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

fix: check all dimentions in can offload

Browse source
This commit is contained in:
hongruichen 2024-07-19 22:56:00 +08:00
parent b1b5cc10b1
commit 1679dcf47e
2 changed files with 12 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

11
ggml/src/ggml-qnn.cpp
View file

@ -359,7 +359,16 @@ GGML_CALL static bool ggml_backend_qnn_supports_op(ggml_backend_t backend, const
GGML_CALL static bool ggml_backend_qnn_offload_op(ggml_backend_t backend, const ggml_tensor *op) {
GGML_UNUSED(backend);
return op->ne[0] > 1 && op->ne[1] > 1;
size_t dims = ggml_n_dims(op);
bool can_offload = false;
for (size_t i = 0; i < dims; i++) {
if (op->ne[i] > 1) {
can_offload = true;
break;
}
}
return can_offload;
}
static ggml_backend_i ggml_backend_qnn_interface = {

2
ggml/src/ggml-qnn/utils.cpp
View file

@ -153,6 +153,8 @@ const char *opname_from_ggmlop(enum ggml_op ggmlop) {
}
const char *get_qnn_error_string(Qnn_ErrorHandle_t error) {
// A complete list of error codes can be found at here:
// https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/api_error_codes.html
switch (error) {
case QNN_SUCCESS:
return "QNN_SUCCESS";