CUDA cudaMemcpu Segmentation fault when copying array of object pointers

Question

I am trying to move an array of pointers to the device, where each pointer is pointing to a class object. However, I get a Segmentation fault at the line using cudaMemcpy. I am trying to follow the lines used in this post.

main.cu

#include "testclass.cuh"
#include <iostream>


__global__ void printtest(Test* test){
    printf("HELLO FROM CUDA\n");
    printf("CUDA1 : %i\n", test->hello);
    Test test2(6);
    printf("CUDA2 : %i\n", test2.hello);
    printf("BYEEE FROM CUDA\n");

}

int main(){
    printf("hello\n");
    Test* test = new Test(512);
    printf("CPU : %i\n", test->hello);
    Test* devtest;
    cudaMalloc(&devtest, sizeof(Test));
    cudaError_t err = cudaMemcpy(devtest, test, sizeof(Test), cudaMemcpyHostToDevice);
    if (err != cudaSuccess) {                                   
        fprintf(stderr, "Error %s at line %d in file %s\n",             
            cudaGetErrorString(err), __LINE__-3, __FILE__);
    }
    printtest<<<1, 1>>>(devtest);
    cudaDeviceSynchronize();



    printf("hello2\n");
    Test** test3 = new Test*[2];
    test3[0] = new Test(12299);
    test3[1] = new Test(234923);
    printf("CPU : %i\n", test3[0]->hello);
    Test** devtest3;
    cudaMalloc(&devtest3, 2*sizeof(Test*));
    printf("CPU2\n");
    err = cudaMemcpy(devtest3[0], test3[0], sizeof(Test), cudaMemcpyHostToDevice);
    if (err != cudaSuccess) {                                   
        fprintf(stderr, "Error %s at line %d in file %s\n",             
            cudaGetErrorString(err), __LINE__-3, __FILE__);
    }
    printf("CPU3\n");
    printtest<<<1, 1>>>(devtest3[0]);
    cudaDeviceSynchronize();
}

testclass.cu

#include "testclass.cuh"

__host__ __device__ Test::Test(int in){
    hello = in;
}

testclass.cuh

class Test {
public: 
    int hello;
    __host__ __device__ Test(int);
};

devtest3[0] is an uninitialized pointer, so you can't copy anything there. You need to allocate space for the Test objects, too, not just the pointers. — molbdnilo
– molbdnilo, Commented Oct 28, 2020 at 16:29
I see, thank you. I fixed the above by changing Test** devtest3; to Test* devtest[2]; and then mallocing cudaMalloc(&devtest3[0], sizeof(Test));. instead of cudaMalloc(&devtest3, 2*sizeof(Test*));. — s4342re41
– s4342re41, Commented Oct 28, 2020 at 16:36

s4342re41 · Accepted Answer · 2020-10-28 16:37:45Z

1

Solved it using @molbdnilo 's comment.

main.cu

...
printf("hello2\n");
    Test** test3 = new Test*[2];
    test3[0] = new Test(12299);
    test3[1] = new Test(234923);
    printf("CPU : %i\n", test3[0]->hello);
    Test* devtest3[2];
    cudaMalloc(&devtest3[0], sizeof(Test));
    printf("CPU2\n");
    err = cudaMemcpy(devtest3[0], test3[0], sizeof(Test), cudaMemcpyHostToDevice);
    if (err != cudaSuccess) {                                   
        fprintf(stderr, "Error %s at line %d in file %s\n",             
            cudaGetErrorString(err), __LINE__-3, __FILE__);
    }
    printf("CPU3\n");
    printtest<<<1, 1>>>(devtest3[0]);
    cudaDeviceSynchronize();
...

answered Oct 28, 2020 at 16:37

s4342re41

1039 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

CUDA cudaMemcpu Segmentation fault when copying array of object pointers

1 Answer 1

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related