OpenCL variable assignment fails when using arithmetic expressions with variable

**Hello, I´m trying to code a kernel using opencl. But Im stuck with a stranger error inside of the kernel function. The flow is more or less like this:

__kernel function1() {
      struct mytype;
      function2(&mytype);
 }

 function2(struct *mytype) {
      uchar *ptr = mytype->value2;
      function3(ptr);
 }

 function3(byte* ptr) {
      uint16 v1 = 10;
      uint16* ptr2 = (uint16*) ptr;
      *ptr2 = v1 >> 8; 
 }


 struct mytype {
       uchar value1[8];
       uchar value2[8];
       uint key[52];
       uint bufleft;
  } 

The code fails when executing the assignment:

*ptr2 = v1 >> 8;

But the only message I receive is "clFlush(): CL_UNKNOWN_ERROR" If I try to assign a value and not an expression, than it works. I´m using OpenCL 1.2 CUDA in Ubuntu

I think this might be undefined behaviour; if you need to reference the same memory as 2 different types, use a union type. Note that uint16 is a vector of 16 uints, not a ushort (16-bit unsigned integer), and needs to be aligned accordingly. value2 is only guaranteed to be aligned on a 4-byte boundary (because of the uint members of the struct) so this won't be enough.