Nastya received a gift on New Year — a magic wardrobe. It is magic because in the end of each month the number of dresses in it doubles (i.e. the number of dresses becomes twice as large as it is in the beginning of the month).

Unfortunately, right after the doubling the wardrobe eats one of the dresses (if any) with the 50% probability. It happens every month except the last one in the year.

Nastya owns x dresses now, so she became interested in the expected number of dresses she will have in one year. Nastya lives in Byteland, so the year lasts for k + 1 months.

Nastya is really busy, so she wants you to solve this problem. You are the programmer, after all. Also, you should find the answer modulo 109 + 7, because it is easy to see that it is always integer.

Input

The only line contains two integers x and k (0 ≤ x, k ≤ 1018), where x is the initial number of dresses and k + 1 is the number of months in a year in Byteland.

Output

In the only line print a single integer — the expected number of dresses Nastya will own one year later modulo 109 + 7.

Examples

Input

2 0

Output

4

Input

2 1

Output

7

Input

3 2

Output

21

Note

In the first example a year consists on only one month, so the wardrobe does not eat dresses at all.

In the second example after the first month there are 3 dresses with 50% probability and 4 dresses with 50% probability. Thus, in the end of the year there are 6 dresses with 50% probability and 8 dresses with 50% probability. This way the answer for this test is (6 + 8) / 2 = 7.

这个题画个图就能看出来,如果不考虑最后一天则,前面是个连续的序列。那么最后要求和取平均的过程,换成等差数列求和再取平均。然后化简完就是(2∗a−1)2b+1(2*a-1)2^{b}+1(2∗a−1)2b+1,害怕快速幂超时,可以十进制快速幂,也可以欧拉降幂。

#include <bits/stdc++.h>
using namespace std;
template <typename t>
void read(t &x)
{
char ch = getchar();
x = 0;
t f = 1;
while (ch < '0' || ch > '9')
f = (ch == '-' ? -1 : f), ch = getchar();
while (ch >= '0' && ch <= '9')
x = x * 10 + ch - '0', ch = getchar();
x *= f;
} #define wi(n) printf("%d ", n)
#define wl(n) printf("%lld ", n)
#define rep(m, n, i) for (int i = m; i < n; ++i)
#define rrep(m, n, i) for (int i = m; i > n; --i)
#define P puts(" ")
typedef long long ll;
#define MOD 1000000007
#define mp(a, b) make_pair(a, b)
#define N 10005
#define fil(a, n) rep(0, n, i) read(a[i])
//---------------https://lunatic.blog.csdn.net/-------------------//
const ll phi = 1000000006; //1e9+7的欧拉函数
ll fast_pow(ll a, ll b, ll p)
{
ll ret = 1;
for (; b; b >>= 1, a = a * a % p)
if (b & 1)
ret = ret * a % p;
return ret;
}
int main()
{
ll a, b,c;
read(a), read(b);
if (b >= phi)
b = b % phi + phi; //欧拉降幂
ll s1 = fast_pow(2, b, MOD);
cout<<((((2*a)%MOD-1)*s1)+1+MOD)%MOD<<endl;
}

CF思维联系– CodeForces -CodeForces - 992C Nastya and a Wardrobe(欧拉降幂+快速幂)的更多相关文章

  1. CodeForces 992C Nastya and a Wardrobe(规律、快速幂)

    http://codeforces.com/problemset/problem/992/C 题意: 给你两个数x,k,k代表有k+1个月,x每个月可以增长一倍,增长后的下一个月开始时x有50%几率减 ...

  2. Codeforces Round #536 (Div. 2) F 矩阵快速幂 + bsgs(新坑) + exgcd(新坑) + 欧拉降幂

    https://codeforces.com/contest/1106/problem/F 题意 数列公式为\(f_i=(f^{b_1}_{i-1}*f^{b_2}_{i-2}*...*f^{b_k} ...

  3. CodeForces - 906D Power Tower(欧拉降幂定理)

    Power Tower CodeForces - 906D 题目大意:有N个数字,然后给你q个区间,要你求每一个区间中所有的数字从左到右依次垒起来的次方的幂对m取模之后的数字是多少. 用到一个新知识, ...

  4. Codeforces Round #454 (Div. 1) CodeForces 906D Power Tower (欧拉降幂)

    题目链接:http://codeforces.com/contest/906/problem/D 题目大意:给定n个整数w[1],w[2],……,w[n],和一个数m,然后有q个询问,每个询问给出一个 ...

  5. Codeforces Round #257 (Div. 2) B. Jzzhu and Sequences (矩阵快速幂)

    题目链接:http://codeforces.com/problemset/problem/450/B 题意很好懂,矩阵快速幂模版题. /* | 1, -1 | | fn | | 1, 0 | | f ...

  6. codeforces E. Okabe and El Psy Kongroo(dp+矩阵快速幂)

    题目链接:http://codeforces.com/contest/821/problem/E 题意:我们现在位于(0,0)处,目标是走到(K,0)处.每一次我们都可以从(x,y)走到(x+1,y- ...

  7. CodeForces 906D (欧拉降幂)

    Power Tower •题意 求$w_{l}^{w_{l+1}^{w_{l+2}^{w_{l+3}^{w_{l+4}^{w_{l+5}^{...^{w_{r}}}}}}}}$ 对m取模的值 •思路 ...

  8. Codeforces 992C Nastya and a Wardrobe (思维)

    <题目链接> 题目大意: 你开始有X个裙子 你有K+1次增长机会 前K次会100%的增长一倍 但是增长后有50%的机会会减少一个 给你X,K(0<=X,K<=1e18), 问你 ...

  9. Codeforces 785D - Anton and School - 2 - [范德蒙德恒等式][快速幂+逆元]

    题目链接:https://codeforces.com/problemset/problem/785/D 题解: 首先很好想的,如果我们预处理出每个 "(" 的左边还有 $x$ 个 ...

随机推荐

  1. 【高并发】你知道吗?大家都在使用Redisson实现分布式锁了!!

    写在前面 忘记之前在哪个群里有朋友在问:有出分布式锁的文章吗-@冰河?我的回答是:这周会有,也是[高并发]专题的.想了想,还是先发一个如何使用Redisson实现分布式锁的文章吧?为啥?因为使用Red ...

  2. gcc/g++堆栈保护技术

      最近学习内存分布,通过gdb调试发现一些问题,栈空间变量地址应该是从高往低分布的,但是调试发现地址虽然是从高往低分布,但是变量地址的顺序是乱的,请教同事他说可能是gcc/g++默认启用了堆栈保护, ...

  3. "浮动按钮"组件:<fab> —— 快应用组件库H-UI

        <import name="fab" src="../Common/ui/h-ui/basic/c_fab"></import ...

  4. Java中的垃圾回收算法详解

    一.前言   前段时间大致看了一下<深入理解Java虚拟机>这本书,对相关的基础知识有了一定的了解,准备写一写JVM的系列博客,这是第二篇.这篇博客就来谈一谈JVM中使用到的垃圾回收算法. ...

  5. tcp长连接、短连接、连接池的思考

    在基于tcp的 rcp实现方式中,有如下几种选择: 1. 长连接:同步和异步方式. 同步方式下客户端所有请求共用同一连接,在获得连接后要对连接加锁,在读写结束后才解锁释放连接,性能低下,基本很少采用, ...

  6. python 性能测试

            python中使用的性能测试模块是memory_profiler , 我们使用它里面的profile这个装饰器即可测试出我们的代码的内存使用情况了.   如果没有安装 memory_p ...

  7. Springboot启动流程简单分析

    springboot启动的类为SpringApplication,执行构造函数初始化属性值后进入run方法: 然后返回ConfigurableApplicationContext(spring应用). ...

  8. 【python实现卷积神经网络】池化层实现

    代码来源:https://github.com/eriklindernoren/ML-From-Scratch 卷积神经网络中卷积层Conv2D(带stride.padding)的具体实现:https ...

  9. Linux U盘启动盘制作工具

    近期由于自己使用的ubuntu系统一直出问题,想做一下启动盘帮助恢复系统和故障检测,以前一直是用ultraiso来进行写盘的,但是发现制作了几次后,失败的机率很高,主要有以下几种情况: 1.引导有问题 ...

  10. Python 开发工具推荐

    对于开发工具,仁者见仁智者见智,关键是自己喜欢,用着顺手就好,不用刻意去追求别人用的是什么工具. 这里给大家主要推荐三款工具,分别是PyCharm.Sublime Text 3.VS Code,因为这 ...