Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-12 13:27:21 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : relax contiguous constraints in activation function (#2371)

Browse Source
This commit is contained in:
Jiahao Li 2023-07-25 20:58:32 +08:00 committed by GitHub
parent da1889834a
commit 875086bdb9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 18 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
ggml.c
View File

@ -4229,6 +4229,15 @@ bool ggml_is_contiguous(const struct ggml_tensor * tensor) {
tensor->nb[3] == tensor->nb[2]*tensor->ne[2]; tensor->nb[3] == tensor->nb[2]*tensor->ne[2];
} }
static inline bool ggml_is_contiguous_except_dim_1(const struct ggml_tensor * tensor) {
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
return
tensor->nb[0] == GGML_TYPE_SIZE[tensor->type] &&
tensor->nb[2] == tensor->nb[1]*tensor->ne[1] &&
tensor->nb[3] == tensor->nb[2]*tensor->ne[2];
}
bool ggml_is_permuted(const struct ggml_tensor * tensor) { bool ggml_is_permuted(const struct ggml_tensor * tensor) {
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
@ -9814,8 +9823,8 @@ static void ggml_compute_forward_gelu_f32(
const struct ggml_compute_params * params, const struct ggml_compute_params * params,
const struct ggml_tensor * src0, const struct ggml_tensor * src0,
struct ggml_tensor * dst) { struct ggml_tensor * dst) {
GGML_ASSERT(ggml_is_contiguous(src0)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0));
GGML_ASSERT(ggml_is_contiguous(dst)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst));
GGML_ASSERT(ggml_are_same_shape(src0, dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst));
if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) {
@ -9873,8 +9882,8 @@ static void ggml_compute_forward_gelu_quick_f32(
const struct ggml_compute_params * params, const struct ggml_compute_params * params,
const struct ggml_tensor * src0, const struct ggml_tensor * src0,
struct ggml_tensor * dst) { struct ggml_tensor * dst) {
GGML_ASSERT(ggml_is_contiguous(src0)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0));
GGML_ASSERT(ggml_is_contiguous(dst)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst));
GGML_ASSERT(ggml_are_same_shape(src0, dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst));
if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) {
@ -9932,8 +9941,8 @@ static void ggml_compute_forward_silu_f32(
const struct ggml_compute_params * params, const struct ggml_compute_params * params,
const struct ggml_tensor * src0, const struct ggml_tensor * src0,
struct ggml_tensor * dst) { struct ggml_tensor * dst) {
GGML_ASSERT(ggml_is_contiguous(src0)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0));
GGML_ASSERT(ggml_is_contiguous(dst)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst));
GGML_ASSERT(ggml_are_same_shape(src0, dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst));
if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) {
@ -9992,9 +10001,9 @@ static void ggml_compute_forward_silu_back_f32(
const struct ggml_tensor * src0, const struct ggml_tensor * src0,
const struct ggml_tensor * grad, const struct ggml_tensor * grad,
struct ggml_tensor * dst) { struct ggml_tensor * dst) {
GGML_ASSERT(ggml_is_contiguous(grad)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(grad));
GGML_ASSERT(ggml_is_contiguous(src0)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0));
GGML_ASSERT(ggml_is_contiguous(dst)); GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst));
GGML_ASSERT(ggml_are_same_shape(src0, dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst));
GGML_ASSERT(ggml_are_same_shape(src0, grad)); GGML_ASSERT(ggml_are_same_shape(src0, grad));