mirror of
https://github.com/saymrwulf/pytorch.git
synced 2026-05-14 20:57:59 +00:00
ee7b3c9b2b
Summary: RFC. This is a naive implementation of Rebatchin Queue for MultiTask effort. Full disclaimer, I'm very new to Caffe/Machine Learning and I'm doing dodge science here (under Dmytros supervision), so please be extra tough on this review so I can learn best practices :) Differential Revision: D4871970 fbshipit-source-id: 924820ef0fce45b5e2bdabeec9885cbafa23a880
68 lines
1.3 KiB
C++
68 lines
1.3 KiB
C++
#pragma once
|
|
|
|
#include <atomic>
|
|
#include <condition_variable>
|
|
#include <memory>
|
|
#include <mutex>
|
|
#include <queue>
|
|
|
|
#include "caffe2/core/logging.h"
|
|
#include "caffe2/core/operator.h"
|
|
#include "caffe2/core/stats.h"
|
|
#include "caffe2/core/tensor.h"
|
|
|
|
namespace caffe2 {
|
|
|
|
// TODO: This is a very naive implementation with a single mutex. We can do the
|
|
// atomic index + circular queue optimizations or pull something more
|
|
// heavy-weight later
|
|
|
|
class RebatchingQueue {
|
|
public:
|
|
RebatchingQueue(size_t capacity, size_t numBlobs);
|
|
|
|
~RebatchingQueue();
|
|
|
|
bool enqueueOne(
|
|
CPUContext& context,
|
|
const std::vector<const TensorCPU*>& inputs);
|
|
|
|
bool enqueueMany(
|
|
CPUContext& context,
|
|
const std::vector<const TensorCPU*>& inputs);
|
|
|
|
bool dequeue(
|
|
CPUContext& context,
|
|
size_t numElements,
|
|
const std::vector<TensorCPU*>& outputs);
|
|
|
|
size_t capacity() const;
|
|
|
|
size_t numBlobs() const;
|
|
|
|
bool isClosed() const;
|
|
|
|
void close();
|
|
|
|
private:
|
|
bool enqueue(std::vector<std::vector<TensorCPU>> splittedInputs);
|
|
|
|
bool canWrite() const;
|
|
bool canRead() const;
|
|
|
|
const size_t capacity_;
|
|
const size_t numBlobs_;
|
|
|
|
mutable std::mutex mutex_;
|
|
|
|
bool isClosed_{false};
|
|
|
|
uint64_t head_{0};
|
|
uint64_t tail_{0};
|
|
|
|
std::condition_variable cvEmpty_;
|
|
std::condition_variable cvOverflow_;
|
|
|
|
std::vector<std::vector<TensorCPU>> queue_;
|
|
};
|
|
} // caffe2
|