CODEWITHSUNDEEP
cuda
asdasd
asdasd

Reputation: 55

I get malloc 3906 error when i try to fill my array

this is my code

using namespace std;
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#define N 8000

void fillArray(int *data, int count){
    for(int i =0; i < count; i++)
        data[i] = (int) rand() / ((int) RAND_MAX);
}

__global__ void add(int* a, int *b){
    int add = 0;

    int tid = threadIdx.x + blockIdx.x * blockDim.x;
    if(tid < N){
        add = a[tid] + b[tid];
    }
}

__global__ void subtract(int* a, int *b){
    int subtract = 0;

    int tid = threadIdx.x + blockIdx.x * blockDim.x;
    if(tid < N){
        subtract = a[tid] - b[tid];
    }
}

float duration(int *devA, int *devB, int blocksPerGrid, int threadsPerBlock){

    cudaEvent_t start, stop;
    cudaEventCreate(&start);
    cudaEventCreate(&stop);
    cudaEventRecord(start,0);
    cudaEventRecord(stop,0);
    cudaEventSynchronize(stop);

    cudaMalloc((void**) &devA, N * sizeof(int));
    cudaMalloc((void**) &devB, N * sizeof(int));

    add<<<blocksPerGrid, threadsPerBlock>>>(devA,devB);

    float elapsedTime;
    cudaEventElapsedTime(&elapsedTime,start,stop);
    cudaEventDestroy(start);
    cudaEventDestroy(stop);

    return elapsedTime;
}



int main(void) {

    int *a = new int(N);
    int *b = new int(N);
    float dur = 0 ;



    fillArray(a, N);
    fillArray(b, N);

    dur =  duration(a,b,N,1);

    cout << "Global memory version:\n";
    cout << "Process completed in " << dur;
    cout << "for a data set of " << N << " integers.";


    return 0;
}

As you can see, i fill my array with fillArray function in CPU side. But fill array function gives error:

malloc.c 3906 : sYSMalloc: Assertion bla bla

What I'm missing here? I just try to fill array. What might me the problem? Event if i remove the add function in duration function i get this error. What is the problem here?

Upvotes: 0

Views: 109

Answers (1)

BenC
BenC

Reputation: 8976

The error is in the creation of your a and b arrays. As @QWR and @talonmies said, using valgrind (or any Windows substitute) can help you find the source of that kind of error:

==8288== Invalid write of size 4
==8288==    at 0x400DD2: fillArray(int*, int) (kernel.cu:11)
==8288==    by 0x400F79: main (kernel.cu:63)
==8288==  Address 0x62783e4 is 0 bytes after a block of size 4 alloc'd
==8288==    at 0x4C2BA77: operator new(unsigned long) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==8288==    by 0x400F41: main (kernel.cu:57)
==8288== 
==8288== Invalid write of size 4
==8288==    at 0x400DD2: fillArray(int*, int) (kernel.cu:11)
==8288==    by 0x400F8A: main (kernel.cu:64)
==8288==  Address 0x6278434 is 0 bytes after a block of size 4 alloc'd
==8288==    at 0x4C2BA77: operator new(unsigned long) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==8288==    by 0x400F55: main (kernel.cu:58)

If you change:

int *a = new int(N);
int *b = new int(N);

To:

int *a = new int[N];
int *b = new int[N];

The error disappears. Indeed, you were not allocating memory for an array, but just for one integer.

When debugging a CUDA code, you need to use both GPU/device debugging tools (cuda-memcheck, cuda-gdb) and CPU/host tools (valgrind), since errors can happen on both GPU and CPU. Do not forget to compile with the two debug flags of nvcc: -G for device code, and -g for host code.

You should also delete your arrays at the end of your main, for good practice:

delete [] a;
delete [] b;

Upvotes: 2

Related Questions