get_global_offset() seems to be broken on GPU devices
I recently stumbled on a problem with the get_global_offset() function.
With Stream SDK 2.3 and Catalyst 10.12, the function did not return the expected results according to the specification. On the CPU it seemed to work. I use the C++ API and a Radeon HD5870.
Nevertheless, in order to support both device classes I had to work around the issue by passing the offset via another parameter to the kernel.
The problem can be easily reproduced by writing a simple kernel which writes the return value of the function to a buffer. Of course, some offset must be specified when the kernel is enqueued
Is anyone able to confirm this issue?