Hi there, this is my first post here, so it will not be perfect but I always like feedback to improve.
Now to the topic:
I recently wrote some OpenCL kernels that used OpenGL VBOs as read and write buffers.
To use these buffers in CL one must acquire (and later release) them, so I did this.
Later I was adding another kernel that used them and forgot to acquire these VBOs.
This lead to the neat error message
Unknown error executing CL_COMMAND_NDRANGE_KERNEL on GeForce GTX 560 (Device 0).
And the returned error code was -1000.
Now I appreciate that there was at least an error message but this one was pretty unhelpful.
I do not know how much information is available (on driver side) at the point the kernel fails,
but I would like to have some more precise error message:
Access to OpenGL memory location that was not acquired.
Or at least
Memory error on ...
To know that the problem is memory related.
Maybe there could even be some kind of debug mode, where the NDRangeKernel command checks all arguments on validity,
which would mean that for OpenGL buffers there would be a test if they are acquired.
I know this will probably never be done (or there are already tools that do this and that I am unaware of),
but I needed to at least write this down somewhere, as the google search did not lead me to anyone with the same error message.
Anyhow I also would like to hear of other unhelpful error messages of OpenCL and what was their real meaning.