I get the following problem on OpenCL platform
(running on Intel HD 4000), when i fill buffer inside kernel with values buf[getGlobalId()] = getGlobalId()
some of the values are not copied back from Device to host. Instead of [0, 1,2,3,4...]
sometimes i get [0,0,0,3,4,...]
.
I tried adding localBarrie
r at the end of kernel, and explicit buffer handling but it did't get expected results.
When I set execution mode to CPU everything works fine, does anybody know any reason for this behavior ?