X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FSupport%2FAPInt.cpp;h=9d1468493d5b4c577de1025146b466466225630d;hb=67c59826bf0c2e980130781d1ea0d6b062e4b659;hp=80747fd12a9d9a249e3d98cbb365d9e8eb0293e6;hpb=fad86b003a839cef40ec8ce8408322f4913368ca;p=oota-llvm.git diff --git a/lib/Support/APInt.cpp b/lib/Support/APInt.cpp index 80747fd12a9..9d1468493d5 100644 --- a/lib/Support/APInt.cpp +++ b/lib/Support/APInt.cpp @@ -14,62 +14,79 @@ #define DEBUG_TYPE "apint" #include "llvm/ADT/APInt.h" +#include "llvm/ADT/StringRef.h" #include "llvm/ADT/FoldingSet.h" #include "llvm/ADT/SmallString.h" #include "llvm/Support/Debug.h" +#include "llvm/Support/ErrorHandling.h" #include "llvm/Support/MathExtras.h" +#include "llvm/Support/raw_ostream.h" #include #include #include #include using namespace llvm; -/// This enumeration just provides for internal constants used in this -/// translation unit. -enum { - MIN_INT_BITS = 1, ///< Minimum number of bits that can be specified - ///< Note that this must remain synchronized with IntegerType::MIN_INT_BITS - MAX_INT_BITS = (1<<23)-1 ///< Maximum number of bits that can be specified - ///< Note that this must remain synchronized with IntegerType::MAX_INT_BITS -}; - /// A utility function for allocating memory, checking for allocation failures, /// and ensuring the contents are zeroed. -inline static uint64_t* getClearedMemory(uint32_t numWords) { +inline static uint64_t* getClearedMemory(unsigned numWords) { uint64_t * result = new uint64_t[numWords]; assert(result && "APInt memory allocation fails!"); memset(result, 0, numWords * sizeof(uint64_t)); return result; } -/// A utility function for allocating memory and checking for allocation +/// A utility function for allocating memory and checking for allocation /// failure. The content is not zeroed. -inline static uint64_t* getMemory(uint32_t numWords) { +inline static uint64_t* getMemory(unsigned numWords) { uint64_t * result = new uint64_t[numWords]; assert(result && "APInt memory allocation fails!"); return result; } -APInt::APInt(uint32_t numBits, uint64_t val, bool isSigned) - : BitWidth(numBits), VAL(0) { - assert(BitWidth >= MIN_INT_BITS && "bitwidth too small"); - assert(BitWidth <= MAX_INT_BITS && "bitwidth too large"); - if (isSingleWord()) - VAL = val; - else { - pVal = getClearedMemory(getNumWords()); - pVal[0] = val; - if (isSigned && int64_t(val) < 0) - for (unsigned i = 1; i < getNumWords(); ++i) - pVal[i] = -1ULL; +/// A utility function that converts a character to a digit. +inline static unsigned getDigit(char cdigit, uint8_t radix) { + unsigned r; + + if (radix == 16) { + r = cdigit - '0'; + if (r <= 9) + return r; + + r = cdigit - 'A'; + if (r <= 5) + return r + 10; + + r = cdigit - 'a'; + if (r <= 5) + return r + 10; } - clearUnusedBits(); + + r = cdigit - '0'; + if (r < radix) + return r; + + return -1U; +} + + +void APInt::initSlowCase(unsigned numBits, uint64_t val, bool isSigned) { + pVal = getClearedMemory(getNumWords()); + pVal[0] = val; + if (isSigned && int64_t(val) < 0) + for (unsigned i = 1; i < getNumWords(); ++i) + pVal[i] = -1ULL; +} + +void APInt::initSlowCase(const APInt& that) { + pVal = getMemory(getNumWords()); + memcpy(pVal, that.pVal, getNumWords() * APINT_WORD_SIZE); } -APInt::APInt(uint32_t numBits, uint32_t numWords, const uint64_t bigVal[]) - : BitWidth(numBits), VAL(0) { - assert(BitWidth >= MIN_INT_BITS && "bitwidth too small"); - assert(BitWidth <= MAX_INT_BITS && "bitwidth too large"); + +APInt::APInt(unsigned numBits, unsigned numWords, const uint64_t bigVal[]) + : BitWidth(numBits), VAL(0) { + assert(BitWidth && "Bitwidth too small"); assert(bigVal && "Null pointer detected!"); if (isSingleWord()) VAL = bigVal[0]; @@ -77,7 +94,7 @@ APInt::APInt(uint32_t numBits, uint32_t numWords, const uint64_t bigVal[]) // Get memory, cleared to 0 pVal = getClearedMemory(getNumWords()); // Calculate the number of words to copy - uint32_t words = std::min(numWords, getNumWords()); + unsigned words = std::min(numWords, getNumWords()); // Copy the words from bigVal to pVal memcpy(pVal, bigVal, words * APINT_WORD_SIZE); } @@ -85,54 +102,31 @@ APInt::APInt(uint32_t numBits, uint32_t numWords, const uint64_t bigVal[]) clearUnusedBits(); } -APInt::APInt(uint32_t numbits, const char StrStart[], uint32_t slen, - uint8_t radix) +APInt::APInt(unsigned numbits, const StringRef& Str, uint8_t radix) : BitWidth(numbits), VAL(0) { - assert(BitWidth >= MIN_INT_BITS && "bitwidth too small"); - assert(BitWidth <= MAX_INT_BITS && "bitwidth too large"); - fromString(numbits, StrStart, slen, radix); + assert(BitWidth && "Bitwidth too small"); + fromString(numbits, Str, radix); } -APInt::APInt(const APInt& that) - : BitWidth(that.BitWidth), VAL(0) { - assert(BitWidth >= MIN_INT_BITS && "bitwidth too small"); - assert(BitWidth <= MAX_INT_BITS && "bitwidth too large"); - if (isSingleWord()) - VAL = that.VAL; - else { - pVal = getMemory(getNumWords()); - memcpy(pVal, that.pVal, getNumWords() * APINT_WORD_SIZE); - } -} - -APInt::~APInt() { - if (!isSingleWord() && pVal) - delete [] pVal; -} - -APInt& APInt::operator=(const APInt& RHS) { +APInt& APInt::AssignSlowCase(const APInt& RHS) { // Don't do anything for X = X if (this == &RHS) return *this; - // If the bitwidths are the same, we can avoid mucking with memory if (BitWidth == RHS.getBitWidth()) { - if (isSingleWord()) - VAL = RHS.VAL; - else - memcpy(pVal, RHS.pVal, getNumWords() * APINT_WORD_SIZE); + // assume same bit-width single-word case is already handled + assert(!isSingleWord()); + memcpy(pVal, RHS.pVal, getNumWords() * APINT_WORD_SIZE); return *this; } - if (isSingleWord()) - if (RHS.isSingleWord()) - VAL = RHS.VAL; - else { - VAL = 0; - pVal = getMemory(RHS.getNumWords()); - memcpy(pVal, RHS.pVal, RHS.getNumWords() * APINT_WORD_SIZE); - } - else if (getNumWords() == RHS.getNumWords()) + if (isSingleWord()) { + // assume case where both are single words is already handled + assert(!RHS.isSingleWord()); + VAL = 0; + pVal = getMemory(RHS.getNumWords()); + memcpy(pVal, RHS.pVal, RHS.getNumWords() * APINT_WORD_SIZE); + } else if (getNumWords() == RHS.getNumWords()) memcpy(pVal, RHS.pVal, RHS.getNumWords() * APINT_WORD_SIZE); else if (RHS.isSingleWord()) { delete [] pVal; @@ -147,7 +141,7 @@ APInt& APInt::operator=(const APInt& RHS) { } APInt& APInt::operator=(uint64_t RHS) { - if (isSingleWord()) + if (isSingleWord()) VAL = RHS; else { pVal[0] = RHS; @@ -159,23 +153,23 @@ APInt& APInt::operator=(uint64_t RHS) { /// Profile - This method 'profiles' an APInt for use with FoldingSet. void APInt::Profile(FoldingSetNodeID& ID) const { ID.AddInteger(BitWidth); - + if (isSingleWord()) { ID.AddInteger(VAL); return; } - uint32_t NumWords = getNumWords(); + unsigned NumWords = getNumWords(); for (unsigned i = 0; i < NumWords; ++i) ID.AddInteger(pVal[i]); } -/// add_1 - This function adds a single "digit" integer, y, to the multiple +/// add_1 - This function adds a single "digit" integer, y, to the multiple /// "digit" integer array, x[]. x[] is modified to reflect the addition and /// 1 is returned if there is a carry out, otherwise 0 is returned. /// @returns the carry of the addition. -static bool add_1(uint64_t dest[], uint64_t x[], uint32_t len, uint64_t y) { - for (uint32_t i = 0; i < len; ++i) { +static bool add_1(uint64_t dest[], uint64_t x[], unsigned len, uint64_t y) { + for (unsigned i = 0; i < len; ++i) { dest[i] = y + x[i]; if (dest[i] < y) y = 1; // Carry one to next digit. @@ -189,24 +183,24 @@ static bool add_1(uint64_t dest[], uint64_t x[], uint32_t len, uint64_t y) { /// @brief Prefix increment operator. Increments the APInt by one. APInt& APInt::operator++() { - if (isSingleWord()) + if (isSingleWord()) ++VAL; else add_1(pVal, pVal, getNumWords(), 1); return clearUnusedBits(); } -/// sub_1 - This function subtracts a single "digit" (64-bit word), y, from -/// the multi-digit integer array, x[], propagating the borrowed 1 value until +/// sub_1 - This function subtracts a single "digit" (64-bit word), y, from +/// the multi-digit integer array, x[], propagating the borrowed 1 value until /// no further borrowing is neeeded or it runs out of "digits" in x. The result /// is 1 if "borrowing" exhausted the digits in x, or 0 if x was not exhausted. /// In other words, if y > x then this function returns 1, otherwise 0. /// @returns the borrow out of the subtraction -static bool sub_1(uint64_t x[], uint32_t len, uint64_t y) { - for (uint32_t i = 0; i < len; ++i) { +static bool sub_1(uint64_t x[], unsigned len, uint64_t y) { + for (unsigned i = 0; i < len; ++i) { uint64_t X = x[i]; x[i] -= y; - if (y > X) + if (y > X) y = 1; // We have to "borrow 1" from next "digit" else { y = 0; // No need to borrow @@ -218,7 +212,7 @@ static bool sub_1(uint64_t x[], uint32_t len, uint64_t y) { /// @brief Prefix decrement operator. Decrements the APInt by one. APInt& APInt::operator--() { - if (isSingleWord()) + if (isSingleWord()) --VAL; else sub_1(pVal, getNumWords(), 1); @@ -226,13 +220,13 @@ APInt& APInt::operator--() { } /// add - This function adds the integer array x to the integer array Y and -/// places the result in dest. +/// places the result in dest. /// @returns the carry out from the addition /// @brief General addition of 64-bit integer arrays -static bool add(uint64_t *dest, const uint64_t *x, const uint64_t *y, - uint32_t len) { +static bool add(uint64_t *dest, const uint64_t *x, const uint64_t *y, + unsigned len) { bool carry = false; - for (uint32_t i = 0; i< len; ++i) { + for (unsigned i = 0; i< len; ++i) { uint64_t limit = std::min(x[i],y[i]); // must come first in case dest == x dest[i] = x[i] + y[i] + carry; carry = dest[i] < limit || (carry && dest[i] == limit); @@ -242,10 +236,10 @@ static bool add(uint64_t *dest, const uint64_t *x, const uint64_t *y, /// Adds the RHS APint to this APInt. /// @returns this, after addition of RHS. -/// @brief Addition assignment operator. +/// @brief Addition assignment operator. APInt& APInt::operator+=(const APInt& RHS) { assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); - if (isSingleWord()) + if (isSingleWord()) VAL += RHS.VAL; else { add(pVal, pVal, RHS.pVal, getNumWords()); @@ -253,13 +247,13 @@ APInt& APInt::operator+=(const APInt& RHS) { return clearUnusedBits(); } -/// Subtracts the integer array y from the integer array x +/// Subtracts the integer array y from the integer array x /// @returns returns the borrow out. /// @brief Generalized subtraction of 64-bit integer arrays. -static bool sub(uint64_t *dest, const uint64_t *x, const uint64_t *y, - uint32_t len) { +static bool sub(uint64_t *dest, const uint64_t *x, const uint64_t *y, + unsigned len) { bool borrow = false; - for (uint32_t i = 0; i < len; ++i) { + for (unsigned i = 0; i < len; ++i) { uint64_t x_tmp = borrow ? x[i] - 1 : x[i]; borrow = y[i] > x_tmp || (borrow && x[i] == 0); dest[i] = x_tmp - y[i]; @@ -269,10 +263,10 @@ static bool sub(uint64_t *dest, const uint64_t *x, const uint64_t *y, /// Subtracts the RHS APInt from this APInt /// @returns this, after subtraction -/// @brief Subtraction assignment operator. +/// @brief Subtraction assignment operator. APInt& APInt::operator-=(const APInt& RHS) { assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); - if (isSingleWord()) + if (isSingleWord()) VAL -= RHS.VAL; else sub(pVal, pVal, RHS.pVal, getNumWords()); @@ -280,16 +274,16 @@ APInt& APInt::operator-=(const APInt& RHS) { } /// Multiplies an integer array, x by a a uint64_t integer and places the result -/// into dest. +/// into dest. /// @returns the carry out of the multiplication. /// @brief Multiply a multi-digit APInt by a single digit (64-bit) integer. -static uint64_t mul_1(uint64_t dest[], uint64_t x[], uint32_t len, uint64_t y) { +static uint64_t mul_1(uint64_t dest[], uint64_t x[], unsigned len, uint64_t y) { // Split y into high 32-bit part (hy) and low 32-bit part (ly) uint64_t ly = y & 0xffffffffULL, hy = y >> 32; uint64_t carry = 0; // For each digit of x. - for (uint32_t i = 0; i < len; ++i) { + for (unsigned i = 0; i < len; ++i) { // Split x into high and low words uint64_t lx = x[i] & 0xffffffffULL; uint64_t hx = x[i] >> 32; @@ -302,28 +296,28 @@ static uint64_t mul_1(uint64_t dest[], uint64_t x[], uint32_t len, uint64_t y) { // Determine if the add above introduces carry. hasCarry = (dest[i] < carry) ? 1 : 0; carry = hx * ly + (dest[i] >> 32) + (hasCarry ? (1ULL << 32) : 0); - // The upper limit of carry can be (2^32 - 1)(2^32 - 1) + + // The upper limit of carry can be (2^32 - 1)(2^32 - 1) + // (2^32 - 1) + 2^32 = 2^64. hasCarry = (!carry && hasCarry) ? 1 : (!carry ? 2 : 0); carry += (lx * hy) & 0xffffffffULL; dest[i] = (carry << 32) | (dest[i] & 0xffffffffULL); - carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? (1ULL << 32) : 0) + + carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? (1ULL << 32) : 0) + (carry >> 32) + ((lx * hy) >> 32) + hx * hy; } return carry; } -/// Multiplies integer array x by integer array y and stores the result into +/// Multiplies integer array x by integer array y and stores the result into /// the integer array dest. Note that dest's size must be >= xlen + ylen. /// @brief Generalized multiplicate of integer arrays. -static void mul(uint64_t dest[], uint64_t x[], uint32_t xlen, uint64_t y[], - uint32_t ylen) { +static void mul(uint64_t dest[], uint64_t x[], unsigned xlen, uint64_t y[], + unsigned ylen) { dest[xlen] = mul_1(dest, x, xlen, y[0]); - for (uint32_t i = 1; i < ylen; ++i) { + for (unsigned i = 1; i < ylen; ++i) { uint64_t ly = y[i] & 0xffffffffULL, hy = y[i] >> 32; uint64_t carry = 0, lx = 0, hx = 0; - for (uint32_t j = 0; j < xlen; ++j) { + for (unsigned j = 0; j < xlen; ++j) { lx = x[j] & 0xffffffffULL; hx = x[j] >> 32; // hasCarry - A flag to indicate if has carry. @@ -340,7 +334,7 @@ static void mul(uint64_t dest[], uint64_t x[], uint32_t xlen, uint64_t y[], resul = (carry << 32) | (resul & 0xffffffffULL); dest[i+j] += resul; carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? (1ULL << 32) : 0)+ - (carry >> 32) + (dest[i+j] < resul ? 1 : 0) + + (carry >> 32) + (dest[i+j] < resul ? 1 : 0) + ((lx * hy) >> 32) + hx * hy; } dest[i+xlen] = carry; @@ -356,15 +350,15 @@ APInt& APInt::operator*=(const APInt& RHS) { } // Get some bit facts about LHS and check for zero - uint32_t lhsBits = getActiveBits(); - uint32_t lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; - if (!lhsWords) + unsigned lhsBits = getActiveBits(); + unsigned lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + if (!lhsWords) // 0 * X ===> 0 return *this; // Get some bit facts about RHS and check for zero - uint32_t rhsBits = RHS.getActiveBits(); - uint32_t rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + unsigned rhsBits = RHS.getActiveBits(); + unsigned rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; if (!rhsWords) { // X * 0 ===> 0 clear(); @@ -372,7 +366,7 @@ APInt& APInt::operator*=(const APInt& RHS) { } // Allocate space for the result - uint32_t destWords = rhsWords + lhsWords; + unsigned destWords = rhsWords + lhsWords; uint64_t *dest = getMemory(destWords); // Perform the long multiply @@ -380,7 +374,7 @@ APInt& APInt::operator*=(const APInt& RHS) { // Copy result back into *this clear(); - uint32_t wordsToCopy = destWords >= getNumWords() ? getNumWords() : destWords; + unsigned wordsToCopy = destWords >= getNumWords() ? getNumWords() : destWords; memcpy(pVal, dest, wordsToCopy * APINT_WORD_SIZE); // delete dest array and return @@ -394,8 +388,8 @@ APInt& APInt::operator&=(const APInt& RHS) { VAL &= RHS.VAL; return *this; } - uint32_t numWords = getNumWords(); - for (uint32_t i = 0; i < numWords; ++i) + unsigned numWords = getNumWords(); + for (unsigned i = 0; i < numWords; ++i) pVal[i] &= RHS.pVal[i]; return *this; } @@ -406,8 +400,8 @@ APInt& APInt::operator|=(const APInt& RHS) { VAL |= RHS.VAL; return *this; } - uint32_t numWords = getNumWords(); - for (uint32_t i = 0; i < numWords; ++i) + unsigned numWords = getNumWords(); + for (unsigned i = 0; i < numWords; ++i) pVal[i] |= RHS.pVal[i]; return *this; } @@ -418,45 +412,33 @@ APInt& APInt::operator^=(const APInt& RHS) { VAL ^= RHS.VAL; this->clearUnusedBits(); return *this; - } - uint32_t numWords = getNumWords(); - for (uint32_t i = 0; i < numWords; ++i) + } + unsigned numWords = getNumWords(); + for (unsigned i = 0; i < numWords; ++i) pVal[i] ^= RHS.pVal[i]; return clearUnusedBits(); } -APInt APInt::operator&(const APInt& RHS) const { - assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); - if (isSingleWord()) - return APInt(getBitWidth(), VAL & RHS.VAL); - - uint32_t numWords = getNumWords(); +APInt APInt::AndSlowCase(const APInt& RHS) const { + unsigned numWords = getNumWords(); uint64_t* val = getMemory(numWords); - for (uint32_t i = 0; i < numWords; ++i) + for (unsigned i = 0; i < numWords; ++i) val[i] = pVal[i] & RHS.pVal[i]; return APInt(val, getBitWidth()); } -APInt APInt::operator|(const APInt& RHS) const { - assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); - if (isSingleWord()) - return APInt(getBitWidth(), VAL | RHS.VAL); - - uint32_t numWords = getNumWords(); +APInt APInt::OrSlowCase(const APInt& RHS) const { + unsigned numWords = getNumWords(); uint64_t *val = getMemory(numWords); - for (uint32_t i = 0; i < numWords; ++i) + for (unsigned i = 0; i < numWords; ++i) val[i] = pVal[i] | RHS.pVal[i]; return APInt(val, getBitWidth()); } -APInt APInt::operator^(const APInt& RHS) const { - assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); - if (isSingleWord()) - return APInt(BitWidth, VAL ^ RHS.VAL); - - uint32_t numWords = getNumWords(); +APInt APInt::XorSlowCase(const APInt& RHS) const { + unsigned numWords = getNumWords(); uint64_t *val = getMemory(numWords); - for (uint32_t i = 0; i < numWords; ++i) + for (unsigned i = 0; i < numWords; ++i) val[i] = pVal[i] ^ RHS.pVal[i]; // 0^0==1 so clear the high bits in case they got set. @@ -467,8 +449,8 @@ bool APInt::operator !() const { if (isSingleWord()) return !VAL; - for (uint32_t i = 0; i < getNumWords(); ++i) - if (pVal[i]) + for (unsigned i = 0; i < getNumWords(); ++i) + if (pVal[i]) return false; return true; } @@ -500,22 +482,18 @@ APInt APInt::operator-(const APInt& RHS) const { return Result.clearUnusedBits(); } -bool APInt::operator[](uint32_t bitPosition) const { - return (maskBit(bitPosition) & +bool APInt::operator[](unsigned bitPosition) const { + return (maskBit(bitPosition) & (isSingleWord() ? VAL : pVal[whichWord(bitPosition)])) != 0; } -bool APInt::operator==(const APInt& RHS) const { - assert(BitWidth == RHS.BitWidth && "Comparison requires equal bit widths"); - if (isSingleWord()) - return VAL == RHS.VAL; - +bool APInt::EqualSlowCase(const APInt& RHS) const { // Get some facts about the number of bits used in the two operands. - uint32_t n1 = getActiveBits(); - uint32_t n2 = RHS.getActiveBits(); + unsigned n1 = getActiveBits(); + unsigned n2 = RHS.getActiveBits(); // If the number of bits isn't the same, they aren't equal - if (n1 != n2) + if (n1 != n2) return false; // If the number of bits fits in a word, we only need to compare the low word. @@ -524,16 +502,13 @@ bool APInt::operator==(const APInt& RHS) const { // Otherwise, compare everything for (int i = whichWord(n1 - 1); i >= 0; --i) - if (pVal[i] != RHS.pVal[i]) + if (pVal[i] != RHS.pVal[i]) return false; return true; } -bool APInt::operator==(uint64_t Val) const { - if (isSingleWord()) - return VAL == Val; - - uint32_t n = getActiveBits(); +bool APInt::EqualSlowCase(uint64_t Val) const { + unsigned n = getActiveBits(); if (n <= APINT_BITS_PER_WORD) return pVal[0] == Val; else @@ -546,8 +521,8 @@ bool APInt::ult(const APInt& RHS) const { return VAL < RHS.VAL; // Get active bit length of both operands - uint32_t n1 = getActiveBits(); - uint32_t n2 = RHS.getActiveBits(); + unsigned n1 = getActiveBits(); + unsigned n2 = RHS.getActiveBits(); // If magnitude of LHS is less than RHS, return true. if (n1 < n2) @@ -562,11 +537,11 @@ bool APInt::ult(const APInt& RHS) const { return pVal[0] < RHS.pVal[0]; // Otherwise, compare all words - uint32_t topWord = whichWord(std::max(n1,n2)-1); + unsigned topWord = whichWord(std::max(n1,n2)-1); for (int i = topWord; i >= 0; --i) { - if (pVal[i] > RHS.pVal[i]) + if (pVal[i] > RHS.pVal[i]) return false; - if (pVal[i] < RHS.pVal[i]) + if (pVal[i] < RHS.pVal[i]) return true; } return false; @@ -604,89 +579,56 @@ bool APInt::slt(const APInt& RHS) const { return true; else if (rhsNeg) return false; - else + else return lhs.ult(rhs); } -APInt& APInt::set(uint32_t bitPosition) { - if (isSingleWord()) +APInt& APInt::set(unsigned bitPosition) { + if (isSingleWord()) VAL |= maskBit(bitPosition); - else + else pVal[whichWord(bitPosition)] |= maskBit(bitPosition); return *this; } -APInt& APInt::set() { - if (isSingleWord()) { - VAL = -1ULL; - return clearUnusedBits(); - } - - // Set all the bits in all the words. - for (uint32_t i = 0; i < getNumWords(); ++i) - pVal[i] = -1ULL; - // Clear the unused ones - return clearUnusedBits(); -} - /// Set the given bit to 0 whose position is given as "bitPosition". /// @brief Set a given bit to 0. -APInt& APInt::clear(uint32_t bitPosition) { - if (isSingleWord()) +APInt& APInt::clear(unsigned bitPosition) { + if (isSingleWord()) VAL &= ~maskBit(bitPosition); - else + else pVal[whichWord(bitPosition)] &= ~maskBit(bitPosition); return *this; } -/// @brief Set every bit to 0. -APInt& APInt::clear() { - if (isSingleWord()) - VAL = 0; - else - memset(pVal, 0, getNumWords() * APINT_WORD_SIZE); - return *this; -} - -/// @brief Bitwise NOT operator. Performs a bitwise logical NOT operation on -/// this APInt. -APInt APInt::operator~() const { - APInt Result(*this); - Result.flip(); - return Result; -} - /// @brief Toggle every bit to its opposite value. -APInt& APInt::flip() { - if (isSingleWord()) { - VAL ^= -1ULL; - return clearUnusedBits(); - } - for (uint32_t i = 0; i < getNumWords(); ++i) - pVal[i] ^= -1ULL; - return clearUnusedBits(); -} -/// Toggle a given bit to its opposite value whose position is given +/// Toggle a given bit to its opposite value whose position is given /// as "bitPosition". /// @brief Toggles a given bit to its opposite value. -APInt& APInt::flip(uint32_t bitPosition) { +APInt& APInt::flip(unsigned bitPosition) { assert(bitPosition < BitWidth && "Out of the bit-width range!"); if ((*this)[bitPosition]) clear(bitPosition); else set(bitPosition); return *this; } -uint32_t APInt::getBitsNeeded(const char* str, uint32_t slen, uint8_t radix) { - assert(str != 0 && "Invalid value string"); - assert(slen > 0 && "Invalid string length"); +unsigned APInt::getBitsNeeded(const StringRef& str, uint8_t radix) { + assert(!str.empty() && "Invalid string length"); + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + + size_t slen = str.size(); - // Each computation below needs to know if its negative - uint32_t isNegative = str[0] == '-'; - if (isNegative) { + // Each computation below needs to know if it's negative. + StringRef::iterator p = str.begin(); + unsigned isNegative = *p == '-'; + if (*p == '-' || *p == '+') { + p++; slen--; - str++; + assert(slen && "String is only a sign, needs a value."); } + // For radixes of power-of-two values, the bits required is accurately and // easily computed if (radix == 2) @@ -696,45 +638,127 @@ uint32_t APInt::getBitsNeeded(const char* str, uint32_t slen, uint8_t radix) { if (radix == 16) return slen * 4 + isNegative; - // Otherwise it must be radix == 10, the hard case - assert(radix == 10 && "Invalid radix"); - // This is grossly inefficient but accurate. We could probably do something // with a computation of roughly slen*64/20 and then adjust by the value of // the first few digits. But, I'm not sure how accurate that could be. // Compute a sufficient number of bits that is always large enough but might - // be too large. This avoids the assertion in the constructor. - uint32_t sufficient = slen*64/18; + // be too large. This avoids the assertion in the constructor. This + // calculation doesn't work appropriately for the numbers 0-9, so just use 4 + // bits in that case. + unsigned sufficient = slen == 1 ? 4 : slen * 64/18; // Convert to the actual binary value. - APInt tmp(sufficient, str, slen, radix); + APInt tmp(sufficient, StringRef(p, slen), radix); + + // Compute how many bits are required. If the log is infinite, assume we need + // just bit. + unsigned log = tmp.logBase2(); + if (log == (unsigned)-1) { + return isNegative + 1; + } else { + return isNegative + log + 1; + } +} + +// From http://www.burtleburtle.net, byBob Jenkins. +// When targeting x86, both GCC and LLVM seem to recognize this as a +// rotate instruction. +#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k)))) + +// From http://www.burtleburtle.net, by Bob Jenkins. +#define mix(a,b,c) \ + { \ + a -= c; a ^= rot(c, 4); c += b; \ + b -= a; b ^= rot(a, 6); a += c; \ + c -= b; c ^= rot(b, 8); b += a; \ + a -= c; a ^= rot(c,16); c += b; \ + b -= a; b ^= rot(a,19); a += c; \ + c -= b; c ^= rot(b, 4); b += a; \ + } + +// From http://www.burtleburtle.net, by Bob Jenkins. +#define final(a,b,c) \ + { \ + c ^= b; c -= rot(b,14); \ + a ^= c; a -= rot(c,11); \ + b ^= a; b -= rot(a,25); \ + c ^= b; c -= rot(b,16); \ + a ^= c; a -= rot(c,4); \ + b ^= a; b -= rot(a,14); \ + c ^= b; c -= rot(b,24); \ + } + +// hashword() was adapted from http://www.burtleburtle.net, by Bob +// Jenkins. k is a pointer to an array of uint32_t values; length is +// the length of the key, in 32-bit chunks. This version only handles +// keys that are a multiple of 32 bits in size. +static inline uint32_t hashword(const uint64_t *k64, size_t length) +{ + const uint32_t *k = reinterpret_cast(k64); + uint32_t a,b,c; + + /* Set up the internal state */ + a = b = c = 0xdeadbeef + (((uint32_t)length)<<2); + + /*------------------------------------------------- handle most of the key */ + while (length > 3) + { + a += k[0]; + b += k[1]; + c += k[2]; + mix(a,b,c); + length -= 3; + k += 3; + } - // Compute how many bits are required. - return isNegative + tmp.logBase2() + 1; + /*------------------------------------------- handle the last 3 uint32_t's */ + switch (length) { /* all the case statements fall through */ + case 3 : c+=k[2]; + case 2 : b+=k[1]; + case 1 : a+=k[0]; + final(a,b,c); + case 0: /* case 0: nothing left to add */ + break; + } + /*------------------------------------------------------ report the result */ + return c; } -uint64_t APInt::getHashValue() const { - // Put the bit width into the low order bits. - uint64_t hash = BitWidth; +// hashword8() was adapted from http://www.burtleburtle.net, by Bob +// Jenkins. This computes a 32-bit hash from one 64-bit word. When +// targeting x86 (32 or 64 bit), both LLVM and GCC compile this +// function into about 35 instructions when inlined. +static inline uint32_t hashword8(const uint64_t k64) +{ + uint32_t a,b,c; + a = b = c = 0xdeadbeef + 4; + b += k64 >> 32; + a += k64 & 0xffffffff; + final(a,b,c); + return c; +} +#undef final +#undef mix +#undef rot - // Add the sum of the words to the hash. +uint64_t APInt::getHashValue() const { + uint64_t hash; if (isSingleWord()) - hash += VAL << 6; // clear separation of up to 64 bits + hash = hashword8(VAL); else - for (uint32_t i = 0; i < getNumWords(); ++i) - hash += pVal[i] << 6; // clear sepration of up to 64 bits + hash = hashword(pVal, getNumWords()*2); return hash; } /// HiBits - This function returns the high "numBits" bits of this APInt. -APInt APInt::getHiBits(uint32_t numBits) const { +APInt APInt::getHiBits(unsigned numBits) const { return APIntOps::lshr(*this, BitWidth - numBits); } /// LoBits - This function returns the low "numBits" bits of this APInt. -APInt APInt::getLoBits(uint32_t numBits) const { - return APIntOps::lshr(APIntOps::shl(*this, BitWidth - numBits), +APInt APInt::getLoBits(unsigned numBits) const { + return APIntOps::lshr(APIntOps::shl(*this, BitWidth - numBits), BitWidth - numBits); } @@ -742,28 +766,24 @@ bool APInt::isPowerOf2() const { return (!!*this) && !(*this & (*this - APInt(BitWidth,1))); } -uint32_t APInt::countLeadingZeros() const { - uint32_t Count = 0; - if (isSingleWord()) - Count = CountLeadingZeros_64(VAL); - else { - for (uint32_t i = getNumWords(); i > 0u; --i) { - if (pVal[i-1] == 0) - Count += APINT_BITS_PER_WORD; - else { - Count += CountLeadingZeros_64(pVal[i-1]); - break; - } +unsigned APInt::countLeadingZerosSlowCase() const { + unsigned Count = 0; + for (unsigned i = getNumWords(); i > 0u; --i) { + if (pVal[i-1] == 0) + Count += APINT_BITS_PER_WORD; + else { + Count += CountLeadingZeros_64(pVal[i-1]); + break; } } - uint32_t remainder = BitWidth % APINT_BITS_PER_WORD; + unsigned remainder = BitWidth % APINT_BITS_PER_WORD; if (remainder) Count -= APINT_BITS_PER_WORD - remainder; return std::min(Count, BitWidth); } -static uint32_t countLeadingOnes_64(uint64_t V, uint32_t skip) { - uint32_t Count = 0; +static unsigned countLeadingOnes_64(uint64_t V, unsigned skip) { + unsigned Count = 0; if (skip) V <<= skip; while (V && (V & (1ULL << 63))) { @@ -773,14 +793,20 @@ static uint32_t countLeadingOnes_64(uint64_t V, uint32_t skip) { return Count; } -uint32_t APInt::countLeadingOnes() const { +unsigned APInt::countLeadingOnes() const { if (isSingleWord()) return countLeadingOnes_64(VAL, APINT_BITS_PER_WORD - BitWidth); - uint32_t highWordBits = BitWidth % APINT_BITS_PER_WORD; - uint32_t shift = (highWordBits == 0 ? 0 : APINT_BITS_PER_WORD - highWordBits); + unsigned highWordBits = BitWidth % APINT_BITS_PER_WORD; + unsigned shift; + if (!highWordBits) { + highWordBits = APINT_BITS_PER_WORD; + shift = 0; + } else { + shift = APINT_BITS_PER_WORD - highWordBits; + } int i = getNumWords() - 1; - uint32_t Count = countLeadingOnes_64(pVal[i], shift); + unsigned Count = countLeadingOnes_64(pVal[i], shift); if (Count == highWordBits) { for (i--; i >= 0; --i) { if (pVal[i] == -1ULL) @@ -794,11 +820,11 @@ uint32_t APInt::countLeadingOnes() const { return Count; } -uint32_t APInt::countTrailingZeros() const { +unsigned APInt::countTrailingZeros() const { if (isSingleWord()) - return std::min(uint32_t(CountTrailingZeros_64(VAL)), BitWidth); - uint32_t Count = 0; - uint32_t i = 0; + return std::min(unsigned(CountTrailingZeros_64(VAL)), BitWidth); + unsigned Count = 0; + unsigned i = 0; for (; i < getNumWords() && pVal[i] == 0; ++i) Count += APINT_BITS_PER_WORD; if (i < getNumWords()) @@ -806,11 +832,9 @@ uint32_t APInt::countTrailingZeros() const { return std::min(Count, BitWidth); } -uint32_t APInt::countTrailingOnes() const { - if (isSingleWord()) - return std::min(uint32_t(CountTrailingOnes_64(VAL)), BitWidth); - uint32_t Count = 0; - uint32_t i = 0; +unsigned APInt::countTrailingOnesSlowCase() const { + unsigned Count = 0; + unsigned i = 0; for (; i < getNumWords() && pVal[i] == -1ULL; ++i) Count += APINT_BITS_PER_WORD; if (i < getNumWords()) @@ -818,11 +842,9 @@ uint32_t APInt::countTrailingOnes() const { return std::min(Count, BitWidth); } -uint32_t APInt::countPopulation() const { - if (isSingleWord()) - return CountPopulation_64(VAL); - uint32_t Count = 0; - for (uint32_t i = 0; i < getNumWords(); ++i) +unsigned APInt::countPopulationSlowCase() const { + unsigned Count = 0; + for (unsigned i = 0; i < getNumWords(); ++i) Count += CountPopulation_64(pVal[i]); return Count; } @@ -832,9 +854,9 @@ APInt APInt::byteSwap() const { if (BitWidth == 16) return APInt(BitWidth, ByteSwap_16(uint16_t(VAL))); else if (BitWidth == 32) - return APInt(BitWidth, ByteSwap_32(uint32_t(VAL))); + return APInt(BitWidth, ByteSwap_32(unsigned(VAL))); else if (BitWidth == 48) { - uint32_t Tmp1 = uint32_t(VAL >> 16); + unsigned Tmp1 = unsigned(VAL >> 16); Tmp1 = ByteSwap_32(Tmp1); uint16_t Tmp2 = uint16_t(VAL); Tmp2 = ByteSwap_16(Tmp2); @@ -844,7 +866,7 @@ APInt APInt::byteSwap() const { else { APInt Result(BitWidth, 0); char *pByte = (char*)Result.pVal; - for (uint32_t i = 0; i < BitWidth / APINT_WORD_SIZE / 2; ++i) { + for (unsigned i = 0; i < BitWidth / APINT_WORD_SIZE / 2; ++i) { char Tmp = pByte[i]; pByte[i] = pByte[BitWidth / APINT_WORD_SIZE - 1 - i]; pByte[BitWidth / APINT_WORD_SIZE - i - 1] = Tmp; @@ -853,7 +875,7 @@ APInt APInt::byteSwap() const { } } -APInt llvm::APIntOps::GreatestCommonDivisor(const APInt& API1, +APInt llvm::APIntOps::GreatestCommonDivisor(const APInt& API1, const APInt& API2) { APInt A = API1, B = API2; while (!!B) { @@ -864,7 +886,7 @@ APInt llvm::APIntOps::GreatestCommonDivisor(const APInt& API1, return A; } -APInt llvm::APIntOps::RoundDoubleToAPInt(double Double, uint32_t width) { +APInt llvm::APIntOps::RoundDoubleToAPInt(double Double, unsigned width) { union { double D; uint64_t I; @@ -886,7 +908,7 @@ APInt llvm::APIntOps::RoundDoubleToAPInt(double Double, uint32_t width) { // If the exponent doesn't shift all bits out of the mantissa if (exp < 52) - return isNeg ? -APInt(width, mantissa >> (52 - exp)) : + return isNeg ? -APInt(width, mantissa >> (52 - exp)) : APInt(width, mantissa >> (52 - exp)); // If the client didn't provide enough bits for us to shift the mantissa into @@ -896,26 +918,27 @@ APInt llvm::APIntOps::RoundDoubleToAPInt(double Double, uint32_t width) { // Otherwise, we have to shift the mantissa bits up to the right location APInt Tmp(width, mantissa); - Tmp = Tmp.shl((uint32_t)exp - 52); + Tmp = Tmp.shl((unsigned)exp - 52); return isNeg ? -Tmp : Tmp; } -/// RoundToDouble - This function convert this APInt to a double. +/// RoundToDouble - This function converts this APInt to a double. /// The layout for double is as following (IEEE Standard 754): /// -------------------------------------- /// | Sign Exponent Fraction Bias | /// |-------------------------------------- | /// | 1[63] 11[62-52] 52[51-00] 1023 | -/// -------------------------------------- +/// -------------------------------------- double APInt::roundToDouble(bool isSigned) const { // Handle the simple case where the value is contained in one uint64_t. + // It is wrong to optimize getWord(0) to VAL; there might be more than one word. if (isSingleWord() || getActiveBits() <= APINT_BITS_PER_WORD) { if (isSigned) { - int64_t sext = (int64_t(VAL) << (64-BitWidth)) >> (64-BitWidth); + int64_t sext = (int64_t(getWord(0)) << (64-BitWidth)) >> (64-BitWidth); return double(sext); } else - return double(VAL); + return double(getWord(0)); } // Determine if the value is negative. @@ -925,7 +948,7 @@ double APInt::roundToDouble(bool isSigned) const { APInt Tmp(isNeg ? -(*this) : (*this)); // Figure out how many bits we're using. - uint32_t n = Tmp.getActiveBits(); + unsigned n = Tmp.getActiveBits(); // The exponent (without bias normalization) is just the number of bits // we are using. Note that the sign bit is gone since we constructed the @@ -936,7 +959,7 @@ double APInt::roundToDouble(bool isSigned) const { if (exp > 1023) { if (!isSigned || !isNeg) return std::numeric_limits::infinity(); - else + else return -std::numeric_limits::infinity(); } exp += 1023; // Increment for 1023 bias @@ -967,12 +990,12 @@ double APInt::roundToDouble(bool isSigned) const { } // Truncate to new width. -APInt &APInt::trunc(uint32_t width) { +APInt &APInt::trunc(unsigned width) { assert(width < BitWidth && "Invalid APInt Truncate request"); - assert(width >= MIN_INT_BITS && "Can't truncate to 0 bits"); - uint32_t wordsBefore = getNumWords(); + assert(width && "Can't truncate to 0 bits"); + unsigned wordsBefore = getNumWords(); BitWidth = width; - uint32_t wordsAfter = getNumWords(); + unsigned wordsAfter = getNumWords(); if (wordsBefore != wordsAfter) { if (wordsAfter == 1) { uint64_t *tmp = pVal; @@ -980,7 +1003,7 @@ APInt &APInt::trunc(uint32_t width) { delete [] tmp; } else { uint64_t *newVal = getClearedMemory(wordsAfter); - for (uint32_t i = 0; i < wordsAfter; ++i) + for (unsigned i = 0; i < wordsAfter; ++i) newVal[i] = pVal[i]; delete [] pVal; pVal = newVal; @@ -990,9 +1013,8 @@ APInt &APInt::trunc(uint32_t width) { } // Sign extend to a new width. -APInt &APInt::sext(uint32_t width) { +APInt &APInt::sext(unsigned width) { assert(width > BitWidth && "Invalid APInt SignExtend request"); - assert(width <= MAX_INT_BITS && "Too many bits"); // If the sign bit isn't set, this is the same as zext. if (!isNegative()) { zext(width); @@ -1000,14 +1022,14 @@ APInt &APInt::sext(uint32_t width) { } // The sign bit is set. First, get some facts - uint32_t wordsBefore = getNumWords(); - uint32_t wordBits = BitWidth % APINT_BITS_PER_WORD; + unsigned wordsBefore = getNumWords(); + unsigned wordBits = BitWidth % APINT_BITS_PER_WORD; BitWidth = width; - uint32_t wordsAfter = getNumWords(); + unsigned wordsAfter = getNumWords(); // Mask the high order word appropriately if (wordsBefore == wordsAfter) { - uint32_t newWordBits = width % APINT_BITS_PER_WORD; + unsigned newWordBits = width % APINT_BITS_PER_WORD; // The extension is contained to the wordsBefore-1th word. uint64_t mask = ~0ULL; if (newWordBits) @@ -1025,11 +1047,11 @@ APInt &APInt::sext(uint32_t width) { if (wordsBefore == 1) newVal[0] = VAL | mask; else { - for (uint32_t i = 0; i < wordsBefore; ++i) + for (unsigned i = 0; i < wordsBefore; ++i) newVal[i] = pVal[i]; newVal[wordsBefore-1] |= mask; } - for (uint32_t i = wordsBefore; i < wordsAfter; i++) + for (unsigned i = wordsBefore; i < wordsAfter; i++) newVal[i] = -1ULL; if (wordsBefore != 1) delete [] pVal; @@ -1038,18 +1060,17 @@ APInt &APInt::sext(uint32_t width) { } // Zero extend to a new width. -APInt &APInt::zext(uint32_t width) { +APInt &APInt::zext(unsigned width) { assert(width > BitWidth && "Invalid APInt ZeroExtend request"); - assert(width <= MAX_INT_BITS && "Too many bits"); - uint32_t wordsBefore = getNumWords(); + unsigned wordsBefore = getNumWords(); BitWidth = width; - uint32_t wordsAfter = getNumWords(); + unsigned wordsAfter = getNumWords(); if (wordsBefore != wordsAfter) { uint64_t *newVal = getClearedMemory(wordsAfter); if (wordsBefore == 1) newVal[0] = VAL; - else - for (uint32_t i = 0; i < wordsBefore; ++i) + else + for (unsigned i = 0; i < wordsBefore; ++i) newVal[i] = pVal[i]; if (wordsBefore != 1) delete [] pVal; @@ -1058,7 +1079,7 @@ APInt &APInt::zext(uint32_t width) { return *this; } -APInt &APInt::zextOrTrunc(uint32_t width) { +APInt &APInt::zextOrTrunc(unsigned width) { if (BitWidth < width) return zext(width); if (BitWidth > width) @@ -1066,7 +1087,7 @@ APInt &APInt::zextOrTrunc(uint32_t width) { return *this; } -APInt &APInt::sextOrTrunc(uint32_t width) { +APInt &APInt::sextOrTrunc(unsigned width) { if (BitWidth < width) return sext(width); if (BitWidth > width) @@ -1077,12 +1098,12 @@ APInt &APInt::sextOrTrunc(uint32_t width) { /// Arithmetic right-shift this APInt by shiftAmt. /// @brief Arithmetic right-shift function. APInt APInt::ashr(const APInt &shiftAmt) const { - return ashr((uint32_t)shiftAmt.getLimitedValue(BitWidth)); + return ashr((unsigned)shiftAmt.getLimitedValue(BitWidth)); } /// Arithmetic right-shift this APInt by shiftAmt. /// @brief Arithmetic right-shift function. -APInt APInt::ashr(uint32_t shiftAmt) const { +APInt APInt::ashr(unsigned shiftAmt) const { assert(shiftAmt <= BitWidth && "Invalid shift amount"); // Handle a degenerate case if (shiftAmt == 0) @@ -1093,8 +1114,8 @@ APInt APInt::ashr(uint32_t shiftAmt) const { if (shiftAmt == BitWidth) return APInt(BitWidth, 0); // undefined else { - uint32_t SignBit = APINT_BITS_PER_WORD - BitWidth; - return APInt(BitWidth, + unsigned SignBit = APINT_BITS_PER_WORD - BitWidth; + return APInt(BitWidth, (((int64_t(VAL) << SignBit) >> SignBit) >> shiftAmt)); } } @@ -1113,17 +1134,17 @@ APInt APInt::ashr(uint32_t shiftAmt) const { uint64_t * val = new uint64_t[getNumWords()]; // Compute some values needed by the following shift algorithms - uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; // bits to shift per word - uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; // word offset for shift - uint32_t breakWord = getNumWords() - 1 - offset; // last word affected - uint32_t bitsInWord = whichBit(BitWidth); // how many bits in last word? + unsigned wordShift = shiftAmt % APINT_BITS_PER_WORD; // bits to shift per word + unsigned offset = shiftAmt / APINT_BITS_PER_WORD; // word offset for shift + unsigned breakWord = getNumWords() - 1 - offset; // last word affected + unsigned bitsInWord = whichBit(BitWidth); // how many bits in last word? if (bitsInWord == 0) bitsInWord = APINT_BITS_PER_WORD; // If we are shifting whole words, just move whole words if (wordShift == 0) { // Move the words containing significant bits - for (uint32_t i = 0; i <= breakWord; ++i) + for (unsigned i = 0; i <= breakWord; ++i) val[i] = pVal[i+offset]; // move whole word // Adjust the top significant word for sign bit fill, if negative @@ -1131,11 +1152,11 @@ APInt APInt::ashr(uint32_t shiftAmt) const { if (bitsInWord < APINT_BITS_PER_WORD) val[breakWord] |= ~0ULL << bitsInWord; // set high bits } else { - // Shift the low order words - for (uint32_t i = 0; i < breakWord; ++i) { + // Shift the low order words + for (unsigned i = 0; i < breakWord; ++i) { // This combines the shifted corresponding word with the low bits from // the next word (shifted into this word's high bits). - val[i] = (pVal[i+offset] >> wordShift) | + val[i] = (pVal[i+offset] >> wordShift) | (pVal[i+offset+1] << (APINT_BITS_PER_WORD - wordShift)); } @@ -1148,17 +1169,17 @@ APInt APInt::ashr(uint32_t shiftAmt) const { if (isNegative()) { if (wordShift > bitsInWord) { if (breakWord > 0) - val[breakWord-1] |= + val[breakWord-1] |= ~0ULL << (APINT_BITS_PER_WORD - (wordShift - bitsInWord)); val[breakWord] |= ~0ULL; - } else + } else val[breakWord] |= (~0ULL << (bitsInWord - wordShift)); } } // Remaining words are 0 or -1, just assign them. uint64_t fillValue = (isNegative() ? -1ULL : 0); - for (uint32_t i = breakWord+1; i < getNumWords(); ++i) + for (unsigned i = breakWord+1; i < getNumWords(); ++i) val[i] = fillValue; return APInt(val, BitWidth).clearUnusedBits(); } @@ -1166,16 +1187,16 @@ APInt APInt::ashr(uint32_t shiftAmt) const { /// Logical right-shift this APInt by shiftAmt. /// @brief Logical right-shift function. APInt APInt::lshr(const APInt &shiftAmt) const { - return lshr((uint32_t)shiftAmt.getLimitedValue(BitWidth)); + return lshr((unsigned)shiftAmt.getLimitedValue(BitWidth)); } /// Logical right-shift this APInt by shiftAmt. /// @brief Logical right-shift function. -APInt APInt::lshr(uint32_t shiftAmt) const { +APInt APInt::lshr(unsigned shiftAmt) const { if (isSingleWord()) { if (shiftAmt == BitWidth) return APInt(BitWidth, 0); - else + else return APInt(BitWidth, this->VAL >> shiftAmt); } @@ -1186,7 +1207,7 @@ APInt APInt::lshr(uint32_t shiftAmt) const { return APInt(BitWidth, 0); // If none of the bits are shifted out, the result is *this. This avoids - // issues with shifting byt he size of the integer type, which produces + // issues with shifting by the size of the integer type, which produces // undefined results in the code below. This is also an optimization. if (shiftAmt == 0) return *this; @@ -1205,28 +1226,28 @@ APInt APInt::lshr(uint32_t shiftAmt) const { } // Compute some values needed by the remaining shift algorithms - uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; - uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; + unsigned wordShift = shiftAmt % APINT_BITS_PER_WORD; + unsigned offset = shiftAmt / APINT_BITS_PER_WORD; // If we are shifting whole words, just move whole words if (wordShift == 0) { - for (uint32_t i = 0; i < getNumWords() - offset; ++i) + for (unsigned i = 0; i < getNumWords() - offset; ++i) val[i] = pVal[i+offset]; - for (uint32_t i = getNumWords()-offset; i < getNumWords(); i++) + for (unsigned i = getNumWords()-offset; i < getNumWords(); i++) val[i] = 0; return APInt(val,BitWidth).clearUnusedBits(); } - // Shift the low order words - uint32_t breakWord = getNumWords() - offset -1; - for (uint32_t i = 0; i < breakWord; ++i) + // Shift the low order words + unsigned breakWord = getNumWords() - offset -1; + for (unsigned i = 0; i < breakWord; ++i) val[i] = (pVal[i+offset] >> wordShift) | (pVal[i+offset+1] << (APINT_BITS_PER_WORD - wordShift)); // Shift the break word. val[breakWord] = pVal[breakWord+offset] >> wordShift; // Remaining words are 0 - for (uint32_t i = breakWord+1; i < getNumWords(); ++i) + for (unsigned i = breakWord+1; i < getNumWords(); ++i) val[i] = 0; return APInt(val, BitWidth).clearUnusedBits(); } @@ -1234,20 +1255,11 @@ APInt APInt::lshr(uint32_t shiftAmt) const { /// Left-shift this APInt by shiftAmt. /// @brief Left-shift function. APInt APInt::shl(const APInt &shiftAmt) const { - // It's undefined behavior in C to shift by BitWidth or greater, but - return shl((uint32_t)shiftAmt.getLimitedValue(BitWidth)); + // It's undefined behavior in C to shift by BitWidth or greater. + return shl((unsigned)shiftAmt.getLimitedValue(BitWidth)); } -/// Left-shift this APInt by shiftAmt. -/// @brief Left-shift function. -APInt APInt::shl(uint32_t shiftAmt) const { - assert(shiftAmt <= BitWidth && "Invalid shift amount"); - if (isSingleWord()) { - if (shiftAmt == BitWidth) - return APInt(BitWidth, 0); // avoid undefined shift results - return APInt(BitWidth, VAL << shiftAmt); - } - +APInt APInt::shlSlowCase(unsigned shiftAmt) const { // If all the bits were shifted out, the result is 0. This avoids issues // with shifting by the size of the integer type, which produces undefined // results. We define these "undefined results" to always be 0. @@ -1266,7 +1278,7 @@ APInt APInt::shl(uint32_t shiftAmt) const { // If we are shifting less than a word, do it the easy way if (shiftAmt < APINT_BITS_PER_WORD) { uint64_t carry = 0; - for (uint32_t i = 0; i < getNumWords(); i++) { + for (unsigned i = 0; i < getNumWords(); i++) { val[i] = pVal[i] << shiftAmt | carry; carry = pVal[i] >> (APINT_BITS_PER_WORD - shiftAmt); } @@ -1274,20 +1286,20 @@ APInt APInt::shl(uint32_t shiftAmt) const { } // Compute some values needed by the remaining shift algorithms - uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; - uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; + unsigned wordShift = shiftAmt % APINT_BITS_PER_WORD; + unsigned offset = shiftAmt / APINT_BITS_PER_WORD; // If we are shifting whole words, just move whole words if (wordShift == 0) { - for (uint32_t i = 0; i < offset; i++) + for (unsigned i = 0; i < offset; i++) val[i] = 0; - for (uint32_t i = offset; i < getNumWords(); i++) + for (unsigned i = offset; i < getNumWords(); i++) val[i] = pVal[i-offset]; return APInt(val,BitWidth).clearUnusedBits(); } // Copy whole words from this to Result. - uint32_t i = getNumWords() - 1; + unsigned i = getNumWords() - 1; for (; i > offset; --i) val[i] = pVal[i-offset] << wordShift | pVal[i-offset-1] >> (APINT_BITS_PER_WORD - wordShift); @@ -1298,10 +1310,10 @@ APInt APInt::shl(uint32_t shiftAmt) const { } APInt APInt::rotl(const APInt &rotateAmt) const { - return rotl((uint32_t)rotateAmt.getLimitedValue(BitWidth)); + return rotl((unsigned)rotateAmt.getLimitedValue(BitWidth)); } -APInt APInt::rotl(uint32_t rotateAmt) const { +APInt APInt::rotl(unsigned rotateAmt) const { if (rotateAmt == 0) return *this; // Don't get too fancy, just use existing shift/or facilities @@ -1313,10 +1325,10 @@ APInt APInt::rotl(uint32_t rotateAmt) const { } APInt APInt::rotr(const APInt &rotateAmt) const { - return rotr((uint32_t)rotateAmt.getLimitedValue(BitWidth)); + return rotr((unsigned)rotateAmt.getLimitedValue(BitWidth)); } -APInt APInt::rotr(uint32_t rotateAmt) const { +APInt APInt::rotr(unsigned rotateAmt) const { if (rotateAmt == 0) return *this; // Don't get too fancy, just use existing shift/or facilities @@ -1333,11 +1345,11 @@ APInt APInt::rotr(uint32_t rotateAmt) const { // values using less than 52 bits, the value is converted to double and then // the libc sqrt function is called. The result is rounded and then converted // back to a uint64_t which is then used to construct the result. Finally, -// the Babylonian method for computing square roots is used. +// the Babylonian method for computing square roots is used. APInt APInt::sqrt() const { // Determine the magnitude of the value. - uint32_t magnitude = getActiveBits(); + unsigned magnitude = getActiveBits(); // Use a fast table for some small values. This also gets rid of some // rounding errors in libc sqrt for small values. @@ -1345,7 +1357,7 @@ APInt APInt::sqrt() const { static const uint8_t results[32] = { /* 0 */ 0, /* 1- 2 */ 1, 1, - /* 3- 6 */ 2, 2, 2, 2, + /* 3- 6 */ 2, 2, 2, 2, /* 7-12 */ 3, 3, 3, 3, 3, 3, /* 13-20 */ 4, 4, 4, 4, 4, 4, 4, 4, /* 21-30 */ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, @@ -1361,10 +1373,10 @@ APInt APInt::sqrt() const { if (magnitude < 52) { #ifdef _MSC_VER // Amazingly, VC++ doesn't have round(). - return APInt(BitWidth, + return APInt(BitWidth, uint64_t(::sqrt(double(isSingleWord()?VAL:pVal[0]))) + 0.5); #else - return APInt(BitWidth, + return APInt(BitWidth, uint64_t(::round(::sqrt(double(isSingleWord()?VAL:pVal[0]))))); #endif } @@ -1373,21 +1385,21 @@ APInt APInt::sqrt() const { // is a classical Babylonian method for computing the square root. This code // was adapted to APINt from a wikipedia article on such computations. // See http://www.wikipedia.org/ and go to the page named - // Calculate_an_integer_square_root. - uint32_t nbits = BitWidth, i = 4; + // Calculate_an_integer_square_root. + unsigned nbits = BitWidth, i = 4; APInt testy(BitWidth, 16); APInt x_old(BitWidth, 1); APInt x_new(BitWidth, 0); APInt two(BitWidth, 2); // Select a good starting value using binary logarithms. - for (;; i += 2, testy = testy.shl(2)) + for (;; i += 2, testy = testy.shl(2)) if (i >= nbits || this->ule(testy)) { x_old = x_old.shl(i / 2); break; } - // Use the Babylonian method to arrive at the integer square root: + // Use the Babylonian method to arrive at the integer square root: for (;;) { x_new = (this->udiv(x_old) + x_old).udiv(two); if (x_old.ule(x_new)) @@ -1396,9 +1408,9 @@ APInt APInt::sqrt() const { } // Make sure we return the closest approximation - // NOTE: The rounding calculation below is correct. It will produce an + // NOTE: The rounding calculation below is correct. It will produce an // off-by-one discrepancy with results from pari/gp. That discrepancy has been - // determined to be a rounding issue with pari/gp as it begins to use a + // determined to be a rounding issue with pari/gp as it begins to use a // floating point representation after 192 bits. There are no discrepancies // between this algorithm and pari/gp for bit widths < 192 bits. APInt square(x_old * x_old); @@ -1413,7 +1425,7 @@ APInt APInt::sqrt() const { else return x_old + 1; } else - assert(0 && "Error in APInt::sqrt computation"); + llvm_unreachable("Error in APInt::sqrt computation"); return x_old + 1; } @@ -1436,7 +1448,7 @@ APInt APInt::multiplicativeInverse(const APInt& modulo) const { APInt r[2] = { modulo, *this }; APInt t[2] = { APInt(BitWidth, 0), APInt(BitWidth, 1) }; APInt q(BitWidth, 0); - + unsigned i; for (i = 0; r[i^1] != 0; i ^= 1) { // An overview of the math without the confusing bit-flipping: @@ -1461,12 +1473,102 @@ APInt APInt::multiplicativeInverse(const APInt& modulo) const { return t[i].isNegative() ? t[i] + modulo : t[i]; } +/// Calculate the magic numbers required to implement a signed integer division +/// by a constant as a sequence of multiplies, adds and shifts. Requires that +/// the divisor not be 0, 1, or -1. Taken from "Hacker's Delight", Henry S. +/// Warren, Jr., chapter 10. +APInt::ms APInt::magic() const { + const APInt& d = *this; + unsigned p; + APInt ad, anc, delta, q1, r1, q2, r2, t; + APInt signedMin = APInt::getSignedMinValue(d.getBitWidth()); + struct ms mag; + + ad = d.abs(); + t = signedMin + (d.lshr(d.getBitWidth() - 1)); + anc = t - 1 - t.urem(ad); // absolute value of nc + p = d.getBitWidth() - 1; // initialize p + q1 = signedMin.udiv(anc); // initialize q1 = 2p/abs(nc) + r1 = signedMin - q1*anc; // initialize r1 = rem(2p,abs(nc)) + q2 = signedMin.udiv(ad); // initialize q2 = 2p/abs(d) + r2 = signedMin - q2*ad; // initialize r2 = rem(2p,abs(d)) + do { + p = p + 1; + q1 = q1<<1; // update q1 = 2p/abs(nc) + r1 = r1<<1; // update r1 = rem(2p/abs(nc)) + if (r1.uge(anc)) { // must be unsigned comparison + q1 = q1 + 1; + r1 = r1 - anc; + } + q2 = q2<<1; // update q2 = 2p/abs(d) + r2 = r2<<1; // update r2 = rem(2p/abs(d)) + if (r2.uge(ad)) { // must be unsigned comparison + q2 = q2 + 1; + r2 = r2 - ad; + } + delta = ad - r2; + } while (q1.ule(delta) || (q1 == delta && r1 == 0)); + + mag.m = q2 + 1; + if (d.isNegative()) mag.m = -mag.m; // resulting magic number + mag.s = p - d.getBitWidth(); // resulting shift + return mag; +} + +/// Calculate the magic numbers required to implement an unsigned integer +/// division by a constant as a sequence of multiplies, adds and shifts. +/// Requires that the divisor not be 0. Taken from "Hacker's Delight", Henry +/// S. Warren, Jr., chapter 10. +APInt::mu APInt::magicu() const { + const APInt& d = *this; + unsigned p; + APInt nc, delta, q1, r1, q2, r2; + struct mu magu; + magu.a = 0; // initialize "add" indicator + APInt allOnes = APInt::getAllOnesValue(d.getBitWidth()); + APInt signedMin = APInt::getSignedMinValue(d.getBitWidth()); + APInt signedMax = APInt::getSignedMaxValue(d.getBitWidth()); + + nc = allOnes - (-d).urem(d); + p = d.getBitWidth() - 1; // initialize p + q1 = signedMin.udiv(nc); // initialize q1 = 2p/nc + r1 = signedMin - q1*nc; // initialize r1 = rem(2p,nc) + q2 = signedMax.udiv(d); // initialize q2 = (2p-1)/d + r2 = signedMax - q2*d; // initialize r2 = rem((2p-1),d) + do { + p = p + 1; + if (r1.uge(nc - r1)) { + q1 = q1 + q1 + 1; // update q1 + r1 = r1 + r1 - nc; // update r1 + } + else { + q1 = q1+q1; // update q1 + r1 = r1+r1; // update r1 + } + if ((r2 + 1).uge(d - r2)) { + if (q2.uge(signedMax)) magu.a = 1; + q2 = q2+q2 + 1; // update q2 + r2 = r2+r2 + 1 - d; // update r2 + } + else { + if (q2.uge(signedMin)) magu.a = 1; + q2 = q2+q2; // update q2 + r2 = r2+r2 + 1; // update r2 + } + delta = d - 1 - r2; + } while (p < d.getBitWidth()*2 && + (q1.ult(delta) || (q1 == delta && r1 == 0))); + magu.m = q2 + 1; // resulting magic number + magu.s = p - d.getBitWidth(); // resulting shift + return magu; +} + /// Implementation of Knuth's Algorithm D (Division of nonnegative integers) /// from "Art of Computer Programming, Volume 2", section 4.3.1, p. 272. The /// variables here have the same names as in the algorithm. Comments explain /// the algorithm and any deviation from it. -static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r, - uint32_t m, uint32_t n) { +static void KnuthDiv(unsigned *u, unsigned *v, unsigned *q, unsigned* r, + unsigned m, unsigned n) { assert(u && "Must provide dividend"); assert(v && "Must provide divisor"); assert(q && "Must provide quotient"); @@ -1478,59 +1580,59 @@ static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r, uint64_t b = uint64_t(1) << 32; #if 0 - DEBUG(cerr << "KnuthDiv: m=" << m << " n=" << n << '\n'); - DEBUG(cerr << "KnuthDiv: original:"); - DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << std::setbase(16) << u[i]); - DEBUG(cerr << " by"); - DEBUG(for (int i = n; i >0; i--) cerr << " " << std::setbase(16) << v[i-1]); - DEBUG(cerr << '\n'); + DEBUG(dbgs() << "KnuthDiv: m=" << m << " n=" << n << '\n'); + DEBUG(dbgs() << "KnuthDiv: original:"); + DEBUG(for (int i = m+n; i >=0; i--) dbgs() << " " << u[i]); + DEBUG(dbgs() << " by"); + DEBUG(for (int i = n; i >0; i--) dbgs() << " " << v[i-1]); + DEBUG(dbgs() << '\n'); #endif - // D1. [Normalize.] Set d = b / (v[n-1] + 1) and multiply all the digits of - // u and v by d. Note that we have taken Knuth's advice here to use a power - // of 2 value for d such that d * v[n-1] >= b/2 (b is the base). A power of - // 2 allows us to shift instead of multiply and it is easy to determine the + // D1. [Normalize.] Set d = b / (v[n-1] + 1) and multiply all the digits of + // u and v by d. Note that we have taken Knuth's advice here to use a power + // of 2 value for d such that d * v[n-1] >= b/2 (b is the base). A power of + // 2 allows us to shift instead of multiply and it is easy to determine the // shift amount from the leading zeros. We are basically normalizing the u // and v so that its high bits are shifted to the top of v's range without // overflow. Note that this can require an extra word in u so that u must // be of length m+n+1. - uint32_t shift = CountLeadingZeros_32(v[n-1]); - uint32_t v_carry = 0; - uint32_t u_carry = 0; + unsigned shift = CountLeadingZeros_32(v[n-1]); + unsigned v_carry = 0; + unsigned u_carry = 0; if (shift) { - for (uint32_t i = 0; i < m+n; ++i) { - uint32_t u_tmp = u[i] >> (32 - shift); + for (unsigned i = 0; i < m+n; ++i) { + unsigned u_tmp = u[i] >> (32 - shift); u[i] = (u[i] << shift) | u_carry; u_carry = u_tmp; } - for (uint32_t i = 0; i < n; ++i) { - uint32_t v_tmp = v[i] >> (32 - shift); + for (unsigned i = 0; i < n; ++i) { + unsigned v_tmp = v[i] >> (32 - shift); v[i] = (v[i] << shift) | v_carry; v_carry = v_tmp; } } u[m+n] = u_carry; #if 0 - DEBUG(cerr << "KnuthDiv: normal:"); - DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << std::setbase(16) << u[i]); - DEBUG(cerr << " by"); - DEBUG(for (int i = n; i >0; i--) cerr << " " << std::setbase(16) << v[i-1]); - DEBUG(cerr << '\n'); + DEBUG(dbgs() << "KnuthDiv: normal:"); + DEBUG(for (int i = m+n; i >=0; i--) dbgs() << " " << u[i]); + DEBUG(dbgs() << " by"); + DEBUG(for (int i = n; i >0; i--) dbgs() << " " << v[i-1]); + DEBUG(dbgs() << '\n'); #endif // D2. [Initialize j.] Set j to m. This is the loop counter over the places. int j = m; do { - DEBUG(cerr << "KnuthDiv: quotient digit #" << j << '\n'); - // D3. [Calculate q'.]. + DEBUG(dbgs() << "KnuthDiv: quotient digit #" << j << '\n'); + // D3. [Calculate q'.]. // Set qp = (u[j+n]*b + u[j+n-1]) / v[n-1]. (qp=qprime=q') // Set rp = (u[j+n]*b + u[j+n-1]) % v[n-1]. (rp=rprime=r') // Now test if qp == b or qp*v[n-2] > b*rp + u[j+n-2]; if so, decrease // qp by 1, inrease rp by v[n-1], and repeat this test if rp < b. The test // on v[n-2] determines at high speed most of the cases in which the trial - // value qp is one too large, and it eliminates all cases where qp is two - // too large. + // value qp is one too large, and it eliminates all cases where qp is two + // too large. uint64_t dividend = ((uint64_t(u[j+n]) << 32) + u[j+n-1]); - DEBUG(cerr << "KnuthDiv: dividend == " << dividend << '\n'); + DEBUG(dbgs() << "KnuthDiv: dividend == " << dividend << '\n'); uint64_t qp = dividend / v[n-1]; uint64_t rp = dividend % v[n-1]; if (qp == b || qp*v[n-2] > b*rp + u[j+n-2]) { @@ -1539,82 +1641,82 @@ static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r, if (rp < b && (qp == b || qp*v[n-2] > b*rp + u[j+n-2])) qp--; } - DEBUG(cerr << "KnuthDiv: qp == " << qp << ", rp == " << rp << '\n'); + DEBUG(dbgs() << "KnuthDiv: qp == " << qp << ", rp == " << rp << '\n'); // D4. [Multiply and subtract.] Replace (u[j+n]u[j+n-1]...u[j]) with // (u[j+n]u[j+n-1]..u[j]) - qp * (v[n-1]...v[1]v[0]). This computation // consists of a simple multiplication by a one-place number, combined with - // a subtraction. + // a subtraction. bool isNeg = false; - for (uint32_t i = 0; i < n; ++i) { + for (unsigned i = 0; i < n; ++i) { uint64_t u_tmp = uint64_t(u[j+i]) | (uint64_t(u[j+i+1]) << 32); uint64_t subtrahend = uint64_t(qp) * uint64_t(v[i]); bool borrow = subtrahend > u_tmp; - DEBUG(cerr << "KnuthDiv: u_tmp == " << u_tmp - << ", subtrahend == " << subtrahend - << ", borrow = " << borrow << '\n'); + DEBUG(dbgs() << "KnuthDiv: u_tmp == " << u_tmp + << ", subtrahend == " << subtrahend + << ", borrow = " << borrow << '\n'); uint64_t result = u_tmp - subtrahend; - uint32_t k = j + i; - u[k++] = (uint32_t)(result & (b-1)); // subtract low word - u[k++] = (uint32_t)(result >> 32); // subtract high word + unsigned k = j + i; + u[k++] = (unsigned)(result & (b-1)); // subtract low word + u[k++] = (unsigned)(result >> 32); // subtract high word while (borrow && k <= m+n) { // deal with borrow to the left borrow = u[k] == 0; u[k]--; k++; } isNeg |= borrow; - DEBUG(cerr << "KnuthDiv: u[j+i] == " << u[j+i] << ", u[j+i+1] == " << - u[j+i+1] << '\n'); + DEBUG(dbgs() << "KnuthDiv: u[j+i] == " << u[j+i] << ", u[j+i+1] == " << + u[j+i+1] << '\n'); } - DEBUG(cerr << "KnuthDiv: after subtraction:"); - DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << u[i]); - DEBUG(cerr << '\n'); - // The digits (u[j+n]...u[j]) should be kept positive; if the result of - // this step is actually negative, (u[j+n]...u[j]) should be left as the + DEBUG(dbgs() << "KnuthDiv: after subtraction:"); + DEBUG(for (int i = m+n; i >=0; i--) dbgs() << " " << u[i]); + DEBUG(dbgs() << '\n'); + // The digits (u[j+n]...u[j]) should be kept positive; if the result of + // this step is actually negative, (u[j+n]...u[j]) should be left as the // true value plus b**(n+1), namely as the b's complement of // the true value, and a "borrow" to the left should be remembered. // if (isNeg) { bool carry = true; // true because b's complement is "complement + 1" - for (uint32_t i = 0; i <= m+n; ++i) { + for (unsigned i = 0; i <= m+n; ++i) { u[i] = ~u[i] + carry; // b's complement carry = carry && u[i] == 0; } } - DEBUG(cerr << "KnuthDiv: after complement:"); - DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << u[i]); - DEBUG(cerr << '\n'); + DEBUG(dbgs() << "KnuthDiv: after complement:"); + DEBUG(for (int i = m+n; i >=0; i--) dbgs() << " " << u[i]); + DEBUG(dbgs() << '\n'); - // D5. [Test remainder.] Set q[j] = qp. If the result of step D4 was + // D5. [Test remainder.] Set q[j] = qp. If the result of step D4 was // negative, go to step D6; otherwise go on to step D7. - q[j] = (uint32_t)qp; + q[j] = (unsigned)qp; if (isNeg) { - // D6. [Add back]. The probability that this step is necessary is very + // D6. [Add back]. The probability that this step is necessary is very // small, on the order of only 2/b. Make sure that test data accounts for - // this possibility. Decrease q[j] by 1 + // this possibility. Decrease q[j] by 1 q[j]--; - // and add (0v[n-1]...v[1]v[0]) to (u[j+n]u[j+n-1]...u[j+1]u[j]). - // A carry will occur to the left of u[j+n], and it should be ignored + // and add (0v[n-1]...v[1]v[0]) to (u[j+n]u[j+n-1]...u[j+1]u[j]). + // A carry will occur to the left of u[j+n], and it should be ignored // since it cancels with the borrow that occurred in D4. bool carry = false; - for (uint32_t i = 0; i < n; i++) { - uint32_t limit = std::min(u[j+i],v[i]); + for (unsigned i = 0; i < n; i++) { + unsigned limit = std::min(u[j+i],v[i]); u[j+i] += v[i] + carry; carry = u[j+i] < limit || (carry && u[j+i] == limit); } u[j+n] += carry; } - DEBUG(cerr << "KnuthDiv: after correction:"); - DEBUG(for (int i = m+n; i >=0; i--) cerr <<" " << u[i]); - DEBUG(cerr << "\nKnuthDiv: digit result = " << q[j] << '\n'); + DEBUG(dbgs() << "KnuthDiv: after correction:"); + DEBUG(for (int i = m+n; i >=0; i--) dbgs() <<" " << u[i]); + DEBUG(dbgs() << "\nKnuthDiv: digit result = " << q[j] << '\n'); // D7. [Loop on j.] Decrease j by one. Now if j >= 0, go back to D3. } while (--j >= 0); - DEBUG(cerr << "KnuthDiv: quotient:"); - DEBUG(for (int i = m; i >=0; i--) cerr <<" " << q[i]); - DEBUG(cerr << '\n'); + DEBUG(dbgs() << "KnuthDiv: quotient:"); + DEBUG(for (int i = m; i >=0; i--) dbgs() <<" " << q[i]); + DEBUG(dbgs() << '\n'); // D8. [Unnormalize]. Now q[...] is the desired quotient, and the desired // remainder may be obtained by dividing u[...] by d. If r is non-null we @@ -1624,50 +1726,50 @@ static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r, // multiplication by d by using a shift left. So, all we have to do is // shift right here. In order to mak if (shift) { - uint32_t carry = 0; - DEBUG(cerr << "KnuthDiv: remainder:"); + unsigned carry = 0; + DEBUG(dbgs() << "KnuthDiv: remainder:"); for (int i = n-1; i >= 0; i--) { r[i] = (u[i] >> shift) | carry; carry = u[i] << (32 - shift); - DEBUG(cerr << " " << r[i]); + DEBUG(dbgs() << " " << r[i]); } } else { for (int i = n-1; i >= 0; i--) { r[i] = u[i]; - DEBUG(cerr << " " << r[i]); + DEBUG(dbgs() << " " << r[i]); } } - DEBUG(cerr << '\n'); + DEBUG(dbgs() << '\n'); } #if 0 - DEBUG(cerr << std::setbase(10) << '\n'); + DEBUG(dbgs() << '\n'); #endif } -void APInt::divide(const APInt LHS, uint32_t lhsWords, - const APInt &RHS, uint32_t rhsWords, +void APInt::divide(const APInt LHS, unsigned lhsWords, + const APInt &RHS, unsigned rhsWords, APInt *Quotient, APInt *Remainder) { assert(lhsWords >= rhsWords && "Fractional result"); - // First, compose the values into an array of 32-bit words instead of + // First, compose the values into an array of 32-bit words instead of // 64-bit words. This is a necessity of both the "short division" algorithm - // and the the Knuth "classical algorithm" which requires there to be native - // operations for +, -, and * on an m bit value with an m*2 bit result. We - // can't use 64-bit operands here because we don't have native results of - // 128-bits. Furthremore, casting the 64-bit values to 32-bit values won't + // and the the Knuth "classical algorithm" which requires there to be native + // operations for +, -, and * on an m bit value with an m*2 bit result. We + // can't use 64-bit operands here because we don't have native results of + // 128-bits. Furthermore, casting the 64-bit values to 32-bit values won't // work on large-endian machines. - uint64_t mask = ~0ull >> (sizeof(uint32_t)*8); - uint32_t n = rhsWords * 2; - uint32_t m = (lhsWords * 2) - n; + uint64_t mask = ~0ull >> (sizeof(unsigned)*CHAR_BIT); + unsigned n = rhsWords * 2; + unsigned m = (lhsWords * 2) - n; // Allocate space for the temporary values we need either on the stack, if // it will fit, or on the heap if it won't. - uint32_t SPACE[128]; - uint32_t *U = 0; - uint32_t *V = 0; - uint32_t *Q = 0; - uint32_t *R = 0; + unsigned SPACE[128]; + unsigned *U = 0; + unsigned *V = 0; + unsigned *Q = 0; + unsigned *R = 0; if ((Remainder?4:3)*n+2*m+1 <= 128) { U = &SPACE[0]; V = &SPACE[m+n+1]; @@ -1675,38 +1777,38 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, if (Remainder) R = &SPACE[(m+n+1) + n + (m+n)]; } else { - U = new uint32_t[m + n + 1]; - V = new uint32_t[n]; - Q = new uint32_t[m+n]; + U = new unsigned[m + n + 1]; + V = new unsigned[n]; + Q = new unsigned[m+n]; if (Remainder) - R = new uint32_t[n]; + R = new unsigned[n]; } // Initialize the dividend - memset(U, 0, (m+n+1)*sizeof(uint32_t)); + memset(U, 0, (m+n+1)*sizeof(unsigned)); for (unsigned i = 0; i < lhsWords; ++i) { uint64_t tmp = (LHS.getNumWords() == 1 ? LHS.VAL : LHS.pVal[i]); - U[i * 2] = (uint32_t)(tmp & mask); - U[i * 2 + 1] = (uint32_t)(tmp >> (sizeof(uint32_t)*8)); + U[i * 2] = (unsigned)(tmp & mask); + U[i * 2 + 1] = (unsigned)(tmp >> (sizeof(unsigned)*CHAR_BIT)); } U[m+n] = 0; // this extra word is for "spill" in the Knuth algorithm. // Initialize the divisor - memset(V, 0, (n)*sizeof(uint32_t)); + memset(V, 0, (n)*sizeof(unsigned)); for (unsigned i = 0; i < rhsWords; ++i) { uint64_t tmp = (RHS.getNumWords() == 1 ? RHS.VAL : RHS.pVal[i]); - V[i * 2] = (uint32_t)(tmp & mask); - V[i * 2 + 1] = (uint32_t)(tmp >> (sizeof(uint32_t)*8)); + V[i * 2] = (unsigned)(tmp & mask); + V[i * 2 + 1] = (unsigned)(tmp >> (sizeof(unsigned)*CHAR_BIT)); } // initialize the quotient and remainder - memset(Q, 0, (m+n) * sizeof(uint32_t)); + memset(Q, 0, (m+n) * sizeof(unsigned)); if (Remainder) - memset(R, 0, n * sizeof(uint32_t)); + memset(R, 0, n * sizeof(unsigned)); - // Now, adjust m and n for the Knuth division. n is the number of words in + // Now, adjust m and n for the Knuth division. n is the number of words in // the divisor. m is the number of words by which the dividend exceeds the - // divisor (i.e. m+n is the length of the dividend). These sizes must not + // divisor (i.e. m+n is the length of the dividend). These sizes must not // contain any zero words or the Knuth algorithm fails. for (unsigned i = n; i > 0 && V[i-1] == 0; i--) { n--; @@ -1723,8 +1825,8 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, // are using base 2^32 instead of base 10. assert(n != 0 && "Divide by zero?"); if (n == 1) { - uint32_t divisor = V[0]; - uint32_t remainder = 0; + unsigned divisor = V[0]; + unsigned remainder = 0; for (int i = m+n-1; i >= 0; i--) { uint64_t partial_dividend = uint64_t(remainder) << 32 | U[i]; if (partial_dividend == 0) { @@ -1732,13 +1834,13 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, remainder = 0; } else if (partial_dividend < divisor) { Q[i] = 0; - remainder = (uint32_t)partial_dividend; + remainder = (unsigned)partial_dividend; } else if (partial_dividend == divisor) { Q[i] = 1; remainder = 0; } else { - Q[i] = (uint32_t)(partial_dividend / divisor); - remainder = (uint32_t)(partial_dividend - (Q[i] * divisor)); + Q[i] = (unsigned)(partial_dividend / divisor); + remainder = (unsigned)(partial_dividend - (Q[i] * divisor)); } } if (R) @@ -1763,10 +1865,10 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, } else Quotient->clear(); - // The quotient is in Q. Reconstitute the quotient into Quotient's low + // The quotient is in Q. Reconstitute the quotient into Quotient's low // order words. if (lhsWords == 1) { - uint64_t tmp = + uint64_t tmp = uint64_t(Q[0]) | (uint64_t(Q[1]) << (APINT_BITS_PER_WORD / 2)); if (Quotient->isSingleWord()) Quotient->VAL = tmp; @@ -1775,7 +1877,7 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, } else { assert(!Quotient->isSingleWord() && "Quotient APInt not large enough"); for (unsigned i = 0; i < lhsWords; ++i) - Quotient->pVal[i] = + Quotient->pVal[i] = uint64_t(Q[i*2]) | (uint64_t(Q[i*2+1]) << (APINT_BITS_PER_WORD / 2)); } } @@ -1797,7 +1899,7 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, // The remainder is in R. Reconstitute the remainder into Remainder's low // order words. if (rhsWords == 1) { - uint64_t tmp = + uint64_t tmp = uint64_t(R[0]) | (uint64_t(R[1]) << (APINT_BITS_PER_WORD / 2)); if (Remainder->isSingleWord()) Remainder->VAL = tmp; @@ -1806,7 +1908,7 @@ void APInt::divide(const APInt LHS, uint32_t lhsWords, } else { assert(!Remainder->isSingleWord() && "Remainder APInt not large enough"); for (unsigned i = 0; i < rhsWords; ++i) - Remainder->pVal[i] = + Remainder->pVal[i] = uint64_t(R[i*2]) | (uint64_t(R[i*2+1]) << (APINT_BITS_PER_WORD / 2)); } } @@ -1830,16 +1932,16 @@ APInt APInt::udiv(const APInt& RHS) const { } // Get some facts about the LHS and RHS number of bits and words - uint32_t rhsBits = RHS.getActiveBits(); - uint32_t rhsWords = !rhsBits ? 0 : (APInt::whichWord(rhsBits - 1) + 1); + unsigned rhsBits = RHS.getActiveBits(); + unsigned rhsWords = !rhsBits ? 0 : (APInt::whichWord(rhsBits - 1) + 1); assert(rhsWords && "Divided by zero???"); - uint32_t lhsBits = this->getActiveBits(); - uint32_t lhsWords = !lhsBits ? 0 : (APInt::whichWord(lhsBits - 1) + 1); + unsigned lhsBits = this->getActiveBits(); + unsigned lhsWords = !lhsBits ? 0 : (APInt::whichWord(lhsBits - 1) + 1); // Deal with some degenerate cases - if (!lhsWords) + if (!lhsWords) // 0 / X ===> 0 - return APInt(BitWidth, 0); + return APInt(BitWidth, 0); else if (lhsWords < rhsWords || this->ult(RHS)) { // X / Y ===> 0, iff X < Y return APInt(BitWidth, 0); @@ -1865,12 +1967,12 @@ APInt APInt::urem(const APInt& RHS) const { } // Get some facts about the LHS - uint32_t lhsBits = getActiveBits(); - uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + unsigned lhsBits = getActiveBits(); + unsigned lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); // Get some facts about the RHS - uint32_t rhsBits = RHS.getActiveBits(); - uint32_t rhsWords = !rhsBits ? 0 : (APInt::whichWord(rhsBits - 1) + 1); + unsigned rhsBits = RHS.getActiveBits(); + unsigned rhsWords = !rhsBits ? 0 : (APInt::whichWord(rhsBits - 1) + 1); assert(rhsWords && "Performing remainder operation by zero ???"); // Check the degenerate cases @@ -1894,33 +1996,33 @@ APInt APInt::urem(const APInt& RHS) const { return Remainder; } -void APInt::udivrem(const APInt &LHS, const APInt &RHS, +void APInt::udivrem(const APInt &LHS, const APInt &RHS, APInt &Quotient, APInt &Remainder) { // Get some size facts about the dividend and divisor - uint32_t lhsBits = LHS.getActiveBits(); - uint32_t lhsWords = !lhsBits ? 0 : (APInt::whichWord(lhsBits - 1) + 1); - uint32_t rhsBits = RHS.getActiveBits(); - uint32_t rhsWords = !rhsBits ? 0 : (APInt::whichWord(rhsBits - 1) + 1); + unsigned lhsBits = LHS.getActiveBits(); + unsigned lhsWords = !lhsBits ? 0 : (APInt::whichWord(lhsBits - 1) + 1); + unsigned rhsBits = RHS.getActiveBits(); + unsigned rhsWords = !rhsBits ? 0 : (APInt::whichWord(rhsBits - 1) + 1); // Check the degenerate cases - if (lhsWords == 0) { + if (lhsWords == 0) { Quotient = 0; // 0 / Y ===> 0 Remainder = 0; // 0 % Y ===> 0 return; - } - - if (lhsWords < rhsWords || LHS.ult(RHS)) { - Quotient = 0; // X / Y ===> 0, iff X < Y + } + + if (lhsWords < rhsWords || LHS.ult(RHS)) { Remainder = LHS; // X % Y ===> X, iff X < Y + Quotient = 0; // X / Y ===> 0, iff X < Y return; - } - + } + if (LHS == RHS) { Quotient = 1; // X / X ===> 1 Remainder = 0; // X % X ===> 0; return; - } - + } + if (lhsWords == 1 && rhsWords == 1) { // There is only one word to consider so use the native versions. uint64_t lhsValue = LHS.isSingleWord() ? LHS.VAL : LHS.pVal[0]; @@ -1934,26 +2036,32 @@ void APInt::udivrem(const APInt &LHS, const APInt &RHS, divide(LHS, lhsWords, RHS, rhsWords, &Quotient, &Remainder); } -void APInt::fromString(uint32_t numbits, const char *str, uint32_t slen, - uint8_t radix) { +void APInt::fromString(unsigned numbits, const StringRef& str, uint8_t radix) { // Check our assumptions here + assert(!str.empty() && "Invalid string length"); assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && "Radix should be 2, 8, 10, or 16!"); - assert(str && "String is null?"); - bool isNeg = str[0] == '-'; - if (isNeg) - str++, slen--; + + StringRef::iterator p = str.begin(); + size_t slen = str.size(); + bool isNeg = *p == '-'; + if (*p == '-' || *p == '+') { + p++; + slen--; + assert(slen && "String is only a sign, needs a value."); + } assert((slen <= numbits || radix != 2) && "Insufficient bit width"); - assert((slen*3 <= numbits || radix != 8) && "Insufficient bit width"); - assert((slen*4 <= numbits || radix != 16) && "Insufficient bit width"); - assert(((slen*64)/22 <= numbits || radix != 10) && "Insufficient bit width"); + assert(((slen-1)*3 <= numbits || radix != 8) && "Insufficient bit width"); + assert(((slen-1)*4 <= numbits || radix != 16) && "Insufficient bit width"); + assert((((slen-1)*64)/22 <= numbits || radix != 10) + && "Insufficient bit width"); // Allocate memory if (!isSingleWord()) pVal = getClearedMemory(getNumWords()); // Figure out if we can shift instead of multiply - uint32_t shift = (radix == 16 ? 4 : radix == 8 ? 3 : radix == 2 ? 1 : 0); + unsigned shift = (radix == 16 ? 4 : radix == 8 ? 3 : radix == 2 ? 1 : 0); // Set up an APInt for the digit to add outside the loop so we don't // constantly construct/destruct it. @@ -1961,36 +2069,17 @@ void APInt::fromString(uint32_t numbits, const char *str, uint32_t slen, APInt apradix(getBitWidth(), radix); // Enter digit traversal loop - for (unsigned i = 0; i < slen; i++) { - // Get a digit - uint32_t digit = 0; - char cdigit = str[i]; - if (radix == 16) { - if (!isxdigit(cdigit)) - assert(0 && "Invalid hex digit in string"); - if (isdigit(cdigit)) - digit = cdigit - '0'; - else if (cdigit >= 'a') - digit = cdigit - 'a' + 10; - else if (cdigit >= 'A') - digit = cdigit - 'A' + 10; - else - assert(0 && "huh? we shouldn't get here"); - } else if (isdigit(cdigit)) { - digit = cdigit - '0'; - assert((radix == 10 || - (radix == 8 && digit != 8 && digit != 9) || - (radix == 2 && (digit == 0 || digit == 1))) && - "Invalid digit in string for given radix"); - } else { - assert(0 && "Invalid character in digit string"); - } + for (StringRef::iterator e = str.end(); p != e; ++p) { + unsigned digit = getDigit(*p, radix); + assert(digit < radix && "Invalid character in digit string"); // Shift or multiply the value by the radix - if (shift) - *this <<= shift; - else - *this *= apradix; + if (slen > 1) { + if (shift) + *this <<= shift; + else + *this *= apradix; + } // Add in the digit we just interpreted if (apdigit.isSingleWord()) @@ -2010,19 +2099,19 @@ void APInt::toString(SmallVectorImpl &Str, unsigned Radix, bool Signed) const { assert((Radix == 10 || Radix == 8 || Radix == 16 || Radix == 2) && "Radix should be 2, 8, 10, or 16!"); - + // First, check for a zero value and just short circuit the logic below. if (*this == 0) { Str.push_back('0'); return; } - + static const char Digits[] = "0123456789ABCDEF"; - + if (isSingleWord()) { char Buffer[65]; char *BufPtr = Buffer+65; - + uint64_t N; if (Signed) { int64_t I = getSExtValue(); @@ -2034,7 +2123,7 @@ void APInt::toString(SmallVectorImpl &Str, unsigned Radix, } else { N = getZExtValue(); } - + while (N) { *--BufPtr = Digits[N % Radix]; N /= Radix; @@ -2044,7 +2133,7 @@ void APInt::toString(SmallVectorImpl &Str, unsigned Radix, } APInt Tmp(*this); - + if (Signed && isNegative()) { // They want to print the signed version and it is a negative value // Flip the bits and add one to turn it into the equivalent positive @@ -2053,18 +2142,18 @@ void APInt::toString(SmallVectorImpl &Str, unsigned Radix, Tmp++; Str.push_back('-'); } - + // We insert the digits backward, then reverse them to get the right order. unsigned StartDig = Str.size(); - - // For the 2, 8 and 16 bit cases, we can just shift instead of divide - // because the number of bits per digit (1, 3 and 4 respectively) divides + + // For the 2, 8 and 16 bit cases, we can just shift instead of divide + // because the number of bits per digit (1, 3 and 4 respectively) divides // equaly. We just shift until the value is zero. if (Radix != 10) { // Just shift tmp right for each digit width until it becomes zero unsigned ShiftAmt = (Radix == 16 ? 4 : (Radix == 8 ? 3 : 1)); unsigned MaskAmt = Radix - 1; - + while (Tmp != 0) { unsigned Digit = unsigned(Tmp.getRawData()[0]) & MaskAmt; Str.push_back(Digits[Digit]); @@ -2075,15 +2164,15 @@ void APInt::toString(SmallVectorImpl &Str, unsigned Radix, while (Tmp != 0) { APInt APdigit(1, 0); APInt tmp2(Tmp.getBitWidth(), 0); - divide(Tmp, Tmp.getNumWords(), divisor, divisor.getNumWords(), &tmp2, + divide(Tmp, Tmp.getNumWords(), divisor, divisor.getNumWords(), &tmp2, &APdigit); - uint32_t Digit = (uint32_t)APdigit.getZExtValue(); + unsigned Digit = (unsigned)APdigit.getZExtValue(); assert(Digit < Radix && "divide failed"); Str.push_back(Digits[Digit]); Tmp = tmp2; } } - + // Reverse the digits before returning. std::reverse(Str.begin()+StartDig, Str.end()); } @@ -2094,7 +2183,7 @@ void APInt::toString(SmallVectorImpl &Str, unsigned Radix, std::string APInt::toString(unsigned Radix = 10, bool Signed = true) const { SmallString<40> S; toString(S, Radix, Signed); - return S.c_str(); + return S.str(); } @@ -2102,21 +2191,21 @@ void APInt::dump() const { SmallString<40> S, U; this->toStringUnsigned(U); this->toStringSigned(S); - fprintf(stderr, "APInt(%db, %su %ss)", BitWidth, U.c_str(), S.c_str()); + dbgs() << "APInt(" << BitWidth << "b, " + << U.str() << "u " << S.str() << "s)"; } -void APInt::print(std::ostream &OS, bool isSigned) const { +void APInt::print(raw_ostream &OS, bool isSigned) const { SmallString<40> S; this->toString(S, 10, isSigned); - OS << S.c_str(); + OS << S.str(); } - // This implements a variety of operations on a representation of // arbitrary precision, two's-complement, bignum integer values. -/* Assumed by lowHalf, highHalf, partMSB and partLSB. A fairly safe - and unrestricting assumption. */ +// Assumed by lowHalf, highHalf, partMSB and partLSB. A fairly safe +// and unrestricting assumption. #define COMPILE_TIME_ASSERT(cond) extern int CTAssert[(cond) ? 1 : -1] COMPILE_TIME_ASSERT(integerPartWidth % 2 == 0); @@ -2293,7 +2382,7 @@ APInt::tcMSB(const integerPart *parts, unsigned int n) the least significant bit of DST. All high bits above srcBITS in DST are zero-filled. */ void -APInt::tcExtract(integerPart *dst, unsigned int dstCount, const integerPart *src, +APInt::tcExtract(integerPart *dst, unsigned int dstCount,const integerPart *src, unsigned int srcBits, unsigned int srcLSB) { unsigned int firstSrcPart, dstParts, shift, n;