/js/lib/Socket.IO-node/support/socket.io-client/lib/vendor/web-socket-js/flash-src/com/hurlant/math/BigInteger.as
ActionScript | 1543 lines | 1136 code | 66 blank | 341 comment | 285 complexity | 7eb5efb1c2cc2a19d158053ed8c12e40 MD5 | raw file
Possible License(s): GPL-2.0, LGPL-2.1, MPL-2.0-no-copyleft-exception, BSD-3-Clause
- /**
- * BigInteger
- *
- * An ActionScript 3 implementation of BigInteger (light version)
- * Copyright (c) 2007 Henri Torgemane
- *
- * Derived from:
- * The jsbn library, Copyright (c) 2003-2005 Tom Wu
- *
- * See LICENSE.txt for full license information.
- */
- package com.hurlant.math
- {
- import com.hurlant.crypto.prng.Random;
- import com.hurlant.util.Hex;
- import com.hurlant.util.Memory;
-
- import flash.utils.ByteArray;
- use namespace bi_internal;
- public class BigInteger
- {
- public static const DB:int = 30; // number of significant bits per chunk
- public static const DV:int = (1<<DB);
- public static const DM:int = (DV-1); // Max value in a chunk
-
- public static const BI_FP:int = 52;
- public static const FV:Number = Math.pow(2, BI_FP);
- public static const F1:int = BI_FP - DB;
- public static const F2:int = 2*DB - BI_FP;
-
- public static const ZERO:BigInteger = nbv(0);
- public static const ONE:BigInteger = nbv(1);
-
- /*bi_internal */public var t:int; // number of chunks.
- bi_internal var s:int; // sign
- bi_internal var a:Array; // chunks
-
- /**
- *
- * @param value
- * @param radix WARNING: If value is ByteArray, this holds the number of bytes to use.
- * @param unsigned
- *
- */
- public function BigInteger(value:* = null, radix:int = 0, unsigned:Boolean = false) {
- a = new Array;
- if (value is String) {
- if (radix&&radix!=16) throw new Error("BigInteger construction with radix!=16 is not supported.");
- value = Hex.toArray(value);
- radix=0;
- }
- if (value is ByteArray) {
- var array:ByteArray = value as ByteArray;
- var length:int = radix || (array.length - array.position);
- fromArray(array, length, unsigned);
- }
- }
- public function dispose():void {
- var r:Random = new Random;
- for (var i:uint=0;i<a.length;i++) {
- a[i] = r.nextByte();
- delete a[i];
- }
- a=null;
- t=0;
- s=0;
- Memory.gc();
- }
-
- public function toString(radix:Number=16):String {
- if (s<0) return "-"+negate().toString(radix);
- var k:int;
- switch (radix) {
- case 2: k=1; break;
- case 4: k=2; break;
- case 8: k=3; break;
- case 16: k=4; break;
- case 32: k=5; break;
- default:
- // return toRadix(radix);
- }
- var km:int = (1<<k)-1;
- var d:int = 0;
- var m:Boolean = false;
- var r:String = "";
- var i:int = t;
- var p:int = DB-(i*DB)%k;
- if (i-->0) {
- if (p<DB && (d=a[i]>>p)>0) {
- m = true;
- r = d.toString(36);
- }
- while (i >= 0) {
- if (p<k) {
- d = (a[i]&((1<<p)-1))<<(k-p);
- d|= a[--i]>>(p+=DB-k);
- } else {
- d = (a[i]>>(p-=k))&km;
- if (p<=0) {
- p += DB;
- --i;
- }
- }
- if (d>0) {
- m = true;
- }
- if (m) {
- r += d.toString(36);
- }
- }
- }
- return m?r:"0";
- }
- public function toArray(array:ByteArray):uint {
- const k:int = 8;
- const km:int = (1<<8)-1;
- var d:int = 0;
- var i:int = t;
- var p:int = DB-(i*DB)%k;
- var m:Boolean = false;
- var c:int = 0;
- if (i-->0) {
- if (p<DB && (d=a[i]>>p)>0) {
- m = true;
- array.writeByte(d);
- c++;
- }
- while (i >= 0) {
- if (p<k) {
- d = (a[i]&((1<<p)-1))<<(k-p);
- d|= a[--i]>>(p+=DB-k);
- } else {
- d = (a[i]>>(p-=k))&km;
- if (p<=0) {
- p += DB;
- --i;
- }
- }
- if (d>0) {
- m = true;
- }
- if (m) {
- array.writeByte(d);
- c++;
- }
- }
- }
- return c;
- }
- /**
- * best-effort attempt to fit into a Number.
- * precision can be lost if it just can't fit.
- */
- public function valueOf():Number {
- if (s==-1) {
- return -negate().valueOf();
- }
- var coef:Number = 1;
- var value:Number = 0;
- for (var i:uint=0;i<t;i++) {
- value += a[i]*coef;
- coef *= DV;
- }
- return value;
- }
- /**
- * -this
- */
- public function negate():BigInteger {
- var r:BigInteger = nbi();
- ZERO.subTo(this, r);
- return r;
- }
- /**
- * |this|
- */
- public function abs():BigInteger {
- return (s<0)?negate():this;
- }
- /**
- * return + if this > v, - if this < v, 0 if equal
- */
- public function compareTo(v:BigInteger):int {
- var r:int = s - v.s;
- if (r!=0) {
- return r;
- }
- var i:int = t;
- r = i-v.t;
- if (r!=0) {
- return r;
- }
- while (--i >=0) {
- r=a[i]-v.a[i];
- if (r != 0) return r;
- }
- return 0;
- }
- /**
- * returns bit length of the integer x
- */
- bi_internal function nbits(x:int):int {
- var r:int = 1;
- var t:int;
- if ((t=x>>>16) != 0) { x = t; r += 16; }
- if ((t=x>>8) != 0) { x = t; r += 8; }
- if ((t=x>>4) != 0) { x = t; r += 4; }
- if ((t=x>>2) != 0) { x = t; r += 2; }
- if ((t=x>>1) != 0) { x = t; r += 1; }
- return r;
- }
- /**
- * returns the number of bits in this
- */
- public function bitLength():int {
- if (t<=0) return 0;
- return DB*(t-1)+nbits(a[t-1]^(s&DM));
- }
- /**
- *
- * @param v
- * @return this % v
- *
- */
- public function mod(v:BigInteger):BigInteger {
- var r:BigInteger = nbi();
- abs().divRemTo(v,null,r);
- if (s<0 && r.compareTo(ZERO)>0) {
- v.subTo(r,r);
- }
- return r;
- }
- /**
- * this^e % m, 0 <= e < 2^32
- */
- public function modPowInt(e:int, m:BigInteger):BigInteger {
- var z:IReduction;
- if (e<256 || m.isEven()) {
- z = new ClassicReduction(m);
- } else {
- z = new MontgomeryReduction(m);
- }
- return exp(e, z);
- }
- /**
- * copy this to r
- */
- bi_internal function copyTo(r:BigInteger):void {
- for (var i:int = t-1; i>=0; --i) {
- r.a[i] = a[i];
- }
- r.t = t;
- r.s = s;
- }
- /**
- * set from integer value "value", -DV <= value < DV
- */
- bi_internal function fromInt(value:int):void {
- t = 1;
- s = (value<0)?-1:0;
- if (value>0) {
- a[0] = value;
- } else if (value<-1) {
- a[0] = value+DV;
- } else {
- t = 0;
- }
- }
- /**
- * set from ByteArray and length,
- * starting a current position
- * If length goes beyond the array, pad with zeroes.
- */
- bi_internal function fromArray(value:ByteArray, length:int, unsigned:Boolean = false):void {
- var p:int = value.position;
- var i:int = p+length;
- var sh:int = 0;
- const k:int = 8;
- t = 0;
- s = 0;
- while (--i >= p) {
- var x:int = i<value.length?value[i]:0;
- if (sh == 0) {
- a[t++] = x;
- } else if (sh+k > DB) {
- a[t-1] |= (x&((1<<(DB-sh))-1))<<sh;
- a[t++] = x>>(DB-sh);
- } else {
- a[t-1] |= x<<sh;
- }
- sh += k;
- if (sh >= DB) sh -= DB;
- }
- if (!unsigned && (value[0]&0x80)==0x80) {
- s = -1;
- if (sh > 0) {
- a[t-1] |= ((1<<(DB-sh))-1)<<sh;
- }
- }
- clamp();
- value.position = Math.min(p+length,value.length);
- }
- /**
- * clamp off excess high words
- */
- bi_internal function clamp():void {
- var c:int = s&DM;
- while (t>0 && a[t-1]==c) {
- --t;
- }
- }
- /**
- * r = this << n*DB
- */
- bi_internal function dlShiftTo(n:int, r:BigInteger):void {
- var i:int;
- for (i=t-1; i>=0; --i) {
- r.a[i+n] = a[i];
- }
- for (i=n-1; i>=0; --i) {
- r.a[i] = 0;
- }
- r.t = t+n;
- r.s = s;
- }
- /**
- * r = this >> n*DB
- */
- bi_internal function drShiftTo(n:int, r:BigInteger):void {
- var i:int;
- for (i=n; i<t; ++i) {
- r.a[i-n] = a[i];
- }
- r.t = Math.max(t-n,0);
- r.s = s;
- }
- /**
- * r = this << n
- */
- bi_internal function lShiftTo(n:int, r:BigInteger):void {
- var bs:int = n%DB;
- var cbs:int = DB-bs;
- var bm:int = (1<<cbs)-1;
- var ds:int = n/DB;
- var c:int = (s<<bs)&DM;
- var i:int;
- for (i=t-1; i>=0; --i) {
- r.a[i+ds+1] = (a[i]>>cbs)|c;
- c = (a[i]&bm)<<bs;
- }
- for (i=ds-1; i>=0; --i) {
- r.a[i] = 0;
- }
- r.a[ds] = c;
- r.t = t+ds+1;
- r.s = s;
- r.clamp();
- }
- /**
- * r = this >> n
- */
- bi_internal function rShiftTo(n:int, r:BigInteger):void {
- r.s = s;
- var ds:int = n/DB;
- if (ds >= t) {
- r.t = 0;
- return;
- }
- var bs:int = n%DB;
- var cbs:int = DB-bs;
- var bm:int = (1<<bs)-1;
- r.a[0] = a[ds]>>bs;
- var i:int;
- for (i=ds+1; i<t; ++i) {
- r.a[i-ds-1] |= (a[i]&bm)<<cbs;
- r.a[i-ds] = a[i]>>bs;
- }
- if (bs>0) {
- r.a[t-ds-1] |= (s&bm)<<cbs;
- }
- r.t = t-ds;
- r.clamp();
- }
- /**
- * r = this - v
- */
- bi_internal function subTo(v:BigInteger, r:BigInteger):void {
- var i:int = 0;
- var c:int = 0;
- var m:int = Math.min(v.t, t);
- while (i<m) {
- c += a[i] - v.a[i];
- r.a[i++] = c & DM;
- c >>= DB;
- }
- if (v.t < t) {
- c -= v.s;
- while (i< t) {
- c+= a[i];
- r.a[i++] = c&DM;
- c >>= DB;
- }
- c += s;
- } else {
- c += s;
- while (i < v.t) {
- c -= v.a[i];
- r.a[i++] = c&DM;
- c >>= DB;
- }
- c -= v.s;
- }
- r.s = (c<0)?-1:0;
- if (c<-1) {
- r.a[i++] = DV+c;
- } else if (c>0) {
- r.a[i++] = c;
- }
- r.t = i;
- r.clamp();
- }
- /**
- * am: Compute w_j += (x*this_i), propagates carries,
- * c is initial carry, returns final carry.
- * c < 3*dvalue, x < 2*dvalue, this_i < dvalue
- */
- bi_internal function am(i:int,x:int,w:BigInteger,j:int,c:int,n:int):int {
- var xl:int = x&0x7fff;
- var xh:int = x>>15;
- while(--n >= 0) {
- var l:int = a[i]&0x7fff;
- var h:int = a[i++]>>15;
- var m:int = xh*l + h*xl;
- l = xl*l + ((m&0x7fff)<<15)+w.a[j]+(c&0x3fffffff);
- c = (l>>>30)+(m>>>15)+xh*h+(c>>>30);
- w.a[j++] = l&0x3fffffff;
- }
- return c;
- }
- /**
- * r = this * v, r != this,a (HAC 14.12)
- * "this" should be the larger one if appropriate
- */
- bi_internal function multiplyTo(v:BigInteger, r:BigInteger):void {
- var x:BigInteger = abs();
- var y:BigInteger = v.abs();
- var i:int = x.t;
- r.t = i+y.t;
- while (--i >= 0) {
- r.a[i] = 0;
- }
- for (i=0; i<y.t; ++i) {
- r.a[i+x.t] = x.am(0, y.a[i], r, i, 0, x.t);
- }
- r.s = 0;
- r.clamp();
- if (s!=v.s) {
- ZERO.subTo(r, r);
- }
- }
- /**
- * r = this^2, r != this (HAC 14.16)
- */
- bi_internal function squareTo(r:BigInteger):void {
- var x:BigInteger = abs();
- var i:int = r.t = 2*x.t;
- while (--i>=0) r.a[i] = 0;
- for (i=0; i<x.t-1; ++i) {
- var c:int = x.am(i, x.a[i], r, 2*i, 0, 1);
- if ((r.a[i+x.t] += x.am(i+1, 2*x.a[i], r, 2*i+1, c, x.t-i-1)) >= DV) {
- r.a[i+x.t] -= DV;
- r.a[i+x.t+1] = 1;
- }
- }
- if (r.t>0) {
- r.a[r.t-1] += x.am(i, x.a[i], r, 2*i, 0, 1);
- }
- r.s = 0;
- r.clamp();
- }
- /**
- * divide this by m, quotient and remainder to q, r (HAC 14.20)
- * r != q, this != m. q or r may be null.
- */
- bi_internal function divRemTo(m:BigInteger, q:BigInteger = null, r:BigInteger = null):void {
- var pm:BigInteger = m.abs();
- if (pm.t <= 0) return;
- var pt:BigInteger = abs();
- if (pt.t < pm.t) {
- if (q!=null) q.fromInt(0);
- if (r!=null) copyTo(r);
- return;
- }
- if (r==null) r = nbi();
- var y:BigInteger = nbi();
- var ts:int = s;
- var ms:int = m.s;
- var nsh:int = DB-nbits(pm.a[pm.t-1]); // normalize modulus
- if (nsh>0) {
- pm.lShiftTo(nsh, y);
- pt.lShiftTo(nsh, r);
- } else {
- pm.copyTo(y);
- pt.copyTo(r);
- }
- var ys:int = y.t;
- var y0:int = y.a[ys-1];
- if (y0==0) return;
- var yt:Number = y0*(1<<F1)+((ys>1)?y.a[ys-2]>>F2:0);
- var d1:Number = FV/yt;
- var d2:Number = (1<<F1)/yt;
- var e:Number = 1<<F2;
- var i:int = r.t;
- var j:int = i-ys;
- var t:BigInteger = (q==null)?nbi():q;
- y.dlShiftTo(j,t);
- if (r.compareTo(t)>=0) {
- r.a[r.t++] = 1;
- r.subTo(t,r);
- }
- ONE.dlShiftTo(ys,t);
- t.subTo(y,y); // "negative" y so we can replace sub with am later.
- while(y.t<ys) y.(y.t++, 0);
- while(--j >= 0) {
- // Estimate quotient digit
- var qd:int = (r.a[--i]==y0)?DM:Number(r.a[i])*d1+(Number(r.a[i-1])+e)*d2;
- if ((r.a[i]+= y.am(0, qd, r, j, 0, ys))<qd) { // Try it out
- y.dlShiftTo(j, t);
- r.subTo(t,r);
- while (r.a[i]<--qd) {
- r.subTo(t,r);
- }
- }
- }
- if (q!=null) {
- r.drShiftTo(ys,q);
- if (ts!=ms) {
- ZERO.subTo(q,q);
- }
- }
- r.t = ys;
- r.clamp();
- if (nsh>0) {
- r.rShiftTo(nsh, r); // Denormalize remainder
- }
- if (ts<0) {
- ZERO.subTo(r,r);
- }
- }
- /**
- * return "-1/this % 2^DB"; useful for Mont. reduction
- * justification:
- * xy == 1 (mod n)
- * xy = 1+km
- * xy(2-xy) = (1+km)(1-km)
- * x[y(2-xy)] = 1-k^2.m^2
- * x[y(2-xy)] == 1 (mod m^2)
- * if y is 1/x mod m, then y(2-xy) is 1/x mod m^2
- * should reduce x and y(2-xy) by m^2 at each step to keep size bounded
- * [XXX unit test the living shit out of this.]
- */
- bi_internal function invDigit():int {
- if (t<1) return 0;
- var x:int = a[0];
- if ((x&1)==0) return 0;
- var y:int = x&3; // y == 1/x mod 2^2
- y = (y*(2-(x&0xf )*y)) &0xf; // y == 1/x mod 2^4
- y = (y*(2-(x&0xff)*y)) &0xff; // y == 1/x mod 2^8
- y = (y*(2-(((x&0xffff)*y)&0xffff)))&0xffff; // y == 1/x mod 2^16
- // last step - calculate inverse mod DV directly;
- // assumes 16 < DB <= 32 and assumes ability to handle 48-bit ints
- // XXX 48 bit ints? Whaaaa? is there an implicit float conversion in here?
- y = (y*(2-x*y%DV))%DV; // y == 1/x mod 2^dbits
- // we really want the negative inverse, and -DV < y < DV
- return (y>0)?DV-y:-y;
- }
- /**
- * true iff this is even
- */
- bi_internal function isEven():Boolean {
- return ((t>0)?(a[0]&1):s) == 0;
- }
- /**
- * this^e, e < 2^32, doing sqr and mul with "r" (HAC 14.79)
- */
- bi_internal function exp(e:int, z:IReduction):BigInteger {
- if (e > 0xffffffff || e < 1) return ONE;
- var r:BigInteger = nbi();
- var r2:BigInteger = nbi();
- var g:BigInteger = z.convert(this);
- var i:int = nbits(e)-1;
- g.copyTo(r);
- while(--i>=0) {
- z.sqrTo(r, r2);
- if ((e&(1<<i))>0) {
- z.mulTo(r2,g,r);
- } else {
- var t:BigInteger = r;
- r = r2;
- r2 = t;
- }
-
- }
- return z.revert(r);
- }
- bi_internal function intAt(str:String, index:int):int {
- return parseInt(str.charAt(index), 36);
- }
- protected function nbi():* {
- return new BigInteger;
- }
- /**
- * return bigint initialized to value
- */
- public static function nbv(value:int):BigInteger {
- var bn:BigInteger = new BigInteger;
- bn.fromInt(value);
- return bn;
- }
- // Functions above are sufficient for RSA encryption.
- // The stuff below is useful for decryption and key generation
- public static const lowprimes:Array = [2,3,5,7,11,13,17,19,23,29,31,37,41,43,47,53,59,61,67,71,73,79,83,89,97,101,103,107,109,113,127,131,137,139,149,151,157,163,167,173,179,181,191,193,197,199,211,223,227,229,233,239,241,251,257,263,269,271,277,281,283,293,307,311,313,317,331,337,347,349,353,359,367,373,379,383,389,397,401,409,419,421,431,433,439,443,449,457,461,463,467,479,487,491,499,503,509];
- public static const lplim:int = (1<<26)/lowprimes[lowprimes.length-1];
- public function clone():BigInteger {
- var r:BigInteger = new BigInteger;
- this.copyTo(r);
- return r;
- }
-
- /**
- *
- * @return value as integer
- *
- */
- public function intValue():int {
- if (s<0) {
- if (t==1) {
- return a[0]-DV;
- } else if (t==0) {
- return -1;
- }
- } else if (t==1) {
- return a[0];
- } else if (t==0) {
- return 0;
- }
- // assumes 16 < DB < 32
- return ((a[1]&((1<<(32-DB))-1))<<DB)|a[0];
- }
-
- /**
- *
- * @return value as byte
- *
- */
- public function byteValue():int {
- return (t==0)?s:(a[0]<<24)>>24;
- }
-
- /**
- *
- * @return value as short (assumes DB>=16)
- *
- */
- public function shortValue():int {
- return (t==0)?s:(a[0]<<16)>>16;
- }
-
- /**
- *
- * @param r
- * @return x s.t. r^x < DV
- *
- */
- protected function chunkSize(r:Number):int {
- return Math.floor(Math.LN2*DB/Math.log(r));
- }
-
- /**
- *
- * @return 0 if this ==0, 1 if this >0
- *
- */
- public function sigNum():int {
- if (s<0) {
- return -1;
- } else if (t<=0 || (t==1 && a[0]<=0)) {
- return 0;
- } else{
- return 1;
- }
- }
-
- /**
- *
- * @param b: radix to use
- * @return a string representing the integer converted to the radix.
- *
- */
- protected function toRadix(b:uint=10):String {
- if (sigNum()==0 || b<2 || b>32) return "0";
- var cs:int = chunkSize(b);
- var a:Number = Math.pow(b, cs);
- var d:BigInteger = nbv(a);
- var y:BigInteger = nbi();
- var z:BigInteger = nbi();
- var r:String = "";
- divRemTo(d, y, z);
- while (y.sigNum()>0) {
- r = (a+z.intValue()).toString(b).substr(1) + r;
- y.divRemTo(d,y,z);
- }
- return z.intValue().toString(b) + r;
- }
-
- /**
- *
- * @param s a string to convert from using radix.
- * @param b a radix
- *
- */
- protected function fromRadix(s:String, b:int = 10):void {
- fromInt(0);
- var cs:int = chunkSize(b);
- var d:Number = Math.pow(b, cs);
- var mi:Boolean = false;
- var j:int = 0;
- var w:int = 0;
- for (var i:int=0;i<s.length;++i) {
- var x:int = intAt(s, i);
- if (x<0) {
- if (s.charAt(i) == "-" && sigNum() == 0) {
- mi = true;
- }
- continue;
- }
- w = b*w+x;
- if (++j >= cs) {
- dMultiply(d);
- dAddOffset(w,0);
- j=0;
- w=0;
- }
- }
- if (j>0) {
- dMultiply(Math.pow(b,j));
- dAddOffset(w,0);
- }
- if (mi) {
- BigInteger.ZERO.subTo(this, this);
- }
- }
-
- // XXX function fromNumber not written yet.
-
- /**
- *
- * @return a byte array.
- *
- */
- public function toByteArray():ByteArray {
- var i:int = t;
- var r:ByteArray = new ByteArray;
- r[0] = s;
- var p:int = DB-(i*DB)%8;
- var d:int;
- var k:int=0;
- if (i-->0) {
- if (p<DB && (d=a[i]>>p)!=(s&DM)>>p) {
- r[k++] = d|(s<<(DB-p));
- }
- while (i>=0) {
- if(p<8) {
- d = (a[i]&((1<<p)-1))<<(8-p);
- d|= a[--i]>>(p+=DB-8);
- } else {
- d = (a[i]>>(p-=8))&0xff;
- if (p<=0) {
- p += DB;
- --i;
- }
- }
- if ((d&0x80)!=0) d|=-256;
- if (k==0 && (s&0x80)!=(d&0x80)) ++k;
- if (k>0 || d!=s) r[k++] = d;
- }
- }
- return r;
- }
- public function equals(a:BigInteger):Boolean {
- return compareTo(a)==0;
- }
- public function min(a:BigInteger):BigInteger {
- return (compareTo(a)<0)?this:a;
- }
- public function max(a:BigInteger):BigInteger {
- return (compareTo(a)>0)?this:a;
- }
-
- /**
- *
- * @param a a BigInteger to perform the operation with
- * @param op a Function implementing the operation
- * @param r a BigInteger to store the result of the operation
- *
- */
- protected function bitwiseTo(a:BigInteger, op:Function, r:BigInteger):void {
- var i:int;
- var f:int;
- var m:int = Math.min(a.t, t);
- for (i=0; i<m; ++i) {
- r.a[i] = op(this.a[i],a.a[i]);
- }
- if (a.t<t) {
- f = a.s&DM;
- for (i=m;i<t;++i) {
- r.a[i] = op(this.a[i],f);
- }
- r.t = t;
- } else {
- f = s&DM;
- for (i=m;i<a.t;++i) {
- r.a[i] = op(f,a.a[i]);
- }
- r.t = a.t;
- }
- r.s = op(s, a.s);
- r.clamp();
- }
-
- private function op_and(x:int, y:int):int {return x&y;}
- public function and(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- bitwiseTo(a, op_and, r);
- return r;
- }
-
- private function op_or(x:int, y:int):int {return x|y;}
- public function or(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- bitwiseTo(a, op_or, r);
- return r;
- }
-
- private function op_xor(x:int, y:int):int {return x^y;}
- public function xor(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- bitwiseTo(a, op_xor, r);
- return r;
- }
-
- private function op_andnot(x:int, y:int):int { return x&~y;}
- public function andNot(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- bitwiseTo(a, op_andnot, r);
- return r;
- }
-
- public function not():BigInteger {
- var r:BigInteger = new BigInteger;
- for (var i:int=0;i<t;++i) {
- r[i] = DM&~a[i];
- }
- r.t = t;
- r.s = ~s;
- return r;
- }
-
- public function shiftLeft(n:int):BigInteger {
- var r:BigInteger = new BigInteger;
- if (n<0) {
- rShiftTo(-n, r);
- } else {
- lShiftTo(n, r);
- }
- return r;
- }
- public function shiftRight(n:int):BigInteger {
- var r:BigInteger = new BigInteger;
- if (n<0) {
- lShiftTo(-n, r);
- } else {
- rShiftTo(n, r);
- }
- return r;
- }
-
- /**
- *
- * @param x
- * @return index of lowet 1-bit in x, x < 2^31
- *
- */
- private function lbit(x:int):int {
- if (x==0) return -1;
- var r:int = 0;
- if ((x&0xffff)==0) { x>>= 16; r += 16; }
- if ((x&0xff) == 0) { x>>= 8; r += 8; }
- if ((x&0xf) == 0) { x>>= 4; r += 4; }
- if ((x&0x3) == 0) { x>>= 2; r += 2; }
- if ((x&0x1) == 0) ++r;
- return r;
- }
-
- /**
- *
- * @return index of lowest 1-bit (or -1 if none)
- *
- */
- public function getLowestSetBit():int {
- for (var i:int=0;i<t;++i) {
- if (a[i]!=0) return i*DB+lbit(a[i]);
- }
- if (s<0) return t*DB;
- return -1;
- }
-
- /**
- *
- * @param x
- * @return number of 1 bits in x
- *
- */
- private function cbit(x:int):int {
- var r:uint =0;
- while (x!=0) { x &= x-1; ++r }
- return r;
- }
-
- /**
- *
- * @return number of set bits
- *
- */
- public function bitCount():int {
- var r:int=0;
- var x:int = s&DM;
- for (var i:int=0;i<t;++i) {
- r += cbit(a[i]^x);
- }
- return r;
- }
-
- /**
- *
- * @param n
- * @return true iff nth bit is set
- *
- */
- public function testBit(n:int):Boolean {
- var j:int = Math.floor(n/DB);
- if (j>=t) {
- return s!=0;
- }
- return ((a[j]&(1<<(n%DB)))!=0);
- }
-
- /**
- *
- * @param n
- * @param op
- * @return this op (1<<n)
- *
- */
- protected function changeBit(n:int,op:Function):BigInteger {
- var r:BigInteger = BigInteger.ONE.shiftLeft(n);
- bitwiseTo(r, op, r);
- return r;
- }
-
- /**
- *
- * @param n
- * @return this | (1<<n)
- *
- */
- public function setBit(n:int):BigInteger { return changeBit(n, op_or); }
- /**
- *
- * @param n
- * @return this & ~(1<<n)
- *
- */
- public function clearBit(n:int):BigInteger { return changeBit(n, op_andnot); }
- /**
- *
- * @param n
- * @return this ^ (1<<n)
- *
- */
- public function flipBit(n:int):BigInteger { return changeBit(n, op_xor); }
- /**
- *
- * @param a
- * @param r = this + a
- *
- */
- protected function addTo(a:BigInteger, r:BigInteger):void {
- var i:int = 0;
- var c:int = 0;
- var m:int = Math.min(a.t, t);
- while (i<m) {
- c += this.a[i] + a.a[i];
- r.a[i++] = c&DM;
- c>>=DB;
- }
- if (a.t < t) {
- c += a.s;
- while (i<t) {
- c += this.a[i];
- r.a[i++] = c&DM;
- c >>= DB;
- }
- c += s;
- } else {
- c += s;
- while (i<a.t) {
- c += a.a[i];
- r.a[i++] = c&DM;
- c >>= DB;
- }
- c += a.s;
- }
- r.s = (c<0)?-1:0;
- if (c>0) {
- r.a[i++] = c;
- } else if (c<-1) {
- r.a[i++] = DV+c;
- }
- r.t = i;
- r.clamp();
- }
-
- /**
- *
- * @param a
- * @return this + a
- *
- */
- public function add(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- addTo(a,r);
- return r;
- }
- /**
- *
- * @param a
- * @return this - a
- *
- */
- public function subtract(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- subTo(a,r);
- return r;
- }
-
- /**
- *
- * @param a
- * @return this * a
- *
- */
- public function multiply(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- multiplyTo(a,r);
- return r;
- }
-
- /**
- *
- * @param a
- * @return this / a
- *
- */
- public function divide(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- divRemTo(a, r, null);
- return r;
- }
-
- public function remainder(a:BigInteger):BigInteger {
- var r:BigInteger = new BigInteger;
- divRemTo(a, null, r);
- return r;
- }
-
- /**
- *
- * @param a
- * @return [this/a, this%a]
- *
- */
- public function divideAndRemainder(a:BigInteger):Array {
- var q:BigInteger = new BigInteger;
- var r:BigInteger = new BigInteger;
- divRemTo(a, q, r);
- return [q,r];
- }
-
- /**
- *
- * this *= n, this >=0, 1 < n < DV
- *
- * @param n
- *
- */
- bi_internal function dMultiply(n:int):void {
- a[t] = am(0, n-1, this, 0, 0, t);
- ++t;
- clamp();
- }
-
- /**
- *
- * this += n << w words, this >= 0
- *
- * @param n
- * @param w
- *
- */
- bi_internal function dAddOffset(n:int, w:int):void {
- while (t<=w) {
- a[t++] = 0;
- }
- a[w] += n;
- while (a[w] >= DV) {
- a[w] -= DV;
- if (++w >= t) {
- a[t++] = 0;
- }
- ++a[w];
- }
- }
- /**
- *
- * @param e
- * @return this^e
- *
- */
- public function pow(e:int):BigInteger {
- return exp(e, new NullReduction);
- }
-
- /**
- *
- * @param a
- * @param n
- * @param r = lower n words of "this * a", a.t <= n
- *
- */
- bi_internal function multiplyLowerTo(a:BigInteger, n:int, r:BigInteger):void {
- var i:int = Math.min(t+a.t, n);
- r.s = 0; // assumes a, this >= 0
- r.t = i;
- while (i>0) {
- r.a[--i]=0;
- }
- var j:int;
- for (j=r.t-t;i<j;++i) {
- r.a[i+t] = am(0, a.a[i], r, i, 0, t);
- }
- for (j=Math.min(a.t,n);i<j;++i) {
- am(0, a.a[i], r, i, 0, n-i);
- }
- r.clamp();
- }
-
- /**
- *
- * @param a
- * @param n
- * @param r = "this * a" without lower n words, n > 0
- *
- */
- bi_internal function multiplyUpperTo(a:BigInteger, n:int, r:BigInteger):void {
- --n;
- var i:int = r.t = t+a.t-n;
- r.s = 0; // assumes a,this >= 0
- while (--i>=0) {
- r.a[i] = 0;
- }
- for (i=Math.max(n-t,0);i<a.t;++i) {
- r.a[t+i-n] = am(n-i, a.a[i], r, 0, 0, t+i-n);
- }
- r.clamp();
- r.drShiftTo(1,r);
- }
-
- /**
- *
- * @param e
- * @param m
- * @return this^e % m (HAC 14.85)
- *
- */
- public function modPow(e:BigInteger, m:BigInteger):BigInteger {
- var i:int = e.bitLength();
- var k:int;
- var r:BigInteger = nbv(1);
- var z:IReduction;
-
- if (i<=0) {
- return r;
- } else if (i<18) {
- k=1;
- } else if (i<48) {
- k=3;
- } else if (i<144) {
- k=4;
- } else if (i<768) {
- k=5;
- } else {
- k=6;
- }
- if (i<8) {
- z = new ClassicReduction(m);
- } else if (m.isEven()) {
- z = new BarrettReduction(m);
- } else {
- z = new MontgomeryReduction(m);
- }
- // precomputation
- var g:Array = [];
- var n:int = 3;
- var k1:int = k-1;
- var km:int = (1<<k)-1;
- g[1] = z.convert(this);
- if (k > 1) {
- var g2:BigInteger = new BigInteger;
- z.sqrTo(g[1], g2);
- while (n<=km) {
- g[n] = new BigInteger;
- z.mulTo(g2, g[n-2], g[n]);
- n += 2;
- }
- }
-
- var j:int = e.t-1;
- var w:int;
- var is1:Boolean = true;
- var r2:BigInteger = new BigInteger;
- var t:BigInteger;
- i = nbits(e.a[j])-1;
- while (j>=0) {
- if (i>=k1) {
- w = (e.a[j]>>(i-k1))&km;
- } else {
- w = (e.a[j]&((1<<(i+1))-1))<<(k1-i);
- if (j>0) {
- w |= e.a[j-1]>>(DB+i-k1);
- }
- }
- n = k;
- while ((w&1)==0) {
- w >>= 1;
- --n;
- }
- if ((i -= n) <0) {
- i += DB;
- --j;
- }
- if (is1) { // ret == 1, don't bother squaring or multiplying it
- g[w].copyTo(r);
- is1 = false;
- } else {
- while (n>1) {
- z.sqrTo(r, r2);
- z.sqrTo(r2, r);
- n -= 2;
- }
- if (n>0) {
- z.sqrTo(r, r2);
- } else {
- t = r;
- r = r2;
- r2 = t;
- }
- z.mulTo(r2, g[w], r);
- }
- while (j>=0 && (e.a[j]&(1<<i)) == 0) {
- z.sqrTo(r, r2);
- t = r;
- r = r2;
- r2 = t;
- if (--i<0) {
- i = DB-1;
- --j;
- }
-
- }
- }
- return z.revert(r);
- }
-
- /**
- *
- * @param a
- * @return gcd(this, a) (HAC 14.54)
- *
- */
- public function gcd(a:BigInteger):BigInteger {
- var x:BigInteger = (s<0)?negate():clone();
- var y:BigInteger = (a.s<0)?a.negate():a.clone();
- if (x.compareTo(y)<0) {
- var t:BigInteger=x;
- x=y;
- y=t;
- }
- var i:int = x.getLowestSetBit();
- var g:int = y.getLowestSetBit();
- if (g<0) return x;
- if (i<g) g= i;
- if (g>0) {
- x.rShiftTo(g, x);
- y.rShiftTo(g, y);
- }
- while (x.sigNum()>0) {
- if ((i = x.getLowestSetBit()) >0) {
- x.rShiftTo(i, x);
- }
- if ((i = y.getLowestSetBit()) >0) {
- y.rShiftTo(i, y);
- }
- if (x.compareTo(y) >= 0) {
- x.subTo(y, x);
- x.rShiftTo(1, x);
- } else {
- y.subTo(x, y);
- y.rShiftTo(1, y);
- }
- }
- if (g>0) {
- y.lShiftTo(g, y);
- }
- return y;
- }
- /**
- *
- * @param n
- * @return this % n, n < 2^DB
- *
- */
- protected function modInt(n:int):int {
- if (n<=0) return 0;
- var d:int = DV%n;
- var r:int = (s<0)?n-1:0;
- if (t>0) {
- if (d==0) {
- r = a[0]%n;
- } else {
- for (var i:int=t-1;i>=0;--i) {
- r = (d*r+a[i])%n;
- }
- }
- }
- return r;
- }
-
- /**
- *
- * @param m
- * @return 1/this %m (HAC 14.61)
- *
- */
- public function modInverse(m:BigInteger):BigInteger {
- var ac:Boolean = m.isEven();
- if ((isEven()&&ac) || m.sigNum()==0) {
- return BigInteger.ZERO;
- }
- var u:BigInteger = m.clone();
- var v:BigInteger = clone();
- var a:BigInteger = nbv(1);
- var b:BigInteger = nbv(0);
- var c:BigInteger = nbv(0);
- var d:BigInteger = nbv(1);
- while (u.sigNum()!=0) {
- while (u.isEven()) {
- u.rShiftTo(1,u);
- if (ac) {
- if (!a.isEven() || !b.isEven()) {
- a.addTo(this,a);
- b.subTo(m,b);
- }
- a.rShiftTo(1,a);
- } else if (!b.isEven()) {
- b.subTo(m,b);
- }
- b.rShiftTo(1,b);
- }
- while (v.isEven()) {
- v.rShiftTo(1,v);
- if (ac) {
- if (!c.isEven() || !d.isEven()) {
- c.addTo(this,c);
- d.subTo(m,d);
- }
- c.rShiftTo(1,c);
- } else if (!d.isEven()) {
- d.subTo(m,d);
- }
- d.rShiftTo(1,d);
- }
- if (u.compareTo(v)>=0) {
- u.subTo(v,u);
- if (ac) {
- a.subTo(c,a);
- }
- b.subTo(d,b);
- } else {
- v.subTo(u,v);
- if (ac) {
- c.subTo(a,c);
- }
- d.subTo(b,d);
- }
- }
- if (v.compareTo(BigInteger.ONE) != 0) {
- return BigInteger.ZERO;
- }
- if (d.compareTo(m) >= 0) {
- return d.subtract(m);
- }
- if (d.sigNum()<0) {
- d.addTo(m,d);
- } else {
- return d;
- }
- if (d.sigNum()<0) {
- return d.add(m);
- } else {
- return d;
- }
- }
- /**
- *
- * @param t
- * @return primality with certainty >= 1-.5^t
- *
- */
- public function isProbablePrime(t:int):Boolean {
- var i:int;
- var x:BigInteger = abs();
- if (x.t == 1 && x.a[0]<=lowprimes[lowprimes.length-1]) {
- for (i=0;i<lowprimes.length;++i) {
- if (x[0]==lowprimes[i]) return true;
- }
- return false;
- }
- if (x.isEven()) return false;
- i = 1;
- while (i<lowprimes.length) {
- var m:int = lowprimes[i];
- var j:int = i+1;
- while (j<lowprimes.length && m<lplim) {
- m *= lowprimes[j++];
- }
- m = x.modInt(m);
- while (i<j) {
- if (m%lowprimes[i++]==0) {
- return false;
- }
- }
- }
- return x.millerRabin(t);
- }
-
- /**
- *
- * @param t
- * @return true if probably prime (HAC 4.24, Miller-Rabin)
- *
- */
- protected function millerRabin(t:int):Boolean {
- var n1:BigInteger = subtract(BigInteger.ONE);
- var k:int = n1.getLowestSetBit();
- if (k<=0) {
- return false;
- }
- var r:BigInteger = n1.shiftRight(k);
- t = (t+1)>>1;
- if (t>lowprimes.length) {
- t = lowprimes.length;
- }
- var a:BigInteger = new BigInteger;
- for (var i:int=0;i<t;++i) {
- a.fromInt(lowprimes[i]);
- var y:BigInteger = a.modPow(r, this);
- if (y.compareTo(BigInteger.ONE)!=0 && y.compareTo(n1)!=0) {
- var j:int = 1;
- while (j++<k && y.compareTo(n1)!=0) {
- y = y.modPowInt(2, this);
- if (y.compareTo(BigInteger.ONE)==0) {
- return false;
- }
- }
- if (y.compareTo(n1)!=0) {
- return false;
- }
- }
- }
- return true;
- }
- /**
- * Tweak our BigInteger until it looks prime enough
- *
- * @param bits
- * @param t
- *
- */
- public function primify(bits:int, t:int):void {
- if (!testBit(bits-1)) { // force MSB set
- bitwiseTo(BigInteger.ONE.shiftLeft(bits-1), op_or, this);
- }
- if (isEven()) {
- dAddOffset(1,0); // force odd
- }
- while (!isProbablePrime(t)) {
- dAddOffset(2,0);
- while(bitLength()>bits) subTo(BigInteger.ONE.shiftLeft(bits-1),this);
- }
- }
- }
- }