From fe680e3d1080a765e5d3150ffd7bab189742898d Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Thu, 7 Dec 2023 22:26:54 +0200 Subject: sync : ggml (new ops, tests, backend, etc.) (#4359) * sync : ggml (part 1) * sync : ggml (part 2, CUDA) * sync : ggml (part 3, Metal) * ggml : build fixes ggml-ci * cuda : restore lost changes * cuda : restore lost changes (StableLM rope) * cmake : enable separable compilation for CUDA ggml-ci * ggml-cuda : remove device side dequantize * Revert "cmake : enable separable compilation for CUDA" This reverts commit 09e35d04b1c4ca67f9685690160b35bc885a89ac. * cuda : remove assert for rope * tests : add test-backend-ops * ggml : fix bug in ggml_concat * ggml : restore `ggml_get_n_tasks()` logic in `ggml_graph_plan()` * ci : try to fix macOS * ggml-backend : remove backend self-registration * ci : disable Metal for macOS cmake build ggml-ci * metal : fix "supports family" call * metal : fix assert * metal : print resource path ggml-ci --------- Co-authored-by: slaren --- ggml-backend.h | 79 +++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 62 insertions(+), 17 deletions(-) (limited to 'ggml-backend.h') diff --git a/ggml-backend.h b/ggml-backend.h index 96668732..58d5ccae 100644 --- a/ggml-backend.h +++ b/ggml-backend.h @@ -7,41 +7,44 @@ extern "C" { #endif + typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t; + typedef struct ggml_backend_buffer * ggml_backend_buffer_t; + typedef struct ggml_backend * ggml_backend_t; + typedef void * ggml_backend_graph_plan_t; + // // Backend buffer // - struct ggml_backend_buffer; - typedef struct ggml_backend_buffer * ggml_backend_buffer_t; + // buffer type + GGML_API ggml_backend_buffer_t ggml_backend_buft_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size); + GGML_API size_t ggml_backend_buft_get_alignment (ggml_backend_buffer_type_t buft); + GGML_API 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); - // backend buffer functions + // buffer GGML_API void ggml_backend_buffer_free (ggml_backend_buffer_t buffer); - GGML_API size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer); GGML_API void * ggml_backend_buffer_get_base (ggml_backend_buffer_t buffer); GGML_API size_t ggml_backend_buffer_get_size (ggml_backend_buffer_t buffer); - GGML_API size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); GGML_API void ggml_backend_buffer_init_tensor (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); - GGML_API void ggml_backend_buffer_free_tensor (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); + GGML_API size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer); + GGML_API size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); + GGML_API ggml_backend_buffer_type_t ggml_backend_buffer_type(ggml_backend_buffer_t buffer); // // Backend // - struct ggml_backend; - typedef struct ggml_backend * ggml_backend_t; - typedef void * ggml_backend_graph_plan_t; - - GGML_API ggml_backend_t ggml_get_backend(const struct ggml_tensor * tensor); GGML_API const char * ggml_backend_name(ggml_backend_t backend); GGML_API void ggml_backend_free(ggml_backend_t backend); - GGML_API ggml_backend_buffer_t ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size); - - GGML_API size_t ggml_backend_get_alignment(ggml_backend_t backend); + GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend); + GGML_API ggml_backend_buffer_t ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size); + GGML_API size_t ggml_backend_get_alignment(ggml_backend_t backend); - GGML_API void ggml_backend_tensor_set_async( struct ggml_tensor * tensor, const void * data, size_t offset, size_t size); - GGML_API void ggml_backend_tensor_get_async(const struct ggml_tensor * tensor, void * data, size_t offset, size_t size); + GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size); + GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size); GGML_API void ggml_backend_tensor_set( struct ggml_tensor * tensor, const void * data, size_t offset, size_t size); GGML_API void ggml_backend_tensor_get(const struct ggml_tensor * tensor, void * data, size_t offset, size_t size); @@ -57,6 +60,7 @@ extern "C" { // tensor copy between different backends GGML_API void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst); + GGML_API void ggml_backend_tensor_copy_async(ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst); // automatic fallback to sync copy // // CPU backend @@ -68,8 +72,23 @@ extern "C" { GGML_API void ggml_backend_cpu_set_n_threads(ggml_backend_t backend_cpu, int n_threads); // Create a backend buffer from an existing pointer - GGML_API ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(ggml_backend_t backend_cpu, void * ptr, size_t size); + GGML_API ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size); + + GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void); + // + // Backend registry + // + + // The backend registry is a registry of all the available backends, and allows initializing backends in a generic way + + 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 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); + GGML_API ggml_backend_buffer_t ggml_backend_reg_alloc_buffer(size_t i, size_t size); // // Backend scheduler @@ -131,6 +150,32 @@ extern "C" { ggml_backend_sched_t sched, struct ggml_cgraph * graph); + + // + // Utils + // + + struct ggml_backend_graph_copy { + ggml_backend_buffer_t buffer; + struct ggml_context * ctx_allocated; + struct ggml_context * ctx_unallocated; + struct ggml_cgraph * graph; + }; + + // Copy a graph to a different backend + GGML_API struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, struct ggml_cgraph * graph); + GGML_API void ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy); + + typedef bool (*ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data); + + // Compare the output of two backends + GGML_API void ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data); + + // Tensor initialization + GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr); + GGML_API void ggml_backend_view_init(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); + + #ifdef __cplusplus } #endif -- cgit v1.2.3