hdu 1208 Pascal's Travels (DP记忆化搜索)

Pascal's Travels

Time Limit: 2000/1000 MS (Java/Others)    Memory Limit: 65536/32768 K (Java/Others)
Total Submission(s): 1115    Accepted Submission(s): 488


Problem Description
An n x n game board is populated with integers, one nonnegative integer per square. The goal is to travel along any legitimate path from the upper left corner to the lower right corner of the board. The integer in any one square dictates how large a step away from that location must be. If the step size would advance travel off the game board, then a step in that particular direction is forbidden. All steps must be either to the right or toward the bottom. Note that a 0 is a dead end which prevents any further progress.


Consider the 4 x 4 board shown in Figure 1, where the solid circle identifies the start position and the dashed circle identifies the target. Figure 2 shows the three paths from the start to the target, with the irrelevant numbers in each removed.


Figure 1


Figure 2


 

Input
The input contains data for one to thirty boards, followed by a final line containing only the integer -1. The data for a board starts with a line containing a single positive integer n, 4 <= n <= 34, which is the number of rows in this board. This is followed by n rows of data. Each row contains n single digits, 0-9, with no spaces between them.


 

Output
The output consists of one line for each board, containing a single integer, which is the number of paths from the upper left corner to the lower right corner. There will be fewer than 2^63 paths for any board.


 

Sample Input
  
4 2331 1213 1231 3110 4 3332 1213 1232 2120 5 11101 01111 11111 11101 11101 -1


 

Sample Output
  
3 0 7
Hint
Hint
Brute force methods examining every path will likely exceed the allotted time limit. 64-bit integer values are available as "__int64" values using the Visual C/C++ or "long long" values using GNU C/C++ or "int64" values using Free Pascal compilers.


 

Source


 

Recommend
mcqsmall

思路:没啥好说的,目标点记为1,从开始点开始,走右和下,和加起来即可,具体代码一看便知。

#include<iostream>
#include<cstring>
using namespace std;
const int mm=37;
long long dp[mm][mm];
char f[mm][mm];
bool vis[mm][mm];
int m;
long long DP(int x,int y)
{
  if(x<0||x>=m||y<0||y>=m)return 0;
  if(vis[x][y])return dp[x][y];
  if(f[x][y]=='0')return 0;
  dp[x][y]=DP(x+f[x][y]-'0',y)+DP(x,y+f[x][y]-'0');
  vis[x][y]=1;
  return dp[x][y];
}
int main()
{
  while(cin>>m)
  { if(m==-1)break;
   memset(dp,0,sizeof(dp));
   memset(vis,0,sizeof(vis));
    for(int i=0;i<m;i++)
      {
         cin>>f[i];
      }
      dp[m-1][m-1]=1;vis[m-1][m-1]=1;
    cout<<DP(0,0)<<"\n";
  }
}


 

 

 

### RK3588平台NPU调用方法 #### 创建和初始化NPU环境 为了在RK3588平台上成功调用NPU进行神经网络推理或加速,首先需要确保设备已正确配置并加载了相应的驱动程序。Rockchip的官方固件通常已经预装了RKNPU驱动[^3]。 一旦确认硬件准备就绪,可以通过以下方式创建和初始化NPU环境: ```cpp #include "rknn_api.h" // 初始化模型路径和其他参数 const char* model_path = "./model.rknn"; int ret; rknn_context ctx; ret = rknn_init(&ctx, model_path, 0, 0, NULL); if (ret < 0) { printf("Failed to initialize rknn context\n"); } ``` 这段代码展示了如何使用`rknn_api.h`库来初始化一个RKNN上下文对象,这一步骤对于后续的操作至关重要[^2]。 #### 加载和编译模型 接下来,在实际运行之前还需要加载预先训练好的神经网络模型文件(通常是`.rknn`格式)。此过程涉及读取模型二进制数据,并将其传递给RKNN API以便内部处理和优化。 ```cpp // 假设模型已经被转换成 .rknn 文件格式 char *model_data; // 模型的数据指针 size_t model_size; // 模型大小 FILE *fp = fopen(model_path, "rb+"); fseek(fp, 0L, SEEK_END); model_size = ftell(fp); rewind(fp); model_data = (char *)malloc(sizeof(char)*model_size); fread(model_data, sizeof(unsigned char), model_size, fp); fclose(fp); // 将模型数据传入RKNN API ret = rknn_load_rknn(ctx, &model_data, &model_size); free(model_data); if(ret != 0){ printf("Load Model Failed!\n"); } else{ printf("Model Loaded Successfully.\n"); } ``` 这里说明了从磁盘读取模型文件的具体操作流程,并通过API函数将这些信息提交给了底层框架去解析和设置好用于推断所需的资源[^1]。 #### 执行前向传播计算 当一切准备工作完成后就可以开始真正的预测工作——即让NPU执行一次完整的前向传播运算。这个阶段主要是构建输入张量、启动异步任务以及收集输出结果。 ```cpp float input_tensor[INPUT_SIZE]; // 输入特征图数组 float output_tensors[MAX_OUTPUTS][OUTPUT_SIZE]; // 输出特征图数组 struct rknn_input inputs[] = {{input_tensor}}; struct rknn_output outputs[MAX_OUTPUTS]; for(int i=0;i<NUM_ITERATIONS;++i){ memset(inputs, 0 ,sizeof(struct rknn_input)); memcpy(input_tensor, inputData[i], INPUT_SIZE*sizeof(float)); // 启动推理任务 ret = rknn_run(ctx, nullptr); if(ret!=0){ printf("Inference failed at iteration %d", i); break; } // 获取输出结果 for(size_t j=0;j<num_outputs;++j){ struct rknn_output& out = outputs[j]; size_t bufSize = OUTPUT_SIZE * sizeof(float); void* buffer = malloc(bufSize); ret = rknn_get_output(ctx, j, &out.datatype, &buffer, &bufSize, false); if(!ret && buffer){ memcpy(output_tensors[j], buffer, bufSize); free(buffer); } } } printf("All iterations completed successfully."); ``` 上述片段体现了典型的基于RKNN SDK的应用场景:先准备好待测样本作为输入;接着触发内核中的计算逻辑;最后获取到经过变换后的响应值供下一步分析所用[^4]。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值