CudaMemcpyToSymbol does not copy data

Question

I want to use memory __constant__to which all threads in all my cores will be available.

The ad looks something like this:

extern __constant__ float smooth [8 * 1024];

I copy data to this variable using

cudaMemcpyToSymbol("smooth", smooth_local, smooth_size, 0, cudaMemcpyHostToDevice);

smooth_size = 7K bytes

It gave me the wrong conclusion

but when I ran it in mode -deviceemuand tried to print the contents of both of these variables inside the kernel, I got all zeros for smooth and smooth_local was correct.

I tried to print the output right after cudaMemcpyToSymbol, but it gave me 0.

Can you shed light on my problem?

Answer 1

To declare CUDA read-only memory, it will look like this:

__constant__ float smooth[8 * 1024];

, CUDA (.. ​​). CUDA, , separete.cpp/.cu , .cpp/.cu, . cudaMemCopyToSymbol. , , 64k CUDA.