ggml : adapt Metal to new ggml_backend interface

ggml-backend.c

@@ -145,6 +145,9 @@ void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst
         return;
     }
 
+    //printf("src->data = %p, src->extra = %p\n", src->data, src->extra);
+    //printf("dst->data = %p, dst->extra = %p\n", dst->data, dst->extra);
+
     if (dst->backend->interface.cpy_tensor_from != NULL) {
         dst->backend->interface.cpy_tensor_from(dst->backend->context, src, dst);
     } else if (src->backend->interface.cpy_tensor_to != NULL) {
@@ -311,8 +314,9 @@ struct ggml_backend * ggml_backend_cpu_init(void) {
     struct ggml_backend * cpu_backend = malloc(sizeof(struct ggml_backend));
 
     *cpu_backend = (struct ggml_backend) {
-        /* .interface = */ cpu_backend_interface,
-        /* .context   = */ ctx
+        /* .interface     = */ cpu_backend_interface,
+        /* .context       = */ ctx,
+        /* .is_ram_shared = */ true,
     };
     return cpu_backend;
 }

ggml-backend.h

@@ -96,6 +96,8 @@ extern "C" {
     struct ggml_backend {
         struct ggml_backend_interface interface;
         ggml_backend_context_t context;
+
+        bool is_ram_shared;
     };
 
     // backend helper functions

ggml-cuda.cu

@@ -1810,8 +1810,9 @@ ggml_backend * ggml_backend_cuda_init(void) {
 
     ggml_backend * cuda_backend = new ggml_backend;
     *cuda_backend = (ggml_backend){
-        /* .interface = */ cuda_backend_interface,
-        /* .context   = */ ctx
+        /* .interface     = */ cuda_backend_interface,
+        /* .context       = */ ctx
+        /* .is_ram_shared = */ false,
     };
     return cuda_backend;
 }

ggml-metal.h

@@ -22,48 +22,49 @@
 #include <stddef.h>
 #include <stdbool.h>
 
-// max memory buffers that can be mapped to the device
-#define GGML_METAL_MAX_BUFFERS 16
-
-struct ggml_tensor;
-struct ggml_cgraph;
+//struct ggml_tensor;
+//struct ggml_cgraph;
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-struct ggml_metal_context;
-
-// number of command buffers to use
-struct ggml_metal_context * ggml_metal_init(int n_cb);
-void ggml_metal_free(struct ggml_metal_context * ctx);
+struct ggml_backend;
 
-// set the number of command buffers to use
-void ggml_metal_set_n_cb(struct ggml_metal_context * ctx, int n_cb);
+struct ggml_backend * ggml_backend_metal_init(void);
 
-// creates a mapping between a host memory buffer and a device memory buffer
-// - make sure to map all buffers used in the graph before calling ggml_metal_graph_compute
-// - the mapping is used during computation to determine the arguments of the compute kernels
-// - you don't need to keep the host memory buffer allocated as it is never accessed by Metal
-// - max_size specifies the maximum size of a tensor and is used to create shared views such
-//   that it is guaranteed that the tensor will fit in at least one of the views
+//struct ggml_metal_context;
 //
-bool ggml_metal_add_buffer(
-        struct ggml_metal_context * ctx,
-                       const char * name,
-                             void * data,
-                           size_t   size,
-                           size_t   max_size);
-
-// set data from host memory into the device
-void ggml_metal_set_tensor(struct ggml_metal_context * ctx, struct ggml_tensor * t);
-
-// get data from the device into host memory
-void ggml_metal_get_tensor(struct ggml_metal_context * ctx, struct ggml_tensor * t);
-
-// same as ggml_graph_compute but uses Metal
-// creates gf->n_threads command buffers in parallel
-void ggml_metal_graph_compute(struct ggml_metal_context * ctx, struct ggml_cgraph * gf);
+//// number of command buffers to use
+//struct ggml_metal_context * ggml_metal_init(int n_cb);
+//void ggml_metal_free(struct ggml_metal_context * ctx);
+//
+//// set the number of command buffers to use
+//void ggml_metal_set_n_cb(struct ggml_metal_context * ctx, int n_cb);
+//
+//// creates a mapping between a host memory buffer and a device memory buffer
+//// - make sure to map all buffers used in the graph before calling ggml_metal_graph_compute
+//// - the mapping is used during computation to determine the arguments of the compute kernels
+//// - you don't need to keep the host memory buffer allocated as it is never accessed by Metal
+//// - max_size specifies the maximum size of a tensor and is used to create shared views such
+////   that it is guaranteed that the tensor will fit in at least one of the views
+////
+//bool ggml_metal_add_buffer(
+//        struct ggml_metal_context * ctx,
+//                       const char * name,
+//                             void * data,
+//                           size_t   size,
+//                           size_t   max_size);
+//
+//// set data from host memory into the device
+//void ggml_metal_set_tensor(struct ggml_metal_context * ctx, struct ggml_tensor * t);
+//
+//// get data from the device into host memory
+//void ggml_metal_get_tensor(struct ggml_metal_context * ctx, struct ggml_tensor * t);
+//
+//// same as ggml_graph_compute but uses Metal
+//// creates gf->n_threads command buffers in parallel
+//void ggml_metal_graph_compute(struct ggml_metal_context * ctx, struct ggml_cgraph * gf);
 
 #ifdef __cplusplus
 }