ggml/kompute: Add multi-GPU support#9328

Open

flynnjiang wants to merge 10 commits intoggml-org:masterfrom

flynnjiang:kompute-mutli-gpu-upstream

Commits on Sep 6, 2024

ggml/kompute: Rename ggml_kompute_context to ggml_backend_kompute_context

Ming Xie
authored and
flynnjiang
committed
ggml/kompute: Introducing struct ggml_backend_kompute_buffer_context

Ming Xie
authored and
flynnjiang
committed
ggml/kompute: Move butf into struct ggml_backend_kompute_context

wishlee1124
authored and
flynnjiang
committed
ggml/kompute: Use the kp::Manager in ggml_backend_kompute_context instead of global

wishlee1124
authored and
flynnjiang
committed
ggml/kompute: Implement ggml_backend_i.offload_op interface

Cong Liu
authored and
flynnjiang
committed
ggml/kompute: Reimplement kompute_manager

Cong Liu
authored and
flynnjiang
committed
ggml/kompute: Remove unused ggml_backend_kompute_device_{ref, unref}()
flynnjiang
committed
ggml/kompute: Introduce ggml_backend_kompute_get_device_count()
flynnjiang
committed
ggml/kompute: Introduce ggml_backend_kompute_get_device_memory()
flynnjiang
committed
llama/kompute: Add multi-GPU support
flynnjiang
committed