[quick start of Digital IC Verification] 17. Basic grammar of SystemVerilog learning 4 (randomization)

Reading guide ： The author has the honor to be a pioneer in the field of electronic information in China “ University of electronic technology ” During postgraduate study , Touch the cutting edge Numbers IC Verification knowledge , I heard something like Huawei Hisilicon 、 Tsinghua purple light 、 MediaTek technology And other top IC related enterprises in the industry , Pairs of numbers IC Verify some knowledge accumulation and learning experience . Want to get started for help IC Verified friends , After one or two thoughts , This column is specially opened , In order to spend the shortest time , Take the least detours , Most learned IC Verify technical knowledge .

One 、 Description of content

• Why use a randomized validation strategy
  • 1、 There is a lot of room for verification （ Time for space , Time is sacrificed ; Different time covers different space ）
  • 2、 Boundary conditions are unexpected , Randomization motivates rich , Ensure the completeness of verification
• SystemVerilog Randomization in
  • Range 、 The weight
• In the class （class） Declare random variables in
• Set constraints for random variables （Constraints）
• The weight distribution of random variables （weighted distribution）
• Turn off or on randomization and constraints （rand_mode() and constraint_mode()）

Two 、 Randomized validation strategy

2.1、 Why use a randomized validation strategy （ important ）

• After the design complexity increases , Direct tests （ Directional test ）（directed testcase）, There is no way to verify all vectors by exhaustive method
• Directional test cases are used to check the determined design properties , Only used to check for predictable errors
• The directional test method and test time are linear Relationship
• Directional test cases Invisible errors cannot be checked

programme ： Randomized test cases with constraints , Excitation randomization of input

The randomized verification strategy can detect unexpected errors and hidden errors in the design

2.2、 What are the random contents （ abstract , understand ）

notes ： Random content needs to be based on reality DUT Requirements to determine ！

• RTL Design configuration information

  • Different design configurations （ randomization ）
  • For example, verify a router , Different numbers of input and output ports can be configured

• Verify the configuration of the environment

  • Configure the whole environment randomly
    • Input data quantity
    • Enter the data type

• Main input data

  • The input data is processed randomization
  • In the input data of Within the effective range Randomize

• Encapsulated input data

  • If the data is encapsulated layer by layer , Different layers of packaging Can be randomized , such as TCP/IP（ Network message ）

• Agreement exceptions , Fault tolerant processing and protocol violations （DFX）

  • Verify how the system handles errors
  • Expected error type , And introduce the error into the system , Ensure that the system design can correctly handle these errors
  • Insert these errors randomly

• Time delay （ Clock cycle ）

  • Randomly insert the time delay according to the Protocol （latency）
  • Check the sensitivity of the design to clock cycles
  • There is no need to verify the establishment and holding time

2.3、SystemVerilog The randomization of

• randomization Allow users to automatically generate random input incentives , Used to verify functions
• SystemVerilog Allow users Use specific constraints , Keep the number of randomly entered data pages in a valid range
• must stay OOP Specify random constraints in （ In a Class in , Grammatical rules ）

2.3.1、rand A random variable

• The value of the random variable is within the specified range Uniform distribution
• If Do not add constraints , The value of a random variable can be Specify any value within the valid range

2.3.2、randc A random variable

• Periodic random variables use keywords randc Make a statement
  • Its value depends on the declared valid range Periodically
  • The data type can only be bit perhaps enum
  • randc The random variable repeats all values in the range , In the process of sequential circulation , The same value will not appear repeatedly （ After a circular introduction , The new cycle starts automatically ）

2.3.3、 Classes with random variables

• Bus Class contains two random variables ：addr and data
• constraint be known as range1 It specifies addr The numerical range of
  • Ensure that the constraints do not conflict

2.3.4、randomize() function （ Start random variables ）

• call randomize() Functions can be all random variables in the object assignment
  • Start random generation ： Handle to class .randomize();, Each call produces a random result
• The value of random variable should Compliance constraint
• randomize When the function succeeds, it returns 1, Return... On failure 0
• If the random variable No constraints added , Then its random value can be any value within the effective range

• produce 50 individual addr and data
• b.randomize() The call time is right b The value of the memory space variable referred to by this handle is randomized

2.3.5、 Constraint interpreter

• Analyze the relationship of constraints

• The same seed （seed） Generate the same random number （ Pseudorandom ）

  • Use different seeds , You can generate a different set of random numbers

• Different EDA The constraint parser of tool manufacturers is customized

2.3.6、constraint Constraint statement block

• For random variables Value Limit
  • You can also be right about The relationship between different variables Constraint
  • The constraint statement block is Class members , Be similar to task,function And variables
  • Constraint statement block Statement
    • Identifier of constraint statement ： Indicates the name of the constraint statement block
    • Constraint statement block ： Is a list of expression statements , Limit the value range of variables or the relationship between variables

• Above addr It's not a random variable , So the constraint will fail

Q： For large space validation , Even if random verification is adopted , There must be places that cannot be covered , For areas not covered , How to ensure its correctness

• A： Random coverage + Directed use cases . Look at the function coverage to judge whether the random points have been reached , If the function coverage does not reach , Adjust the random range or constraint, If not , Then add directed use cases ！ Functional coverage is achieved through feature To decompose .（ Random + Functional coverage ）
• Functional coverage CDV（coverage driven verification）

2.3.7、 Randomized constraints ： Simple expressions

• Constraint variables must have a fixed order （fixed order）
  • Only One Relational operator ：<, <=, ==, >=, >
  • Multiple variables use multiple expressions

2.3.8、 Randomized constraints ： Set the range operator

• If there are no other constraints ,inside The probability that any value within the range of values represented by the operator will be selected is the same
• inside The negation operator of the operator is !, Indicates that the value range is no longer inside Within the range indicated by the operator

• ! inside Means not in inside Within the scope of

2.3.9、 Randomized constraints ： Weight distribution

• Numerical distribution operator ：dist
  • Add weight to some values （ Purpose ）
    • attribute 1： Test cases need relevant values
    • attribute 2： Specify a certain numerical distribution for the test results
  • Two operators ：:= and :/
    • :=： The operator means that the specified value has the same distribution weight
    • :/： The operator means that the specified value divides the weight equally , If the weight is w, The value has n individual , Then the weight of each value is w/n
  • The value can be a number , It can also be a numerical range ：[lo:hi]
  • The weight is not a percentage , It doesn't have to be 100
  • randc Keyword declared random variables You can't Set weight

2.3.10、 Randomized constraints ： Bidirectional constraint

• Constraint statements are not procedural statements （procedural） But declarative statements （declarative）
  • All constraint statements take effect at the same time

2.3.11、 Randomized constraints ： Conditionality

• Constraint statements provide two syntaxes for declaring conditional relationships ：
  • ->
  • if ... else ...

2.3.12、 The result possibility of randomized constraints ： Unconstrained situation

• Possible distribution of values

• Equal probability distribution of each combination

2.3.13、 The result possibility of randomized constraints ： There are constraints

Conditional operators ：->

• -> Operators will affect the distribution of values
• -> Operators are bidirectional operators

• y The value of depends on x, When x = 0 when ,y = 0;
• therefore , When x = 0 when ,y It is impossible to take other values , namely x = 0 also y != 0 The odds are 0

-> It is the understanding of bidirectional constraints ：

• The newly added constraints will affect the distribution

• When x=0 when ,y=0, But when y=0 when , dissatisfaction y>0 Conditions , therefore x Not for 0,x Only for 1！

Conditional operators ：solve … before …

• solve ... before ... grammar The valid range of the value will not be changed , But it will change the probability of occurrence of the value

• solve y before x Don't write this sentence , Except that it won't appear 3 In this case , rest 5 The probability of this situation is 1/5;
• Yes solve y before x This sentence , Will give priority to y The combination group that appears , here y Yes 4 Species value , So it can be divided into 4 Group , Two in each group . And the probability of each group is 1/4, If a group of two situations will occur , Then the probability of each case is 1/8, If there is only one case in a group , Then the probability of this case is 1/4！

Iterative constraints ：foreach

• about Array Variable , have access to Loop variable Constraint

• Remember the core ： From left to right

2.3.14、 Randomized constraints ：functions

Functions in constraint statements

• Some attributes cannot be constrained by simple expressions
  • For example, use a loop to calculate the sum of values in an array
  • If you don't use loops , Then you must expand the loop
• SystemVerilog Expressions in constraint statements can call functions
  • Function It cannot contain output perhaps ref Parameters
  • function Must be automatic
  • Constraint statements in functions You cannot modify other constraint statements
  • Must be in Call the function before the constraint is valid Count , And function The return value must be treated as a state variable
  • Random variables as parameters of a function must create an implicit variable order or priority
    • As in the following code y, You must create a constraint ！

2.3.15、 Randomized constraints ： Restraint protection

• Constraint protection is expressed as prediction , Its function is to create protection constraints

  • Restraint protection It is not a logical relationship that the constraint interpreter must satisfy
  • Restraint protection can Prevent the constraint interpreter from generating error data

• Constraint protection is implemented before constraint resolution , It mainly includes ：

  • constant
  • State variables
  • Object handle comparison

2.3.16、 Enable or disable random variables （randomize Large switch of ）

• Use rand_mode() Function can turn off random variables
  • You can control random variables to start （active） Or shut down （inactive）
    • When the random variable is inactive , Indicates that the variable is not declared rand or randc
    • rand_mode() The function is SV Built in functions , Can't be covered

2.3.17、 Activate or deactivate constraints （constraint Large switch of ）

• Activate or deactivate constraints
  • Inactive constraints cannot be invoked randomize() Function implementation randomization
  • All constraint initial states are active
  • constraint_mode() The function is SV Built in functions , Can't be covered

• Use format ： Handle . Constraint name .constraint_mode()

2.3.18、 Test questions

• The pointer may be null, Need to add a judgment protection