* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file parallel_tool.cpp
* \brief
*/
#include "passes/pass_utils/parallel_tool.h"
#include "tilefwk/tilefwk.h"
#include "interface/inner/tilefwk.h"
#include "interface/program/program.h"
#include "interface/configs/config_manager.h"
namespace npu::tile_fwk {
void ParallelQueue::Clear() { workItems.clear(); }
void ParallelQueue::Insert(int start, int end, std::function<void(int, int, int)>* bodyPtr)
{
std::unique_lock lock(get_mtx);
workItems.emplace_back(start, end, bodyPtr);
}
bool ParallelQueue::Empty() { return workItems.empty(); }
bool ParallelQueue::Get(WorkItem* workItem)
{
std::unique_lock lock(get_mtx);
if (Empty()) {
return false;
}
*workItem = workItems.front();
workItems.pop_front();
return true;
}
void ParallelTool::Init()
{
if (parallelToolPtr == nullptr) {
const int default_concurrency = config::GetPassGlobalConfig(KEY_PASS_THREAD_NUM, 1);
parallelToolPtr = std::make_unique<ParallelTool>(default_concurrency);
}
}
ParallelTool& ParallelTool::Instance()
{
Init();
return *parallelToolPtr;
}
ParallelTool::ParallelTool(unsigned int num_threads)
: taskReadys(num_threads), numThread(num_threads), waiting_mtx(num_threads), waiting_cv(num_threads)
{
if (num_threads <= 1) {
return;
}
for (int i = 0; i < static_cast<int>(num_threads); i++) {
taskReadys[i] = false;
workers.emplace_back([this, i]() {
while (true) {
std::unique_lock lock(waiting_mtx[i]);
waiting_cv[i].wait(lock, [this, i] { return taskReadys[i] || killThreads; });
if (killThreads) {
break;
}
ExecTaskVec(i);
std::unique_lock completion_lock(completion_mtx);
taskReadys[i] = false;
workingThreads--;
if (workingThreads == 0) {
completion_cv.notify_one();
}
}
});
}
}
ParallelTool::~ParallelTool()
{
if (numThread > 1) {
killThreads = true;
for (int i = 0; i < numThread; i++) {
waiting_cv[i].notify_one();
}
for (int i = 0; i < numThread; i++) {
workers[i].join();
}
}
}
void ParallelTool::ExecTaskVec(int threadIdx)
{
(void)threadIdx;
WorkItem workItem;
while (workQueue.Get(&workItem)) {
(*workItem.bodyPtr_)(workItem.start_, workItem.end_, threadIdx);
}
}
int ParallelTool::GetThreadNum() { return numThread; }
void ParallelTool::Parallel_for(int start, int end, int step, std::function<void(int, int, int)> body)
{
Parallel_for(start, end, step, numThread, body);
}
void ParallelTool::Parallel_for(int start, int end, int step, int numWork, std::function<void(int, int, int)> body)
{
if (numThread <= 1) {
body(start, end, 0);
return;
}
if (end >= start && step > 0) {
int numTask = (end - start) / step;
int numTaskSingle = (numTask + numWork - 1) / numWork;
int stepSize = step * numTaskSingle;
for (int i = start; i < end; i += stepSize) {
int boundEnd = std::min(end, i + stepSize);
workQueue.Insert(i, boundEnd, &body);
}
} else if (end <= start && step < 0) {
int numTask = (start - end) / (-step);
int numTaskSingle = (numTask + numWork - 1) / numWork;
int stepSize = step * numTaskSingle;
for (int i = start; i > end; i += stepSize) {
int boundEnd = std::max(end, i + stepSize);
workQueue.Insert(i, boundEnd, &body);
}
}
std::unique_lock lock(completion_mtx);
workingThreads = numThread;
for (int i = 0; i < numThread; i++) {
taskReadys[i] = true;
}
for (int i = 0; i < numThread; i++) {
waiting_cv[i].notify_one();
}
completion_cv.wait(lock, [this]() { return workingThreads == 0; });
}
std::unique_ptr<ParallelTool> ParallelTool::parallelToolPtr = nullptr;
}