metal : make the backend async v2 #15906
metal : make the backend async
97b96c1a
cont : add comments, extend op offload, clean up
c5637cf3
metal : fix batch size for MUL_MAT_ID
bdff7729
metal : remove deprecated ggml_backend_metal_buffer_from_ptr
d91ba85d
metal : create only metal buffers, no wrapping of host memory
85aaf52b
metal : restore .alloc_buffer for buffer_from_ptr_type
7fc2b3d5
metal : remove broken implementation of GGML_OP_SET
f288225d
metal : clean-up loose ends, ready for tests
0926cb49
metal : support both private and shared buffers
7b59f0f7
metal : enable private buffers + add global device queue
040844cf
metal : disable host buffer to prevent races
afd95d20
metal : avoid extra copy during set_tensor
e65c53e9
metal : use separate buffer types for shread and private Metal buffers
523750a8
metal : simplify synchronization logic
9248aecd
metal : fix build
5fdce91b
slaren
commented
on 2025-09-10
metal : do not implement cpy_tensor
c9a5ba47
metal : separate implementations for shared and private buffers
e796f669
ggerganov
merged
0f0a3c28
into master 30 days ago
ggerganov
deleted the gg/metal-async-v2 branch 30 days ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub