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

CANN: ROPE operator optimization (#10540)

Browse source 
* [cann] ROPE operator optimization

Co-authored-by: noemotiovon <noemotiovon@gmail.com>
This commit is contained in:
Chenguang Li 2024-11-28 14:24:46 +08:00 • committed by GitHub
parent 9f912511bc
commit b7420131bf
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 211 additions and 106 deletions
Download patch file Download diff file Expand all files Collapse all files

9
ggml/src/ggml-cann/ggml-cann.cpp
View file

@ -1738,13 +1738,8 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
}
case GGML_OP_ROPE: {
// TODO: with ops-test v == 1
float * freq_scale = (float*)((int32_t*)op->op_params + 6);
float * ext_factor = (float*)((int32_t*)op->op_params + 7);
float * attn_factor = (float*)((int32_t*)op->op_params + 8);
// TODO: with freq_factors
if (op->src[2] != NULL) {
return false;
}
// TODO: n_dims <= ne0
if (op->src[0]->ne[0] != op->op_params[1]) {
return false;
@ -1753,10 +1748,6 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
if (*ext_factor != 0) {
return false;
}
// TODO: freq_scale != 1
if (*freq_scale != 1) {
return false;
}
// TODO: attn_factor != 1
if (*attn_factor != 1) {
return false;