OpenCL 1.0 bug when casting to (float4 *)

Accelerated Computing CUDA CUDA Programming and Performance
1

This kernel works perfectly fine until I uncomment one of the last 2 lines. Then, the kernel compiles without warnings and runs but returns “CL_INVALID_COMMAND_QUEUE”.

I’m using toolkit_4.0.13 on Ubuntu.

__kernel void vector_add(__global float* a, __global float* b, __global float* c, 

                         int count, int local_buf_size, int loop

                         )

{

    float4 * f4 = (float4*) (& c[4]);

    const int gsize = get_local_size(0);  // group size

    const int lid = get_local_id(0);

    for(int k = 0; k < loop; k++) {

        for(int i = lid; i < count; i += gsize) 

            c[i] = a[i] + b[i];

        barrier(CLK_LOCAL_MEM_FENCE | CLK_GLOBAL_MEM_FENCE);

    }

// (*f4).x=9;

    // f4->x = 9;

}
2

Try specifying __global when declaring f4;

3

Oh, thank you so much Maxim. Works great.

Reza.