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

this code

using namespace std; #include <iostream> #include <stdio.h> #include <stdlib.h> #define n 8000  void fillarray(int *data, int count){     for(int =0; < 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 data set of " << n << " integers.";       return 0; } 

as can see, fill array fillarray function in cpu side. fill array function gives error:

malloc.c 3906 : sysmalloc: assertion bla bla 

what i'm missing here? try fill array. might me problem? event if remove add function in duration function error. problem here?

the error in creation of a , b arrays. @qwr , @talonmies said, using valgrind (or windows substitute) can find source of kind of error:

==8288== invalid write of size 4 ==8288==    @ 0x400dd2: fillarray(int*, int) (kernel.cu:11) ==8288==    0x400f79: main (kernel.cu:63) ==8288==  address 0x62783e4 0 bytes after block of size 4 alloc'd ==8288==    @ 0x4c2ba77: operator new(unsigned long) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==8288==    0x400f41: main (kernel.cu:57) ==8288==  ==8288== invalid write of size 4 ==8288==    @ 0x400dd2: fillarray(int*, int) (kernel.cu:11) ==8288==    0x400f8a: main (kernel.cu:64) ==8288==  address 0x6278434 0 bytes after block of size 4 alloc'd ==8288==    @ 0x4c2ba77: operator new(unsigned long) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==8288==    0x400f55: main (kernel.cu:58) 

if 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, not allocating memory array, 1 integer.

when debugging cuda code, need use both gpu/device debugging tools (cuda-memcheck, cuda-gdb) , cpu/host tools (valgrind), since errors can happen on both gpu , cpu. not forget compile 2 debug flags of nvcc: -g device code, , -g host code.

you should delete arrays @ end of main, good practice:

delete [] a; delete [] b;