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

review: fix a memory leak introduced by review modification which explained in https://github.com/zhouwg/llama.cpp/pull/1

Browse source
This commit is contained in:
zhou.weiguo 2024-06-09 09:06:44 +08:00
parent fdf0272dfb
commit 3e8b61f970
No known key found for this signature in database
GPG key ID: 952EA81D18BB2FA8
1 changed files with 7 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

12
ggml-qnn.cpp
View file

@ -2771,6 +2771,7 @@ GGML_CALL static bool ggml_backend_buffer_is_qnn(ggml_backend_buffer_t buffer) {
GGML_CALL static void ggml_backend_qnn_buffer_free_buffer(ggml_backend_buffer_t buffer) { GGML_CALL static void ggml_backend_qnn_buffer_free_buffer(ggml_backend_buffer_t buffer) {
ggml_backend_qnn_buffer_context * ctx = (ggml_backend_qnn_buffer_context *) buffer->context; ggml_backend_qnn_buffer_context * ctx = (ggml_backend_qnn_buffer_context *) buffer->context;
delete ctx; delete ctx;
} }
@ -3105,12 +3106,14 @@ ggml_backend_buffer_type_t ggml_backend_qnn_buffer_type(size_t device) {
return nullptr; return nullptr;
} }
//ref:https://github.com/zhouwg/llama.cpp/pull/1
static ggml_backend_qnn_buffer_type_context ggml_backend_qnn_buffer_type_contexts[GGML_QNN_MAX_DEVICES];
static ggml_backend_buffer_type ggml_backend_qnn_buffer_types[GGML_QNN_MAX_DEVICES]; static ggml_backend_buffer_type ggml_backend_qnn_buffer_types[GGML_QNN_MAX_DEVICES];
static bool ggml_backend_qnn_buffer_type_initialized = false; static bool ggml_backend_qnn_buffer_type_initialized = false;
if (!ggml_backend_qnn_buffer_type_initialized) { if (!ggml_backend_qnn_buffer_type_initialized) {
for (int i = 0; i < GGML_QNN_MAX_DEVICES; i++) { for (size_t i = 0; i < GGML_QNN_MAX_DEVICES; i++) {
auto & context = ggml_backend_qnn_buffer_type_contexts[i];
context = { i, std::string(GGML_QNN_NAME) + std::to_string(i) };
ggml_backend_qnn_buffer_types[i] = { ggml_backend_qnn_buffer_types[i] = {
/* .iface = */ { /* .iface = */ {
/* .get_name = */ ggml_backend_qnn_buffer_type_name, /* .get_name = */ ggml_backend_qnn_buffer_type_name,
@ -3121,8 +3124,7 @@ ggml_backend_buffer_type_t ggml_backend_qnn_buffer_type(size_t device) {
/* .supports_backend = */ ggml_backend_qnn_buffer_type_supports_backend, /* .supports_backend = */ ggml_backend_qnn_buffer_type_supports_backend,
/* .is_host = */ ggml_backend_qnn_buffer_is_host /* .is_host = */ ggml_backend_qnn_buffer_is_host
}, },
/* .context = */ new ggml_backend_qnn_buffer_type_context { device, /* .context = */ & context,
GGML_QNN_NAME + std::to_string(device)},
}; };
} }
ggml_backend_qnn_buffer_type_initialized = true; ggml_backend_qnn_buffer_type_initialized = true;