Hi,
I would like to notify to the AMD team that I have several "access violations" once I start to play with float8.
So, it will be fine to have a better support of float8/16 on the next version. More test cases are needed.
NB : My code works with Intel's OpenCL and what I do with float8 is quite simple (=,+,/... pass float8 by pointer, ...).
Regards