Neo Zhang Jianyu commited on
Commit
b1865d2
·
unverified ·
1 Parent(s): 472195f

Support multiple GPUs (split mode) on SYCL backend (llama/5806)

Browse files

* suport multiple cards: split-mode - layer|row

* rm warning

* rebase with master, support tow new OPs, close feature for -sm=row, fix for unit test

* update news

* fix merge error

* update according to review comments

Files changed (2) hide show
  1. ggml-sycl.cpp +0 -0
  2. ggml-sycl.h +5 -0
ggml-sycl.cpp CHANGED
The diff for this file is too large to render. See raw diff
 
ggml-sycl.h CHANGED
@@ -24,6 +24,11 @@ GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type(void);
24
  GGML_API void ggml_backend_sycl_print_sycl_devices(void);
25
  GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
26
  GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
 
 
 
 
 
27
  #ifdef __cplusplus
28
  }
29
  #endif
 
24
  GGML_API void ggml_backend_sycl_print_sycl_devices(void);
25
  GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
26
  GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
27
+ GGML_API GGML_CALL int ggml_backend_sycl_get_device_count();
28
+ GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split);
29
+ GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
30
+ GGML_API GGML_CALL int ggml_backend_sycl_get_device_index(int device_id);
31
+
32
  #ifdef __cplusplus
33
  }
34
  #endif