CUDA invalid device symbol error on cudaMemcpyToSymbol

Problem

In your CUDA code, you have a cudaMemcpyToSymbol call of this form:

cudaMemcpyToSymbol( "deviceVar", hostVar, hostVarSize, 0, cudaMemcpyHostToDevice );

This gives a runtime error of invalid device symbol.

Solution

Check if you are using CUDA 5.0 or later. The technique of passing the device symbol by using its name string is gone in CUDA 5.0. Instead, just remove the double quotes to specify the device variable itself like this:

cudaMemcpyToSymbol( deviceVar, &hostVar, hostVarSize, 0, cudaMemcpyHostToDevice );

Tried with: CUDA 5.0