All Downloads are FREE. Search and download functionalities are using the official Maven repository.

com.bouncycastle.crypto.digests.RIPEMD160Digest Maven / Gradle / Ivy

The newest version!
package com.bouncycastle.crypto.digests;


/**
 * implementation of RIPEMD see,
 * http://www.esat.kuleuven.ac.be/~bosselae/ripemd160.html
 */
public class RIPEMD160Digest
    extends GeneralDigest
{
    private static final int DIGEST_LENGTH = 20;

    private int H0, H1, H2, H3, H4; // IV's

    private int[] X = new int[16];
    private int xOff;

    /**
     * Standard constructor
     */
    public RIPEMD160Digest()
    {
        reset();
    }

    /**
     * Copy constructor.  This will copy the state of the provided
     * message digest.
     */
    public RIPEMD160Digest(RIPEMD160Digest t)
    {
        super(t);

        H0 = t.H0;
        H1 = t.H1;
        H2 = t.H2;
        H3 = t.H3;
        H4 = t.H4;

        System.arraycopy(t.X, 0, X, 0, t.X.length);
        xOff = t.xOff;
    }

    public String getAlgorithmName()
    {
        return "RIPEMD160";
    }

    public int getDigestSize()
    {
        return DIGEST_LENGTH;
    }

    protected void processWord(
        byte[] in,
        int inOff)
    {
        X[xOff++] = (in[inOff] & 0xff) | ((in[inOff + 1] & 0xff) << 8)
            | ((in[inOff + 2] & 0xff) << 16) | ((in[inOff + 3] & 0xff) << 24); 

        if (xOff == 16)
        {
            processBlock();
        }
    }

    protected void processLength(
        long bitLength)
    {
        if (xOff > 14)
        {
        processBlock();
        }

        X[14] = (int)(bitLength & 0xffffffff);
        X[15] = (int)(bitLength >>> 32);
    }

    private void unpackWord(
        int word,
        byte[] out,
        int outOff)
    {
        out[outOff]     = (byte)word;
        out[outOff + 1] = (byte)(word >>> 8);
        out[outOff + 2] = (byte)(word >>> 16);
        out[outOff + 3] = (byte)(word >>> 24);
    }

    public int doFinal(
        byte[] out,
        int outOff)
    {
        finish();

        unpackWord(H0, out, outOff);
        unpackWord(H1, out, outOff + 4);
        unpackWord(H2, out, outOff + 8);
        unpackWord(H3, out, outOff + 12);
        unpackWord(H4, out, outOff + 16);

        reset();

        return DIGEST_LENGTH;
    }

    /**
    * reset the chaining variables to the IV values.
    */
    public void reset()
    {
        super.reset();

        H0 = 0x67452301;
        H1 = 0xefcdab89;
        H2 = 0x98badcfe;
        H3 = 0x10325476;
        H4 = 0xc3d2e1f0;

        xOff = 0;

        for (int i = 0; i != X.length; i++)
        {
            X[i] = 0;
        }
    }

    /*
     * rotate int x left n bits.
     */
    private int RL(
        int x,
        int n)
    {
        return (x << n) | (x >>> (32 - n));
    }

    /*
     * f1,f2,f3,f4,f5 are the basic RIPEMD160 functions.
     */

    /*
     * rounds 0-15
     */
    private int f1(
        int x,
        int y,
        int z)
    {
        return x ^ y ^ z;
    }

    /*
     * rounds 16-31
     */
    private int f2(
        int x,
        int y,
        int z)
    {
        return (x & y) | (~x & z);
    }

    /*
     * rounds 32-47
     */
    private int f3(
        int x,
        int y,
        int z)
    {
        return (x | ~y) ^ z;
    }

    /*
     * rounds 48-63
     */
    private int f4(
        int x,
        int y,
        int z)
    {
        return (x & z) | (y & ~z);
    }

    /*
     * rounds 64-79
     */
    private int f5(
        int x,
        int y,
        int z)
    {
        return x ^ (y | ~z);
    }

    protected void processBlock()
    {
        int a, aa;
        int b, bb;
        int c, cc;
        int d, dd;
        int e, ee;

        a = aa = H0;
        b = bb = H1;
        c = cc = H2;
        d = dd = H3;
        e = ee = H4;

        //
        // Rounds 1 - 16
        //
        // left
        a = RL(a + f1(b,c,d) + X[ 0], 11) + e; c = RL(c, 10);
        e = RL(e + f1(a,b,c) + X[ 1], 14) + d; b = RL(b, 10);
        d = RL(d + f1(e,a,b) + X[ 2], 15) + c; a = RL(a, 10);
        c = RL(c + f1(d,e,a) + X[ 3], 12) + b; e = RL(e, 10);
        b = RL(b + f1(c,d,e) + X[ 4],  5) + a; d = RL(d, 10);
        a = RL(a + f1(b,c,d) + X[ 5],  8) + e; c = RL(c, 10);
        e = RL(e + f1(a,b,c) + X[ 6],  7) + d; b = RL(b, 10);
        d = RL(d + f1(e,a,b) + X[ 7],  9) + c; a = RL(a, 10);
        c = RL(c + f1(d,e,a) + X[ 8], 11) + b; e = RL(e, 10);
        b = RL(b + f1(c,d,e) + X[ 9], 13) + a; d = RL(d, 10);
        a = RL(a + f1(b,c,d) + X[10], 14) + e; c = RL(c, 10);
        e = RL(e + f1(a,b,c) + X[11], 15) + d; b = RL(b, 10);
        d = RL(d + f1(e,a,b) + X[12],  6) + c; a = RL(a, 10);
        c = RL(c + f1(d,e,a) + X[13],  7) + b; e = RL(e, 10);
        b = RL(b + f1(c,d,e) + X[14],  9) + a; d = RL(d, 10);
        a = RL(a + f1(b,c,d) + X[15],  8) + e; c = RL(c, 10);

        // right
        aa = RL(aa + f5(bb,cc,dd) + X[ 5] + 0x50a28be6,  8) + ee; cc = RL(cc, 10);
        ee = RL(ee + f5(aa,bb,cc) + X[14] + 0x50a28be6,  9) + dd; bb = RL(bb, 10);
        dd = RL(dd + f5(ee,aa,bb) + X[ 7] + 0x50a28be6,  9) + cc; aa = RL(aa, 10);
        cc = RL(cc + f5(dd,ee,aa) + X[ 0] + 0x50a28be6, 11) + bb; ee = RL(ee, 10);
        bb = RL(bb + f5(cc,dd,ee) + X[ 9] + 0x50a28be6, 13) + aa; dd = RL(dd, 10);
        aa = RL(aa + f5(bb,cc,dd) + X[ 2] + 0x50a28be6, 15) + ee; cc = RL(cc, 10);
        ee = RL(ee + f5(aa,bb,cc) + X[11] + 0x50a28be6, 15) + dd; bb = RL(bb, 10);
        dd = RL(dd + f5(ee,aa,bb) + X[ 4] + 0x50a28be6,  5) + cc; aa = RL(aa, 10);
        cc = RL(cc + f5(dd,ee,aa) + X[13] + 0x50a28be6,  7) + bb; ee = RL(ee, 10);
        bb = RL(bb + f5(cc,dd,ee) + X[ 6] + 0x50a28be6,  7) + aa; dd = RL(dd, 10);
        aa = RL(aa + f5(bb,cc,dd) + X[15] + 0x50a28be6,  8) + ee; cc = RL(cc, 10);
        ee = RL(ee + f5(aa,bb,cc) + X[ 8] + 0x50a28be6, 11) + dd; bb = RL(bb, 10);
        dd = RL(dd + f5(ee,aa,bb) + X[ 1] + 0x50a28be6, 14) + cc; aa = RL(aa, 10);
        cc = RL(cc + f5(dd,ee,aa) + X[10] + 0x50a28be6, 14) + bb; ee = RL(ee, 10);
        bb = RL(bb + f5(cc,dd,ee) + X[ 3] + 0x50a28be6, 12) + aa; dd = RL(dd, 10);
        aa = RL(aa + f5(bb,cc,dd) + X[12] + 0x50a28be6,  6) + ee; cc = RL(cc, 10);

        //
        // Rounds 16-31
        //
        // left
        e = RL(e + f2(a,b,c) + X[ 7] + 0x5a827999,  7) + d; b = RL(b, 10);
        d = RL(d + f2(e,a,b) + X[ 4] + 0x5a827999,  6) + c; a = RL(a, 10);
        c = RL(c + f2(d,e,a) + X[13] + 0x5a827999,  8) + b; e = RL(e, 10);
        b = RL(b + f2(c,d,e) + X[ 1] + 0x5a827999, 13) + a; d = RL(d, 10);
        a = RL(a + f2(b,c,d) + X[10] + 0x5a827999, 11) + e; c = RL(c, 10);
        e = RL(e + f2(a,b,c) + X[ 6] + 0x5a827999,  9) + d; b = RL(b, 10);
        d = RL(d + f2(e,a,b) + X[15] + 0x5a827999,  7) + c; a = RL(a, 10);
        c = RL(c + f2(d,e,a) + X[ 3] + 0x5a827999, 15) + b; e = RL(e, 10);
        b = RL(b + f2(c,d,e) + X[12] + 0x5a827999,  7) + a; d = RL(d, 10);
        a = RL(a + f2(b,c,d) + X[ 0] + 0x5a827999, 12) + e; c = RL(c, 10);
        e = RL(e + f2(a,b,c) + X[ 9] + 0x5a827999, 15) + d; b = RL(b, 10);
        d = RL(d + f2(e,a,b) + X[ 5] + 0x5a827999,  9) + c; a = RL(a, 10);
        c = RL(c + f2(d,e,a) + X[ 2] + 0x5a827999, 11) + b; e = RL(e, 10);
        b = RL(b + f2(c,d,e) + X[14] + 0x5a827999,  7) + a; d = RL(d, 10);
        a = RL(a + f2(b,c,d) + X[11] + 0x5a827999, 13) + e; c = RL(c, 10);
        e = RL(e + f2(a,b,c) + X[ 8] + 0x5a827999, 12) + d; b = RL(b, 10);

        // right
        ee = RL(ee + f4(aa,bb,cc) + X[ 6] + 0x5c4dd124,  9) + dd; bb = RL(bb, 10);
        dd = RL(dd + f4(ee,aa,bb) + X[11] + 0x5c4dd124, 13) + cc; aa = RL(aa, 10);
        cc = RL(cc + f4(dd,ee,aa) + X[ 3] + 0x5c4dd124, 15) + bb; ee = RL(ee, 10);
        bb = RL(bb + f4(cc,dd,ee) + X[ 7] + 0x5c4dd124,  7) + aa; dd = RL(dd, 10);
        aa = RL(aa + f4(bb,cc,dd) + X[ 0] + 0x5c4dd124, 12) + ee; cc = RL(cc, 10);
        ee = RL(ee + f4(aa,bb,cc) + X[13] + 0x5c4dd124,  8) + dd; bb = RL(bb, 10);
        dd = RL(dd + f4(ee,aa,bb) + X[ 5] + 0x5c4dd124,  9) + cc; aa = RL(aa, 10);
        cc = RL(cc + f4(dd,ee,aa) + X[10] + 0x5c4dd124, 11) + bb; ee = RL(ee, 10);
        bb = RL(bb + f4(cc,dd,ee) + X[14] + 0x5c4dd124,  7) + aa; dd = RL(dd, 10);
        aa = RL(aa + f4(bb,cc,dd) + X[15] + 0x5c4dd124,  7) + ee; cc = RL(cc, 10);
        ee = RL(ee + f4(aa,bb,cc) + X[ 8] + 0x5c4dd124, 12) + dd; bb = RL(bb, 10);
        dd = RL(dd + f4(ee,aa,bb) + X[12] + 0x5c4dd124,  7) + cc; aa = RL(aa, 10);
        cc = RL(cc + f4(dd,ee,aa) + X[ 4] + 0x5c4dd124,  6) + bb; ee = RL(ee, 10);
        bb = RL(bb + f4(cc,dd,ee) + X[ 9] + 0x5c4dd124, 15) + aa; dd = RL(dd, 10);
        aa = RL(aa + f4(bb,cc,dd) + X[ 1] + 0x5c4dd124, 13) + ee; cc = RL(cc, 10);
        ee = RL(ee + f4(aa,bb,cc) + X[ 2] + 0x5c4dd124, 11) + dd; bb = RL(bb, 10);

        //
        // Rounds 32-47
        //
        // left
        d = RL(d + f3(e,a,b) + X[ 3] + 0x6ed9eba1, 11) + c; a = RL(a, 10);
        c = RL(c + f3(d,e,a) + X[10] + 0x6ed9eba1, 13) + b; e = RL(e, 10);
        b = RL(b + f3(c,d,e) + X[14] + 0x6ed9eba1,  6) + a; d = RL(d, 10);
        a = RL(a + f3(b,c,d) + X[ 4] + 0x6ed9eba1,  7) + e; c = RL(c, 10);
        e = RL(e + f3(a,b,c) + X[ 9] + 0x6ed9eba1, 14) + d; b = RL(b, 10);
        d = RL(d + f3(e,a,b) + X[15] + 0x6ed9eba1,  9) + c; a = RL(a, 10);
        c = RL(c + f3(d,e,a) + X[ 8] + 0x6ed9eba1, 13) + b; e = RL(e, 10);
        b = RL(b + f3(c,d,e) + X[ 1] + 0x6ed9eba1, 15) + a; d = RL(d, 10);
        a = RL(a + f3(b,c,d) + X[ 2] + 0x6ed9eba1, 14) + e; c = RL(c, 10);
        e = RL(e + f3(a,b,c) + X[ 7] + 0x6ed9eba1,  8) + d; b = RL(b, 10);
        d = RL(d + f3(e,a,b) + X[ 0] + 0x6ed9eba1, 13) + c; a = RL(a, 10);
        c = RL(c + f3(d,e,a) + X[ 6] + 0x6ed9eba1,  6) + b; e = RL(e, 10);
        b = RL(b + f3(c,d,e) + X[13] + 0x6ed9eba1,  5) + a; d = RL(d, 10);
        a = RL(a + f3(b,c,d) + X[11] + 0x6ed9eba1, 12) + e; c = RL(c, 10);
        e = RL(e + f3(a,b,c) + X[ 5] + 0x6ed9eba1,  7) + d; b = RL(b, 10);
        d = RL(d + f3(e,a,b) + X[12] + 0x6ed9eba1,  5) + c; a = RL(a, 10);

        // right
        dd = RL(dd + f3(ee,aa,bb) + X[15] + 0x6d703ef3,  9) + cc; aa = RL(aa, 10);
        cc = RL(cc + f3(dd,ee,aa) + X[ 5] + 0x6d703ef3,  7) + bb; ee = RL(ee, 10);
        bb = RL(bb + f3(cc,dd,ee) + X[ 1] + 0x6d703ef3, 15) + aa; dd = RL(dd, 10);
        aa = RL(aa + f3(bb,cc,dd) + X[ 3] + 0x6d703ef3, 11) + ee; cc = RL(cc, 10);
        ee = RL(ee + f3(aa,bb,cc) + X[ 7] + 0x6d703ef3,  8) + dd; bb = RL(bb, 10);
        dd = RL(dd + f3(ee,aa,bb) + X[14] + 0x6d703ef3,  6) + cc; aa = RL(aa, 10);
        cc = RL(cc + f3(dd,ee,aa) + X[ 6] + 0x6d703ef3,  6) + bb; ee = RL(ee, 10);
        bb = RL(bb + f3(cc,dd,ee) + X[ 9] + 0x6d703ef3, 14) + aa; dd = RL(dd, 10);
        aa = RL(aa + f3(bb,cc,dd) + X[11] + 0x6d703ef3, 12) + ee; cc = RL(cc, 10);
        ee = RL(ee + f3(aa,bb,cc) + X[ 8] + 0x6d703ef3, 13) + dd; bb = RL(bb, 10);
        dd = RL(dd + f3(ee,aa,bb) + X[12] + 0x6d703ef3,  5) + cc; aa = RL(aa, 10);
        cc = RL(cc + f3(dd,ee,aa) + X[ 2] + 0x6d703ef3, 14) + bb; ee = RL(ee, 10);
        bb = RL(bb + f3(cc,dd,ee) + X[10] + 0x6d703ef3, 13) + aa; dd = RL(dd, 10);
        aa = RL(aa + f3(bb,cc,dd) + X[ 0] + 0x6d703ef3, 13) + ee; cc = RL(cc, 10);
        ee = RL(ee + f3(aa,bb,cc) + X[ 4] + 0x6d703ef3,  7) + dd; bb = RL(bb, 10);
        dd = RL(dd + f3(ee,aa,bb) + X[13] + 0x6d703ef3,  5) + cc; aa = RL(aa, 10);

        //
        // Rounds 48-63
        //
        // left
        c = RL(c + f4(d,e,a) + X[ 1] + 0x8f1bbcdc, 11) + b; e = RL(e, 10);
        b = RL(b + f4(c,d,e) + X[ 9] + 0x8f1bbcdc, 12) + a; d = RL(d, 10);
        a = RL(a + f4(b,c,d) + X[11] + 0x8f1bbcdc, 14) + e; c = RL(c, 10);
        e = RL(e + f4(a,b,c) + X[10] + 0x8f1bbcdc, 15) + d; b = RL(b, 10);
        d = RL(d + f4(e,a,b) + X[ 0] + 0x8f1bbcdc, 14) + c; a = RL(a, 10);
        c = RL(c + f4(d,e,a) + X[ 8] + 0x8f1bbcdc, 15) + b; e = RL(e, 10);
        b = RL(b + f4(c,d,e) + X[12] + 0x8f1bbcdc,  9) + a; d = RL(d, 10);
        a = RL(a + f4(b,c,d) + X[ 4] + 0x8f1bbcdc,  8) + e; c = RL(c, 10);
        e = RL(e + f4(a,b,c) + X[13] + 0x8f1bbcdc,  9) + d; b = RL(b, 10);
        d = RL(d + f4(e,a,b) + X[ 3] + 0x8f1bbcdc, 14) + c; a = RL(a, 10);
        c = RL(c + f4(d,e,a) + X[ 7] + 0x8f1bbcdc,  5) + b; e = RL(e, 10);
        b = RL(b + f4(c,d,e) + X[15] + 0x8f1bbcdc,  6) + a; d = RL(d, 10);
        a = RL(a + f4(b,c,d) + X[14] + 0x8f1bbcdc,  8) + e; c = RL(c, 10);
        e = RL(e + f4(a,b,c) + X[ 5] + 0x8f1bbcdc,  6) + d; b = RL(b, 10);
        d = RL(d + f4(e,a,b) + X[ 6] + 0x8f1bbcdc,  5) + c; a = RL(a, 10);
        c = RL(c + f4(d,e,a) + X[ 2] + 0x8f1bbcdc, 12) + b; e = RL(e, 10);

        // right
        cc = RL(cc + f2(dd,ee,aa) + X[ 8] + 0x7a6d76e9, 15) + bb; ee = RL(ee, 10);
        bb = RL(bb + f2(cc,dd,ee) + X[ 6] + 0x7a6d76e9,  5) + aa; dd = RL(dd, 10);
        aa = RL(aa + f2(bb,cc,dd) + X[ 4] + 0x7a6d76e9,  8) + ee; cc = RL(cc, 10);
        ee = RL(ee + f2(aa,bb,cc) + X[ 1] + 0x7a6d76e9, 11) + dd; bb = RL(bb, 10);
        dd = RL(dd + f2(ee,aa,bb) + X[ 3] + 0x7a6d76e9, 14) + cc; aa = RL(aa, 10);
        cc = RL(cc + f2(dd,ee,aa) + X[11] + 0x7a6d76e9, 14) + bb; ee = RL(ee, 10);
        bb = RL(bb + f2(cc,dd,ee) + X[15] + 0x7a6d76e9,  6) + aa; dd = RL(dd, 10);
        aa = RL(aa + f2(bb,cc,dd) + X[ 0] + 0x7a6d76e9, 14) + ee; cc = RL(cc, 10);
        ee = RL(ee + f2(aa,bb,cc) + X[ 5] + 0x7a6d76e9,  6) + dd; bb = RL(bb, 10);
        dd = RL(dd + f2(ee,aa,bb) + X[12] + 0x7a6d76e9,  9) + cc; aa = RL(aa, 10);
        cc = RL(cc + f2(dd,ee,aa) + X[ 2] + 0x7a6d76e9, 12) + bb; ee = RL(ee, 10);
        bb = RL(bb + f2(cc,dd,ee) + X[13] + 0x7a6d76e9,  9) + aa; dd = RL(dd, 10);
        aa = RL(aa + f2(bb,cc,dd) + X[ 9] + 0x7a6d76e9, 12) + ee; cc = RL(cc, 10);
        ee = RL(ee + f2(aa,bb,cc) + X[ 7] + 0x7a6d76e9,  5) + dd; bb = RL(bb, 10);
        dd = RL(dd + f2(ee,aa,bb) + X[10] + 0x7a6d76e9, 15) + cc; aa = RL(aa, 10);
        cc = RL(cc + f2(dd,ee,aa) + X[14] + 0x7a6d76e9,  8) + bb; ee = RL(ee, 10);

        //
        // Rounds 64-79
        //
        // left
        b = RL(b + f5(c,d,e) + X[ 4] + 0xa953fd4e,  9) + a; d = RL(d, 10);
        a = RL(a + f5(b,c,d) + X[ 0] + 0xa953fd4e, 15) + e; c = RL(c, 10);
        e = RL(e + f5(a,b,c) + X[ 5] + 0xa953fd4e,  5) + d; b = RL(b, 10);
        d = RL(d + f5(e,a,b) + X[ 9] + 0xa953fd4e, 11) + c; a = RL(a, 10);
        c = RL(c + f5(d,e,a) + X[ 7] + 0xa953fd4e,  6) + b; e = RL(e, 10);
        b = RL(b + f5(c,d,e) + X[12] + 0xa953fd4e,  8) + a; d = RL(d, 10);
        a = RL(a + f5(b,c,d) + X[ 2] + 0xa953fd4e, 13) + e; c = RL(c, 10);
        e = RL(e + f5(a,b,c) + X[10] + 0xa953fd4e, 12) + d; b = RL(b, 10);
        d = RL(d + f5(e,a,b) + X[14] + 0xa953fd4e,  5) + c; a = RL(a, 10);
        c = RL(c + f5(d,e,a) + X[ 1] + 0xa953fd4e, 12) + b; e = RL(e, 10);
        b = RL(b + f5(c,d,e) + X[ 3] + 0xa953fd4e, 13) + a; d = RL(d, 10);
        a = RL(a + f5(b,c,d) + X[ 8] + 0xa953fd4e, 14) + e; c = RL(c, 10);
        e = RL(e + f5(a,b,c) + X[11] + 0xa953fd4e, 11) + d; b = RL(b, 10);
        d = RL(d + f5(e,a,b) + X[ 6] + 0xa953fd4e,  8) + c; a = RL(a, 10);
        c = RL(c + f5(d,e,a) + X[15] + 0xa953fd4e,  5) + b; e = RL(e, 10);
        b = RL(b + f5(c,d,e) + X[13] + 0xa953fd4e,  6) + a; d = RL(d, 10);

        // right
        bb = RL(bb + f1(cc,dd,ee) + X[12],  8) + aa; dd = RL(dd, 10);
        aa = RL(aa + f1(bb,cc,dd) + X[15],  5) + ee; cc = RL(cc, 10);
        ee = RL(ee + f1(aa,bb,cc) + X[10], 12) + dd; bb = RL(bb, 10);
        dd = RL(dd + f1(ee,aa,bb) + X[ 4],  9) + cc; aa = RL(aa, 10);
        cc = RL(cc + f1(dd,ee,aa) + X[ 1], 12) + bb; ee = RL(ee, 10);
        bb = RL(bb + f1(cc,dd,ee) + X[ 5],  5) + aa; dd = RL(dd, 10);
        aa = RL(aa + f1(bb,cc,dd) + X[ 8], 14) + ee; cc = RL(cc, 10);
        ee = RL(ee + f1(aa,bb,cc) + X[ 7],  6) + dd; bb = RL(bb, 10);
        dd = RL(dd + f1(ee,aa,bb) + X[ 6],  8) + cc; aa = RL(aa, 10);
        cc = RL(cc + f1(dd,ee,aa) + X[ 2], 13) + bb; ee = RL(ee, 10);
        bb = RL(bb + f1(cc,dd,ee) + X[13],  6) + aa; dd = RL(dd, 10);
        aa = RL(aa + f1(bb,cc,dd) + X[14],  5) + ee; cc = RL(cc, 10);
        ee = RL(ee + f1(aa,bb,cc) + X[ 0], 15) + dd; bb = RL(bb, 10);
        dd = RL(dd + f1(ee,aa,bb) + X[ 3], 13) + cc; aa = RL(aa, 10);
        cc = RL(cc + f1(dd,ee,aa) + X[ 9], 11) + bb; ee = RL(ee, 10);
        bb = RL(bb + f1(cc,dd,ee) + X[11], 11) + aa; dd = RL(dd, 10);

        dd += c + H1;
        H1 = H2 + d + ee;
        H2 = H3 + e + aa;
        H3 = H4 + a + bb;
        H4 = H0 + b + cc;
        H0 = dd;

        //
        // reset the offset and clean out the word buffer.
        //
        xOff = 0;
        for (int i = 0; i != X.length; i++)
        {
            X[i] = 0;
        }
    }
}




© 2015 - 2024 Weber Informatics LLC | Privacy Policy