToMove = getNumWords() - WordShift;
931 if (WordsToMove != 0) {
932 // Sign extend the last word to fill in the unused bits.
933 U.pVal[getNumWords() - 1] = SignExtend64(
934 U.pVal[getNumWords() - 1], ((BitWidth - 1) % APINT_BITS_PER_WORD) + 1);
935
936 // Fastpath for moving by whole words.
937 if (BitShift == 0) {
938 std::memmove(U.pVal, U.pVal + WordShift, WordsToMove * APINT_WORD_SIZE);
939 } else {
940 // Move the words containing significant bits.
941 for (unsigned i = 0; i != WordsToMove - 1; ++i)
942 U.pVal[i] = (U.pVal[i + WordShift] >> BitShift) |
943 (U.pVal[i + WordShift + 1] << (APINT_BITS_PER_WORD - BitShift));
944
945 // Handle the last word which has no high bits to copy.
946 U.pVal[WordsToMove - 1] = U.pVal[WordShift + WordsToMove - 1] >> BitShift;
947 // Sign extend one more time.
948 U.pVal[WordsToMove - 1] =
949 SignExtend64(U.pVal[WordsToMove - 1], APINT_BITS_PER_WORD - BitShift);
950 }
951 }
952
953 // Fill in the remainder based on the original sign.
954 std::memset(U.pVal + WordsToMove, Negative ? -1 : 0,
955 WordShift * APINT_WORD_SIZE);
956 clearUnusedBits();
957 }
958
959 /// Logical right-shift this APInt by shiftAmt.
960 /// Logical right-shift function.
lshrInPlace(const APInt & shiftAmt)961 void APInt::lshrInPlace(const APInt &shiftAmt) {
962 lshrInPlace((unsigned)shiftAmt.getLimitedValue(BitWidth));
963 }
964
965 /// Logical right-shift this APInt by shiftAmt.
966 /// Logical right-shift function.
lshrSlowCase(unsigned ShiftAmt)967 void APInt::lshrSlowCase(unsigned ShiftAmt) {
968 tcShiftRight(U.pVal, getNumWords(), ShiftAmt);
969 }
970
971 /// Left-shift this APInt by shiftAmt.
972 /// Left-shift function.
operator <<=(const APInt & shiftAmt)973 APInt &APInt::operator<<=(const APInt &shiftAmt) {
974 // It's undefined behavior in C to shift by BitWidth or greater.
975 *this <<= (unsigned)shiftAmt.getLimitedValue(BitWidth);
976 return *this;
977 }
978
shlSlowCase(unsigned ShiftAmt)979 void APInt::shlSlowCase(unsigned ShiftAmt) {
980 tcShiftLeft(U.pVal, getNumWords(), ShiftAmt);
981 clearUnusedBits();
982 }
983
984 // Calculate the rotate amount modulo the bit width.
rotateModulo(unsigned BitWidth,const APInt & rotateAmt)985 static unsigned rotateModulo(unsigned BitWidth, const APInt &rotateAmt) {
986 unsigned rotBitWidth = rotateAmt.getBitWidth();
987 APInt rot = rotateAmt;
988 if (rotBitWidth < BitWidth) {
989 // Extend the rotate APInt, so that the urem doesn't divide by 0.
990 // e.g. APInt(1, 32) would give APInt(1, 0).
991 rot = rotateAmt.zext(BitWidth);
992 }
993 rot = rot.urem(APInt(rot.getBitWidth(), BitWidth));
994 return rot.getLimitedValue(BitWidth);
995 }
996
rotl(const APInt & rotateAmt) const997 APInt APInt::rotl(const APInt &rotateAmt) const {
998 return rotl(rotateModulo(BitWidth, rotateAmt));
999 }
1000
rotl(unsigned rotateAmt) const1001 APInt APInt::rotl(unsigned rotateAmt) const {
1002 rotateAmt %= BitWidth;
1003 if (rotateAmt == 0)
1004 return *this;
1005 return shl(rotateAmt) | lshr(BitWidth - rotateAmt);
1006 }
1007
rotr(const APInt & rotateAmt) const1008 APInt APInt::rotr(const APInt &rotateAmt) const {
1009 return rotr(rotateModulo(BitWidth, rotateAmt));
1010 }
1011
rotr(unsigned rotateAmt) const1012 APInt APInt::rotr(unsigned rotateAmt) const {
1013 rotateAmt %= BitWidth;
1014 if (rotateAmt == 0)
1015 return *this;
1016 return lshr(rotateAmt) | shl(BitWidth - rotateAmt);
1017 }
1018
1019 // Square Root - this method computes and returns the square root of "this".
1020 // Three mechanisms are used for computation. For small values (<= 5 bits),
1021 // a table lookup is done. This gets some performance for common cases. For
1022 // values using less than 52 bits, the value is converted to double and then
1023 // the libc sqrt function is called. The result is rounded and then converted
1024 // back to a uint64_t which is then used to construct the result. Finally,
1025 // the Babylonian method for computing square roots is used.
sqrt() const1026 APInt APInt::sqrt() const {
1027
1028 // Determine the magnitude of the value.
1029 unsigned magnitude = getActiveBits();
1030
1031 // Use a fast table for some small values. This also gets rid of some
1032 // rounding errors in libc sqrt for small values.
1033 if (magnitude <= 5) {
1034 static const uint8_t results[32] = {
1035 /* 0 */ 0,
1036 /* 1- 2 */ 1, 1,
1037 /* 3- 6 */ 2, 2, 2, 2,
1038 /* 7-12 */ 3, 3, 3, 3, 3, 3,
1039 /* 13-20 */ 4, 4, 4, 4, 4, 4, 4, 4,
1040 /* 21-30 */ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
1041 /* 31 */ 6
1042 };
1043 return APInt(BitWidth, results[ (isSingleWord() ? U.VAL : U.pVal[0]) ]);
1044 }
1045
1046 // If the magnitude of the value fits in less than 52 bits (the precision of
1047 // an IEEE double precision floating point value), then we can use the
1048 // libc sqrt function which will probably use a hardware sqrt computation.
1049 // This should be faster than the algorithm below.
1050 if (magnitude < 52) {
1051 return APInt(BitWidth,
1052 uint64_t(::round(::sqrt(double(isSingleWord() ? U.VAL
1053 : U.pVal[0])))));
1054 }
1055
1056 // Okay, all the short cuts are exhausted. We must compute it. The following
1057 // is a classical Babylonian method for computing the square root. This code
1058 // was adapted to APInt from a wikipedia article on such computations.
1059 // See http://www.wikipedia.org/ and go to the page named
1060 // Calculate_an_integer_square_root.
1061 unsigned nbits = BitWidth, i = 4;
1062 APInt testy(BitWidth, 16);
1063 APInt x_old(BitWidth, 1);
1064 APInt x_new(BitWidth, 0);
1065 APInt two(BitWidth, 2);
1066
1067 // Select a good starting value using binary logarithms.
1068 for (;; i += 2, testy = testy.shl(2))
1069 if (i >= nbits || this->ule(testy)) {
1070 x_old = x_old.shl(i / 2);
1071 break;
1072 }
1073
1074 // Use the Babylonian method to arrive at the integer square root:
1075 for (;;) {
1076 x_new = (this->udiv(x_old) + x_old).udiv(two);
1077 if (x_old.ule(x_new))
1078 break;
1079 x_old = x_new;
1080 }
1081
1082 // Make sure we return the closest approximation
1083 // NOTE: The rounding calculation below is correct. It will produce an
1084 // off-by-one discrepancy with results from pari/gp. That discrepancy has been
1085 // determined to be a rounding issue with pari/gp as it begins to use a
1086 // floating point representation after 192 bits. There are no discrepancies
1087 // between this algorithm and pari/gp for bit widths < 192 bits.
1088 APInt square(x_old * x_old);
1089 APInt nextSquare((x_old + 1) * (x_old +1));
1090 if (this->ult(square))
1091 return x_old;
1092 assert(this->ule(nextSquare) && "Error in APInt::sqrt computation");
1093 APInt midpoint((nextSquare - square).udiv(two));
1094 APInt offset(*this - square);
1095 if (offset.ult(midpoint))
1096 return x_old;
1097 return x_old + 1;
1098 }
1099
1100 /// Computes the multiplicative inverse of this APInt for a given modulo. The
1101 /// iterative extended Euclidean algorithm is used to solve for this value,
1102 /// however we simplify it to speed up calculating only the inverse, and take
1103 /// advantage of div+rem calculations. We also use some tricks to avoid copying
1104 /// (potentially large) APInts around.
multiplicativeInverse(const APInt & modulo) const1105 APInt APInt::multiplicativeInverse(const APInt& modulo) const {
1106 assert(ult(modulo) && "This APInt must be smaller than the modulo");
1107
1108 // Using the properties listed at the following web page (accessed 06/21/08):
1109 // http://www.numbertheory.org/php/euclid.html
1110 // (especially the properties numbered 3, 4 and 9) it can be proved that
1111 // BitWidth bits suffice for all the computations in the algorithm implemented
1112 // below. More precisely, this number of bits suffice if the multiplicative
1113 // inverse exists, but may not suffice for the general extended Euclidean
1114 // algorithm.
1115
1116 APInt r[2] = { modulo, *this };
1117 APInt t[2] = { APInt(BitWidth, 0), APInt(BitWidth, 1) };
1118 APInt q(BitWidth, 0);
1119
1120 unsigned i;
1121 for (i = 0; r[i^1] != 0; i ^= 1) {
1122 // An overview of the math without the confusing bit-flipping:
1123 // q = r[i-2] / r[i-1]
1124 // r[i] = r[i-2] % r[i-1]
1125 // t[i] = t[i-2] - t[i-1] * q
1126 udivrem(r[i], r[i^1], q, r[i]);
1127 t[i] -= t[i^1] * q;
1128 }
1129
1130 // If this APInt and the modulo are not coprime, there is no multiplicative
1131 // inverse, so return 0. We check this by looking at the next-to-last
1132 // remainder, which is the gcd(*this,modulo) as calculated by the Euclidean
1133 // algorithm.
1134 if (r[i] != 1)
1135 return APInt(BitWidth, 0);
1136
1137 // The next-to-last t is the multiplicative inverse. However, we are
1138 // interested in a positive inverse. Calculate a positive one from a negative
1139 // one if necessary. A simple addition of the modulo suffices because
1140 // abs(t[i]) is known to be less than *this/2 (see the link above).
1141 if (t[i].isNegative())
1142 t[i] += modulo;
1143
1144 return std::move(t[i]);
1145 }
1146
1147 /// Calculate the magic numbers required to implement a signed integer division
1148 /// by a constant as a sequence of multiplies, adds and shifts. Requires that
1149 /// the divisor not be 0, 1, or -1. Taken from "Hacker's Delight", Henry S.
1150 /// Warren, Jr., chapter 10.
magic() const1151 APInt::ms APInt::magic() const {
1152 const APInt& d = *this;
1153 unsigned p;
1154 APInt ad, anc, delta, q1, r1, q2, r2, t;
1155 APInt signedMin = APInt::getSignedMinValue(d.getBitWidth());
1156 struct ms mag;
1157
1158 ad = d.abs();
1159 t = signedMin + (d.lshr(d.getBitWidth() - 1));
1160 anc = t - 1 - t.urem(ad); // absolute value of nc
1161 p = d.getBitWidth() - 1; // initialize p
1162 q1 = signedMin.udiv(anc); // initialize q1 = 2p/abs(nc)
1163 r1 = signedMin - q1*anc; // initialize r1 = rem(2p,abs(nc))
1164 q2 = signedMin.udiv(ad); // initialize q2 = 2p/abs(d)
1165 r2 = signedMin - q2*ad; // initialize r2 = rem(2p,abs(d))
1166 do {
1167 p = p + 1;
1168 q1 = q1<<1; // update q1 = 2p/abs(nc)
1169 r1 = r1<<1; // update r1 = rem(2p/abs(nc))
1170 if (r1.uge(anc)) { // must be unsigned comparison
1171 q1 = q1 + 1;
1172 r1 = r1 - anc;
1173 }
1174 q2 = q2<<1; // update q2 = 2p/abs(d)
1175 r2 = r2<<1; // update r2 = rem(2p/abs(d))
1176 if (r2.uge(ad)) { // must be unsigned comparison
1177 q2 = q2 + 1;
1178 r2 = r2 - ad;
1179 }
1180 delta = ad - r2;
1181 } while (q1.ult(delta) || (q1 == delta && r1 == 0));
1182
1183 mag.m = q2 + 1;
1184 if (d.isNegative()) mag.m = -mag.m; // resulting magic number
1185 mag.s = p - d.getBitWidth(); // resulting shift
1186 return mag;
1187 }
1188
1189 /// Calculate the magic numbers required to implement an unsigned integer
1190 /// division by a constant as a sequence of multiplies, adds and shifts.
1191 /// Requires that the divisor not be 0. Taken from "Hacker's Delight", Henry
1192 /// S. Warren, Jr., chapter 10.
1193 /// LeadingZeros can be used to simplify the calculation if the upper bits
1194 /// of the divided value are known zero.
magicu(unsigned LeadingZeros) const1195 APInt::mu APInt::magicu(unsigned LeadingZeros) const {
1196 const APInt& d = *this;
1197 unsigned p;
1198 APInt nc, delta, q1, r1, q2, r2;
1199 struct mu magu;
1200 magu.a = 0; // initialize "add" indicator
1201 APInt allOnes = APInt::getAllOnesValue(d.getBitWidth()).lshr(LeadingZeros);
1202 APInt signedMin = APInt::getSignedMinValue(d.getBitWidth());
1203 APInt signedMax = APInt::getSignedMaxValue(d.getBitWidth());
1204
1205 nc = allOnes - (allOnes - d).urem(d);
1206 p = d.getBitWidth() - 1; // initialize p
1207 q1 = signedMin.udiv(nc); // initialize q1 = 2p/nc
1208 r1 = signedMin - q1*nc; // initialize r1 = rem(2p,nc)
1209 q2 = signedMax.udiv(d); // initialize q2 = (2p-1)/d
1210 r2 = signedMax - q2*d; // initialize r2 = rem((2p-1),d)
1211 do {
1212 p = p + 1;
1213 if (r1.uge(nc - r1)) {
1214 q1 = q1 + q1 + 1; // update q1
1215 r1 = r1 + r1 - nc; // update r1
1216 }
1217 else {
1218 q1 = q1+q1; // update q1
1219 r1 = r1+r1; // update r1
1220 }
1221 if ((r2 + 1).uge(d - r2)) {
1222 if (q2.uge(signedMax)) magu.a = 1;
1223 q2 = q2+q2 + 1; // update q2
1224 r2 = r2+r2 + 1 - d; // update r2
1225 }
1226 else {
1227 if (q2.uge(signedMin)) magu.a = 1;
1228 q2 = q2+q2; // update q2
1229 r2 = r2+r2 + 1; // update r2
1230 }
1231 delta = d - 1 - r2;
1232 } while (p < d.getBitWidth()*2 &&
1233 (q1.ult(delta) || (q1 == delta && r1 == 0)));
1234 magu.m = q2 + 1; // resulting magic number
1235 magu.s = p - d.getBitWidth(); // resulting shift
1236 return magu;
1237 }
1238
1239 /// Implementation of Knuth's Algorithm D (Division of nonnegative integers)
1240 /// from "Art of Computer Programming, Volume 2", section 4.3.1, p. 272. The
1241 /// variables here have the same names as in the algorithm. Comments explain
1242 /// the algorithm and any deviation from it.
KnuthDiv(uint32_t * u,uint32_t * v,uint32_t * q,uint32_t * r,unsigned m,unsigned n)1243 static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r,
1244 unsigned m, unsigned n) {
1245 assert(u && "Must provide dividend");
1246 assert(v && "Must provide divisor");
1247 assert(q && "Must provide quotient");
1248 assert(u != v && u != q && v != q && "Must use different memory");
1249 assert(n>1 && "n must be > 1");
1250
1251 // b denotes the base of the number system. In our case b is 2^32.
1252 const uint64_t b = uint64_t(1) << 32;
1253
1254 // The DEBUG macros here tend to be spam in the debug output if you're not
1255 // debugging this code. Disable them unless KNUTH_DEBUG is defined.
1256 #pragma push_macro("LLVM_DEBUG")
1257 #ifndef KNUTH_DEBUG
1258 #undef LLVM_DEBUG
1259 #define LLVM_DEBUG(X) \
1260 do { \
1261 } while (false)
1262 #endif
1263
1264 LLVM_DEBUG(dbgs() << "KnuthDiv: m=" << m << " n=" << n << '\n');
1265 LLVM_DEBUG(dbgs() << "KnuthDiv: original:");
1266 LLVM_DEBUG(for (int i = m + n; i >= 0; i--) dbgs() << " " << u[i]);
1267 LLVM_DEBUG(dbgs() << " by");
1268 LLVM_DEBUG(for (int i = n; i > 0; i--) dbgs() << " " << v[i - 1]);
1269 LLVM_DEBUG(dbgs() << '\n');
1270 // D1. [Normalize.] Set d = b / (v[n-1] + 1) and multiply all the digits of
1271 // u and v by d. Note that we have taken Knuth's advice here to use a power
1272 // of 2 value for d such that d * v[n-1] >= b/2 (b is the base). A power of
1273 // 2 allows us to shift instead of multiply and it is easy to determine the
1274 // shift amount from the leading zeros. We are basically normalizing the u
1275 // and v so that its high bits are shifted to the top of v's range without
1276 // overflow. Note that this can require an extra word in u so that u must
1277 // be of length m+n+1.
1278 unsigned shift = countLeadingZeros(v[n-1]);
1279 uint32_t v_carry = 0;
1280 uint32_t u_carry = 0;
1281 if (shift) {
1282 for (unsigned i = 0; i < m+n; ++i) {
1283 uint32_t u_tmp = u[i] >> (32 - shift);
1284 u[i] = (u[i] << shift) | u_carry;
1285 u_carry = u_tmp;
1286 }
1287 for (unsigned i = 0; i < n; ++i) {
1288 uint32_t v_tmp = v[i] >> (32 - shift);
1289 v[i] = (v[i] << shift) | v_carry;
1290 v_carry = v_tmp;
1291 }
1292 }
1293 u[m+n] = u_carry;
1294
1295 LLVM_DEBUG(dbgs() << "KnuthDiv: normal:");
1296 LLVM_DEBUG(for (int i = m + n; i >= 0; i--) dbgs() << " " << u[i]);
1297 LLVM_DEBUG(dbgs() << " by");
1298 LLVM_DEBUG(for (int i = n; i > 0; i--) dbgs() << " " << v[i - 1]);
1299 LLVM_DEBUG(dbgs() << '\n');
1300
1301 // D2. [Initialize j.] Set j to m. This is the loop counter over the places.
1302 int j = m;
1303 do {
1304 LLVM_DEBUG(dbgs() << "KnuthDiv: quotient digit #" << j << '\n');
1305 // D3. [Calculate q'.].
1306 // Set qp = (u[j+n]*b + u[j+n-1]) / v[n-1]. (qp=qprime=q')
1307 // Set rp = (u[j+n]*b + u[j+n-1]) % v[n-1]. (rp=rprime=r')
1308 // Now test if qp == b or qp*v[n-2] > b*rp + u[j+n-2]; if so, decrease
1309 // qp by 1, increase rp by v[n-1], and repeat this test if rp < b. The test
1310 // on v[n-2] determines at high speed most of the cases in which the trial
1311 // value qp is one too large, and it eliminates all cases where qp is two
1312 // too large.
1313 uint64_t dividend = Make_64(u[j+n], u[j+n-1]);
1314 LLVM_DEBUG(dbgs() << "KnuthDiv: dividend == " << dividend << '\n');
1315 uint64_t qp = dividend / v[n-1];
1316 uint64_t rp = dividend % v[n-1];
1317 if (qp == b || qp*v[n-2] > b*rp + u[j+n-2]) {
1318 qp--;
1319 rp += v[n-1];
1320 if (rp < b && (qp == b || qp*v[n-2] > b*rp + u[j+n-2]))
1321 qp--;
1322 }
1323 LLVM_DEBUG(dbgs() << "KnuthDiv: qp == " << qp << ", rp == " << rp << '\n');
1324
1325 // D4. [Multiply and subtract.] Replace (u[j+n]u[j+n-1]...u[j]) with
1326 // (u[j+n]u[j+n-1]..u[j]) - qp * (v[n-1]...v[1]v[0]). This computation
1327 // consists of a simple multiplication by a one-place number, combined with
1328 // a subtraction.
1329 // The digits (u[j+n]...u[j]) should be kept positive; if the result of
1330 // this step is actually negative, (u[j+n]...u[j]) should be left as the
1331 // true value plus b**(n+1), namely as the b's complement of
1332 // the true value, and a "borrow" to the left should be remembered.
1333 int64_t borrow = 0;
1334 for (unsigned i = 0; i < n; ++i) {
1335 uint64_t p = uint64_t(qp) * uint64_t(v[i]);
1336 int64_t subres = int64_t(u[j+i]) - borrow - Lo_32(p);
1337 u[j+i] = Lo_32(subres);
1338 borrow = Hi_32(p) - Hi_32(subres);
1339 LLVM_DEBUG(dbgs() << "KnuthDiv: u[j+i] = " << u[j + i]
1340 << ", borrow = " << borrow << '\n');
1341 }
1342 bool isNeg = u[j+n] < borrow;
1343 u[j+n] -= Lo_32(borrow);
1344
1345 LLVM_DEBUG(dbgs() << "KnuthDiv: after subtraction:");
1346 LLVM_DEBUG(for (int i = m + n; i >= 0; i--) dbgs() << " " << u[i]);
1347 LLVM_DEBUG(dbgs() << '\n');
1348
1349 // D5. [Test remainder.] Set q[j] = qp. If the result of step D4 was
1350 // negative, go to step D6; otherwise go on to step D7.
1351 q[j] = Lo_32(qp);
1352 if (isNeg) {
1353 // D6. [Add back]. The probability that this step is necessary is very
1354 // small, on the order of only 2/b. Make sure that test data accounts for
1355 // this possibility. Decrease q[j] by 1
1356 q[j]--;
1357 // and add (0v[n-1]...v[1]v[0]) to (u[j+n]u[j+n-1]...u[j+1]u[j]).
1358 // A carry will occur to the left of u[j+n], and it should be ignored
1359 // since it cancels with the borrow that occurred in D4.
1360 bool carry = false;
1361 for (unsigned i = 0; i < n; i++) {
1362 uint32_t limit = std::min(u[j+i],v[i]);
1363 u[j+i] += v[i] + carry;
1364 carry = u[j+i] < limit || (carry && u[j+i] == limit);
1365 }
1366 u[j+n] += carry;
1367 }
1368 LLVM_DEBUG(dbgs() << "KnuthDiv: after correction:");
1369 LLVM_DEBUG(for (int i = m + n; i >= 0; i--) dbgs() << " " << u[i]);
1370 LLVM_DEBUG(dbgs() << "\nKnuthDiv: digit result = " << q[j] << '\n');
1371
1372 // D7. [Loop on j.] Decrease j by one. Now if j >= 0, go back to D3.
1373 } while (--j >= 0);
1374
1375 LLVM_DEBUG(dbgs() << "KnuthDiv: quotient:");
1376 LLVM_DEBUG(for (int i = m; i >= 0; i--) dbgs() << " " << q[i]);
1377 LLVM_DEBUG(dbgs() << '\n');
1378
1379 // D8. [Unnormalize]. Now q[...] is the desired quotient, and the desired
1380 // remainder may be obtained by dividing u[...] by d. If r is non-null we
1381 // compute the remainder (urem uses this).
1382 if (r) {
1383 // The value d is expressed by the "shift" value above since we avoided
1384 // multiplication by d by using a shift left. So, all we have to do is
1385 // shift right here.
1386 if (shift) {
1387 uint32_t carry = 0;
1388 LLVM_DEBUG(dbgs() << "KnuthDiv: remainder:");
1389 for (int i = n-1; i >= 0; i--) {
1390 r[i] = (u[i] >> shift) | carry;
1391 carry = u[i] << (32 - shift);
1392 LLVM_DEBUG(dbgs() << " " << r[i]);
1393 }
1394 } else {
1395 for (int i = n-1; i >= 0; i--) {
1396 r[i] = u[i];
1397 LLVM_DEBUG(dbgs() << " " << r[i]);
1398 }
1399 }
1400 LLVM_DEBUG(dbgs() << '\n');
1401 }
1402 LLVM_DEBUG(dbgs() << '\n');
1403
1404 #pragma pop_macro("LLVM_DEBUG")
1405 }
1406
divide(const WordType * LHS,unsigned lhsWords,const WordType * RHS,unsigned rhsWords,WordType * Quotient,WordType * Remainder)1407 void APInt::divide(const WordType *LHS, unsigned lhsWords, const WordType *RHS,
1408 unsigned rhsWords, WordType *Quotient, WordType *Remainder) {
1409 assert(lhsWords >= rhsWords && "Fractional result");
1410
1411 // First, compose the values into an array of 32-bit words instead of
1412 // 64-bit words. This is a necessity of both the "short division" algorithm
1413 // and the Knuth "classical algorithm" which requires there to be native
1414 // operations for +, -, and * on an m bit value with an m*2 bit result. We
1415 // can't use 64-bit operands here because we don't have native results of
1416 // 128-bits. Furthermore, casting the 64-bit values to 32-bit values won't
1417 // work on large-endian machines.
1418 unsigned n = rhsWords * 2;
1419 unsigned m = (lhsWords * 2) - n;
1420
1421 // Allocate space for the temporary values we need either on the stack, if
1422 // it will fit, or on the heap if it won't.
1423 uint32_t SPACE[128];
1424 uint32_t *U = nullptr;
1425 uint32_t *V = nullptr;
1426 uint32_t *Q = nullptr;
1427 uint32_t *R = nullptr;
1428 if ((Remainder?4:3)*n+2*m+1 <= 128) {
1429 U = &SPACE[0];
1430 V = &SPACE[m+n+1];
1431 Q = &SPACE[(m+n+1) + n];
1432 if (Remainder)
1433 R = &SPACE[(m+n+1) + n + (m+n)];
1434 } else {
1435 U = new uint32_t[m + n + 1];
1436 V = new uint32_t[n];
1437 Q = new uint32_t[m+n];
1438 if (Remainder)
1439 R = new uint32_t[n];
1440 }
1441
1442 // Initialize the dividend
1443 memset(U, 0, (m+n+1)*sizeof(uint32_t));
1444 for (unsigned i = 0; i < lhsWords; ++i) {
1445 uint64_t tmp = LHS[i];
1446 U[i * 2] = Lo_32(tmp);
1447 U[i * 2 + 1] = Hi_32(tmp);
1448 }
1449 U[m+n] = 0; // this extra word is for "spill" in the Knuth algorithm.
1450
1451 // Initialize the divisor
1452 memset(V, 0, (n)*sizeof(uint32_t));
1453 for (unsigned i = 0; i < rhsWords; ++i) {
1454 uint64_t tmp = RHS[i];
1455 V[i * 2] = Lo_32(tmp);
1456 V[i * 2 + 1] = Hi_32(tmp);
1457 }
1458
1459 // initialize the quotient and remainder
1460 memset(Q, 0, (m+n) * sizeof(uint32_t));
1461 if (Remainder)
1462 memset(R, 0, n * sizeof(uint32_t));
1463
1464 // Now, adjust m and n for the Knuth division. n is the number of words in
1465 // the divisor. m is the number of words by which the dividend exceeds the
1466 // divisor (i.e. m+n is the length of the dividend). These sizes must not
1467 // contain any zero words or the Knuth algorithm fails.
1468 for (unsigned i = n; i > 0 && V[i-1] == 0; i--) {
1469 n--;
1470 m++;
1471 }
1472 for (unsigned i = m+n; i > 0 && U[i-1] == 0; i--)
1473 m--;
1474
1475 // If we're left with only a single word for the divisor, Knuth doesn't work
1476 // so we implement the short division algorithm here. This is much simpler
1477 // and faster because we are certain that we can divide a 64-bit quantity
1478 // by a 32-bit quantity at hardware speed and short division is simply a
1479 // series of such operations. This is just like doing short division but we
1480 // are using base 2^32 instead of base 10.
1481 assert(n != 0 && "Divide by zero?");
1482 if (n == 1) {
1483 uint32_t divisor = V[0];
1484 uint32_t remainder = 0;
1485 for (int i = m; i >= 0; i--) {
1486 uint64_t partial_dividend = Make_64(remainder, U[i]);
1487 if (partial_dividend == 0) {
1488 Q[i] = 0;
1489 remainder = 0;
1490 } else if (partial_dividend < divisor) {
1491 Q[i] = 0;
1492 remainder = Lo_32(partial_dividend);
1493 } else if (partial_dividend == divisor) {
1494 Q[i] = 1;
1495 remainder = 0;
1496 } else {
1497 Q[i] = Lo_32(partial_dividend / divisor);
1498 remainder = Lo_32(partial_dividend - (Q[i] * divisor));
1499 }
1500 }
1501 if (R)
1502 R[0] = remainder;
1503 } else {
1504 // Now we're ready to invoke the Knuth classical divide algorithm. In this
1505 // case n > 1.
1506 KnuthDiv(U, V, Q, R, m, n);
1507 }
1508
1509 // If the caller wants the quotient
1510 if (Quotient) {
1511 for (unsigned i = 0; i < lhsWords; ++i)
1512 Quotient[i] = Make_64(Q[i*2+1], Q[i*2]);
1513 }
1514
1515 // If the caller wants the remainder
1516 if (Remainder) {
1517 for (unsigned i = 0; i < rhsWords; ++i)
1518 Remainder[i] = Make_64(R[i*2+1], R[i*2]);
1519 }
1520
1521 // Clean up the memory we allocated.
1522 if (U != &SPACE[0]) {
1523 delete [] U;
1524 delete [] V;
1525 delete [] Q;
1526 delete [] R;
1527 }
1528 }
1529
udiv(const APInt & RHS) const1530 APInt APInt::udiv(const APInt &RHS) const {
1531 assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
1532
1533 // First, deal with the easy case
1534 if (isSingleWord()) {
1535 assert(RHS.U.VAL != 0 && "Divide by zero?");
1536 return APInt(BitWidth, U.VAL / RHS.U.VAL);
1537 }
1538
1539 // Get some facts about the LHS and RHS number of bits and words
1540 unsigned lhsWords = getNumWords(getActiveBits());
1541 unsigned rhsBits = RHS.getActiveBits();
1542 unsigned rhsWords = getNumWords(rhsBits);
1543 assert(rhsWords && "Divided by zero???");
1544
1545 // Deal with some degenerate cases
1546 if (!lhsWords)
1547 // 0 / X ===> 0
1548 return APInt(BitWidth, 0);
1549 if (rhsBits == 1)
1550 // X / 1 ===> X
1551 return *this;
1552 if (lhsWords < rhsWords || this->ult(RHS))
1553 // X / Y ===> 0, iff X < Y
1554 return APInt(BitWidth, 0);
1555 if (*this == RHS)
1556 // X / X ===> 1
1557 return APInt(BitWidth, 1);
1558 if (lhsWords == 1) // rhsWords is 1 if lhsWords is 1.
1559 // All high words are zero, just use native divide
1560 return APInt(BitWidth, this->U.pVal[0] / RHS.U.pVal[0]);
1561
1562 // We have to compute it the hard way. Invoke the Knuth divide algorithm.
1563 APInt Quotient(BitWidth, 0); // to hold result.
1564 divide(U.pVal, lhsWords, RHS.U.pVal, rhsWords, Quotient.U.pVal, nullptr);
1565 return Quotient;
1566 }
1567
udiv(uint64_t RHS) const1568 APInt APInt::udiv(uint64_t RHS) const {
1569 assert(RHS != 0 && "Divide by zero?");
1570
1571 // First, deal with the easy case
1572 if (isSingleWord())
1573 return APInt(BitWidth, U.VAL / RHS);
1574
1575 // Get some facts about the LHS words.
1576 unsigned lhsWords = getNumWords(getActiveBits());
1577
1578 // Deal with some degenerate cases
1579 if (!lhsWords)
1580 // 0 / X ===> 0
1581 return APInt(BitWidth, 0);
1582 if (RHS == 1)
1583 // X / 1 ===> X
1584 return *this;
1585 if (this->ult(RHS))
1586 // X / Y ===> 0, iff X < Y
1587 return APInt(BitWidth, 0);
1588 if (*this == RHS)
1589 // X / X ===> 1
1590 return APInt(BitWidth, 1);
1591 if (lhsWords == 1) // rhsWords is 1 if lhsWords is 1.
1592 // All high words are zero, just use native divide
1593 return APInt(BitWidth, this->U.pVal[0] / RHS);
1594
1595 // We have to compute it the hard way. Invoke the Knuth divide algorithm.
1596 APInt Quotient(BitWidth, 0); // to hold result.
1597 divide(U.pVal, lhsWords, &RHS, 1, Quotient.U.pVal, nullptr);
1598 return Quotient;
1599 }
1600
sdiv(const APInt & RHS) const1601 APInt APInt::sdiv(const APInt &RHS) const {
1602 if (isNegative()) {
1603 if (RHS.isNegative())
1604 return (-(*this)).udiv(-RHS);
1605 return -((-(*this)).udiv(RHS));
1606 }
1607 if (RHS.isNegative())
1608 return -(this->udiv(-RHS));
1609 return this->udiv(RHS);
1610 }
1611
sdiv(int64_t RHS) const1612 APInt APInt::sdiv(int64_t RHS) const {
1613 if (isNegative()) {
1614 if (RHS < 0)
1615 return (-(*this)).udiv(-RHS);
1616 return -((-(*this)).udiv(RHS));
1617 }
1618 if (RHS < 0)
1619 return -(this->udiv(-RHS));
1620 return this->udiv(RHS);
1621 }
1622
urem(const APInt & RHS) const1623 APInt APInt::urem(const APInt &RHS) const {
1624 assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
1625 if (isSingleWord()) {
1626 assert(RHS.U.VAL != 0 && "Remainder by zero?");
1627 return APInt(BitWidth, U.VAL % RHS.U.VAL);
1628 }
1629
1630 // Get some facts about the LHS
1631 unsigned lhsWords = getNumWords(getActiveBits());
1632
1633 // Get some facts about the RHS
1634 unsigned rhsBits = RHS.getActiveBits();
1635 unsigned rhsWords = getNumWords(rhsBits);
1636 assert(rhsWords && "Performing remainder operation by zero ???");
1637
1638 // Check the degenerate cases
1639 if (lhsWords == 0)
1640 // 0 % Y ===> 0
1641 return APInt(BitWidth, 0);
1642 if (rhsBits == 1)
1643 // X % 1 ===> 0
1644 return APInt(BitWidth, 0);
1645 if (lhsWords < rhsWords || this->ult(RHS))
1646 // X % Y ===> X, iff X < Y
1647 return *this;
1648 if (*this == RHS)
1649 // X % X == 0;
1650 return APInt(BitWidth, 0);
1651 if (lhsWords == 1)
1652 // All high words are zero, just use native remainder
1653 return APInt(BitWidth, U.pVal[0] % RHS.U.pVal[0]);
1654
1655 // We have to compute it the hard way. Invoke the Knuth divide algorithm.
1656 APInt Remainder(BitWidth, 0);
1657 divide(U.pVal, lhsWords, RHS.U.pVal, rhsWords, nullptr, Remainder.U.pVal);
1658 return Remainder;
1659 }
1660
urem(uint64_t RHS) const1661 uint64_t APInt::urem(uint64_t RHS) const {
1662 assert(RHS != 0 && "Remainder by zero?");
1663
1664 if (isSingleWord())
1665 return U.VAL % RHS;
1666
1667 // Get some facts about the LHS
1668 unsigned lhsWords = getNumWords(getActiveBits());
1669
1670 // Check the degenerate cases
1671 if (lhsWords == 0)
1672 // 0 % Y ===> 0
1673 return 0;
1674 if (RHS == 1)
1675 // X % 1 ===> 0
1676 return 0;
1677 if (this->ult(RHS))
1678 // X % Y ===> X, iff X < Y
1679 return getZExtValue();
1680 if (*this == RHS)
1681 // X % X == 0;
1682 return 0;
1683 if (lhsWords == 1)
1684 // All high words are zero, just use native remainder
1685 return U.pVal[0] % RHS;
1686
1687 // We have to compute it the hard way. Invoke the Knuth divide algorithm.
1688 uint64_t Remainder;
1689 divide(U.pVal, lhsWords, &RHS, 1, nullptr, &Remainder);
1690 return Remainder;
1691 }
1692
srem(const APInt & RHS) const1693 APInt APInt::srem(const APInt &RHS) const {
1694 if (isNegative()) {
1695 if (RHS.isNegative())
1696 return -((-(*this)).urem(-RHS));
1697 return -((-(*this)).urem(RHS));
1698 }
1699 if (RHS.isNegative())
1700 return this->urem(-RHS);
1701 return this->urem(RHS);
1702 }
1703
srem(int64_t RHS) const1704 int64_t APInt::srem(int64_t RHS) const {
1705 if (isNegative()) {
1706 if (RHS < 0)
1707 return -((-(*this)).urem(-RHS));
1708 return -((-(*this)).urem(RHS));
1709 }
1710 if (RHS < 0)
1711 return this->urem(-RHS);
1712 return this->urem(RHS);
1713 }
1714
udivrem(const APInt & LHS,const APInt & RHS,APInt & Quotient,APInt & Remainder)1715 void APInt::udivrem(const APInt &LHS, const APInt &RHS,
1716 APInt &Quotient, APInt &Remainder) {
1717 assert(LHS.BitWidth == RHS.BitWidth && "Bit widths must be the same");
1718 unsigned BitWidth = LHS.BitWidth;
1719
1720 // First, deal with the easy case
1721 if (LHS.isSingleWord()) {
1722 assert(RHS.U.VAL != 0 && "Divide by zero?");
1723 uint64_t QuotVal = LHS.U.VAL / RHS.U.VAL;
1724 uint64_t RemVal = LHS.U.VAL % RHS.U.VAL;
1725 Quotient = APInt(BitWidth, QuotVal);
1726 Remainder = APInt(BitWidth, RemVal);
1727 return;
1728 }
1729
1730 // Get some size facts about the dividend and divisor
1731 unsigned lhsWords = getNumWords(LHS.getActiveBits());
1732 unsigned rhsBits = RHS.getActiveBits();
1733 unsigned rhsWords = getNumWords(rhsBits);
1734 assert(rhsWords && "Performing divrem operation by zero ???");
1735
1736 // Check the degenerate cases
1737 if (lhsWords == 0) {
1738 Quotient = APInt(BitWidth, 0); // 0 / Y ===> 0
1739 Remainder = APInt(BitWidth, 0); // 0 % Y ===> 0
1740 return;
1741 }
1742
1743 if (rhsBits == 1) {
1744 Quotient = LHS; // X / 1 ===> X
1745 Remainder = APInt(BitWidth, 0); // X % 1 ===> 0
1746 }
1747
1748 if (lhsWords < rhsWords || LHS.ult(RHS)) {
1749 Remainder = LHS; // X % Y ===> X, iff X < Y
1750 Quotient = APInt(BitWidth, 0); // X / Y ===> 0, iff X < Y
1751 return;
1752 }
1753
1754 if (LHS == RHS) {
1755 Quotient = APInt(BitWidth, 1); // X / X ===> 1
1756 Remainder = APInt(BitWidth, 0); // X % X ===> 0;
1757 return;
1758 }
1759
1760 // Make sure there is enough space to hold the results.
1761 // NOTE: This assumes that reallocate won't affect any bits if it doesn't
1762 // change the size. This is necessary if Quotient or Remainder is aliased
1763 // with LHS or RHS.
1764 Quotient.reallocate(BitWidth);
1765 Remainder.reallocate(BitWidth);
1766
1767 if (lhsWords == 1) { // rhsWords is 1 if lhsWords is 1.
1768 // There is only one word to consider so use the native versions.
1769 uint64_t lhsValue = LHS.U.pVal[0];
1770 uint64_t rhsValue = RHS.U.pVal[0];
1771 Quotient = lhsValue / rhsValue;
1772 Remainder = lhsValue % rhsValue;
1773 return;
1774 }
1775
1776 // Okay, lets do it the long way
1777 divide(LHS.U.pVal, lhsWords, RHS.U.pVal, rhsWords, Quotient.U.pVal,
1778 Remainder.U.pVal);
1779 // Clear the rest of the Quotient and Remainder.
1780 std::memset(Quotient.U.pVal + lhsWords, 0,
1781 (getNumWords(BitWidth) - lhsWords) * APINT_WORD_SIZE);
1782 std::memset(Remainder.U.pVal + rhsWords, 0,
1783 (getNumWords(BitWidth) - rhsWords) * APINT_WORD_SIZE);
1784 }
1785
udivrem(const APInt & LHS,uint64_t RHS,APInt & Quotient,uint64_t & Remainder)1786 void APInt::udivrem(const APInt &LHS, uint64_t RHS, APInt &Quotient,
1787 uint64_t &Remainder) {
1788 assert(RHS != 0 && "Divide by zero?");
1789 unsigned BitWidth = LHS.BitWidth;
1790
1791 // First, deal with the easy case
1792 if (LHS.isSingleWord()) {
1793 uint64_t QuotVal = LHS.U.VAL / RHS;
1794 Remainder = LHS.U.VAL % RHS;
1795 Quotient = APInt(BitWidth, QuotVal);
1796 return;
1797 }
1798
1799 // Get some size facts about the dividend and divisor
1800 unsigned lhsWords = getNumWords(LHS.getActiveBits());
1801
1802 // Check the degenerate cases
1803 if (lhsWords == 0) {
1804 Quotient = APInt(BitWidth, 0); // 0 / Y ===> 0
1805 Remainder = 0; // 0 % Y ===> 0
1806 return;
1807 }
1808
1809 if (RHS == 1) {
1810 Quotient = LHS; // X / 1 ===> X
1811 Remainder = 0; // X % 1 ===> 0
1812 return;
1813 }
1814
1815 if (LHS.ult(RHS)) {
1816 Remainder = LHS.getZExtValue(); // X % Y ===> X, iff X < Y
1817 Quotient = APInt(BitWidth, 0); // X / Y ===> 0, iff X < Y
1818 return;
1819 }
1820
1821 if (LHS == RHS) {
1822 Quotient = APInt(BitWidth, 1); // X / X ===> 1
1823 Remainder = 0; // X % X ===> 0;
1824 return;
1825 }
1826
1827 // Make sure there is enough space to hold the results.
1828 // NOTE: This assumes that reallocate won't affect any bits if it doesn't
1829 // change the size. This is necessary if Quotient is aliased with LHS.
1830 Quotient.reallocate(BitWidth);
1831
1832 if (lhsWords == 1) { // rhsWords is 1 if lhsWords is 1.
1833 // There is only one word to consider so use the native versions.
1834 uint64_t lhsValue = LHS.U.pVal[0];
1835 Quotient = lhsValue / RHS;
1836 Remainder = lhsValue % RHS;
1837 return;
1838 }
1839
1840 // Okay, lets do it the long way
1841 divide(LHS.U.pVal, lhsWords, &RHS, 1, Quotient.U.pVal, &Remainder);
1842 // Clear the rest of the Quotient.
1843 std::memset(Quotient.U.pVal + lhsWords, 0,
1844 (getNumWords(BitWidth) - lhsWords) * APINT_WORD_SIZE);
1845 }
1846
sdivrem(const APInt & LHS,const APInt & RHS,APInt & Quotient,APInt & Remainder)1847 void APInt::sdivrem(const APInt &LHS, const APInt &RHS,
1848 APInt &Quotient, APInt &Remainder) {
1849 if (LHS.isNegative()) {
1850 if (RHS.isNegative())
1851 APInt::udivrem(-LHS, -RHS, Quotient, Remainder);
1852 else {
1853 APInt::udivrem(-LHS, RHS, Quotient, Remainder);
1854 Quotient.negate();
1855 }
1856 Remainder.negate();
1857 } else if (RHS.isNegative()) {
1858 APInt::udivrem(LHS, -RHS, Quotient, Remainder);
1859 Quotient.negate();
1860 } else {
1861 APInt::udivrem(LHS, RHS, Quotient, Remainder);
1862 }
1863 }
1864
sdivrem(const APInt & LHS,int64_t RHS,APInt & Quotient,int64_t & Remainder)1865 void APInt::sdivrem(const APInt &LHS, int64_t RHS,
1866 APInt &Quotient, int64_t &Remainder) {
1867 uint64_t R = Remainder;
1868 if (LHS.isNegative()) {
1869 if (RHS < 0)
1870 APInt::udivrem(-LHS, -RHS, Quotient, R);
1871 else {
1872 APInt::udivrem(-LHS, RHS, Quotient, R);
1873 Quotient.negate();
1874 }
1875 R = -R;
1876 } else if (RHS < 0) {
1877 APInt::udivrem(LHS, -RHS, Quotient, R);
1878 Quotient.negate();
1879 } else {
1880 APInt::udivrem(LHS, RHS, Quotient, R);
1881 }
1882 Remainder = R;
1883 }
1884
sadd_ov(const APInt & RHS,bool & Overflow) const1885 APInt APInt::sadd_ov(const APInt &RHS, bool &Overflow) const {
1886 APInt Res = *this+RHS;
1887 Overflow = isNonNegative() == RHS.isNonNegative() &&
1888 Res.isNonNegative() != isNonNegative();
1889 return Res;
1890 }
1891
uadd_ov(const APInt & RHS,bool & Overflow) const1892 APInt APInt::uadd_ov(const APInt &RHS, bool &Overflow) const {
1893 APInt Res = *this+RHS;
1894 Overflow = Res.ult(RHS);
1895 return Res;
1896 }
1897
ssub_ov(const APInt & RHS,bool & Overflow) const1898 APInt APInt::ssub_ov(const APInt &RHS, bool &Overflow) const {
1899 APInt Res = *this - RHS;
1900 Overflow = isNonNegative() != RHS.isNonNegative() &&
1901 Res.isNonNegative() != isNonNegative();
1902 return Res;
1903 }
1904
usub_ov(const APInt & RHS,bool & Overflow) const1905 APInt APInt::usub_ov(const APInt &RHS, bool &Overflow) const {
1906 APInt Res = *this-RHS;
1907 Overflow = Res.ugt(*this);
1908 return Res;
1909 }
1910
sdiv_ov(const APInt & RHS,bool & Overflow) const1911 APInt APInt::sdiv_ov(const APInt &RHS, bool &Overflow) const {
1912 // MININT/-1 --> overflow.
1913 Overflow = isMinSignedValue() && RHS.isAllOnesValue();
1914 return sdiv(RHS);
1915 }
1916
smul_ov(const APInt & RHS,bool & Overflow) const1917 APInt APInt::smul_ov(const APInt &RHS, bool &Overflow) const {
1918 APInt Res = *this * RHS;
1919
1920 if (*this != 0 && RHS != 0)
1921 Overflow = Res.sdiv(RHS) != *this || Res.sdiv(*this) != RHS;
1922 else
1923 Overflow = false;
1924 return Res;
1925 }
1926
umul_ov(const APInt & RHS,bool & Overflow) const1927 APInt APInt::umul_ov(const APInt &RHS, bool &Overflow) const {
1928 APInt Res = *this * RHS;
1929
1930 if (*this != 0 && RHS != 0)
1931 Overflow = Res.udiv(RHS) != *this || Res.udiv(*this) != RHS;
1932 else
1933 Overflow = false;
1934 return Res;
1935 }
1936
sshl_ov(const APInt & ShAmt,bool & Overflow) const1937 APInt APInt::sshl_ov(const APInt &ShAmt, bool &Overflow) const {
1938 Overflow = ShAmt.uge(getBitWidth());
1939 if (Overflow)
1940 return APInt(BitWidth, 0);
1941
1942 if (isNonNegative()) // Don't allow sign change.
1943 Overflow = ShAmt.uge(countLeadingZeros());
1944 else
1945 Overflow = ShAmt.uge(countLeadingOnes());
1946
1947 return *this << ShAmt;
1948 }
1949
ushl_ov(const APInt & ShAmt,bool & Overflow) const1950 APInt APInt::ushl_ov(const APInt &ShAmt, bool &Overflow) const {
1951 Overflow = ShAmt.uge(getBitWidth());
1952 if (Overflow)
1953 return APInt(BitWidth, 0);
1954
1955 Overflow = ShAmt.ugt(countLeadingZeros());
1956
1957 return *this << ShAmt;
1958 }
1959
1960
1961
1962
fromString(unsigned numbits,StringRef str,uint8_t radix)1963 void APInt::fromString(unsigned numbits, StringRef str, uint8_t radix) {
1964 // Check our assumptions here
1965 assert(!str.empty() && "Invalid string length");
1966 assert((radix == 10 || radix == 8 || radix == 16 || radix == 2 ||
1967 radix == 36) &&
1968 "Radix should be 2, 8, 10, 16, or 36!");
1969
1970 StringRef::iterator p = str.begin();
1971 size_t slen = str.size();
1972 bool isNeg = *p == '-';
1973 if (*p == '-' || *p == '+') {
1974 p++;
1975 slen--;
1976 assert(slen && "String is only a sign, needs a value.");
1977 }
1978 assert((slen <= numbits || radix != 2) && "Insufficient bit width");
1979 assert(((slen-1)*3 <= numbits || radix != 8) && "Insufficient bit width");
1980 assert(((slen-1)*4 <= numbits || radix != 16) && "Insufficient bit width");
1981 assert((((slen-1)*64)/22 <= numbits || radix != 10) &&
1982 "Insufficient bit width");
1983
1984 // Allocate memory if needed
1985 if (isSingleWord())
1986 U.VAL = 0;
1987 else
1988 U.pVal = getClearedMemory(getNumWords());
1989
1990 // Figure out if we can shift instead of multiply
1991 unsigned shift = (radix == 16 ? 4 : radix == 8 ? 3 : radix == 2 ? 1 : 0);
1992
1993 // Enter digit traversal loop
1994 for (StringRef::iterator e = str.end(); p != e; ++p) {
1995 unsigned digit = getDigit(*p, radix);
1996 assert(digit < radix && "Invalid character in digit string");
1997
1998 // Shift or multiply the value by the radix
1999 if (slen > 1) {
2000 if (shift)
2001 *this <<= shift;
2002 else
2003 *this *= radix;
2004 }
2005
2006 // Add in the digit we just interpreted
2007 *this += digit;
2008 }
2009 // If its negative, put it in two's complement form
2010 if (isNeg)
2011 this->negate();
2012 }
2013
toString(SmallVectorImpl<char> & Str,unsigned Radix,bool Signed,bool formatAsCLiteral) const2014 void APInt::toString(SmallVectorImpl<char> &Str, unsigned Radix,
2015 bool Signed, bool formatAsCLiteral) const {
2016 assert((Radix == 10 || Radix == 8 || Radix == 16 || Radix == 2 ||
2017 Radix == 36) &&
2018 "Radix should be 2, 8, 10, 16, or 36!");
2019
2020 const char *Prefix = "";
2021 if (formatAsCLiteral) {
2022 switch (Radix) {
2023 case 2:
2024 // Binary literals are a non-standard extension added in gcc 4.3:
2025 // http://gcc.gnu.org/onlinedocs/gcc-4.3.0/gcc/Binary-constants.html
2026 Prefix = "0b";
2027 break;
2028 case 8:
2029 Prefix = "0";
2030 break;
2031 case 10:
2032 break; // No prefix
2033 case 16:
2034 Prefix = "0x";
2035 break;
2036 default:
2037 llvm_unreachable("Invalid radix!");
2038 }
2039 }
2040
2041 // First, check for a zero value and just short circuit the logic below.
2042 if (*this == 0) {
2043 while (*Prefix) {
2044 Str.push_back(*Prefix);
2045 ++Prefix;
2046 };
2047 Str.push_back('0');
2048 return;
2049 }
2050
2051 static const char Digits[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
2052
2053 if (isSingleWord()) {
2054 char Buffer[65];
2055 char *BufPtr = std::end(Buffer);
2056
2057 uint64_t N;
2058 if (!Signed) {
2059 N = getZExtValue();
2060 } else {
2061 int64_t I = getSExtValue();
2062 if (I >= 0) {
2063 N = I;
2064 } else {
2065 Str.push_back('-');
2066 N = -(uint64_t)I;
2067 }
2068 }
2069
2070 while (*Prefix) {
2071 Str.push_back(*Prefix);
2072 ++Prefix;
2073 };
2074
2075 while (N) {
2076 *--BufPtr = Digits[N % Radix];
2077 N /= Radix;
2078 }
2079 Str.append(BufPtr, std::end(Buffer));
2080 return;
2081 }
2082
2083 APInt Tmp(*this);
2084
2085 if (Signed && isNegative()) {
2086 // They want to print the signed version and it is a negative value
2087 // Flip the bits and add one to turn it into the equivalent positive
2088 // value and put a '-' in the result.
2089 Tmp.negate();
2090 Str.push_back('-');
2091 }
2092
2093 while (*Prefix) {
2094 Str.push_back(*Prefix);
2095 ++Prefix;
2096 };
2097
2098 // We insert the digits backward, then reverse them to get the right order.
2099 unsigned StartDig = Str.size();
2100
2101 // For the 2, 8 and 16 bit cases, we can just shift instead of divide
2102 // because the number of bits per digit (1, 3 and 4 respectively) divides
2103 // equally. We just shift until the value is zero.
2104 if (Radix == 2 || Radix == 8 || Radix == 16) {
2105 // Just shift tmp right for each digit width until it becomes zero
2106 unsigned ShiftAmt = (Radix == 16 ? 4 : (Radix == 8 ? 3 : 1));
2107 unsigned MaskAmt = Radix - 1;
2108
2109 while (Tmp.getBoolValue()) {
2110 unsigned Digit = unsigned(Tmp.getRawData()[0]) & MaskAmt;
2111 Str.push_back(Digits[Digit]);
2112 Tmp.lshrInPlace(ShiftAmt);
2113 }
2114 } else {
2115 while (Tmp.getBoolValue()) {
2116 uint64_t Digit;
2117 udivrem(Tmp, Radix, Tmp, Digit);
2118 assert(Digit < Radix && "divide failed");
2119 Str.push_back(Digits[Digit]);
2120 }
2121 }
2122
2123 // Reverse the digits before returning.
2124 std::reverse(Str.begin()+StartDig, Str.end());
2125 }
2126
2127 /// Returns the APInt as a std::string. Note that this is an inefficient method.
2128 /// It is better to pass in a SmallVector/SmallString to the methods above.
toString(unsigned Radix=10,bool Signed=true) const2129 std::string APInt::toString(unsigned Radix = 10, bool Signed = true) const {
2130 SmallString<40> S;
2131 toString(S, Radix, Signed, /* formatAsCLiteral = */false);
2132 return S.str();
2133 }
2134
2135 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
dump() const2136 LLVM_DUMP_METHOD void APInt::dump() const {
2137 SmallString<40> S, U;
2138 this->toStringUnsigned(U);
2139 this->toStringSigned(S);
2140 dbgs() << "APInt(" << BitWidth << "b, "
2141 << U << "u " << S << "s)\n";
2142 }
2143 #endif
2144
print(raw_ostream & OS,bool isSigned) const2145 void APInt::print(raw_ostream &OS, bool isSigned) const {
2146 SmallString<40> S;
2147 this->toString(S, 10, isSigned, /* formatAsCLiteral = */false);
2148 OS << S;
2149 }
2150
2151 // This implements a variety of operations on a representation of
2152 // arbitrary precision, two's-complement, bignum integer values.
2153
2154 // Assumed by lowHalf, highHalf, partMSB and partLSB. A fairly safe
2155 // and unrestricting assumption.
2156 static_assert(APInt::APINT_BITS_PER_WORD % 2 == 0,
2157 "Part width must be divisible by 2!");
2158
2159 /* Some handy functions local to this file. */
2160
2161 /* Returns the integer part with the least significant BITS set.
2162 BITS cannot be zero. */
lowBitMask(unsigned bits)2163 static inline APInt::WordType lowBitMask(unsigned bits) {
2164 assert(bits != 0 && bits <= APInt::APINT_BITS_PER_WORD);
2165
2166 return ~(APInt::WordType) 0 >> (APInt::APINT_BITS_PER_WORD - bits);
2167 }
2168
2169 /* Returns the value of the lower half of PART. */
lowHalf(APInt::WordType part)2170 static inline APInt::WordType lowHalf(APInt::WordType part) {
2171 return part & lowBitMask(APInt::APINT_BITS_PER_WORD / 2);
2172 }
2173
2174 /* Returns the value of the upper half of PART. */
highHalf(APInt::WordType part)2175 static inline APInt::WordType highHalf(APInt::WordType part) {
2176 return part >> (APInt::APINT_BITS_PER_WORD / 2);
2177 }
2178
2179 /* Returns the bit number of the most significant set bit of a part.
2180 If the input number has no bits set -1U is returned. */
partMSB(APInt::WordType value)2181 static unsigned partMSB(APInt::WordType value) {
2182 return findLastSet(value, ZB_Max);
2183 }
2184
2185 /* Returns the bit number of the least significant set bit of a
2186 part. If the input number has no bits set -1U is returned. */
partLSB(APInt::WordType value)2187 static unsigned partLSB(APInt::WordType value) {
2188 return findFirstSet(value, ZB_Max);
2189 }
2190
2191 /* Sets the least significant part of a bignum to the input value, and
2192 zeroes out higher parts. */
tcSet(WordType * dst,WordType part,unsigned parts)2193 void APInt::tcSet(WordType *dst, WordType part, unsigned parts) {
2194 assert(parts > 0);
2195
2196 dst[0] = part;
2197 for (unsigned i = 1; i < parts; i++)
2198 dst[i] = 0;
2199 }
2200
2201 /* Assign one bignum to another. */
tcAssign(WordType * dst,const WordType * src,unsigned parts)2202 void APInt::tcAssign(WordType *dst, const WordType *src, unsigned parts) {
2203 for (unsigned i = 0; i < parts; i++)
2204 dst[i] = src[i];
2205 }
2206
2207 /* Returns true if a bignum is zero, false otherwise. */
tcIsZero(const WordType * src,unsigned parts)2208 bool APInt::tcIsZero(const WordType *src, unsigned parts) {
2209 for (unsigned i = 0; i < parts; i++)
2210 if (src[i])
2211 return false;
2212
2213 return true;
2214 }
2215
2216 /* Extract the given bit of a bignum; returns 0 or 1. */
tcExtractBit(const WordType * parts,unsigned bit)2217 int APInt::tcExtractBit(const WordType *parts, unsigned bit) {
2218 return (parts[whichWord(bit)] & maskBit(bit)) != 0;
2219 }
2220
2221 /* Set the given bit of a bignum. */
tcSetBit(WordType * parts,unsigned bit)2222 void APInt::tcSetBit(WordType *parts, unsigned bit) {
2223 parts[whichWord(bit)] |= maskBit(bit);
2224 }
2225
2226 /* Clears the given bit of a bignum. */
tcClearBit(WordType * parts,unsigned bit)2227 void APInt::tcClearBit(WordType *parts, unsigned bit) {
2228 parts[whichWord(bit)] &= ~maskBit(bit);
2229 }
2230
2231 /* Returns the bit number of the least significant set bit of a
2232 number. If the input number has no bits set -1U is returned. */
tcLSB(const WordType * parts,unsigned n)2233 unsigned APInt::tcLSB(const WordType *parts, unsigned n) {
2234 for (unsigned i = 0; i < n; i++) {
2235 if (parts[i] != 0) {
2236 unsigned lsb = partLSB(parts[i]);
2237
2238 return lsb + i * APINT_BITS_PER_WORD;
2239 }
2240 }
2241
2242 return -1U;
2243 }
2244
2245 /* Returns the bit number of the most significant set bit of a number.
2246 If the input number has no bits set -1U is returned. */
tcMSB(const WordType * parts,unsigned n)2247 unsigned APInt::tcMSB(const WordType *parts, unsigned n) {
2248 do {
2249 --n;
2250
2251 if (parts[n] != 0) {
2252 unsigned msb = partMSB(parts[n]);
2253
2254 return msb + n * APINT_BITS_PER_WORD;
2255 }
2256 } while (n);
2257
2258 return -1U;
2259 }
2260
2261 /* Copy the bit vector of width srcBITS from SRC, starting at bit
2262 srcLSB, to DST, of dstCOUNT parts, such that the bit srcLSB becomes
2263 the least significant bit of DST. All high bits above srcBITS in
2264 DST are zero-filled. */
2265 void
tcExtract(WordType * dst,unsigned dstCount,const WordType * src,unsigned srcBits,unsigned srcLSB)2266 APInt::tcExtract(WordType *dst, unsigned dstCount, const WordType *src,
2267 unsigned srcBits, unsigned srcLSB) {
2268 unsigned dstParts = (srcBits + APINT_BITS_PER_WORD - 1) / APINT_BITS_PER_WORD;
2269 assert(dstParts <= dstCount);
2270
2271 unsigned firstSrcPart = srcLSB / APINT_BITS_PER_WORD;
2272 tcAssign (dst, src + firstSrcPart, dstParts);
2273
2274 unsigned shift = srcLSB % APINT_BITS_PER_WORD;
2275 tcShiftRight (dst, dstParts, shift);
2276
2277 /* We now have (dstParts * APINT_BITS_PER_WORD - shift) bits from SRC
2278 in DST. If this is less that srcBits, append the rest, else
2279 clear the high bits. */
2280 unsigned n = dstParts * APINT_BITS_PER_WORD - shift;
2281 if (n < srcBits) {
2282 WordType mask = lowBitMask (srcBits - n);
2283 dst[dstParts - 1] |= ((src[firstSrcPart + dstParts] & mask)
2284 << n % APINT_BITS_PER_WORD);
2285 } else if (n > srcBits) {
2286 if (srcBits % APINT_BITS_PER_WORD)
2287 dst[dstParts - 1] &= lowBitMask (srcBits % APINT_BITS_PER_WORD);
2288 }
2289
2290 /* Clear high parts. */
2291 while (dstParts < dstCount)
2292 dst[dstParts++] = 0;
2293 }
2294
2295 /* DST += RHS + C where C is zero or one. Returns the carry flag. */
tcAdd(WordType * dst,const WordType * rhs,WordType c,unsigned parts)2296 APInt::WordType APInt::tcAdd(WordType *dst, const WordType *rhs,
2297 WordType c, unsigned parts) {
2298 assert(c <= 1);
2299
2300 for (unsigned i = 0; i < parts; i++) {
2301 WordType l = dst[i];
2302 if (c) {
2303 dst[i] += rhs[i] + 1;
2304 c = (dst[i] <= l);
2305 } else {
2306 dst[i] += rhs[i];
2307 c = (dst[i] < l);
2308 }
2309 }
2310
2311 return c;
2312 }
2313
2314 /// This function adds a single "word" integer, src, to the multiple
2315 /// "word" integer array, dst[]. dst[] is modified to reflect the addition and
2316 /// 1 is returned if there is a carry out, otherwise 0 is returned.
2317 /// @returns the carry of the addition.
tcAddPart(WordType * dst,WordType src,unsigned parts)2318 APInt::WordType APInt::tcAddPart(WordType *dst, WordType src,
2319 unsigned parts) {
2320 for (unsigned i = 0; i < parts; ++i) {
2321 dst[i] += src;
2322 if (dst[i] >= src)
2323 return 0; // No need to carry so exit early.
2324 src = 1; // Carry one to next digit.
2325 }
2326
2327 return 1;
2328 }
2329
2330 /* DST -= RHS + C where C is zero or one. Returns the carry flag. */
tcSubtract(WordType * dst,const WordType * rhs,WordType c,unsigned parts)2331 APInt::WordType APInt::tcSubtract(WordType *dst, const WordType *rhs,
2332 WordType c, unsigned parts) {
2333 assert(c <= 1);
2334
2335 for (unsigned i = 0; i < parts; i++) {
2336 WordType l = dst[i];
2337 if (c) {
2338 dst[i] -= rhs[i] + 1;
2339 c = (dst[i] >= l);
2340 } else {
2341 dst[i] -= rhs[i];
2342 c = (dst[i] > l);
2343 }
2344 }
2345
2346 return c;
2347 }
2348
2349 /// This function subtracts a single "word" (64-bit word), src, from
2350 /// the multi-word integer array, dst[], propagating the borrowed 1 value until
2351 /// no further borrowing is needed or it runs out of "words" in dst. The result
2352 /// is 1 if "borrowing" exhausted the digits in dst, or 0 if dst was not
2353 /// exhausted. In other words, if src > dst then this function returns 1,
2354 /// otherwise 0.
2355 /// @returns the borrow out of the subtraction
tcSubtractPart(WordType * dst,WordType src,unsigned parts)2356 APInt::WordType APInt::tcSubtractPart(WordType *dst, WordType src,
2357 unsigned parts) {
2358 for (unsigned i = 0; i < parts; ++i) {
2359 WordType Dst = dst[i];
2360 dst[i] -= src;
2361 if (src <= Dst)
2362 return 0; // No need to borrow so exit early.
2363 src = 1; // We have to "borrow 1" from next "word"
2364 }
2365
2366 return 1;
2367 }
2368
2369 /* Negate a bignum in-place. */
tcNegate(WordType * dst,unsigned parts)2370 void APInt::tcNegate(WordType *dst, unsigned parts) {
2371 tcComplement(dst, parts);
2372 tcIncrement(dst, parts);
2373 }
2374
2375 /* DST += SRC * MULTIPLIER + CARRY if add is true
2376 DST = SRC * MULTIPLIER + CARRY if add is false
2377
2378 Requires 0 <= DSTPARTS <= SRCPARTS + 1. If DST overlaps SRC
2379 they must start at the same point, i.e. DST == SRC.
2380
2381 If DSTPARTS == SRCPARTS + 1 no overflow occurs and zero is
2382 returned. Otherwise DST is filled with the least significant
2383 DSTPARTS parts of the result, and if all of the omitted higher
2384 parts were zero return zero, otherwise overflow occurred and
2385 return one. */
tcMultiplyPart(WordType * dst,const WordType * src,WordType multiplier,WordType carry,unsigned srcParts,unsigned dstParts,bool add)2386 int APInt::tcMultiplyPart(WordType *dst, const WordType *src,
2387 WordType multiplier, WordType carry,
2388 unsigned srcParts, unsigned dstParts,
2389 bool add) {
2390 /* Otherwise our writes of DST kill our later reads of SRC. */
2391 assert(dst <= src || dst >= src + srcParts);
2392 assert(dstParts <= srcParts + 1);
2393
2394 /* N loops; minimum of dstParts and srcParts. */
2395 unsigned n = std::min(dstParts, srcParts);
2396
2397 for (unsigned i = 0; i < n; i++) {
2398 WordType low, mid, high, srcPart;
2399
2400 /* [ LOW, HIGH ] = MULTIPLIER * SRC[i] + DST[i] + CARRY.
2401
2402 This cannot overflow, because
2403
2404 (n - 1) * (n - 1) + 2 (n - 1) = (n - 1) * (n + 1)
2405
2406 which is less than n^2. */
2407
2408 srcPart = src[i];
2409
2410 if (multiplier == 0 || srcPart == 0) {
2411 low = carry;
2412 high = 0;
2413 } else {
2414 low = lowHalf(srcPart) * lowHalf(multiplier);
2415 high = highHalf(srcPart) * highHalf(multiplier);
2416
2417 mid = lowHalf(srcPart) * highHalf(multiplier);
2418 high += highHalf(mid);
2419 mid <<= APINT_BITS_PER_WORD / 2;
2420 if (low + mid < low)
2421 high++;
2422 low += mid;
2423
2424 mid = highHalf(srcPart) * lowHalf(multiplier);
2425 high += highHalf(mid);
2426 mid <<= APINT_BITS_PER_WORD / 2;
2427 if (low + mid < low)
2428 high++;
2429 low += mid;
2430
2431 /* Now add carry. */
2432 if (low + carry < low)
2433 high++;
2434 low += carry;
2435 }
2436
2437 if (add) {
2438 /* And now DST[i], and store the new low part there. */
2439 if (low + dst[i] < low)
2440 high++;
2441 dst[i] += low;
2442 } else
2443 dst[i] = low;
2444
2445 carry = high;
2446 }
2447
2448 if (srcParts < dstParts) {
2449 /* Full multiplication, there is no overflow. */
2450 assert(srcParts + 1 == dstParts);
2451 dst[srcParts] = carry;
2452 return 0;
2453 }
2454
2455 /* We overflowed if there is carry. */
2456 if (carry)
2457 return 1;
2458
2459 /* We would overflow if any significant unwritten parts would be
2460 non-zero. This is true if any remaining src parts are non-zero
2461 and the multiplier is non-zero. */
2462 if (multiplier)
2463 for (unsigned i = dstParts; i < srcParts; i++)
2464 if (src[i])
2465 return 1;
2466
2467 /* We fitted in the narrow destination. */
2468 return 0;
2469 }
2470
2471 /* DST = LHS * RHS, where DST has the same width as the operands and
2472 is filled with the least significant parts of the result. Returns
2473 one if overflow occurred, otherwise zero. DST must be disjoint
2474 from both operands. */
tcMultiply(WordType * dst,const WordType * lhs,const WordType * rhs,unsigned parts)2475 int APInt::tcMultiply(WordType *dst, const WordType *lhs,
2476 const WordType *rhs, unsigned parts) {
2477 assert(dst != lhs && dst != rhs);
2478
2479 int overflow = 0;
2480 tcSet(dst, 0, parts);
2481
2482 for (unsigned i = 0; i < parts; i++)
2483 overflow |= tcMultiplyPart(&dst[i], lhs, rhs[i], 0, parts,
2484 parts - i, true);
2485
2486 return overflow;
2487 }
2488
2489 /// DST = LHS * RHS, where DST has width the sum of the widths of the
2490 /// operands. No overflow occurs. DST must be disjoint from both operands.
tcFullMultiply(WordType * dst,const WordType * lhs,const WordType * rhs,unsigned lhsParts,unsigned rhsParts)2491 void APInt::tcFullMultiply(WordType *dst, const WordType *lhs,
2492 const WordType *rhs, unsigned lhsParts,
2493 unsigned rhsParts) {
2494 /* Put the narrower number on the LHS for less loops below. */
2495 if (lhsParts > rhsParts)
2496 return tcFullMultiply (dst, rhs, lhs, rhsParts, lhsParts);
2497
2498 assert(dst != lhs && dst != rhs);
2499
2500 tcSet(dst, 0, rhsParts);
2501
2502 for (unsigned i = 0; i < lhsParts; i++)
2503 tcMultiplyPart(&dst[i], rhs, lhs[i], 0, rhsParts, rhsParts + 1, true);
2504 }
2505
2506 /* If RHS is zero LHS and REMAINDER are left unchanged, return one.
2507 Otherwise set LHS to LHS / RHS with the fractional part discarded,
2508 set REMAINDER to the remainder, return zero. i.e.
2509
2510 OLD_LHS = RHS * LHS + REMAINDER
2511
2512 SCRATCH is a bignum of the same size as the operands and result for
2513 use by the routine; its contents need not be initialized and are
2514 destroyed. LHS, REMAINDER and SCRATCH must be distinct.
2515 */
tcDivide(WordType * lhs,const WordType * rhs,WordType * remainder,WordType * srhs,unsigned parts)2516 int APInt::tcDivide(WordType *lhs, const WordType *rhs,
2517 WordType *remainder, WordType *srhs,
2518 unsigned parts) {
2519 assert(lhs != remainder && lhs != srhs && remainder != srhs);
2520
2521 unsigned shiftCount = tcMSB(rhs, parts) + 1;
2522 if (shiftCount == 0)
2523 return true;
2524
2525 shiftCount = parts * APINT_BITS_PER_WORD - shiftCount;
2526 unsigned n = shiftCount / APINT_BITS_PER_WORD;
2527 WordType mask = (WordType) 1 << (shiftCount % APINT_BITS_PER_WORD);
2528
2529 tcAssign(srhs, rhs, parts);
2530 tcShiftLeft(srhs, parts, shiftCount);
2531 tcAssign(remainder, lhs, parts);
2532 tcSet(lhs, 0, parts);
2533
2534 /* Loop, subtracting SRHS if REMAINDER is greater and adding that to
2535 the total. */
2536 for (;;) {
2537 int compare = tcCompare(remainder, srhs, parts);
2538 if (compare >= 0) {
2539 tcSubtract(remainder, srhs, 0, parts);
2540 lhs[n] |= mask;
2541 }
2542
2543 if (shiftCount == 0)
2544 break;
2545 shiftCount--;
2546 tcShiftRight(srhs, parts, 1);
2547 if ((mask >>= 1) == 0) {
2548 mask = (WordType) 1 << (APINT_BITS_PER_WORD - 1);
2549 n--;
2550 }
2551 }
2552
2553 return false;
2554 }
2555
2556 /// Shift a bignum left Cound bits in-place. Shifted in bits are zero. There are
2557 /// no restrictions on Count.
tcShiftLeft(WordType * Dst,unsigned Words,unsigned Count)2558 void APInt::tcShiftLeft(WordType *Dst, unsigned Words, unsigned Count) {
2559 // Don't bother performing a no-op shift.
2560 if (!Count)
2561 return;
2562
2563 // WordShift is the inter-part shift; BitShift is the intra-part shift.
2564 unsigned WordShift = std::min(Count / APINT_BITS_PER_WORD, Words);
2565 unsigned BitShift = Count % APINT_BITS_PER_WORD;
2566
2567 // Fastpath for moving by whole words.
2568 if (BitShift == 0) {
2569 std::memmove(Dst + WordShift, Dst, (Words - WordShift) * APINT_WORD_SIZE);
2570 } else {
2571 while (Words-- > WordShift) {
2572 Dst[Words] = Dst[Words - WordShift] << BitShift;
2573 if (Words > WordShift)
2574 Dst[Words] |=
2575 Dst[Words - WordShift - 1] >> (APINT_BITS_PER_WORD - BitShift);
2576 }
2577 }
2578
2579 // Fill in the remainder with 0s.
2580 std::memset(Dst, 0, WordShift * APINT_WORD_SIZE);
2581 }
2582
2583 /// Shift a bignum right Count bits in-place. Shifted in bits are zero. There
2584 /// are no restrictions on Count.
tcShiftRight(WordType * Dst,unsigned Words,unsigned Count)2585 void APInt::tcShiftRight(WordType *Dst, unsigned Words, unsigned Count) {
2586 // Don't bother performing a no-op shift.
2587 if (!Count)
2588 return;
2589
2590 // WordShift is the inter-part shift; BitShift is the intra-part shift.
2591 unsigned WordShift = std::min(Count / APINT_BITS_PER_WORD, Words);
2592 unsigned BitShift = Count % APINT_BITS_PER_WORD;
2593
2594 unsigned WordsToMove = Words - WordShift;
2595 // Fastpath for moving by whole words.
2596 if (BitShift == 0) {
2597 std::memmove(Dst, Dst + WordShift, WordsToMove * APINT_WORD_SIZE);
2598 } else {
2599 for (unsigned i = 0; i != WordsToMove; ++i) {
2600 Dst[i] = Dst[i + WordShift] >> BitShift;
2601 if (i + 1 != WordsToMove)
2602 Dst[i] |= Dst[i + WordShift + 1] << (APINT_BITS_PER_WORD - BitShift);
2603 }
2604 }
2605
2606 // Fill in the remainder with 0s.
2607 std::memset(Dst + WordsToMove, 0, WordShift * APINT_WORD_SIZE);
2608 }
2609
2610 /* Bitwise and of two bignums. */
tcAnd(WordType * dst,const WordType * rhs,unsigned parts)2611 void APInt::tcAnd(WordType *dst, const WordType *rhs, unsigned parts) {
2612 for (unsigned i = 0; i < parts; i++)
2613 dst[i] &= rhs[i];
2614 }
2615
2616 /* Bitwise inclusive or of two bignums. */
tcOr(WordType * dst,const WordType * rhs,unsigned parts)2617 void APInt::tcOr(WordType *dst, const WordType *rhs, unsigned parts) {
2618 for (unsigned i = 0; i < parts; i++)
2619 dst[i] |= rhs[i];
2620 }
2621
2622 /* Bitwise exclusive or of two bignums. */
tcXor(WordType * dst,const WordType * rhs,unsigned parts)2623 void APInt::tcXor(WordType *dst, const WordType *rhs, unsigned parts) {
2624 for (unsigned i = 0; i < parts; i++)
2625 dst[i] ^= rhs[i];
2626 }
2627
2628 /* Complement a bignum in-place. */
tcComplement(WordType * dst,unsigned parts)2629 void APInt::tcComplement(WordType *dst, unsigned parts) {
2630 for (unsigned i = 0; i < parts; i++)
2631 dst[i] = ~dst[i];
2632 }
2633
2634 /* Comparison (unsigned) of two bignums. */
tcCompare(const WordType * lhs,const WordType * rhs,unsigned parts)2635 int APInt::tcCompare(const WordType *lhs, const WordType *rhs,
2636 unsigned parts) {
2637 while (parts) {
2638 parts--;
2639 if (lhs[parts] != rhs[parts])
2640 return (lhs[parts] > rhs[parts]) ? 1 : -1;
2641 }
2642
2643 return 0;
2644 }
2645
2646 /* Set the least significant BITS bits of a bignum, clear the
2647 rest. */
tcSetLeastSignificantBits(WordType * dst,unsigned parts,unsigned bits)2648 void APInt::tcSetLeastSignificantBits(WordType *dst, unsigned parts,
2649 unsigned bits) {
2650 unsigned i = 0;
2651 while (bits > APINT_BITS_PER_WORD) {
2652 dst[i++] = ~(WordType) 0;
2653 bits -= APINT_BITS_PER_WORD;
2654 }
2655
2656 if (bits)
2657 dst[i++] = ~(WordType) 0 >> (APINT_BITS_PER_WORD - bits);
2658
2659 while (i < parts)
2660 dst[i++] = 0;
2661 }
2662
RoundingUDiv(const APInt & A,const APInt & B,APInt::Rounding RM)2663 APInt llvm::APIntOps::RoundingUDiv(const APInt &A, const APInt &B,
2664 APInt::Rounding RM) {
2665 // Currently udivrem always rounds down.
2666 switch (RM) {
2667 case APInt::Rounding::DOWN:
2668 case APInt::Rounding::TOWARD_ZERO:
2669 return A.udiv(B);
2670 case APInt::Rounding::UP: {
2671 APInt Quo, Rem;
2672 APInt::udivrem(A, B, Quo, Rem);
2673 if (Rem == 0)
2674 return Quo;
2675 return Quo + 1;
2676 }
2677 }
2678 llvm_unreachable("Unknown APInt::Rounding enum");
2679 }
2680
RoundingSDiv(const APInt & A,const APInt & B,APInt::Rounding RM)2681 APInt llvm::APIntOps::RoundingSDiv(const APInt &A, const APInt &B,
2682 APInt::Rounding RM) {
2683 switch (RM) {
2684 case APInt::Rounding::DOWN:
2685 case APInt::Rounding::UP: {
2686 APInt Quo, Rem;
2687 APInt::sdivrem(A, B, Quo, Rem);
2688 if (Rem == 0)
2689 return Quo;
2690 // This algorithm deals with arbitrary rounding mode used by sdivrem.
2691 // We want to check whether the non-integer part of the mathematical value
2692 // is negative or not. If the non-integer part is negative, we need to round
2693 // down from Quo; otherwise, if it's positive or 0, we return Quo, as it's
2694 // already rounded down.
2695 if (RM == APInt::Rounding::DOWN) {
2696 if (Rem.isNegative() != B.isNegative())
2697 return Quo - 1;
2698 return Quo;
2699 }
2700 if (Rem.isNegative() != B.isNegative())
2701 return Quo;
2702 return Quo + 1;
2703 }
2704 // Currently sdiv rounds twards zero.
2705 case APInt::Rounding::TOWARD_ZERO:
2706 return A.sdiv(B);
2707 }
2708 llvm_unreachable("Unknown APInt::Rounding enum");
2709 }
2710