Memory access violation when accessing constant variable within a kernel

I have a constant variable which is a

int*

.

I allocate and copy the value to this variable in a host function.

Cuda Memcheck reports access violations.

code:

__constant__ int *dev_gridx; // outside any functions in this .cu file

cudaError_t Renderer::renderWithCuda(double *T) {
    // ...
    cudaStatus = cudaMalloc((void**)&dev_gridx, sizeof(int));
	    if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
	    }

    cudaStatus = cudaMemcpyToSymbol(dev_gridx, &gridx, sizeof(int)); // gridx is a int* variable on cpu
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed! gridx");
	}
    // ...
}

__global__ kernel() {
        int x = blockIdx.x*blockDim.x + threadIdx.x;
	int y = blockIdx.y*blockDim.y + threadIdx.y;
	int z = blockIdx.z*blockDim.z + threadIdx.z;

	/* 3D grid with 1D block:*/

	int gridx = dev_gridx; // dev_gridx is stored in constant memory, reported access violation
}

error message:

================================================================================
CUDA Memory Checker detected 32 threads caused an access violation:
Launch Parameters
    CUcontext    = 00fcf578
    CUstream     = 043ec108
    CUmodule     = 078f64f0
    CUfunction   = 0bcb38e8
    FunctionName = _Z14radianceKernelPdS_S_S_
    GridId       = 17
    gridDim      = {16,1,1}
    blockDim     = {89,1,1}
    sharedSize   = 256
    Parameters:
        dev_le = 0x08dd0000  0
        dev_l = 0x08a21e00  0
        dev_le_mean = 0x08a22200  0
        dev_temperature_grid = 0x08cd0000  -6.27743856220419e+66
    Parameters (raw):
         0x08dd0000 0x08a21e00 0x08a22200 0x08cd0000

Summary of access violations:
path\kernel.cu(900): error MemoryChecker: #misaligned=32  #invalidAddress=0
================================================================================

Memory Checker detected 32 access violations.
error = misaligned load (global memory)
gridid = 17
blockIdx = {0,0,0}
threadIdx = {0,0,0}
address = 0x0000001e
accessSize = 4

You can’t take the address of a device variable in host code.

Here’s a similar question with an appropriate answer:

https://stackoverflow.com/questions/45974010/cuda-cudainvalidvalue-error-when-cudamemcpytosymbol