device memory reallocation

[fermath]

Hi all,

I would like to know if there is a cuda function equivalent to the C realloc() function.

In my code I make several sequential callls to a kernel and in each of these calls I send a vector to it. The size of the vector varies for different calls. I would like to be able to increase the size of the vector maintaining the vector data, just as realloc() does. I cannot define a vector of the biggest possible size, because it does not fit in device memory.

Thanks in advance,

[eyalhir74]

Hi all,

I would like to know if there is a cuda function equivalent to the C realloc() function.

In my code I make several sequential callls to a kernel and in each of these calls I send a vector to it. The size of the vector varies for different calls. I would like to be able to increase the size of the vector maintaining the vector data, just as realloc() does. I cannot define a vector of the biggest possible size, because it does not fit in device memory.

Thanks in advance,

No there is none :(

You’ll have to use the max available memory you can + break the data into chunks if it doesnt fit into device memory.

eyal

[fermath]

No there is none :(

You’ll have to use the max available memory you can + break the data into chunks if it doesnt fit into device memory.

eyal

No luck. Thanks for your reply

[kbam]

No luck. Thanks for your reply

In case it is of help to you, you can copy part of a large host array into a smaller device array

http://forums.nvidia.com/index.php?showtop…rt=#entry998208