mirror of
https://github.com/intel/compute-runtime.git
synced 2025-12-30 09:58:55 +08:00
When implicit scaling is not used, a root device in L0 defaults to sub-device 0. In that case, all allocations need to be made against that sub-device. For list and module creation, we were passing the root NEO device instead, and we were allocating their internal allocations using the least-occupied bank, which meant we were allocating alternately between the two sub-devices. This patch makes sure the NEO device used on explicit mode for the root device is sub-device 0. Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>