diff options
| author | slaren <slarengh@gmail.com> | 2024-06-13 03:11:35 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-06-13 03:11:35 +0200 |
| commit | f578b86b2123d0f92afbaa98a031df4d4464e582 (patch) | |
| tree | 2a21feec089e5fcaa6f9d34be5468a17c3a5ddc7 /ggml-backend.h | |
| parent | 1c641e6aac5c18b964e7b32d9dbbb4bf5301d0d7 (diff) | |
move BLAS to a separate backend (#6210)
* move BLAS to a separate backend
* rename GGML_USE_OPENBLAS to GGML_USE_BLAS
* alloc : reuse same buffer when the same buffer type if used multiple times
* set number of threads automatically for openblas and blis
* sched : print assignments when GGML_SCHED_DEBUG env variable is set
* sched : allow ops with weights on an incompatible buffer type
This will cause the weight to be copied to a backend that supports the
op, which is very costly. The weight should have been stored in a buffer
of a backend that can run the op, but llama.cpp cannot do this
automatically at the moment.
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Diffstat (limited to 'ggml-backend.h')
| -rw-r--r-- | ggml-backend.h | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/ggml-backend.h b/ggml-backend.h index c582b068..47fd8147 100644 --- a/ggml-backend.h +++ b/ggml-backend.h @@ -23,7 +23,6 @@ extern "C" { GGML_API size_t ggml_backend_buft_get_alignment (ggml_backend_buffer_type_t buft); GGML_API size_t ggml_backend_buft_get_max_size (ggml_backend_buffer_type_t buft); GGML_API GGML_CALL size_t ggml_backend_buft_get_alloc_size (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor); - GGML_API bool ggml_backend_buft_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend); GGML_API bool ggml_backend_buft_is_host (ggml_backend_buffer_type_t buft); // buffer @@ -74,6 +73,7 @@ extern "C" { GGML_API enum ggml_status ggml_backend_graph_compute (ggml_backend_t backend, struct ggml_cgraph * cgraph); GGML_API enum ggml_status ggml_backend_graph_compute_async(ggml_backend_t backend, struct ggml_cgraph * cgraph); GGML_API bool ggml_backend_supports_op(ggml_backend_t backend, const struct ggml_tensor * op); + GGML_API bool ggml_backend_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft); GGML_API bool ggml_backend_offload_op(ggml_backend_t backend, const struct ggml_tensor * op); // tensor copy between different backends @@ -90,7 +90,7 @@ extern "C" { GGML_API void ggml_backend_event_free (ggml_backend_event_t event); GGML_API void ggml_backend_event_record (ggml_backend_event_t event); GGML_API void ggml_backend_event_synchronize(ggml_backend_event_t event); - GGML_API void ggml_backend_event_wait (ggml_backend_t backend, ggml_backend_event_t event); // wait async on event + GGML_API void ggml_backend_event_wait (ggml_backend_t backend, ggml_backend_event_t event); // // CPU backend @@ -119,7 +119,7 @@ extern "C" { GGML_API size_t ggml_backend_reg_get_count(void); GGML_API size_t ggml_backend_reg_find_by_name(const char * name); - GGML_API ggml_backend_t ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is name[:params] + GGML_API ggml_backend_t ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is backend_name:params (params is optional) GGML_API const char * ggml_backend_reg_get_name(size_t i); GGML_API ggml_backend_t ggml_backend_reg_init_backend(size_t i, const char * params); // params is backend-specific GGML_API ggml_backend_buffer_type_t ggml_backend_reg_get_default_buffer_type(size_t i); |