File size: 7,614 Bytes
8207382 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 | // Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#pragma once
#include <atomic>
#include <condition_variable>
#include <iostream>
#include <memory>
#include <mutex>
#include <queue>
#include <vector>
#include "absl/status/status.h"
#include "absl/status/statusor.h"
#include "src/base/base_pipeline.h"
#include "thread_pool.h"
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
class AutoParallelSimpleInferencePipeline : public BasePipeline {
private:
struct InferenceInstance {
std::shared_ptr<BasePipeline> pipeline;
std::queue<PipelineInput> task_queue;
std::queue<std::promise<PipelineResult>> promise_queue;
std::mutex queue_mutex;
std::atomic<bool> is_busy{false};
int instance_id;
};
public:
AutoParallelSimpleInferencePipeline(const PipelineParams ¶ms);
absl::Status Init();
std::future<PipelineResult> PredictAsync(const PipelineInput &input);
absl::Status PredictThread(const PipelineInput &input);
absl::StatusOr<PipelineResult> GetResult();
virtual ~AutoParallelSimpleInferencePipeline();
private:
void ProcessInstanceTasks(int instance_id);
PipelineParams params_;
int thread_num_;
std::atomic<int> round_robin_index_{0};
std::unique_ptr<PaddlePool::ThreadPool> pool_;
std::vector<std::unique_ptr<InferenceInstance>> instances_;
std::queue<std::future<PipelineResult>> legacy_results_;
std::mutex legacy_results_mutex_;
};
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
AutoParallelSimpleInferencePipeline<Pipeline, PipelineParams, PipelineInput,
PipelineResult>::
AutoParallelSimpleInferencePipeline(const PipelineParams ¶ms)
: BasePipeline(), params_(params), thread_num_(params.thread_num) {
if (thread_num_ > 1) {
auto status = Init();
if (!status.ok()) {
INFOE("Pipeline pool init error : %s", status.ToString().c_str());
exit(-1);
}
}
}
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
absl::Status
AutoParallelSimpleInferencePipeline<Pipeline, PipelineParams, PipelineInput,
PipelineResult>::Init() {
try {
pool_ = std::unique_ptr<PaddlePool::ThreadPool>(
new PaddlePool::ThreadPool(thread_num_));
for (int i = 0; i < thread_num_; i++) {
auto instance =
std::unique_ptr<InferenceInstance>(new InferenceInstance());
instance->instance_id = i;
instance->pipeline = std::shared_ptr<BasePipeline>(new Pipeline(params_));
instances_.push_back(std::move(instance));
}
} catch (const std::bad_alloc &e) {
return absl::ResourceExhaustedError(std::string("Out of memory: ") +
e.what());
} catch (const std::exception &e) {
return absl::InternalError(std::string("Init failed: ") + e.what());
}
return absl::OkStatus();
}
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
std::future<PipelineResult> AutoParallelSimpleInferencePipeline<
Pipeline, PipelineParams, PipelineInput,
PipelineResult>::PredictAsync(const PipelineInput &input) {
int instance_id = round_robin_index_.fetch_add(1) % thread_num_;
auto &instance = instances_[instance_id];
std::promise<PipelineResult> promise;
auto future = promise.get_future();
{
std::lock_guard<std::mutex> lock(instance->queue_mutex);
instance->task_queue.push(input);
instance->promise_queue.push(std::move(promise));
}
bool expected = false;
if (instance->is_busy.compare_exchange_strong(
expected, true)) { // one instance just process one input
pool_->submit([this, instance_id]() { ProcessInstanceTasks(instance_id); });
}
return future;
}
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
void AutoParallelSimpleInferencePipeline<
Pipeline, PipelineParams, PipelineInput,
PipelineResult>::ProcessInstanceTasks(int instance_id) {
auto &instance = instances_[instance_id];
while (true) {
std::vector<std::string> input;
std::promise<PipelineResult> promise;
{
std::lock_guard<std::mutex> lock(instance->queue_mutex);
if (instance->task_queue.empty()) {
instance->is_busy = false;
if (!instance->task_queue.empty()) {
bool expected = false;
if (instance->is_busy.compare_exchange_strong(expected, true)) {
continue;
}
}
return;
}
input = std::move(instance->task_queue.front());
instance->task_queue.pop();
promise = std::move(instance->promise_queue.front());
instance->promise_queue.pop();
}
try {
PipelineResult result = instance->pipeline->Predict(input);
promise.set_value(std::move(result));
} catch (const std::exception &e) {
promise.set_exception(std::current_exception());
}
}
}
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
absl::Status AutoParallelSimpleInferencePipeline<
Pipeline, PipelineParams, PipelineInput,
PipelineResult>::PredictThread(const PipelineInput &input) {
try {
auto future = PredictAsync(input);
std::lock_guard<std::mutex> lock(legacy_results_mutex_);
legacy_results_.push(std::move(future));
return absl::OkStatus();
} catch (const std::exception &e) {
return absl::InternalError(std::string("Failed to submit inference: ") +
e.what());
}
}
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
absl::StatusOr<PipelineResult>
AutoParallelSimpleInferencePipeline<Pipeline, PipelineParams, PipelineInput,
PipelineResult>::GetResult() {
std::lock_guard<std::mutex> lock(legacy_results_mutex_);
if (legacy_results_.empty())
return absl::NotFoundError("No inference result available");
try {
auto future = std::move(legacy_results_.front());
legacy_results_.pop();
PipelineResult result = future.get();
return result;
} catch (const std::exception &e) {
return absl::InternalError(std::string("Failed to get inference result: ") +
e.what());
}
}
template <typename Pipeline, typename PipelineParams, typename PipelineInput,
typename PipelineResult>
AutoParallelSimpleInferencePipeline<
Pipeline, PipelineParams, PipelineInput,
PipelineResult>::~AutoParallelSimpleInferencePipeline() {
while (!legacy_results_.empty()) {
try {
legacy_results_.front().get();
} catch (...) {
}
legacy_results_.pop();
}
for (auto &instance : instances_) {
while (instance->is_busy.load()) {
std::this_thread::sleep_for(std::chrono::milliseconds(10));
}
}
}
|