mirror of
https://github.com/dashr9230/SA-MP.git
synced 2024-12-22 14:37:29 +08:00
bcdbedc0be
* Add RakNet source files to the VS project
1821 lines
34 KiB
C++
1821 lines
34 KiB
C++
/// \file
|
|
/// \brief \b [Internal] Used for RSA generation.
|
|
///
|
|
/// This file is part of RakNet Copyright 2003 Kevin Jenkins.
|
|
///
|
|
/// (128)2^7-bit to (32768)2^14-bit signed 2's complement & unsigned extended arithmetic
|
|
///
|
|
/// catid(cat02e@fsu.edu)
|
|
///
|
|
/// 7/30/2004 Fixed VS6 compat
|
|
/// 7/28/2004 Fixed macros so they can be used outside of the big namespace
|
|
/// Now using pre-processor definitions from types.h for inline assembly
|
|
/// 7/26/2004 Removed a lot of assembly, made add/sub assembly optional
|
|
/// 7/25/2004 Merged the wrapper class Int<T> from older code
|
|
/// 7/24/2004 Replaced trivial assembly code with std:: functions
|
|
/// Refined some assembly code with Art of Assembly chapter 9
|
|
/// Added binary ops
|
|
/// 7/23/2004 Finished assembly coding
|
|
/// Removed Int<T> class, for now
|
|
/// Added old C++ code back in with USEASSEMBLY
|
|
/// 7/22/2004 Signed arithmetic (needed for ext. Euclidean algo)
|
|
/// Cleaned up coding style
|
|
/// Began rewriting parts in assembly
|
|
/// 7/21/2004 Began writing
|
|
///
|
|
/// Tabs: 4 spaces
|
|
/// Dist: public
|
|
|
|
#ifndef BIGTYPES_H
|
|
#define BIGTYPES_H
|
|
|
|
#if !defined(_COMPATIBILITY_1)
|
|
|
|
#include "Types.h"
|
|
|
|
//#define BIG_USES_STRINGS /* undefining this means you cannot convert bigs to strings or from strings */
|
|
|
|
#ifdef BIG_USES_STRINGS
|
|
# include <string>
|
|
#endif
|
|
|
|
|
|
#ifdef _MSC_VER
|
|
#pragma warning( push )
|
|
#endif
|
|
|
|
namespace big
|
|
{
|
|
|
|
using namespace cat;
|
|
|
|
//// basic definitions ////
|
|
|
|
// word size
|
|
typedef u32 word; // assembly implementation is for 32-bit word size
|
|
const u32 WORDBITS = sizeof( word ) * 8;
|
|
const u32 HALFWORDBITS = sizeof( word ) * 8 / 2;
|
|
const word WORDHIGHBIT = ( word ) 1 << ( WORDBITS - 1 );
|
|
const word WORDALLBITS = ( word ) 0 - 1;
|
|
const word WORDLOBITS = ( ( word ) 1 << HALFWORDBITS ) - 1;
|
|
const word WORDHIBITS = WORDALLBITS ^ WORDLOBITS;
|
|
#define BIGHIGHBIT(n) ((n)[sizeof(n) / sizeof(big::word) - 1] & WORDHIGHBIT)
|
|
|
|
// template operator parameter modes
|
|
#define BIGONETYPE template<class T> /* supports only one class */
|
|
#define BIGTWOTYPES template<class T, class Bigger> /* sizeof Bigger >= sizeof T */
|
|
#define BIGSMALLTYPE template<class Smaller> /* sizeof self >= sizeof Smaller */
|
|
|
|
|
|
//// big types ////
|
|
|
|
#define BIGWORDCOUNT_FROMBITCOUNT(bits) ((bits) / 8 / sizeof(big::word))
|
|
#define BIGWORDCOUNT(T) (sizeof(T) / sizeof(big::word))
|
|
#define BIGBITCOUNT(T) (sizeof(T) * 8)
|
|
|
|
// low words -- [0] < [1] < [2] < [3] -- high words
|
|
typedef word u128[ BIGWORDCOUNT_FROMBITCOUNT( 128 ) ];
|
|
typedef word u256[ BIGWORDCOUNT_FROMBITCOUNT( 256 ) ];
|
|
typedef word u512[ BIGWORDCOUNT_FROMBITCOUNT( 512 ) ];
|
|
typedef word u1024[ BIGWORDCOUNT_FROMBITCOUNT( 1024 ) ];
|
|
typedef word u2048[ BIGWORDCOUNT_FROMBITCOUNT( 2048 ) ];
|
|
typedef word u4096[ BIGWORDCOUNT_FROMBITCOUNT( 4096 ) ];
|
|
typedef word u8192[ BIGWORDCOUNT_FROMBITCOUNT( 8192 ) ];
|
|
typedef word u16384[ BIGWORDCOUNT_FROMBITCOUNT( 16384 ) ];
|
|
typedef word u32768[ BIGWORDCOUNT_FROMBITCOUNT( 32768 ) ];
|
|
|
|
// use these macros to create temporary variables when
|
|
// those variables are to be twice/half the size of another
|
|
// variable of varying size.
|
|
#define BIGDOUBLESIZE(T, var_name) big::word (var_name)[BIGWORDCOUNT(T) * 2] /* WARNING: invalid w/ u32768 */
|
|
#define BIGHALFSIZE(T, var_name) big::word (var_name)[BIGWORDCOUNT(T) / 2] /* WARNING: invalid w/ u128 */
|
|
|
|
|
|
//// library summary ////
|
|
|
|
// assignment
|
|
BIGONETYPE INLINE void zero( T &n ); // n = 0
|
|
BIGONETYPE INLINE void usetw( T &a, word b ); // a = b, zero-extend
|
|
BIGONETYPE INLINE void ssetw( T &a, word b ); // a = b, sign-extend
|
|
|
|
BIGONETYPE INLINE void set ( T &a, T &b )
|
|
|
|
; // a = b
|
|
BIGTWOTYPES INLINE void usetlow( Bigger &a, T &b ); // a_low = b (zero-extend)
|
|
|
|
BIGTWOTYPES INLINE void ssetlow( Bigger &a, T &b ); // a_low = b (sign-extend)
|
|
|
|
BIGTWOTYPES INLINE void sethigh( Bigger &a, T &b ); // a_high = b
|
|
|
|
BIGTWOTYPES INLINE void takelow( T &a, Bigger &b ); // a = b_low
|
|
|
|
BIGTWOTYPES INLINE void takehigh( T &a, Bigger &b ); // a = b_high
|
|
|
|
// comparison
|
|
BIGONETYPE bool ugreater( T &a, T &b ); // a > b (unsigned)
|
|
|
|
BIGONETYPE bool ugreaterOrEqual( T &a, T &b ); // a >= b (unsigned)
|
|
|
|
BIGONETYPE bool sgreater( T &a, T &b ); // a > b (signed)
|
|
|
|
BIGONETYPE bool sgreaterOrEqual( T &a, T &b ); // a >= b (signed)
|
|
|
|
BIGONETYPE INLINE bool equal( T &a, T &b ); // a == b
|
|
|
|
BIGONETYPE INLINE bool isZero( T &n ); // a == 0
|
|
|
|
// binary
|
|
BIGONETYPE void bAND( T &a, T &b ); // a &= b
|
|
|
|
BIGONETYPE void bOR( T &a, T &b ); // a |= b
|
|
|
|
BIGONETYPE void bXOR( T &a, T &b ); // a ^= b
|
|
|
|
BIGONETYPE void bNOT( T &n ); // n = ~n
|
|
|
|
// shifting
|
|
BIGONETYPE void shiftLeft1( T &n ); // n <<= 1
|
|
|
|
BIGONETYPE void shiftLeft( T &n, u32 s ); // n <<= s (s <= WORDBITS)
|
|
|
|
BIGONETYPE void ushiftRight1( T &n ); // n >>= 1 (unsigned)
|
|
|
|
BIGONETYPE void ushiftRight( T &n, u32 s ); // n >>= s (unsigned) (s <= WORDBITS)
|
|
|
|
BIGONETYPE void sshiftRight1( T &n ); // n >>= 1 (signed)
|
|
|
|
BIGONETYPE void sshiftRight( T &n, u32 s ); // n >>= s (signed) (s <= WORDBITS)
|
|
|
|
// addition/subtraction
|
|
BIGONETYPE void add ( T &a, T &b )
|
|
|
|
; // a += b
|
|
BIGONETYPE void increment( T &n ); // ++n
|
|
|
|
BIGONETYPE void subtract( T &a, T &b ); // a -= b
|
|
|
|
BIGONETYPE void decrement( T &n ); // --n
|
|
|
|
// negation
|
|
BIGONETYPE void negate( T &n ); // n = -n
|
|
|
|
// multiplication
|
|
BIGONETYPE void usquare( T &a ); // a *= a, signed
|
|
|
|
BIGTWOTYPES void umultiply( T &a, T &b, Bigger &m ); // m = a * b (&a != &b != &m), unsigned
|
|
|
|
BIGTWOTYPES void umultiply( Bigger &a, T &b ); // a *= b (&a != &b), unsigned
|
|
|
|
BIGONETYPE void ssquare( T &a ); // a *= a, signed
|
|
|
|
BIGTWOTYPES void smultiply( T &a, T &b, Bigger &m ); // m = a * b (&a != &b != &m), signed
|
|
|
|
BIGTWOTYPES void smultiply( Bigger &a, T &b ); // a *= b (&a != &b), signed
|
|
|
|
// division/remainder
|
|
BIGONETYPE void udivide( T &a, T &b, T &q, T &r ); // {q, r} = a / b (&q != &r), unsigned
|
|
|
|
BIGONETYPE void umodulo( T &a, T &b, T &r ); // r = a Mod b, unsigned
|
|
|
|
BIGONETYPE void sdivide( T &a, T &b, T &q, T &r ); // {q, r} = a / b (&q != &r), signed
|
|
|
|
BIGONETYPE void smodulo( T &a, T &b, T &r ); // r = a Mod b, signed
|
|
|
|
#ifdef BIG_USES_STRINGS
|
|
// converting to/from strings
|
|
BIGONETYPE std::string toString( T &n, bool sign, u16 radix ); // n -> string
|
|
|
|
BIGONETYPE void fromString( std::string s, T &n, bool sign, u16 radix ); // s -> n
|
|
|
|
#endif
|
|
|
|
|
|
//////// wrapper class ////////
|
|
|
|
#define BIGINTFAST INLINE Int<T> & /* operation is done to self, returns itself */
|
|
#define BIGINTSLOW Int<T> /* new object is created and returned */
|
|
|
|
BIGONETYPE class Int
|
|
{
|
|
|
|
protected:
|
|
T raw;
|
|
|
|
public:
|
|
operator T &(); // automatic casting to T: you may use BigInt classes as parameters to the functions
|
|
|
|
public:
|
|
Int();
|
|
Int( word n );
|
|
#ifdef BIG_USES_STRINGS
|
|
|
|
Int( std::string &s );
|
|
#endif
|
|
|
|
Int( T &n );
|
|
|
|
public:
|
|
BIGINTFAST zero();
|
|
BIGINTFAST operator=( word n );
|
|
BIGINTFAST operator=( T &n );
|
|
|
|
public:
|
|
BIGINTFAST operator<<=( u32 s );
|
|
BIGINTSLOW operator<<( u32 s );
|
|
BIGINTFAST operator>>=( u32 s );
|
|
BIGINTSLOW operator>>( u32 s );
|
|
|
|
public:
|
|
BIGINTFAST operator+=( T &n );
|
|
BIGINTSLOW operator+( T &n );
|
|
BIGINTFAST operator-=( T &n );
|
|
BIGINTSLOW operator-( T &n );
|
|
BIGINTFAST operator++(); // prefix
|
|
BIGINTSLOW operator++( int ); // postfix
|
|
BIGINTFAST operator--(); // prefix
|
|
BIGINTSLOW operator--( int ); // postfix
|
|
|
|
public:
|
|
BIGINTSLOW operator-( int ); // negation
|
|
|
|
public:
|
|
BIGSMALLTYPE BIGINTFAST operator*=( Smaller &n )
|
|
{
|
|
smultiply( raw, n );
|
|
return *this;
|
|
}
|
|
|
|
BIGINTSLOW operator*( T &n );
|
|
BIGINTFAST square();
|
|
|
|
public:
|
|
BIGINTFAST operator/=( T &n );
|
|
BIGINTSLOW operator/( T &n );
|
|
BIGINTFAST operator%=( T &n );
|
|
BIGINTSLOW operator%( T &n );
|
|
|
|
public:
|
|
/* fast */
|
|
bool operator>( T &n );
|
|
/* fast */
|
|
bool operator>=( T &n );
|
|
/* fast */
|
|
bool operator<( T &n );
|
|
/* fast */
|
|
bool operator<=( T &n );
|
|
/* fast */
|
|
bool operator==( T &n );
|
|
/* fast */
|
|
bool operator!=( T &n );
|
|
/* fast */
|
|
bool operator!();
|
|
|
|
public:
|
|
#ifdef BIG_USES_STRINGS
|
|
/* fast */
|
|
std::string str();
|
|
BIGINTFAST operator=( std::string &s );
|
|
BIGINTFAST operator=( const char *s );
|
|
#endif
|
|
|
|
};
|
|
|
|
|
|
//////// assignment ////////
|
|
|
|
// n = 0
|
|
BIGONETYPE INLINE void zero( T &n )
|
|
{
|
|
memset( n, 0, sizeof( T ) );
|
|
}
|
|
|
|
// a = b, zero-extend
|
|
BIGONETYPE INLINE void usetw( T &a, word b )
|
|
{
|
|
a[ 0 ] = b;
|
|
memset( a + 1, 0, sizeof( T ) - sizeof( word ) );
|
|
}
|
|
|
|
// a = b, sign-extend
|
|
BIGONETYPE INLINE void ssetw( T &a, word b )
|
|
{
|
|
a[ 0 ] = b;
|
|
memset( a + 1, ( b & WORDHIGHBIT ) ? WORDALLBITS : 0, sizeof( T ) - sizeof( word ) );
|
|
}
|
|
|
|
// a = b
|
|
BIGONETYPE INLINE void set ( T &a, T &b )
|
|
{
|
|
memcpy( a, b, sizeof( T ) );
|
|
}
|
|
|
|
// a_low = b (zero-extend)
|
|
BIGTWOTYPES INLINE void usetlow( Bigger &a, T &b )
|
|
{
|
|
memcpy( a, b, sizeof( T ) );
|
|
#ifdef _MSC_VER
|
|
#pragma warning( disable : 4318 ) // warning C4318: passing constant zero as the length to memset
|
|
#endif
|
|
memset( a + BIGWORDCOUNT( T ), 0, sizeof( Bigger ) - sizeof( T ) );
|
|
}
|
|
|
|
// a_low = b (sign-extend)
|
|
BIGTWOTYPES INLINE void ssetlow( Bigger &a, T &b )
|
|
{
|
|
memcpy( a, b, sizeof( T ) );
|
|
memset( a + BIGWORDCOUNT( T ), BIGHIGHBIT( b ) ? WORDALLBITS : 0, sizeof( Bigger ) - sizeof( T ) );
|
|
}
|
|
|
|
// a_high = b
|
|
BIGTWOTYPES INLINE void sethigh( Bigger &a, T &b )
|
|
{
|
|
memcpy( a + BIGWORDCOUNT( Bigger ) - BIGWORDCOUNT( T ), b, sizeof( T ) );
|
|
memset( a, 0, sizeof( Bigger ) - sizeof( T ) );
|
|
}
|
|
|
|
// a = b_low
|
|
BIGTWOTYPES INLINE void takelow( T &a, Bigger &b )
|
|
{
|
|
memcpy( a, b, sizeof( T ) );
|
|
}
|
|
|
|
// a = b_high
|
|
BIGTWOTYPES INLINE void takehigh( T &a, Bigger &b )
|
|
{
|
|
memcpy( a, b + BIGWORDCOUNT( Bigger ) - BIGWORDCOUNT( T ), sizeof( T ) );
|
|
}
|
|
|
|
|
|
//////// comparison ////////
|
|
|
|
// a > b
|
|
BIGONETYPE bool ugreater( T &a, T &b )
|
|
{
|
|
for ( s32 ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
{
|
|
if ( a[ ii ] > b[ ii ] )
|
|
return true;
|
|
|
|
if ( a[ ii ] < b[ ii ] )
|
|
return false;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
// a >= b
|
|
BIGONETYPE bool ugreaterOrEqual( T &a, T &b )
|
|
{
|
|
for ( s32 ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
{
|
|
if ( a[ ii ] > b[ ii ] )
|
|
return true;
|
|
|
|
if ( a[ ii ] < b[ ii ] )
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
// a > b
|
|
BIGONETYPE bool sgreater( T &a, T &b )
|
|
{
|
|
for ( s32 ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
{
|
|
if ( a[ ii ] > b[ ii ] )
|
|
return BIGHIGHBIT( a ) == 0;
|
|
|
|
if ( a[ ii ] < b[ ii ] )
|
|
return BIGHIGHBIT( b ) != 0;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
// a >= b
|
|
BIGONETYPE bool sgreaterOrEqual( T &a, T &b )
|
|
{
|
|
for ( s32 ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
{
|
|
if ( a[ ii ] > b[ ii ] )
|
|
return BIGHIGHBIT( a ) == 0;
|
|
|
|
if ( a[ ii ] < b[ ii ] )
|
|
return BIGHIGHBIT( b ) != 0;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
// a == b
|
|
BIGONETYPE INLINE bool equal( T &a, T &b )
|
|
{
|
|
return memcmp( a, b, sizeof( T ) ) == 0;
|
|
}
|
|
|
|
// a == 0
|
|
BIGONETYPE INLINE bool isZero( T &n )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
if ( n[ ii ] )
|
|
return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
|
|
//////// binary ////////
|
|
|
|
// a &= b
|
|
BIGONETYPE void bAND( T &a, T &b )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
a[ ii ] &= b[ ii ];
|
|
}
|
|
|
|
// a |= b
|
|
BIGONETYPE void bOR( T &a, T &b )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
a[ ii ] |= b[ ii ];
|
|
}
|
|
|
|
// a ^= b
|
|
BIGONETYPE void bXOR( T &a, T &b )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
a[ ii ] ^= b[ ii ];
|
|
}
|
|
|
|
// n = ~n
|
|
BIGONETYPE void bNOT( T &n )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
n[ ii ] = ~n[ ii ];
|
|
}
|
|
|
|
|
|
//////// shifting ////////
|
|
|
|
// n <<= 1
|
|
BIGONETYPE void shiftLeft1( T &n )
|
|
{
|
|
register word w_i, carry = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
w_i = n[ ii ];
|
|
|
|
n[ ii ] = ( w_i << 1 ) | carry;
|
|
carry = w_i >> ( WORDBITS - 1 );
|
|
}
|
|
}
|
|
|
|
// n <<= s (s <= WORDBITS)
|
|
BIGONETYPE void shiftLeft( T &n, u32 s )
|
|
{
|
|
register s32 ii;
|
|
register u32 bases = s / WORDBITS;
|
|
u32 bits = s % WORDBITS;
|
|
|
|
// move whole bases first
|
|
|
|
if ( bases )
|
|
{
|
|
// shift bases
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1 - bases; ii >= 0; --ii )
|
|
n[ ii + bases ] = n[ ii ];
|
|
|
|
// clear the original locii of those bases
|
|
memset( n, 0, bases * sizeof( word ) );
|
|
}
|
|
|
|
if ( bits )
|
|
{
|
|
register word w_i, carry = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
w_i = n[ ii ];
|
|
|
|
n[ ii ] = ( w_i << bits ) | carry;
|
|
carry = w_i >> ( WORDBITS - bits );
|
|
}
|
|
}
|
|
}
|
|
|
|
// n >>= 1 (unsigned)
|
|
BIGONETYPE void ushiftRight1( T &n )
|
|
{
|
|
register word w_i, carry = 0;
|
|
|
|
for ( s32 ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
{
|
|
w_i = n[ ii ];
|
|
|
|
n[ ii ] = carry | ( w_i >> 1 );
|
|
carry = w_i << ( WORDBITS - 1 );
|
|
}
|
|
}
|
|
|
|
// n >>= s (unsigned) (s <= WORDBITS)
|
|
BIGONETYPE void ushiftRight( T &n, u32 s )
|
|
{
|
|
register s32 ii;
|
|
register u32 bases = s / WORDBITS;
|
|
register u32 bits = s % WORDBITS;
|
|
|
|
// move whole bases first
|
|
|
|
if ( bases )
|
|
{
|
|
// shift bases
|
|
|
|
for ( ii = bases; ii < BIGWORDCOUNT( T ); ++ii )
|
|
n[ ii - bases ] = n[ ii ];
|
|
|
|
// clear the original locii of those bases
|
|
memset( n + BIGWORDCOUNT( T ) - bases, 0, bases * sizeof( word ) );
|
|
}
|
|
|
|
if ( bits )
|
|
{
|
|
register word w_i, carry = 0;
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1 - bases; ii >= 0; --ii )
|
|
{
|
|
w_i = n[ ii ];
|
|
|
|
n[ ii ] = carry | ( w_i >> bits );
|
|
carry = w_i << ( WORDBITS - bits );
|
|
}
|
|
}
|
|
}
|
|
|
|
// n >>= 1 (signed)
|
|
BIGONETYPE void sshiftRight1( T &n )
|
|
{
|
|
register word w_i, carry = BIGHIGHBIT( n ) ? 1 : 0;
|
|
|
|
for ( s32 ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
{
|
|
w_i = n[ ii ];
|
|
|
|
n[ ii ] = carry | ( w_i >> 1 );
|
|
carry = w_i << ( WORDBITS - 1 );
|
|
}
|
|
}
|
|
|
|
// n >>= s (signed) (s <= WORDBITS)
|
|
BIGONETYPE void sshiftRight( T &n, u32 s )
|
|
{
|
|
register s32 ii;
|
|
register u32 bases = s / WORDBITS;
|
|
register u32 bits = s % WORDBITS;
|
|
|
|
word filler = BIGHIGHBIT( n ) ? WORDALLBITS : 0;
|
|
|
|
// move whole bases first
|
|
|
|
if ( bases )
|
|
{
|
|
// shift bases
|
|
|
|
for ( ii = bases; ii < BIGWORDCOUNT( T ); ++ii )
|
|
n[ ii - bases ] = n[ ii ];
|
|
|
|
// clear the original locii of those bases
|
|
memset( n + BIGWORDCOUNT( T ) - bases, filler, bases * sizeof( word ) );
|
|
}
|
|
|
|
if ( bits )
|
|
{
|
|
register word w_i, carry = filler << ( WORDBITS - bits );
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1 - bases; ii >= 0; --ii )
|
|
{
|
|
w_i = n[ ii ];
|
|
|
|
n[ ii ] = carry | ( w_i >> bits );
|
|
carry = w_i << ( WORDBITS - bits );
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
//////// addition/subtraction ////////
|
|
|
|
#if defined(NO_TEMPLATE_INLINE_ASSEMBLY) && defined(ASSEMBLY_INTEL_SYNTAX)
|
|
void BorlandAdd( void *a, void *b, u32 c )
|
|
{
|
|
ASSEMBLY_BLOCK // BorlandC, x86, 32-bit words
|
|
{
|
|
mov esi, b
|
|
mov edi, a
|
|
mov ecx, c
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi ]
|
|
xor edx, edx // edx used later to index later words >= 4
|
|
|
|
add [ edi ]
|
|
, eax // and now we can use ADD instead of ADC on the first addition
|
|
mov eax, [ esi + 4 ]
|
|
adc [ edi + 4 ], eax
|
|
mov eax, [ esi + 8 ]
|
|
adc [ edi + 8 ], eax
|
|
mov eax, [ esi + 12 ]
|
|
adc [ edi + 12 ], eax
|
|
|
|
jecxz done_already
|
|
|
|
next_word:
|
|
inc edx
|
|
inc edx
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi + edx * 8 ]
|
|
adc [ edi + edx * 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 4 ]
|
|
adc [ edi + edx * 8 + 4 ], eax
|
|
mov eax, [ esi + edx * 8 + 8 ]
|
|
adc [ edi + edx * 8 + 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 12 ]
|
|
adc [ edi + edx * 8 + 12 ], eax
|
|
|
|
loop next_word
|
|
|
|
done_already:
|
|
}
|
|
}
|
|
|
|
#endif
|
|
|
|
// a += b
|
|
BIGONETYPE void add ( T &a, T &b )
|
|
{
|
|
#if defined(NO_TEMPLATE_INLINE_ASSEMBLY) && defined(ASSEMBLY_INTEL_SYNTAX)
|
|
const u32 qc1 = BIGWORDCOUNT( T ) / 4 - 1;
|
|
|
|
BorlandAdd( a, b, qc1 );
|
|
|
|
#elif defined(ASSEMBLY_INTEL_SYNTAX)
|
|
|
|
const u32 qc1 = BIGWORDCOUNT( T ) / 4 - 1;
|
|
|
|
ASSEMBLY_BLOCK // VS.NET, x86, 32-bit words
|
|
{
|
|
mov esi, b
|
|
mov edi, a
|
|
mov ecx, qc1
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi ]
|
|
xor edx, edx // edx used later to index later words >= 4
|
|
|
|
add [ edi ], eax // and now we can use ADD instead of ADC on the first addition
|
|
mov eax, [ esi + 4 ]
|
|
adc [ edi + 4 ], eax
|
|
mov eax, [ esi + 8 ]
|
|
adc [ edi + 8 ], eax
|
|
mov eax, [ esi + 12 ]
|
|
adc [ edi + 12 ], eax
|
|
|
|
jecxz done_already
|
|
|
|
next_word:
|
|
inc edx
|
|
inc edx
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi + edx * 8 ]
|
|
adc [ edi + edx * 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 4 ]
|
|
adc [ edi + edx * 8 + 4 ], eax
|
|
mov eax, [ esi + edx * 8 + 8 ]
|
|
adc [ edi + edx * 8 + 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 12 ]
|
|
adc [ edi + edx * 8 + 12 ], eax
|
|
|
|
loop next_word
|
|
|
|
done_already:
|
|
}
|
|
#else
|
|
register word carry = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word a_i = a[ ii ];
|
|
word b_i = b[ ii ];
|
|
a[ ii ] += b_i + carry;
|
|
|
|
carry = ( ( a_i & ( WORDALLBITS >> 1 ) ) + ( b_i & ( WORDALLBITS >> 1 ) ) + carry ) >> ( WORDBITS - 1 );
|
|
carry += ( a_i >> ( WORDBITS - 1 ) ) + ( b_i >> ( WORDBITS - 1 ) );
|
|
carry >>= 1;
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
// ++n
|
|
BIGONETYPE void increment( T &n )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
if ( ++n[ ii ] )
|
|
break;
|
|
}
|
|
|
|
#if defined(NO_TEMPLATE_INLINE_ASSEMBLY) && defined(ASSEMBLY_INTEL_SYNTAX)
|
|
void BorlandSubtract( void *a, void *b, u32 c )
|
|
{
|
|
ASSEMBLY_BLOCK // BorlandC, x86, 32-bit words
|
|
{
|
|
mov esi, b
|
|
mov edi, a
|
|
mov ecx, c
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi ]
|
|
xor edx, edx // edx used later to index later words >= 4
|
|
|
|
add [ edi ]
|
|
, eax // and now we can use ADD instead of ADC on the first addition
|
|
mov eax, [ esi + 4 ]
|
|
adc [ edi + 4 ], eax
|
|
mov eax, [ esi + 8 ]
|
|
adc [ edi + 8 ], eax
|
|
mov eax, [ esi + 12 ]
|
|
adc [ edi + 12 ], eax
|
|
|
|
jecxz done_already
|
|
|
|
next_word:
|
|
inc edx
|
|
inc edx
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi + edx * 8 ]
|
|
adc [ edi + edx * 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 4 ]
|
|
adc [ edi + edx * 8 + 4 ], eax
|
|
mov eax, [ esi + edx * 8 + 8 ]
|
|
adc [ edi + edx * 8 + 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 12 ]
|
|
adc [ edi + edx * 8 + 12 ], eax
|
|
|
|
loop next_word
|
|
|
|
done_already:
|
|
}
|
|
}
|
|
|
|
#endif
|
|
|
|
// a -= b
|
|
BIGONETYPE void subtract( T &a, T &b )
|
|
{
|
|
#if defined(NO_TEMPLATE_INLINE_ASSEMBLY) && defined(ASSEMBLY_INTEL_SYNTAX)
|
|
const u32 qc1 = BIGWORDCOUNT( T ) / 4 - 1;
|
|
|
|
BorlandSubtract( a, b, qc1 );
|
|
|
|
#elif defined(ASSEMBLY_INTEL_SYNTAX)
|
|
|
|
const u32 qc1 = BIGWORDCOUNT( T ) / 4 - 1;
|
|
|
|
ASSEMBLY_BLOCK // VS.NET, x86, 32-bit words
|
|
{
|
|
mov esi, b
|
|
mov edi, a
|
|
mov ecx, qc1
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi ]
|
|
xor edx, edx // edx used later to index later words >= 4
|
|
sub [ edi ], eax // first subtraction doesn't need to borrow
|
|
mov eax, [ esi + 4 ]
|
|
sbb [ edi + 4 ], eax
|
|
mov eax, [ esi + 8 ]
|
|
sbb [ edi + 8 ], eax
|
|
mov eax, [ esi + 12 ]
|
|
sbb [ edi + 12 ], eax
|
|
|
|
jecxz done_already
|
|
|
|
next_word:
|
|
inc edx
|
|
inc edx
|
|
|
|
// unrolled loop since word count is a multiple of 4 >= 4
|
|
mov eax, [ esi + edx * 8 ]
|
|
sbb [ edi + edx * 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 4 ]
|
|
sbb [ edi + edx * 8 + 4 ], eax
|
|
mov eax, [ esi + edx * 8 + 8 ]
|
|
sbb [ edi + edx * 8 + 8 ], eax
|
|
mov eax, [ esi + edx * 8 + 12 ]
|
|
sbb [ edi + edx * 8 + 12 ], eax
|
|
|
|
loop next_word
|
|
|
|
done_already:
|
|
}
|
|
#else
|
|
register word borrow = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word a_i = a[ ii ];
|
|
word b_i = b[ ii ];
|
|
a[ ii ] -= b_i + borrow;
|
|
|
|
borrow = ( ( a_i & ( WORDALLBITS >> 1 ) ) - ( b_i & ( WORDALLBITS >> 1 ) ) - borrow ) >> ( WORDBITS - 1 );
|
|
borrow += ( b_i >> ( WORDBITS - 1 ) ) - ( a_i >> ( WORDBITS - 1 ) );
|
|
++borrow;
|
|
borrow >>= 1;
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
// --n
|
|
BIGONETYPE void decrement( T &n )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
if ( n[ ii ] -- )
|
|
break;
|
|
}
|
|
|
|
|
|
//////// negation ////////
|
|
|
|
// n = -n
|
|
BIGONETYPE void negate( T &n )
|
|
{
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
n[ ii ] = ~n[ ii ];
|
|
|
|
increment( n );
|
|
}
|
|
|
|
|
|
//////// multiplication ////////
|
|
|
|
// a *= a, unsigned
|
|
BIGONETYPE void usquare( T &a )
|
|
{
|
|
T a0, a1;
|
|
|
|
set ( a0, a )
|
|
|
|
;
|
|
set ( a1, a )
|
|
|
|
;
|
|
zero( a );
|
|
|
|
u32 shifts = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word w_i = a0[ ii ];
|
|
|
|
u16 ctr = WORDBITS;
|
|
|
|
while ( w_i )
|
|
{
|
|
if ( w_i & 1 )
|
|
{
|
|
if ( shifts )
|
|
{
|
|
shiftLeft( a1, shifts );
|
|
shifts = 0;
|
|
}
|
|
|
|
add ( a, a1 )
|
|
|
|
;
|
|
}
|
|
|
|
w_i >>= 1;
|
|
++shifts;
|
|
--ctr;
|
|
}
|
|
|
|
shifts += ctr;
|
|
}
|
|
}
|
|
|
|
// m = a * b (&a != &b != &m), unsigned
|
|
BIGTWOTYPES void umultiply( T &a0, T &b0, Bigger &m )
|
|
{
|
|
Bigger a;
|
|
usetlow( a, a0 );
|
|
|
|
zero( m );
|
|
|
|
u32 shifts = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word w_i = b0[ ii ];
|
|
|
|
u16 ctr = WORDBITS;
|
|
|
|
while ( w_i )
|
|
{
|
|
if ( w_i & 1 )
|
|
{
|
|
if ( shifts )
|
|
{
|
|
shiftLeft( a, shifts );
|
|
shifts = 0;
|
|
}
|
|
|
|
add ( m, a )
|
|
|
|
;
|
|
}
|
|
|
|
w_i >>= 1;
|
|
++shifts;
|
|
--ctr;
|
|
}
|
|
|
|
shifts += ctr;
|
|
}
|
|
}
|
|
|
|
// a *= b (&a != &b), unsigned
|
|
BIGTWOTYPES void umultiply( Bigger &a0, T &b0 )
|
|
{
|
|
Bigger a;
|
|
|
|
set ( a, a0 )
|
|
|
|
;
|
|
|
|
zero( a0 );
|
|
|
|
u32 shifts = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word w_i = b0[ ii ];
|
|
|
|
u16 ctr = WORDBITS;
|
|
|
|
while ( w_i )
|
|
{
|
|
if ( w_i & 1 )
|
|
{
|
|
if ( shifts )
|
|
{
|
|
shiftLeft( a, shifts );
|
|
shifts = 0;
|
|
}
|
|
|
|
add ( a0, a )
|
|
|
|
;
|
|
}
|
|
|
|
w_i >>= 1;
|
|
++shifts;
|
|
--ctr;
|
|
}
|
|
|
|
shifts += ctr;
|
|
}
|
|
}
|
|
|
|
// a *= a, signed
|
|
BIGONETYPE void ssquare( T &a )
|
|
{
|
|
T a0, a1;
|
|
|
|
if ( BIGHIGHBIT( a ) )
|
|
negate( a );
|
|
|
|
set ( a0, a )
|
|
|
|
;
|
|
set ( a1, a )
|
|
|
|
;
|
|
zero( a );
|
|
|
|
u32 shifts = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word w_i = a0[ ii ];
|
|
|
|
u16 ctr = WORDBITS;
|
|
|
|
while ( w_i )
|
|
{
|
|
if ( w_i & 1 )
|
|
{
|
|
if ( shifts )
|
|
{
|
|
shiftLeft( a1, shifts );
|
|
shifts = 0;
|
|
}
|
|
|
|
add ( a, a1 )
|
|
|
|
;
|
|
}
|
|
|
|
w_i >>= 1;
|
|
++shifts;
|
|
--ctr;
|
|
}
|
|
|
|
shifts += ctr;
|
|
}
|
|
}
|
|
|
|
// m = a * b (&a != &b != &m), signed
|
|
BIGTWOTYPES void smultiply( T &a0, T &b0, Bigger &m )
|
|
{
|
|
Bigger a;
|
|
ssetlow( a, a0 );
|
|
|
|
word sign_a = BIGHIGHBIT( a );
|
|
|
|
if ( sign_a )
|
|
negate( a );
|
|
|
|
T b;
|
|
|
|
set ( b, b0 )
|
|
|
|
;
|
|
|
|
word sign_b = BIGHIGHBIT( b );
|
|
|
|
if ( sign_b )
|
|
negate( b );
|
|
|
|
zero( m );
|
|
|
|
u32 shifts = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word w_i = b[ ii ];
|
|
|
|
u16 ctr = WORDBITS;
|
|
|
|
while ( w_i )
|
|
{
|
|
if ( w_i & 1 )
|
|
{
|
|
if ( shifts )
|
|
{
|
|
shiftLeft( a, shifts );
|
|
shifts = 0;
|
|
}
|
|
|
|
add ( m, a )
|
|
|
|
;
|
|
}
|
|
|
|
w_i >>= 1;
|
|
++shifts;
|
|
--ctr;
|
|
}
|
|
|
|
shifts += ctr;
|
|
}
|
|
|
|
if ( sign_a ^ sign_b )
|
|
negate( m );
|
|
}
|
|
|
|
// a *= b (&a != &b), signed
|
|
BIGTWOTYPES void smultiply( Bigger &a0, T &b0 )
|
|
{
|
|
Bigger a;
|
|
ssetlow( a, a0 );
|
|
|
|
word sign_a = BIGHIGHBIT( a );
|
|
|
|
if ( sign_a )
|
|
negate( a );
|
|
|
|
T b;
|
|
|
|
set ( b, b0 )
|
|
|
|
;
|
|
|
|
word sign_b = BIGHIGHBIT( b );
|
|
|
|
if ( sign_b )
|
|
negate( b );
|
|
|
|
zero( a0 );
|
|
|
|
u32 shifts = 0;
|
|
|
|
for ( u32 ii = 0; ii < BIGWORDCOUNT( T ); ++ii )
|
|
{
|
|
word w_i = b[ ii ];
|
|
|
|
u16 ctr = WORDBITS;
|
|
|
|
while ( w_i )
|
|
{
|
|
if ( w_i & 1 )
|
|
{
|
|
if ( shifts )
|
|
{
|
|
shiftLeft( a, shifts );
|
|
shifts = 0;
|
|
}
|
|
|
|
add ( a0, a )
|
|
|
|
;
|
|
}
|
|
|
|
w_i >>= 1;
|
|
++shifts;
|
|
--ctr;
|
|
}
|
|
|
|
shifts += ctr;
|
|
}
|
|
|
|
if ( sign_a ^ sign_b )
|
|
negate( a0 );
|
|
}
|
|
|
|
|
|
//////// division/remainder ////////
|
|
|
|
// {q, r} = a / b (&q != &r), unsigned
|
|
BIGONETYPE void udivide( T &a, T &b0, T &q, T &r )
|
|
{
|
|
T b;
|
|
|
|
set ( b, b0 )
|
|
|
|
;
|
|
set ( r, a )
|
|
|
|
;
|
|
zero( q );
|
|
|
|
u32 shifts = 1;
|
|
|
|
// sort of: shift b left until b > r, then shift back one
|
|
if ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
s32 ii, jj;
|
|
|
|
// shift by words if possible
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
if ( r[ ii ] )
|
|
break;
|
|
|
|
for ( jj = ii; jj >= 0; --jj )
|
|
if ( b[ jj ] )
|
|
break;
|
|
|
|
if ( ii != jj )
|
|
{
|
|
shifts = ( ii - jj ) * WORDBITS;
|
|
shiftLeft( b, shifts );
|
|
++shifts;
|
|
}
|
|
|
|
while ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
shiftLeft1( b );
|
|
|
|
++shifts;
|
|
}
|
|
|
|
while ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
}
|
|
|
|
else
|
|
if ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
|
|
u32 qshifts = 0;
|
|
|
|
while ( shifts-- )
|
|
{
|
|
++qshifts;
|
|
|
|
if ( !ugreater( b, r ) )
|
|
{
|
|
subtract( r, b );
|
|
|
|
shiftLeft( q, qshifts );
|
|
qshifts = 0;
|
|
|
|
q[ 0 ] |= 1;
|
|
}
|
|
|
|
ushiftRight1( b );
|
|
}
|
|
|
|
shiftLeft( q, qshifts );
|
|
}
|
|
|
|
// r = a Mod b, unsigned
|
|
BIGONETYPE void umodulo( T &a, T &b0, T &r )
|
|
{
|
|
T b;
|
|
u32 shifts = 1;
|
|
|
|
set ( b, b0 )
|
|
|
|
;
|
|
set ( r, a )
|
|
|
|
;
|
|
|
|
if ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
s32 ii, jj;
|
|
|
|
// shift by words if possible
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
if ( r[ ii ] )
|
|
break;
|
|
|
|
for ( jj = ii; jj >= 0; --jj )
|
|
if ( b[ jj ] )
|
|
break;
|
|
|
|
if ( ii != jj )
|
|
{
|
|
shifts = ( ii - jj ) * WORDBITS;
|
|
shiftLeft( b, shifts );
|
|
++shifts;
|
|
}
|
|
|
|
while ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
shiftLeft1( b );
|
|
|
|
++shifts;
|
|
}
|
|
|
|
while ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
}
|
|
|
|
else
|
|
if ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
|
|
while ( shifts-- )
|
|
{
|
|
if ( !ugreater( b, r ) )
|
|
subtract( r, b );
|
|
|
|
ushiftRight1( b );
|
|
}
|
|
}
|
|
|
|
// {q, r} = a / b (&q != &r), signed
|
|
BIGONETYPE void sdivide( T &a, T &b0, T &q, T &r )
|
|
{
|
|
T b;
|
|
|
|
set ( b, b0 )
|
|
|
|
;
|
|
set ( r, a )
|
|
|
|
;
|
|
zero( q );
|
|
|
|
word sign_a = BIGHIGHBIT( a );
|
|
|
|
if ( sign_a )
|
|
negate( r );
|
|
|
|
word sign_b = BIGHIGHBIT( b );
|
|
|
|
if ( sign_b )
|
|
negate( b );
|
|
|
|
u32 shifts = 1;
|
|
|
|
if ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
s32 ii, jj;
|
|
|
|
// shift by words if possible
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
if ( r[ ii ] )
|
|
break;
|
|
|
|
for ( jj = ii; jj >= 0; --jj )
|
|
if ( b[ jj ] )
|
|
break;
|
|
|
|
if ( ii != jj )
|
|
{
|
|
shifts = ( ii - jj ) * WORDBITS;
|
|
shiftLeft( b, shifts );
|
|
++shifts;
|
|
}
|
|
|
|
while ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
shiftLeft1( b );
|
|
|
|
++shifts;
|
|
}
|
|
|
|
while ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
}
|
|
|
|
else
|
|
if ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
|
|
u32 qshifts = 0;
|
|
|
|
while ( shifts-- )
|
|
{
|
|
++qshifts;
|
|
|
|
if ( !ugreater( b, r ) )
|
|
{
|
|
subtract( r, b );
|
|
|
|
shiftLeft( q, qshifts );
|
|
qshifts = 0;
|
|
|
|
q[ 0 ] |= 1;
|
|
}
|
|
|
|
ushiftRight1( b );
|
|
}
|
|
|
|
shiftLeft( q, qshifts );
|
|
|
|
if ( sign_a ^ sign_b )
|
|
negate( q );
|
|
|
|
if ( sign_a )
|
|
negate( r );
|
|
}
|
|
|
|
// r = a Mod b, signed
|
|
BIGONETYPE void smodulo( T &a, T &b0, T &r )
|
|
{
|
|
T b;
|
|
u32 shifts = 1;
|
|
|
|
set ( b, b0 )
|
|
|
|
;
|
|
set ( r, a )
|
|
|
|
;
|
|
|
|
word sign_a = BIGHIGHBIT( a );
|
|
|
|
if ( sign_a )
|
|
negate( r );
|
|
|
|
word sign_b = BIGHIGHBIT( b );
|
|
|
|
if ( sign_b )
|
|
negate( b );
|
|
|
|
if ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
s32 ii, jj;
|
|
|
|
// shift by words if possible
|
|
|
|
for ( ii = BIGWORDCOUNT( T ) - 1; ii >= 0; --ii )
|
|
if ( r[ ii ] )
|
|
break;
|
|
|
|
for ( jj = ii; jj >= 0; --jj )
|
|
if ( b[ jj ] )
|
|
break;
|
|
|
|
if ( ii != jj )
|
|
{
|
|
shifts = ( ii - jj ) * WORDBITS;
|
|
shiftLeft( b, shifts );
|
|
++shifts;
|
|
}
|
|
|
|
while ( !BIGHIGHBIT( b ) && ugreater( r, b ) )
|
|
{
|
|
shiftLeft1( b );
|
|
|
|
++shifts;
|
|
}
|
|
|
|
while ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
}
|
|
|
|
else
|
|
if ( ugreater( b, r ) )
|
|
{
|
|
ushiftRight1( b );
|
|
|
|
--shifts;
|
|
}
|
|
|
|
while ( shifts-- )
|
|
{
|
|
if ( !ugreater( b, r ) )
|
|
subtract( r, b );
|
|
|
|
ushiftRight1( b );
|
|
}
|
|
|
|
if ( sign_a )
|
|
negate( r );
|
|
}
|
|
|
|
|
|
//////// converting to/from strings ////////
|
|
|
|
#ifdef BIG_USES_STRINGS
|
|
|
|
// n -> string
|
|
BIGONETYPE std::string toString( T &n0, bool sign = true, u16 radix = 10 )
|
|
{
|
|
T n, base, r;
|
|
std::string s;
|
|
|
|
set ( n, n0 )
|
|
|
|
;
|
|
usetw( base, radix );
|
|
|
|
word sign_n = 0;
|
|
|
|
if ( sign && ( sign_n = BIGHIGHBIT( n ) ) )
|
|
negate( n );
|
|
|
|
do // always allow first iteration for zero
|
|
{
|
|
// {q, r} = n / base
|
|
udivide( n, base, n, r );
|
|
|
|
char ch = ( char ) r[ 0 ];
|
|
|
|
if ( ch >= 10 )
|
|
ch += 'a' - 10;
|
|
else
|
|
ch += '0';
|
|
|
|
// insert character
|
|
s = ch + s;
|
|
}
|
|
|
|
while ( !isZero( n ) );
|
|
|
|
if ( sign_n )
|
|
s = '-' + s;
|
|
|
|
return s;
|
|
}
|
|
|
|
// s -> n, signed
|
|
BIGONETYPE void fromString( std::string s, T &n, bool sign = true, u16 radix = 10 )
|
|
{
|
|
T acc, base, temp;
|
|
|
|
usetw( acc, 1 );
|
|
usetw( base, radix );
|
|
zero( n );
|
|
|
|
u32 len = ( u32 ) s.length();
|
|
const char *citer = s.c_str() + len;
|
|
|
|
while ( len-- )
|
|
{
|
|
char ch = *( --citer );
|
|
|
|
if ( IS_ALPHA( ch ) ) // assumes alpha characters only up to radix
|
|
ch = TO_LOWER( ch ) - 'a' + 10;
|
|
else
|
|
if ( sign && ch == '-' ) // '-' should be first character
|
|
{
|
|
negate( n );
|
|
break;
|
|
}
|
|
else // assumes it's alphanumeric/-
|
|
ch -= '0';
|
|
|
|
usetw( temp, ch );
|
|
|
|
umultiply( temp, acc );
|
|
|
|
add ( n, temp )
|
|
|
|
;
|
|
|
|
umultiply( acc, base );
|
|
}
|
|
}
|
|
|
|
#endif // BIG_USES_STRINGS
|
|
|
|
|
|
//////// class wrapper ////////
|
|
|
|
BIGONETYPE INLINE Int<T>::Int()
|
|
{
|
|
big::zero( raw );
|
|
}
|
|
|
|
BIGONETYPE INLINE Int<T>::Int( word n )
|
|
{
|
|
ssetw( raw, n );
|
|
}
|
|
|
|
#ifdef BIG_USES_STRINGS
|
|
|
|
BIGONETYPE INLINE Int<T>::Int( std::string &s )
|
|
{
|
|
fromString( s, raw );
|
|
}
|
|
|
|
#endif
|
|
|
|
BIGONETYPE INLINE Int<T>::Int( T &n )
|
|
{
|
|
set ( raw, n )
|
|
|
|
;
|
|
}
|
|
|
|
BIGONETYPE INLINE Int<T>::operator T &()
|
|
{
|
|
return raw;
|
|
}
|
|
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::zero()
|
|
{
|
|
big::zero( raw );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator=( word n )
|
|
{
|
|
ssetw( raw, n );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator=( T &n )
|
|
{
|
|
set ( raw, n )
|
|
|
|
;
|
|
return *this;
|
|
}
|
|
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator<<=( u32 s )
|
|
{
|
|
shiftLeft( raw, s );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator<<( u32 s )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp <<= s;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator>>=( u32 s )
|
|
{
|
|
shiftRight( raw, s );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator>>( u32 s )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp >>= s;
|
|
}
|
|
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator+=( T &n )
|
|
{
|
|
add ( raw, n )
|
|
|
|
;
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator+( T &n )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp += n;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator-=( T &n )
|
|
{
|
|
subtract( raw, n );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator-( T &n )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp -= n;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator++() // prefix
|
|
{
|
|
increment( raw );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator++( int ) // postfix
|
|
{
|
|
Int<T> temp( raw );
|
|
increment( raw );
|
|
return temp;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator--() // prefix
|
|
{
|
|
decrement( raw );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator--( int ) // postfix
|
|
{
|
|
Int<T> temp( raw );
|
|
decrement( raw );
|
|
return temp;
|
|
}
|
|
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator-( int ) // negation
|
|
{
|
|
Int<T> temp( raw );
|
|
negate( temp );
|
|
return temp;
|
|
}
|
|
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator*( T &n )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp *= n;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::square()
|
|
{
|
|
square( raw );
|
|
return *this;
|
|
}
|
|
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator/=( T &n )
|
|
{
|
|
T discard;
|
|
divide( raw, n, raw, discard );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator/( T &n )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp /= n;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator%=( T &n )
|
|
{
|
|
modulus( raw, n, raw );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTSLOW Int<T>::operator%( T &n )
|
|
{
|
|
Int<T> temp( raw );
|
|
return temp %= n;
|
|
}
|
|
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator>( T &n )
|
|
{
|
|
return sgreater( raw, n );
|
|
}
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator>=( T &n )
|
|
{
|
|
return sgreaterOrEqual( raw, n );
|
|
}
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator<( T &n )
|
|
{
|
|
return !sgreaterOrEqual( raw, n );
|
|
}
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator<=( T &n )
|
|
{
|
|
return !sgreater( raw, n );
|
|
}
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator==( T &n )
|
|
{
|
|
return equal( raw, n );
|
|
}
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator!=( T &n )
|
|
{
|
|
return !equal( raw, n );
|
|
}
|
|
|
|
BIGONETYPE INLINE /* fast */ bool Int<T>::operator!()
|
|
{
|
|
return isZero( raw );
|
|
}
|
|
|
|
#ifdef BIG_USES_STRINGS
|
|
|
|
BIGONETYPE INLINE /* fast */ std::string Int<T>::str()
|
|
{
|
|
return toString( raw );
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator=( std::string &s )
|
|
{
|
|
fromString( s, raw );
|
|
return *this;
|
|
}
|
|
|
|
BIGONETYPE BIGINTFAST Int<T>::operator=( const char *s )
|
|
{
|
|
fromString( std::string( s ), raw );
|
|
return *this;
|
|
}
|
|
|
|
#endif // BIG_USES_STRINGS
|
|
}
|
|
|
|
#endif // #if !defined(_COMPATIBILITY_1)
|
|
|
|
#endif // BIGTYPES_H
|
|
|
|
#ifdef _MSC_VER
|
|
#pragma warning( pop )
|
|
#endif
|