llama.cpp

hexagon: improve buffer allocations and mappings (mmap stacking)

#32144

Merged