EngineX-Hygon/enginex-hygon-llama.cpp
5
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

退回到 b7516 版本

Browse Source
This commit is contained in:
xiezhongtao
2026-01-16 18:12:13 +08:00
parent 9d7890f8c6
commit 7e0d40b535
380 changed files with 18454 additions and 38808 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
tools/server/server-queue.cpp
View File

@@ -74,26 +74,11 @@ int server_queue::get_new_id() {
return new_id;
}
void server_queue::pop_deferred_task(int id_slot) {
void server_queue::pop_deferred_task() {
std::unique_lock<std::mutex> lock(mutex_tasks);
if (!queue_tasks_deferred.empty()) {
// try to find a task that uses the specified slot
bool found = false;
for (auto it = queue_tasks_deferred.begin(); it != queue_tasks_deferred.end(); ++it) {
if (it->id_slot == id_slot) {
QUE_DBG("pop deferred task (use slot %d), id_task = %d\n", id_slot, it->id);
queue_tasks.emplace_front(std::move(*it));
queue_tasks_deferred.erase(it);
found = true;
break;
}
}
// if not tasks found using the slot, just pop the first deferred task (default behavior)
if (!found) {
QUE_DBG("pop deferred task, id_task = %d\n", queue_tasks_deferred.front().id);
queue_tasks.emplace_front(std::move(queue_tasks_deferred.front()));
queue_tasks_deferred.pop_front();
}
queue_tasks.emplace_front(std::move(queue_tasks_deferred.front()));
queue_tasks_deferred.pop_front();
}
time_last_task = ggml_time_ms();
condition_tasks.notify_one();
@@ -232,12 +217,12 @@ void server_response::add_waiting_task_id(int id_task) {
waiting_task_ids.insert(id_task);
}
void server_response::add_waiting_task_ids(const std::unordered_set<int> & id_tasks) {
void server_response::add_waiting_tasks(const std::vector<server_task> & tasks) {
std::unique_lock<std::mutex> lock(mutex_results);
for (const auto & id_task : id_tasks) {
RES_DBG("add task %d to waiting list. current waiting = %d (before add)\n", id_task, (int) waiting_task_ids.size());
waiting_task_ids.insert(id_task);
for (const auto & task : tasks) {
RES_DBG("add task %d to waiting list. current waiting = %d (before add)\n", task.id, (int) waiting_task_ids.size());
waiting_task_ids.insert(task.id);
}
}
@@ -342,7 +327,6 @@ void server_response::terminate() {
void server_response_reader::post_task(server_task && task, bool front) {
GGML_ASSERT(id_tasks.empty() && "post_task() can only be called once per reader");
GGML_ASSERT(!task.is_parent() && "not supported, use post_tasks() instead");
task.index = 0;
id_tasks.insert(task.id);
states.push_back(task.create_state());
@@ -354,18 +338,11 @@ void server_response_reader::post_tasks(std::vector<server_task> && tasks, bool
GGML_ASSERT(id_tasks.empty() && "post_tasks() can only be called once per reader");
id_tasks = server_task::get_list_id(tasks);
states.reserve(tasks.size());
size_t index = 0;
for (auto & task : tasks) {
task.index = index++;
states.push_back(task.create_state());
// for child tasks
for (auto & child_task : task.child_tasks) {
child_task.index = index++;
states.push_back(child_task.create_state());
}
for (size_t i = 0; i < tasks.size(); i++) {
tasks[i].index = i;
states.push_back(tasks[i].create_state());
}
GGML_ASSERT(states.size() == id_tasks.size());
queue_results.add_waiting_task_ids(id_tasks);
queue_results.add_waiting_tasks(tasks);
queue_tasks.post(std::move(tasks), front);
}