/*******************************************************************************************/
提高代码运行速度可以采用以下方法：
一是运用好的platform：
Multi-core platforms
Many-core platforms
Cloud platforms
二是优化数据结构，算法，software architecture（这是个啥？？？？）
/*******************************************************************************************/
第一章 简介
Computing platform的变化趋势： transistor integration（晶体集成）指数增长，clock speed，power consumption(功耗)，performance per block逐渐趋于稳定
HPC: high performance computing
multi-core 是 each core optimized for executing a single thread
many-core是针对总输出进行优化的，并不注重单个thread的performance
multicore和manycore parallelism 的共同点：
都increase vector unit width（是啥？）
都increase number of cores per die（又是啥？）
都increase bandwidth to off-cheap memory（是啥？？？）
/*******************************************************************************************/
第二章 Multi-core
load-store architecture：MIPS架构（英语：MIPS architecture，为Microprocessor without Interlocked Pipeline Stages的缩写，亦为Millions of Instructions Per Second的相关语），是一种采取精简指令集（RISC）的处理器架构，1981年出现，由MIPS科技公司开发并授权，广泛被使用在许多电子产品、网络设备、个人娱乐装置与商业装置上。最早的MIPS架构是32位元，最新的版本已经变成64位元。
Instruction level parallelism:
出发点：使得给定的single， sequential stream of instructions运行得as fast as possible
优点：no changes in sequential software necessary.
确定：处理器构架明显更复杂
所需设计时间更长
处理器设计正确性的验证所需时间更长
和single in-order的处理器相比消耗更大
一共有四种Taxonomy：
SISD: 无parallelism
SIMD: explicitly exploits data parallelism
MISD: redundancy（用在控制space shuttle flight control中）
MIMD: distributed computing
16-way SIMD 是指width是16？？？？？？
__128： 128-bit operands
SIMD:
出发点：reduce processor complexity by explicitly representing instruction level parallelism in vector form
优点：power-efficient way to improve instruction throughput
exploitable in many compute-intensive applications
缺点：may require hand-tuning(是啥？？)to exploit full benefit
software requires re-compilation to take advantage of new SIMD capabilities(为啥是缺点？？)
explicit representation in vector instructions(为啥是缺点？？)
simultaneous multithreading：
出发点：当不知一个thread of instruction可以用时提高performance
优点：gain power efficiency by increasing increase pipeline utilization
缺点： requires multithreads available
may trigger conflicts in shared cache during execution
does not improve latency of each thread
Latency: from start to finish, in units of time. e.g. CPU Clock Cycles
Throughtput: tasks accomplished per unit time. e.g. GB/s
/*******************************************************************************************/
先把问题列出来：
What are the differences between multi-core and many-core processors?
What is instruction level parallelism? What is SIMD?
What is simultaneous multi-threading?
What are the three metrics for a memory hierarchy?
What are the different system granularities?
How is this relevant to writing fast code?
recursion可以使code简单，但是占据stack太多可能over flow
static, dynamic?? type checking??
type checking is harder to debug
git merge, git rebase????
rationale database
garbage collector
18645 How to write fast code期中准备,布布扣,bubuko.com
18645 How to write fast code期中准备
原文：http://blog.csdn.net/he_wolf/article/details/20119549
