Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 06:39:25 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : support more contiguous cases

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-06-12 15:12:32 +03:00
parent f2a029bd9d
commit cd026b48ef
No known key found for this signature in database
GPG Key ID: BF970631944C16B7
1 changed files with 26 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
ggml.c
View File

@ -3212,35 +3212,42 @@ GGML_CALL bool ggml_is_transposed(const struct ggml_tensor * tensor) {
return tensor->nb[0] > tensor->nb[1]; return tensor->nb[0] > tensor->nb[1];
} }
GGML_CALL bool ggml_is_contiguous(const struct ggml_tensor * tensor) { static bool ggml_is_contiguous_n(const struct ggml_tensor * tensor, int n) {
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); size_t next_nb = ggml_type_size(tensor->type);
if (tensor->ne[0] != ggml_blck_size(tensor->type) && tensor->nb[0] != next_nb) {
return false;
}
next_nb *= tensor->ne[0]/ggml_blck_size(tensor->type);
for (int i = 1; i < GGML_MAX_DIMS; i++) {
if (tensor->ne[i] != 1) {
if (i > n) {
if (tensor->nb[i] != next_nb) {
return false;
}
next_nb *= tensor->ne[i];
} else {
// this dimension does not need to be contiguous
next_nb = tensor->ne[i]*tensor->nb[i];
}
}
}
return true;
}
return GGML_CALL bool ggml_is_contiguous(const struct ggml_tensor * tensor) {
(tensor->ne[0] == ggml_blck_size(tensor->type) || tensor->nb[0] == ggml_type_size(tensor->type)) && return ggml_is_contiguous_0(tensor);
(tensor->ne[1] == 1 || tensor->nb[1] == (tensor->nb[0]*tensor->ne[0])/ggml_blck_size(tensor->type)) &&
(tensor->ne[2] == 1 || tensor->nb[2] == (tensor->nb[1]*tensor->ne[1])) &&
(tensor->ne[3] == 1 || tensor->nb[3] == (tensor->nb[2]*tensor->ne[2]));
} }
GGML_CALL bool ggml_is_contiguous_0(const struct ggml_tensor * tensor) { GGML_CALL bool ggml_is_contiguous_0(const struct ggml_tensor * tensor) {
return ggml_is_contiguous(tensor); return ggml_is_contiguous_n(tensor, 0);
} }
GGML_CALL bool ggml_is_contiguous_1(const struct ggml_tensor * tensor) { GGML_CALL bool ggml_is_contiguous_1(const struct ggml_tensor * tensor) {
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); return ggml_is_contiguous_n(tensor, 1);
return
(tensor->ne[0] == ggml_blck_size(tensor->type) || tensor->nb[0] == ggml_type_size(tensor->type)) &&
(tensor->ne[2] == 1 || tensor->nb[2] == tensor->nb[1]*tensor->ne[1]) &&
(tensor->ne[3] == 1 || tensor->nb[3] == tensor->nb[2]*tensor->ne[2]);
} }
GGML_CALL bool ggml_is_contiguous_2(const struct ggml_tensor * tensor) { GGML_CALL bool ggml_is_contiguous_2(const struct ggml_tensor * tensor) {
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); return ggml_is_contiguous_n(tensor, 2);
return
(tensor->ne[0] == ggml_blck_size(tensor->type) || tensor->nb[0] == ggml_type_size(tensor->type)) &&
(tensor->ne[3] == 1 || tensor->nb[3] == tensor->nb[2]*tensor->ne[2]);
} }
GGML_CALL bool ggml_is_permuted(const struct ggml_tensor * tensor) { GGML_CALL bool ggml_is_permuted(const struct ggml_tensor * tensor) {