happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

CANN: Rename `get_env` to `get_env_as_lowercase` (#18624)

This commit is contained in:
Raul Torres 2026-01-07 02:01:25 +00:00 committed by GitHub
parent 95ea9e0861
commit 193ee38a1b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
ggml/src/ggml-cann/ggml-cann.cpp
View File

@ -105,10 +105,10 @@ int32_t ggml_cann_get_device() {
} }
/** /**
* @brief Get the value of the specified environment variable (name). * @brief Get the value of the specified environment variable (name) as lowercase.
* if not empty, return a std::string object * if not empty, return a std::string object
*/ */
std::optional<std::string> get_env(const std::string & name) { std::optional<std::string> get_env_as_lowercase(const std::string & name) {
const char * val = std::getenv(name.c_str()); const char * val = std::getenv(name.c_str());
if (!val) { if (!val) {
return std::nullopt; return std::nullopt;
@ -259,7 +259,7 @@ struct ggml_cann_pool_buf_prio : public ggml_cann_pool {
* @param device The device ID to associate with this buffer pool. * @param device The device ID to associate with this buffer pool.
*/ */
explicit ggml_cann_pool_buf_prio(int device) : device(device) { explicit ggml_cann_pool_buf_prio(int device) : device(device) {
disable_clean = parse_bool(get_env("GGML_CANN_DISABLE_BUF_POOL_CLEAN").value_or("")); disable_clean = parse_bool(get_env_as_lowercase("GGML_CANN_DISABLE_BUF_POOL_CLEAN").value_or(""));
} }
/** /**
@ -452,7 +452,7 @@ struct ggml_cann_pool_buf : public ggml_cann_pool {
* @param device The device ID to associate with this buffer pool. * @param device The device ID to associate with this buffer pool.
*/ */
explicit ggml_cann_pool_buf(int device) : device(device) { explicit ggml_cann_pool_buf(int device) : device(device) {
disable_clean = parse_bool(get_env("GGML_CANN_DISABLE_BUF_POOL_CLEAN").value_or("")); disable_clean = parse_bool(get_env_as_lowercase("GGML_CANN_DISABLE_BUF_POOL_CLEAN").value_or(""));
} }
/** /**
@ -764,7 +764,7 @@ struct ggml_cann_pool_vmm : public ggml_cann_pool {
* @return A unique pointer to the created CANN pool. * @return A unique pointer to the created CANN pool.
*/ */
std::unique_ptr<ggml_cann_pool> ggml_backend_cann_context::new_pool_for_device(int device) { std::unique_ptr<ggml_cann_pool> ggml_backend_cann_context::new_pool_for_device(int device) {
std::string mem_pool_type = get_env("GGML_CANN_MEM_POOL").value_or(""); std::string mem_pool_type = get_env_as_lowercase("GGML_CANN_MEM_POOL").value_or("");
if (mem_pool_type == "prio") { if (mem_pool_type == "prio") {
GGML_LOG_INFO("%s: device %d use buffer pool with priority queue\n", __func__, device); GGML_LOG_INFO("%s: device %d use buffer pool with priority queue\n", __func__, device);
@ -1217,7 +1217,7 @@ static void ggml_backend_cann_buffer_set_tensor(ggml_backend_buffer_t buffer,
// Why aclrtSynchronizeDevice? // Why aclrtSynchronizeDevice?
// Only check env once. // Only check env once.
static bool weight_to_nz = parse_bool(get_env("GGML_CANN_WEIGHT_NZ").value_or("on")); static bool weight_to_nz = parse_bool(get_env_as_lowercase("GGML_CANN_WEIGHT_NZ").value_or("on"));
if (!need_transform(tensor->type)) { if (!need_transform(tensor->type)) {
ACL_CHECK(aclrtMemcpy((char *) tensor->data + offset, size, data, size, ACL_MEMCPY_HOST_TO_DEVICE)); ACL_CHECK(aclrtMemcpy((char *) tensor->data + offset, size, data, size, ACL_MEMCPY_HOST_TO_DEVICE));
if (weight_to_nz && is_matmul_weight((const ggml_tensor *) tensor)) { if (weight_to_nz && is_matmul_weight((const ggml_tensor *) tensor)) {
@ -1442,7 +1442,7 @@ static size_t ggml_backend_cann_buffer_type_get_alloc_size(ggml_backend_buffer_t
int64_t ne0 = tensor->ne[0]; int64_t ne0 = tensor->ne[0];
// Only check env once. // Only check env once.
static bool weight_to_nz = parse_bool(get_env("GGML_CANN_WEIGHT_NZ").value_or("on")); static bool weight_to_nz = parse_bool(get_env_as_lowercase("GGML_CANN_WEIGHT_NZ").value_or("on"));
// last line must bigger than 32, because every single op deal at // last line must bigger than 32, because every single op deal at
// least 32 bytes. // least 32 bytes.
@ -2136,7 +2136,7 @@ static void evaluate_and_capture_cann_graph(ggml_backend_cann_context * cann_ctx
#endif // USE_ACL_GRAPH #endif // USE_ACL_GRAPH
// Only perform the graph execution if CANN graphs are not enabled, or we are capturing the graph. // Only perform the graph execution if CANN graphs are not enabled, or we are capturing the graph.
// With the use of CANN graphs, the execution will be performed by the graph launch. // With the use of CANN graphs, the execution will be performed by the graph launch.
static bool opt_fusion = parse_bool(get_env("GGML_CANN_OPERATOR_FUSION").value_or("")); static bool opt_fusion = parse_bool(get_env_as_lowercase("GGML_CANN_OPERATOR_FUSION").value_or(""));
if (!use_cann_graph || cann_graph_capture_required) { if (!use_cann_graph || cann_graph_capture_required) {
for (int i = 0; i < cgraph->n_nodes; i++) { for (int i = 0; i < cgraph->n_nodes; i++) {
@ -2201,7 +2201,7 @@ static enum ggml_status ggml_backend_cann_graph_compute(ggml_backend_t backend,
#ifdef USE_ACL_GRAPH #ifdef USE_ACL_GRAPH
bool use_cann_graph = true; bool use_cann_graph = true;
static bool prefill_use_graph = parse_bool(get_env("GGML_CANN_PREFILL_USE_GRAPH").value_or("")); static bool prefill_use_graph = parse_bool(get_env_as_lowercase("GGML_CANN_PREFILL_USE_GRAPH").value_or(""));
if (!prefill_use_graph) { if (!prefill_use_graph) {
// Do not use acl_graph for prefill. // Do not use acl_graph for prefill.
for (int i = 0; i < cgraph->n_nodes; i++) { for (int i = 0; i < cgraph->n_nodes; i++) {