vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

metal : enable ggml-alloc (#2627)

Browse source 
* metal: enable ggml-alloc

Make ggml-alloc work with concurrently dispatch.

* style-fix

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Shouzheng Liu 2023-08-16 16:08:28 -04:00 • committed by GitHub
parent bf83bff674
commit fc8ef549e5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 61 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

25
ggml-alloc.c
View file

@ -67,6 +67,8 @@ struct ggml_allocr {
struct hash_node hash_table[GGML_GRAPH_HASHTABLE_SIZE];
size_t max_size;
bool measure;
int parse_seq[GGML_MAX_NODES];
bool has_parse_seq;
#ifdef GGML_ALLOCATOR_DEBUG
struct ggml_tensor * allocated_tensors[1024];
@ -229,6 +231,17 @@ static void ggml_allocator_free_tensor(struct ggml_allocr * alloc, struct ggml_t
alloc->n_free_blocks++;
}
void ggml_allocr_set_parse_seq(struct ggml_allocr * alloc, int * list, int n) {
int pos = 0;
for (int i = 0; i < n; i++) {
if (list[i] != -1) {
alloc->parse_seq[pos] = list[i];
pos++;
}
}
alloc->has_parse_seq = true;
}
void ggml_allocr_reset(struct ggml_allocr * alloc) {
alloc->n_free_blocks = 1;
size_t align_offset = aligned_offset(alloc->data, 0, alloc->alignment);
@ -248,6 +261,8 @@ struct ggml_allocr * ggml_allocr_new(void * data, size_t size, size_t alignment)
/*.hash_table = */ {{0}},
/*.max_size = */ 0,
/*.measure = */ false,
/*.parse_seq = */ {0},
/*.has_parse_seq = */ false,
#ifdef GGML_ALLOCATOR_DEBUG
/*.allocated_tensors = */ = {0},
#endif
@ -275,6 +290,8 @@ struct ggml_allocr * ggml_allocr_new_measure(size_t alignment) {
/*.hash_table = */ {{0}},
/*.max_size = */ 0,
/*.measure = */ true,
/*.parse_seq = */ {0},
/*.has_parse_seq = */ false,
#ifdef GGML_ALLOCATOR_DEBUG
/*.allocated_tensors = */ = {0},
#endif
@ -473,7 +490,13 @@ static size_t ggml_allocator_alloc_graph_tensors_n(
allocate_node(alloc, input);
}
}
for (int i = 0; i < gf->n_nodes; i++) {
for (int ind = 0; ind < gf->n_nodes; ind++) {
int i;
if (alloc->has_parse_seq) {
i = alloc->parse_seq[ind];
} else {
i = ind;
}
struct ggml_tensor * node = gf->nodes[i];
// allocate parents (leafs)