llama.cpp
opencl: add optimized q8_0 mm kernel for adreno
#18871

Merged

Commits

Add Q8_0 OpenCL kernel

lhez committed 142 days ago
opencl: fix build for non-adreno

lhez committed 142 days ago
opencl: refactor q8_0

lhez committed 142 days ago
opencl: enforce subgroup size of 64 for adreno for q8_0

lhez committed 142 days ago
opencl: suppress warning when adreno kernels are disabled

lhez committed 140 days ago