CUDA中使用VisualProfiler进行优化遇到这样一个问题，该怎么解决？

Question

对于VisualProfiler的使用，新建NewSession，在File处选择已经编译好后出现的.exe文件，
然后点next，点finish后出现如图所示错误：

            问题描述为：The application being profiled returned a non-zero return code.
            百度翻译为：正在被配置的应用程序返回非零返回代码。
            通过网上搜寻，可能的解决办法是：
            1.内存没释放完全
            2.主函数末尾需要添加cudaThreadExit();
            3.主函数末尾添加cudaFree(0);
            如果上述解决办法，解决了，我是不会来这里提问的。
            问题就是没有解决！！！
            其中，我用一段HelloWorld.cu测试，可以使用VisualProfiler

 //使用CUDA的第一个独立编程。简要使用CUDA。
//功能：从主机向设备传入数组a、b，计算a、b数组的和，传入数组c中，再传回主机。

#include<stdio.h>
#include<stdlib.h>
#include<cuda_runtime.h>
#include "device_launch_parameters.h"

#define N 200000

__global__ void Add(int *d_a, int *d_b, int *d_c);

int main()
{
    int *a = (int*)malloc(sizeof(int)*N);
    int *b = (int*)malloc(sizeof(int)*N);
    int *c = (int*)malloc(sizeof(int)*N);
    //CPU计算
    for (int i = 0; i < N; i++)
    {
        a[i] = i;
        b[i] = i;
        c[i] = a[i] + b[i];
    }
    printf("CPU  :  c[N-1] = %d\n", c[N - 1]);
    for (int i = 0; i < N; i++)
    {
        c[i] = 0;
    }
    printf("c[N-1] = %d\n", c[N - 1]);
    int *d_a;
    int *d_b;
    int *d_c;
    cudaMalloc((void**)&d_a, sizeof(int)*N);
    cudaMalloc((void**)&d_b, sizeof(int)*N);
    cudaMalloc((void**)&d_c, sizeof(int)*N);
    dim3 block_add((N + 256 - 1) / 256, 1);
    dim3 thread_add(256, 1);

    cudaEvent_t start, stop;
    cudaEventCreate(&start);
    cudaEventCreate(&stop);
    cudaEventRecord(start, 0);
    cudaMemcpy(d_a, a, sizeof(int)*N, cudaMemcpyHostToDevice);
    cudaMemcpy(d_b, b, sizeof(int)*N, cudaMemcpyHostToDevice);
    Add << <block_add, thread_add >> > (d_a, d_b, d_c);
    cudaMemcpy(c, d_c, sizeof(int)*N, cudaMemcpyDeviceToHost);
    cudaEventRecord(stop, 0);
    cudaEventSynchronize(stop);
    float tm;   //创建时间变量


    cudaEventElapsedTime(&tm, start, stop);//把事件的时间记录下来
    cudaEventDestroy(start);    //销毁事件
    cudaEventDestroy(stop);     //销毁事件
    printf("GPU Elapsed time:%.6f ms.\n", tm);    //屏幕显示时间
    printf("GPU  :  c[N-1] = %d", c[N - 1]);
    free(a);
    free(b);
    free(c);
    cudaFree(d_a);
    cudaFree(d_b);
    cudaFree(d_c);
    return 0;
}

__global__ void Add(int *d_a, int *d_b, int *d_c)
{
    const int tidx = blockIdx.x*blockDim.x + threadIdx.x;
    if (tidx < N)
        d_c[tidx] = d_a[tidx] + d_b[tidx];
}

但是，使用我需要完成任务后，进行调试编译完成之后，不可以使用VisualProfiler
非常奇怪。这是我下面的一段代码：

 #include <stdlib.h>
#include <stdio.h>
#include <math.h>
#include <cuda_runtime.h>
#include "device_launch_parameters.h"

#define C              3e8
#define pi             3.141592653589793
#define uchar unsigned char

float B = 100e6;  //带宽
float fc = 10e9;  //载频
float Fs = 140e6;  //采样频率
float PRF = 500; //脉冲重复周期
float Rs = 10e3;  //最近斜距
float Tp = 20e-6;  //脉冲时宽
float H = 4e3;   //平台高度
int Nan = 8192;    //方位向采样点数
int Nrn = 4096;    //距离向采样点数
int Nz;
float vx = 70.0, vy = 0.0, vz = 0.0, angle_equal = 0;//俯冲参数
float BeamWide_azimuth = 3.0 / 180 * pi;       //雷达波束宽度
int data_nrn_new = 512, data_nan_new = 512;//从原图中截取的图像大小
float DeltaR = C / 2 / Fs;  //距离采样间隔
float x_interval = DeltaR;
float y_interval = DeltaR;
float theta = 3.0 / 180 * pi;


__global__ void pos(float *d_pos_x, float *d_pos_y, float *d_pos_z, int nan, 
    float PRF, float vx, float vy, float vz, float Rs, float angle_equal, float H);

int main()
{
    //雷达坐标
    float* pos_x;
    pos_x = (float*)malloc(sizeof(float)*Nan);
    float* pos_y;
    pos_y = (float*)malloc(sizeof(float)*Nan);
    float* pos_z;
    pos_z = (float*)malloc(sizeof(float)*Nan);

    float *d_pos_x;    //平台坐标x
    cudaMalloc((void **)&d_pos_x, Nan * sizeof(float));
    float *d_pos_y;    //平台坐标y
    cudaMalloc((void **)&d_pos_y, Nan * sizeof(float));
    float *d_pos_z;    //平台坐标z
    cudaMalloc((void **)&d_pos_z, Nan * sizeof(float));
    printf("分配空间完毕\n");
    //线程分配
    dim3 blocks_Pos((Nan + 512 - 1) / 512, 1);//雷达位置线程
    dim3 threads_Pos(512, 1);
    cudaEvent_t start, stop;
    cudaEventCreate(&start);
    cudaEventCreate(&stop);
    cudaEventRecord(start,0);
    pos<< <blocks_Pos, threads_Pos >> >(d_pos_x, d_pos_y, d_pos_z, Nan, PRF, vx, vy, vz, Rs, angle_equal, H);
    printf("核函数计算完毕\n");
    cudaEventRecord(stop, 0);
    cudaEventSynchronize(stop);
    float dt;
    cudaEventElapsedTime(&dt, start, stop);
    cudaEventDestroy(start);
    cudaEventDestroy(stop);
    printf("time: %.2f ms\n", dt);

    cudaMemcpy(pos_x, d_pos_x, sizeof(float)*Nan, cudaMemcpyHostToDevice);
    cudaMemcpy(pos_y, d_pos_y, sizeof(float)*Nan, cudaMemcpyHostToDevice);
    cudaMemcpy(pos_z, d_pos_z, sizeof(float)*Nan, cudaMemcpyHostToDevice);
    printf("数据传输完毕\n");
    cudaFree(d_pos_x);
    cudaFree(d_pos_y);
    cudaFree(d_pos_z);
    return 1;
}

//计算传感器实时位置
__global__ void pos(float *d_pos_x, float *d_pos_y, float *d_pos_z, int nan,
    float PRF, float vx, float vy, float vz, float Rs, float angle_equal, float H)
{
    const int tid = blockIdx.x*blockDim.x + threadIdx.x;

    if (tid<nan)
    {
        d_pos_x[tid] = (tid - nan / 2.0) / PRF*vx;
        d_pos_y[tid] = (tid - nan / 2.0) / PRF*vy - float(sqrt(pow(double(Rs*cos(angle_equal)), 2.0) - pow(double(H), 2.0)));
        d_pos_z[tid] = H;
    }
}

恳请哪位大神可以帮助我看下出现了什么问题

Answer 1

https://blog.csdn.net/qq_25819827/article/details/52107570