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

ggml : use ggml_row_size where possible (#4472)

Browse source 
* ggml : use ggml_row_size where possible

ggml-ci

* ggml : move ggml_nbytes_split to ggml-cuda.cu
This commit is contained in:
slaren 2023-12-14 20:05:21 +01:00 • committed by GitHub
parent cafcd4f895
commit 6744dbe924
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 24 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

18
ggml.c
View file

@ -1997,12 +1997,6 @@ size_t ggml_nbytes_pad(const struct ggml_tensor * tensor) {
return GGML_PAD(ggml_nbytes(tensor), GGML_MEM_ALIGN);
}
size_t ggml_nbytes_split(const struct ggml_tensor * tensor, int nrows_split) {
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
return (nrows_split*tensor->ne[0]*ggml_type_size(tensor->type))/ggml_blck_size(tensor->type);
}
int ggml_blck_size(enum ggml_type type) {
return type_traits[type].blck_size;
}
@ -2491,7 +2485,7 @@ static struct ggml_tensor * ggml_new_tensor_impl(
view_src = view_src->view_src;
}
size_t data_size = ggml_type_size(type)*(ne[0]/ggml_blck_size(type));
size_t data_size = ggml_row_size(type, ne[0]);
for (int i = 1; i < n_dims; i++) {
data_size *= ne[i];
}
@ -9698,7 +9692,7 @@ static void ggml_compute_forward_mul_mat(
if (params->type == GGML_TASK_INIT) {
if (src1->type != vec_dot_type) {
char * wdata = params->wdata;
const size_t row_size = ne10*ggml_type_size(vec_dot_type)/ggml_blck_size(vec_dot_type);
const size_t row_size = ggml_row_size(vec_dot_type, ne10);
assert(params->wsize >= ne11*ne12*ne13*row_size);
assert(src1->type == GGML_TYPE_F32);
@ -9721,7 +9715,7 @@ static void ggml_compute_forward_mul_mat(
}
const void * wdata = (src1->type == vec_dot_type) ? src1->data : params->wdata;
const size_t row_size = ne10*ggml_type_size(vec_dot_type)/ggml_blck_size(vec_dot_type);
const size_t row_size = ggml_row_size(vec_dot_type, ne10);
const int64_t nr0 = ne01; // src0 rows
const int64_t nr1 = cne1*ne12*ne13; // src1 rows
@ -16326,7 +16320,7 @@ struct ggml_cplan ggml_graph_plan(struct ggml_cgraph * cgraph, int n_threads) {
} else
#endif
if (node->src[1]->type != vec_dot_type) {
cur = ggml_type_size(vec_dot_type)*ggml_nelements(node->src[1])/ggml_blck_size(vec_dot_type);
cur = ggml_row_size(vec_dot_type, ggml_nelements(node->src[1]));
}
} break;
case GGML_OP_MUL_MAT_ID:
@ -16343,7 +16337,7 @@ struct ggml_cplan ggml_graph_plan(struct ggml_cgraph * cgraph, int n_threads) {
} else
#endif
if (b->type != vec_dot_type) {
cur = ggml_type_size(vec_dot_type)*ggml_nelements(b)/ggml_blck_size(vec_dot_type);
cur = ggml_row_size(vec_dot_type, ggml_nelements(b));
}
} break;
case GGML_OP_OUT_PROD:
@ -18703,7 +18697,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
return NULL;
}
const size_t size_cur = (ne*ggml_type_size(info->type))/ggml_blck_size(info->type);
const size_t size_cur = ggml_row_size(info->type, ne);
ctx->size += GGML_PAD(size_cur, ctx->alignment);
}