Hello , I'm Xiao Huang , A unicorn company Java Development Engineer . Thank you for meeting us in the vast sea of people , It is said that ： When your talent and ability , When it's not enough to support your dreams , Please calm down and learn I hope you can study with me , Work together , Realize your dreams .

One 、 introduction

about Java For developers , About the underlying knowledge , We usually use it as a black box , There's no need to open this black box .

But with the current development of the programmer industry , We need to open this black box , To explore the mystery .

This series of articles , Will take you to explore the mystery of the bottom black box .

Two 、 Recommended books

The principles of reading ： natural , Look at it

If you go into Lushan Mountain , Without saying , Get your head down , Bend down , Instead of studying the overall context of Lushan Mountain first , Then your learning method must be extremely inefficient and painful .

The most important thing is to slow down your enthusiasm , How can my study be so bad happy ah , Why is it so boring? That , Because your learning method is wrong , Read it in general , Use it first , Use to use , You can see a lot of truth .

• 《 code ： Language hidden behind computer hardware and software 》
• 《 Deep understanding of computer systems 》（ Not recommended ）
• 《 Introduction to algorithms 》、《Java Data structures and algorithms 》、《 The finger of the sword offer》
• 《30 Self made operating system 》
• 《TCP/IP Detailed explanation 》 Volume one
• Long Shu 《 Compiler principle 》

3、 ... and 、 Hardware Basics

1、CPU Production process of

CPU How it was made ？

I believe everyone will have such a question mark , Let me tell you today , be-all CPU It all comes from ： sand

about CPU The production process of , Here's an article , If you are interested, please have a look ：CPU How it was made

If you are not interested, look directly at the summary ：

• First step ： We provide... From the sand monocrystalline silicon Crystal
• The second step ： Cut the crystal into thin sheets , obtain wafer
• The third step ： Bombard metal particles onto wafers , Proceed again electroplate
• Step four ： On a wafer Photolith , Complete the wire interconnection between different transistors
• Step five ： Quality inspection , Remove poor quality CPU

2、CPU Principle

The first problem the computer has to solve ： How to represent numbers ？

The most primitive computers used light bulbs , When we calculate 0100 + 1010 when , We use 8 A light bulb , With The state of the bulb represents 0 and 1, So our first version of the computer has OK 了 .

• ENIAC weighing 27 Tons of , Covers an area of 1800 Square feet （ About us 167.2 Square meters ）. Born during World War II , It was originally used as a tool to assist artillery in calculating shell trajectory .

The first edition of the computer , There is a flaw that makes people laugh and cry .

When we do high-speed computing on computers , The light bulb flashes more frequently , It may cause damage to the bulb , It requires staff to replace the bulb in time , And that affects efficiency .

As the first computer , His appearance is enough to shock people on earth .

At present, most computers use transistors to calculate , utilize And gate 、 Or gate 、 Not gate 、 Or not State to represent different calculation methods .

We often hear in our lives that ,CPU32 position 、CPU64 position , Simply speaking , The difference between them is ： How many bits are read at one time （bit） The number of .

Any of our calculations , Can be obtained by logical operation , Let's look at the following logical operation ：0 && 1 = 0, How did it happen ？

First , Let's look at the circuit diagram ： This is a And gate Circuit diagram For this circuit ,A and B As input ,Q As the output .

for example A Input low level 、B Output high level , that Q It will output low level , Converting to binary is A Input 0、B Output 1, that Q Will be output 0, The corresponding logical operation expression is 0 && 1 = 0

Here's a story about BUG Source story ： Once upon a time there was a man , When doing computer calculations , Find that the numbers are always incorrect , I haven't found the reason for looking for it for a long time , Later, it was found that the computer had a hole covered with worms （BUG） Corroded , It makes it impossible to carry out low level 、 High level switching , From now on , Our programming mistakes are called ：BUG

3、 The execution process of assembly language

Let's think about , In the above circuit , It happened. 0 && 1 = 0 Such an event , How do my users know that this kind of event happened to the machine ？

We can't just take the machine apart , Look at the level conversion inside .

therefore , Here it appears assembly language , And the essence of assembly language is also as The mnemonic of machine language The emergence of

such as , Let's say to the computer , You go and calculate for me 1 + 2 This operation , The computer needs to carry out the difference between high and low levels and output the calculation results

And our assembly language ：mov、add、sub......

We can clearly understand the current operation state of the computer at a glance

Composition diagram of computer ：

Let's take a look at the whole process of computer calculation ：

Here's a description Java and C The difference of language ：

• C Language ： Let... Directly CPU Compile
• Java Language ： Need make JVM translate , To make CPU compile , This is what Java The key to cross platform

4、 Quantum computers

For quantum computers , At present, the world is exploring , No results yet

In our ordinary computers , A bit represents 1 perhaps 0,32 A bit , Can represent 2^32 Any number of

And our qubits , The highlight is , He can also say 1 and 0

• One qubit ：1、0
• Two qubits ：00、01、10、11
• Three qubits ：000、001、011、111......
• Thirty two qubits ： One time means 2 ^ 32 The number of

This description may not be intuitive , Let's look at an example ：

Now there is a number , We know that the number range is ： 1~2^32, How can we find the number quickly ？

For ordinary bits , Only one... Can be represented at a time , So we need to loop through 2^32 Time , To find the number

And for qubits , Use it directly 32 Bit operating system can complete

5、CPU Basic composition of

• PC（Programme Counter）： The address of the current instruction of the program technician
• Registers： register , Temporarily store CPU Calculate the data you need
• ALU（Arithmetic & logic Unit）： Arithmetic unit , Do calculations using
• CU（Control Unit）： control unit
• MMU（memory Mangagement unit）： Memory processing unit
• Cache： cache

5.1 ALU

Previous CPU It belongs to the case of single core , In this case , There will be only one Registers, our PC Will constantly switch to point to New Threads , Store the corresponding data in Registers, For handover （context switch） for , It will seriously affect our efficiency .

current CPU It's usually a multi-core state , When you do the calculation , We will have more than two Registers, In this case , our PC There is no need to switch frequently , our ALU Switch between processing and calculation .

5.2 register

5.3 Cache

We can see from the picture above , Our computer for the convenience of obtaining data , Added L3 cache , For different caches , The length of acquisition time is also different

For multi-core CPU Come on , As shown in the figure below ：

• L1、L2 Stored in different cores
• L3 Stored in the same CPU in

5.3.1 Locality principle

Simply speaking , our CPU While reading the data , Press fast to read the data , Do not take a single byte , As shown in the figure below ： Current CPU need X This goal is worth , Steps are as follows ：

• First step ： Look in the register for , Is there any X This field
• The second step ： Go to L1、L2、L3 Cache Go looking for X This field
• The third step ： De memory 、 Disk, etc X This field
• Step four ： After finding , Will be with X At the beginning 64 Bytes Form a block
• Step five ： stay L3、L2、L1 Store this data separately in , It's convenient to get the cache next time
• Step six ： take X Write to register , Data processing 、

5.3.2 MESI Cache Agreement of conformity

We can see , For the above two nuclei L1、L2 Cache lines should be consistent , A consistent agreement is called ：MESI Cache consistency protocol

CPU Every Cache line Mark four states

• M（ The modified ）： The Cache line It works , The data has been modified , Not consistent with the data in memory , Data only exists in this Cache in .
• E（ Monopoly ）： The Cache line It works , Data is consistent with data in memory , Data only exists in this Cache in .
• S（ share ）： The Cache line It works , Data is consistent with data in memory , Data exists in a lot of Cache in .
• I（ Invalid ）：Cache line It's invalid

Intel —— Cache line

• The larger the cache line , The higher the efficiency of local space , But read time is slow
• The smaller the cache line , The lower the efficiency of local space , But read time is fast
• Intel stipulated by experiment , The size of the cache row is ：64 byte

Bus lock （ When the cache line cannot be loaded , You have to lock the bus ）

One of the cache lock implementations , Some data that cannot be cached or data that spans multiple cache rows , The bus lock must still be used

How did we test our guess ？

We tested two programs

Limited space , If the source code is not displayed here for the time being , Interested parties can pay attention to the official account. , reply ： Algorithm source code

• The first program ： The value changes in the same cache line

• Second procedure ： The change of value is not in the same cache line

The above program verifies our conjecture , Two threads are frequently faster, and the cache in the cache is faster , Resulting in longer running time

5.3.3 Cache line alignment

For some particularly sensitive numbers , There will be highly competitive access by threads , In order to ensure that there is no false sharing , We generally do not require cache line alignment

Simply speaking , We don't want us to get X Numbers at the same time , hold Y Also get in

In us JDK7 and disruptor All take long cache line padding

public long p1, p2, p3, p4, p5, p6, p7; // cache line padding
private volatile long cursor = INITIAL_CURSOR_VALUE;
public long p8, p9, p10, p11, p12, p13, p14; //cache line padding
 Copy code 

In this case , When we cache row fetches , It will cursor Ahead long perhaps hinder long Load into cache block , avoid cursor Cache line alignment

In our JDK8 in , We can add... To this parameter @Contended（ According to the bottom CPU To set , Ensure that two parameters do not share a cache line ）, Need to add -XX:-RestrictContended take effect