ggml: new optimization interface (ggml/988)

This commit is contained in:
Johannes Gäßler 2024-11-16 22:17:59 +02:00 committed by Georgi Gerganov
parent 5c9a8b22b1
commit 8a43e940ab
15 changed files with 2663 additions and 1633 deletions

View file

@ -196,7 +196,7 @@ void ggml_hash_set_reset(struct ggml_hash_set * hash_set);
static bool ggml_hash_contains(const struct ggml_hash_set * hash_set, struct ggml_tensor * key);
// returns GGML_HASHSET_FULL if table is full, otherwise the current index of the key or where it should be inserted
static size_t ggml_hash_find(const struct ggml_hash_set * hash_set, struct ggml_tensor * key);
static size_t ggml_hash_find(const struct ggml_hash_set * hash_set, const struct ggml_tensor * key);
// returns GGML_HASHSET_ALREADY_EXISTS if key already exists, index otherwise, asserts if table is full
static size_t ggml_hash_insert(struct ggml_hash_set * hash_set, struct ggml_tensor * key);
@ -210,7 +210,7 @@ static inline size_t ggml_hash(const struct ggml_tensor * p) {
return (size_t)(uintptr_t)p >> 4;
}
static size_t ggml_hash_find(const struct ggml_hash_set * hash_set, struct ggml_tensor * key) {
static size_t ggml_hash_find(const struct ggml_hash_set * hash_set, const struct ggml_tensor * key) {
size_t h = ggml_hash(key) % hash_set->size;
// linear probing
@ -281,13 +281,14 @@ enum ggml_cgraph_eval_order {
};
struct ggml_cgraph {
int size;
int n_nodes;
int n_leafs;
int size; // maximum number of nodes/leafs/grads/grad_accs
int n_nodes; // number of nodes currently in use
int n_leafs; // number of leafs currently in use
struct ggml_tensor ** nodes;
struct ggml_tensor ** grads;
struct ggml_tensor ** leafs;
struct ggml_tensor ** nodes; // tensors with data that can change if the graph is evaluated
struct ggml_tensor ** grads; // the outputs of these tensors are the gradients of the nodes
struct ggml_tensor ** grad_accs; // accumulators for node gradients
struct ggml_tensor ** leafs; // tensors with constant data
struct ggml_hash_set visited_hash_set;