Update cl_img_swap_ops.asciidoc

Defining behavior as undefined for cases when the number of work-items is not evenly divisible by four and if some work-items in the block of four are inactive, defining 1-dimensional local ID as a base for grouping work-items.

extensions/cl_img_swap_ops.asciidoc

@@ -75,6 +75,12 @@ gentype img_swap_y(gentype value);
 * In the third work-item, `img_swap_x` returns `value` passed as an argument in the fourth work-item.
 * In the fourth work-item, `img_swap_x` returns `value` passed as an argument in the third work-item.
 
+The work-items are assigned into blocks based on their 1-dimensional local ID (see `get_local_linear_id`).
+
+The number of work-items that make up a work-group must be evenly divisible by four; otherwise, the behaviour is undefined.
+
+The function must be called in all four work-items of the block; otherwise, the behaviour is undefined.
+
 Requires that the `__opencl_img_swap` feature macro is defined.
 | gentype *img_swap_y*(gentype value)
     a| `img_swap_y` swaps `values` between work-items in the following way: all work-items are divided into blocks of four consecutive elements. For each block: 
@@ -84,6 +90,12 @@ Requires that the `__opencl_img_swap` feature macro is defined.
 * In the second work-item, `img_swap_y` returns `value` passed as an argument in the fourth work-item.
 * In the fourth work-item, `img_swap_y` returns `value` passed as an argument in the second work-item.
 
+The work-items are assigned into blocks based on their 1-dimensional local ID (see `get_local_linear_id`).
+
+The number of work-items that make up a work-group must be evenly divisible by four; otherwise, the behaviour is undefined.
+
+The function must be called in all four work-items of the block; otherwise, the behaviour is undefined.
+
 Requires that the `__opencl_img_swap` feature macro is defined.
 |====
 --