unspecified launch error from cudaMemcpy

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I’ve allocated a bunch of arrays, and I’m copying them to the device. Everything goes fine until the 10th copy, at which point the program stalls and the cudaMemcpy function returns 3, which is an “unspecified launch error”

a10=cudaMemcpy(Gdevice.A,Ghost.A,sizeof(float)(1023)(1023),cudaMemcpyHostToDevice);

printf(cudaGetErrorString(a10));

It seems like there is some sort of memory limitation, since this cudaMemcpy will make the total amount of memory copied surpass 8MB. I’m using an 8800GTS and running Windows XP.

Thanks for your help,

Joe

2

There shouldn’t be any such memory limitation. I’ve copied hundreds of MB’s to my GTX.

Double check all cudaMallocs to make sure they are the right size. Unspecified lauch error sometimes means you are writing past the end of the allocated data.

3

Thanks MisterAnderson42,

That was exactly the problem. In cudamalloc I forgot to multiply by sizeof(float).

Joe