Methods to reduce the number of pipeline stages
Several techniques have been proposed to reduce the number of pipeline stages.
We categorize them into two types based on their design philosophy: stage merging and stage bypassing.
Stage merging combines multiple stages of the traditional design into a single stage by using faster components or through speculation.
Faster components: allows the merging of VA and SA, LT and ST into the same stage
Speculation: With it, a flit ignores the dependence between various steps and speculatively（预测） executes multiple steps in parallel . If speculation fails, which rarely happens, the router needs to roll back and reexecute the steps sequentially.
Stage bypassing skips pipeline stages in intermediate routers when there is no load ahead or by setting up downstream routers before a flit arrives.
for example, ST and LT are merged.
Multiple intermediate routers are bypassed when there is no contentions in an advanced design SMART NOC.
referenced in
Reducing wire and energy overheads of the SMART NoC using a setup request network
Methods to reduce the number of pipeline stages的更多相关文章
- Hierarchical Tree Traversal in Graphics Pipeline Stages
BACKGROUND Many algorithms on a graphics processing unit (GPU) may benefit from doing a query in a h ...
- 【LeetCode】1404. 将二进制表示减到 1 的步骤数 Number of Steps to Reduce a Number in Binary Representation to One
作者: 负雪明烛 id: fuxuemingzhu 个人博客:http://fuxuemingzhu.cn/ 目录 题目描述 题目大意 解题方法 转成十进制模拟 修改二进制字符串 日期 题目地址:ht ...
- Snoop resynchronization mechanism to preserve read ordering
A processor employing a post-cache (LS2) buffer. Loads are stored into the LS2buffer after probing t ...
- Satisfying memory ordering requirements between partial reads and non-snoop accesses
A method and apparatus for preserving memory ordering in a cache coherent link based interconnect in ...
- (转) Ensemble Methods for Deep Learning Neural Networks to Reduce Variance and Improve Performance
Ensemble Methods for Deep Learning Neural Networks to Reduce Variance and Improve Performance 2018-1 ...
- A trip through the Graphics Pipeline 2011_13 Compute Shaders, UAV, atomic, structured buffer
Welcome back to what’s going to be the last “official” part of this series – I’ll do more GPU-relate ...
- A trip through the Graphics Pipeline 2011_07_Z/Stencil processing, 3 different ways
In this installment, I’ll be talking about the (early) Z pipeline and how it interacts with rasteriz ...
- Spark.ML之PipeLine学习笔记
地址: http://spark.apache.org/docs/2.0.0/ml-pipeline.html Spark PipeLine 是基于DataFrames的高层的API,可以方便用户 ...
- Methods and Systems for Enhancing Hardware Transactions Using Hardware Transactions in Software Slow-Path
Hybrid transaction memory systems and accompanying methods. A transaction to be executed is received ...
随机推荐
- [leetcode]114. Flatten Binary Tree to Linked List将二叉树展成一个链表
Given a binary tree, flatten it to a linked list in-place. For example, given the following tree: 1 ...
- 集成bug统计链接
http://crab.baidu.com/http://bugly.qq.com/ http://bughd.com/ http://www.umeng.com/analyticshttp://tr ...
- android的Afinal框架下的数据库更新
项目需要,版本升级时给表添加了一些新的字段,发现出现异常. 解决方法:监听数据库的版本号,更新数据库. 创建FinalDb对象时使用如下的构造函数,监听版本号的变化: db = FinalDb.cre ...
- illegal multibyte sequence python3
错误信息如下: page_data = m.decode('gbk') UnicodeDecodeError: 'gbk' codec can't decode byte 0xa7 in positi ...
- node.js中对同步，异步，阻塞与非阻塞的理解
我们都知道javascript是单线程的,node.js是一个基于Chrome V8 引擎的 javascript 运行时环境,注意 node.js 不是一门语言,别搞错了. javascript为什 ...
- JFinal Web开发学习(三)前后台路由设计
效果图: 一.写控制器 1.在controller包中新建AdminController后台控制器,继承Controller,实现一个index方法,作为的处理方法. /admin 后面,这个控制器中 ...
- pom.xml中坐标的组成
坐标=组织(也就是所谓的公司名称)+项目名称+版本(如果不加范围默认为compile)
- c#devexpress GridContorl添加进度条
demo 的实现图 下边是步骤和代码 1定义 时钟事件,定时的增加进度条的增量. 2: 添加进度条 3;定义字段属性 using System; using System.Collections.G ...
- C# Contains 包含空字符串的问题
一个基本的条件判断,之前没有遇到,这次遇到后,感觉真是这些年白写程序了. if(("1,2,3").Contains("")) { MessageBox.Sho ...
- Junit4知识梳理
一.junit官网 junit4:http://junit.org/junit4/ junit5:http://junit.org/junit5/ 二.github junit4: https://g ...