pointer not working

Accelerated Computing CUDA CUDA Programming and Performance
1

I am using a similar code in my program…

typedef struct _Node Node;

device Node * globalHead1;
device Node * globalHead2;

device xyz(Node **temp1, Node **temp2){
*temp1=globalHead1;
*temp2=globalHead2;
}

global kernel(){
Node * node1;
Node * node2;

     xyz(&node1,&node2);

}

I am checking correctness through cuda-gdb, and in which I found that it is not working as I was expecting. after calling xyz() from global *temp1 should be globalHead1. but debugger shows that temp1 cannot be dereferenced , illegal memory reference. So can anybody point out that where exactly I am missing something.

Thanks

2

Hum…

The code is incomplete, so maybe I am wrong, but I don’t see the point where you will assign any value (reference) to globalHead1 or globalHead2.
As the partial code is, globalHead1 is an non-initialized pointer to Node structure, containing anything but a correct memory reference (probably pointing to adress 0).

it lacks something like
device Node variableName;

globalHead1 = &variableName;

for globalHead1 pointing on something

3

Here i am not able to dereference temp1 or temp2. the effect of call xyz() should be like, temp1 = &node1, so *temp1 will automatically points to node1. but when I debug it, temp1 remains some garbage value like 0x02 instead &temp1. so it cannot dereference it. i am not able to see what is the problem. i guess it is somehow related to double pointer, though not sure.

4

node1 and node2 are local variables and hence will be mapped to registers…

I am not sure, what &node1 would translate in PTX…

Examine the PTX. You will have more clue.

5

Can I explicitly allocate that variable to a global memory? if I use device then only one shared copy will be created, so is there any support like volatile in ptx.

6

Are you allocaitng memory for globalHead1 and globalHead2 ??

btw,
Instead of XYZ() function, you could just say “node1 = globalHead1; node2 = globalHead2;”