chacha

#include <iostream>
#include <fstream>
#include <vector>
#include <cstring>
#include <cstdio>
#include <string>
#include <iomanip>
#include <stdint.h>
using namespace std;

uint32_t rotl(uint32_t n, size_t shift)
{return (n << shift) | (n >> (32 - shift));}

vector<uint32_t> tmp(4);

uint8_t read(istream& s)
{
    return s.get();
}

void quarterround(vector<uint32_t> &a){
    a[0]=a[0]+a[1];
    a[3]^=a[0];
    a[3]=rotl(a[3],16);
    a[2]=a[2]+a[3];
    a[1]^=a[2];
    a[1]=rotl(a[1],12);
    a[0]=a[0]+a[1];
    a[3]^=a[0];
    a[3]=rotl(a[3],8);
    a[2]=a[2]+a[3];
    a[1]^=a[2];
    a[1]=rotl(a[1],7);
}

void columnround(vector<uint32_t> &a){
    tmp[0]=a[0];tmp[1]=a[4];tmp[2]=a[8];tmp[3]=a[12];
    quarterround(tmp);
    a[0]=tmp[0];a[4]=tmp[1];a[8]=tmp[2];a[12]=tmp[3];
    tmp[0]=a[1];tmp[1]=a[5];tmp[2]=a[9];tmp[3]=a[13];
    quarterround(tmp);
    a[1]=tmp[0];a[5]=tmp[1];a[9]=tmp[2];a[13]=tmp[3];
    tmp[0]=a[2];tmp[1]=a[6];tmp[2]=a[10];tmp[3]=a[14];
    quarterround(tmp);
    a[2]=tmp[0];a[6]=tmp[1];a[10]=tmp[2];a[14]=tmp[3];
    tmp[0]=a[3];tmp[1]=a[7];tmp[2]=a[11];tmp[3]=a[15];
    quarterround(tmp);
    a[3]=tmp[0];a[7]=tmp[1];a[11]=tmp[2];a[15]=tmp[3];
}

void diagonalround(vector<uint32_t> &a){
    tmp[0]=a[0];tmp[1]=a[5];tmp[2]=a[10];tmp[3]=a[15];
    quarterround(tmp);
    a[0]=tmp[0];a[5]=tmp[1];a[10]=tmp[2];a[15]=tmp[3];
    tmp[0]=a[1];tmp[1]=a[6];tmp[2]=a[11];tmp[3]=a[12];
    quarterround(tmp);
    a[1]=tmp[0];a[6]=tmp[1];a[11]=tmp[2];a[12]=tmp[3];
    tmp[0]=a[2];tmp[1]=a[7];tmp[2]=a[8];tmp[3]=a[13];
    quarterround(tmp);
    a[2]=tmp[0];a[7]=tmp[1];a[8]=tmp[2];a[13]=tmp[3];
    tmp[0]=a[3];tmp[1]=a[4];tmp[2]=a[9];tmp[3]=a[14];
    quarterround(tmp);
    a[3]=tmp[0];a[4]=tmp[1];a[9]=tmp[2];a[14]=tmp[3];
}

void doubleround(vector<uint32_t> &a){
    columnround(a);
    diagonalround(a);
}

uint32_t u32(uint8_t b[4]){
  uint32_t u;
  u = b[0];
  u = (u  << 8) + b[1];
  u = (u  << 8) + b[2];
  u = (u  << 8) + b[3];
  return u;
}

int main(){
    int i1;
    uint8_t b[4];
    vector<uint8_t> key(32);
    vector<uint8_t> nonce(12);
    vector<uint32_t> x(16);
    vector<uint32_t> y(16);
    vector<uint8_t> keystream(114);
    vector<uint8_t> input(114);
    vector<uint8_t> output(114);
    uint32_t k = 0x00000001;
    ifstream in("C:\\Users\\lesha\\key.bin",ios::binary|ios::in);
    if (in.is_open())
    {
        for (int i=0; i<32; i++){
            in.read((char*)&key[i],sizeof(key[i]));
        }
    }
    for (int i=0; i<32; i++){
        cout<<hex<<(int)key[i]<<endl;
    }
    in.close();
    ifstream in1("C:\\Users\\lesha\\nonce.bin",ios::binary|ios::in);
    if (in1.is_open())
    {
        for (int i=0; i<12; i++){
            in1.read((char*)&nonce[i],sizeof(nonce[i]));
        }
    }
    in1.close();

    x[0]=0x61707865; x[1]=0x3320646e; x[2]=0x79622d32; x[3]=0x6b206574; x[13]=0x09000000; x[14]=0x4a000000; x[15]=0x00000000;
    for (int i=0; i<8; i++){
        for (int j=0; j<4; j++){
            b[j]=key[4*i+3-j];
        }
        x[4+i]=u32(b);
    }
    for (int i=0; i<3; i++){
        for (int j=0; j<4; j++){
            b[j]=nonce[4*i+3-j];
        }
        x[13+i]=u32(b);
    }
    for (int i=0; i<16; i++){
        y[i]=x[i];
    }
    ifstream in2;
    in2.open("C:\\Users\\lesha\\input.bin",ios::binary|ios::in);
    if (in2.is_open()){
        for (int i=0; i<114; i++){
            in2.read((char*)&input[i],sizeof(input[i]));
        }
    }
    in2.close();
    ofstream out("C:\\Users\\lesha\\output.bin",ios::binary|ios::in);
    for (int i=0; i<114/64+1; i++){
        for (int j=0; j<16; j++){
            x[j]=y[j];
        }
        x[12]=k;
        for (int j=0; j<16; j++){
            y[j]=x[j];
        }
        for (int i=0; i<4; i++){
            for (int j=0; j<4; j++){
                cout<<hex<<(uint32_t)x[4*i+j]<<" ";
            }
            cout<<endl;
        }
        cout<<endl;
        for (int j=0; j<10; j++){
            doubleround(x);
        }
        for (int j=0; j<16; j++){
            x[j]+=y[j];
        }
        for (int i=0; i<4; i++){
            for (int j=0; j<4; j++){
                cout<<hex<<(uint32_t)x[4*i+j]<<" ";
            }
            cout<<endl;
        }
        cout<<endl;
        for (int j=0; j<64; j++){
            if(64*i+j<114){
                output[j+64*i] = input[j+64*i] ^ (uint8_t)(x[j/4]>>(j%4)*8);
                out.write((char*)&output[j+64*i], sizeof(output[j+64*i]));
            }
        }
        k++;
    }
    out.close();
    return 0;
}