并行执行

Based on a lot of experiment results, a conclusion is drawn: comparing with other factors, the performance of branch handling strategy is the key limits of processor to exploit the instruction level parallelism existed in nonscientific code, cache miss have severe effect on superscalar processor's performance when it runs scientific code. Second, in order to reduce the branch penalty and improve the performance of superscalar processor, a new branch handling strategy—a classification based hierarchical branch handling strategy, CHBHS is proposed. It first expands the traditional processor architecture to support multiple condition code, conditional execution and Mbranch instruction, as a result, compiler can reduce the number of static conditional branch when the code is generated. Then, CHBHS tries to use the best suitable mechanism to deal with different branch base on their different behavior. CHBHS can predict the target address of unconditional branch, subroutine call and conditional branch by buffering their target address in branch target buffer, a newly proposed high efficient return address stack is used to reduce the penalty of subroutine return instruction, a new Counter Register Stack is also proposed to reduce the penalty of loop-closing branch to zero, and dynamic branch predictor is incorporate with branch target buffer to predict the outcome of conditional branch.

基于上述结论，为了尽量消除转移指令对处理器开发指令级并行性能力的影响，进一步提高处理器性能，在详尽分析目前已存在的转移处理策略的特点与局限性的基础上，首次提出了一种新的转移处理策略即基于分类的层次转移处理策略CHBHS(Classification Based Hierarchical Branch Handling Strategy)，它首先通过扩展传统的体系结构，支持多条件码、条件式执行及多分支转移技术，以使编译程序在进行代码生成时可尽量少生成条件转移指令，从而减少静态条件转移指令的数目；其次，基于不同的转移指令的行为不同这一事实，提出了对不同的转移指令采用不同的机制进行处理的思想，即对无条件转移指令和函数调用指令以及条件转移指令的目标地址，采用转移目标缓冲器来预测，对于函数返回指令，采用所提出一种的高效返回地址栈来预测其目标地址，对于大多数循环控制转移指令，采用所提出的Counter Register Stack来将其所可能带来的损失减少为0，对于其他的条件转移指令采用动态预测机制来预测其方向。

In recent years, with the increasing of the processor's frequency and the parallel degrees in processor's instruction executing, the processor's memory access operations expand greatly and bring great pressure to computer systems.

近年来，处理器内核的工作频率及处理器内部指令执行并行度的不断增加，使单位时间内处理器内核的访存频度不断增加，导致处理器对访存效率的要求也越来越高。

The high productivity and efficiency is supported by both fast matching algorithm and scalable fault-tolerant computer cluster software, which allows to perform fast parallel fingerprint matching, processes high number of identification requests and handles databases with practically unlimited size.

高生产力和效率是都支持快速匹配算法和可扩展的容错计算机群集软件，它允许执行快速并行的指纹匹配，过程高的数目鉴定的要求和处理数据库与可能性实际上是无限的大小。

One might want to perform parallel compiling, but not do parallel linking, because linking is i/o bound and only gets slower.

你可能想执行并行编译，但不可以并行链接，因为链接是 i/o 限制的，只会更慢。

The paper suggests that during the process optimization, in order to reduce the number of tasks, limiting the parataxis routes is effective, as well as replace that with the combined entity of sequence route and optional route that behaves in special conditions.

本文认为，在流程优化中，为了减少整个流程的任务数量，必须尽量减少并行路由，或者将并行的路由改为顺序路由和在一定条件下才能执行的选择路由的结合体。

In this paper, a specification method and performance evaluation model of large-scale parallel application systems is advanced, based on modules and their composition.

该文提出了一种基于模块及其组合关系的描述方法和系统执行代价计算模型，它不仅能描述并行应用系统的多范例特征，而且将不同并行范例模块的组合时产生的代价引入模型。

Core can speculatively move the blue load instructions before the red stores, and can make full use multiple execution units.

Core可以投机的将所有兰色的读操作提前到所有写操作之前执行，从而并行利用多个执行单元。

The uncertainty of running time of randomized algorithm s provides a better opportunity for asynchronized parallelization.

随机算法的执行时间具有不确定性，这种不确定性为随机算法的异步并行提供了良好的基础，已有许多计算实验表明了随机算法的异步并行可以达到线性甚至超线性的加速。

Task allocation and workload dispatchment become the key factors of whether the redundance resource and potentiality of the system can be made the most of or not.

任务调度是并行计算中最基本、最为关键，也最具挑战性的问题之一，是影响并行计算执行效率的一个关键因素。

Computation and data decompositions are key factors of affecting the performance of parallel programs running on distributed memory multicomputers.

计算与数据划分问题是影响并行程序在分布主存多处理机中执行性能的重要因素，也是并行编译优化的重点。

Do not gum up the work by telling mother what we are going to do .

别告诉母亲我们要干的事，那会把事情搞糟的。

In recent years, application of the partial prestressed concrete beam s is developed.

近年来，随着部分预应力砼梁应用的推广，发现我国规范所采用的名义拉应力法的计算结果不稳定。