Setting an array in device memory with a pointer to a struct; in clucks
I am trying to initialize an array in memory with a pointer to a structure that I am creating inside the kernel. Here is the code that I still don't know what I am doing wrong. I get a segmentation error if I try to do cudaMalloc for each element in the array, if I don't I get an "unspecified start" error.
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
typedef struct {
int status;
int location;
double distance;
} Point;
//Macro for checking cuda errors following a cuda launch or api call
#define cudaCheckError() {\
cudaError_t e=cudaGetLastError();\
if(e!=cudaSuccess) {\
printf("\nCuda failure %s:%d: '%s'\n",__FILE__,__LINE__,cudaGetErrorString(e));\
exit(0); \
}\
}
__global__ void kernel1(Point** d_memory, int limit){
int idx = blockIdx.x * blockDim.x * blockDim.y * blockDim.z
+ threadIdx.z * blockDim.y * blockDim.x
+ threadIdx.y * blockDim.x + threadIdx.x;
if(idx < limit) {
Point* pt = ( Point *) malloc( sizeof(Point) );
pt->distance = 10;
pt->location = -1;
pt->status = -1;
d_memory[idx] = pt;
}
}
__global__ void kernel2(Point** d_memory, int limit){
int i;
for (i=0; i<limit;i++){
printf("%f \n",d_memory[i]->distance);
}
}
int main(int argc, char *argv[])
{
int totalGrid = 257*193*129;
size_t size = sizeof(Point) * totalGrid;
Point ** d_memory;
cudaMalloc((void **)&d_memory, size);
/*
for(int i=0; i<totalGrid; i++){
printf("%d\n",i);
cudaMalloc((void **)&d_memory[i], sizeof(Point));
}*/
dim3 bs(16,8,8);
kernel1<<<6249, bs>>>(d_memory, totalGrid);
cudaCheckError();
cudaDeviceSynchronize();
kernel2<<<1,1>>>(d_memory, totalGrid);
cudaCheckError();
cudaFree(d_memory);
return 0;
}
This is what I used to compile the code
nvcc -arch=sm_20 test.cu
source to share
I believe your problem is
Point **d_memory;
he should be
Point *d_memory;
and you don't need the cast void **
, you need it in the code because your pointer is as passed Point ***
, not Point **
.
Note that cudaMalloc()
will allocate contiguous memory, Point **
assumes you need an array of pointers for which I believe you need something like
Point **d_memory;
cudaMalloc((void **)&d_memory, rows);
for (row = 0 ; row < rows ; ++row)
cudaMalloc(&d_memory[row], columns * sizeof(Point));
But then you will need to check that other objects that take d_memory
as a parameter will refer to accordingly d_memory
.
It also cudaMalloc()
returns cudaSuccess
when the distribution was successful, you never check for that.
source to share