Click to enter 【 Chip design verification 】 Community , See more

---

  Statement ：

• 🥭 Author URI ：【MangoPapa Of CSDN Home page 】.
• ️ This article was first published in CSDN, Please indicate the source of reprint or quotation 【 Click to view the original 】.
• ️ The purpose of this column is Academic paper sharing , This paper is non-profit , Only used for Personal learning record And Knowledge sharing .
• ️ If the pictures or relevant references used in this article infringe upon your legitimate rights and interests , Please contact me to delete .
• Welcome to point out the mistakes in the article , Welcome colleagues to communicate with me ~
• mailbox ：[email protected]

---

---

Overview of the paper

   The title of this paper is Saving and Restoring Simulation Methodology using UVM Factory Overriding to Reduce Simulation Turnaround Time, The author is a Korean engineer of Samsung Electronics .

   This paper proposes and implements a dynamic SnR（save and restore） Method , It can accelerate the simulation turnaround time （simulation turnaround time）.

research objective

Warm up for two hours , Play for a minute

  SoC Simulation verification , Are you worried about Initialize the configuration for two hours , Personal test sequence Run for a minute The status quo of ？ For large SoC for , Running what we wrote test sequence Before , Complicated and time-consuming boot、initialization、phy training The process （ Hereinafter referred to as common sequence）, What we really care about test sequence A few minutes may be enough to run . stay case During commissioning , Every little bit case modify rerun You have to run it again when you are common sequence,common sequence Delete or not delete , Let us who love our work and pursue efficiency languish .

The original optimization method has many limitations

   There are many ways to speed up simulation ：

• simulation Speed up , There are many ways .
  • First, optimization code、flow, But this does not apply to good verification engineers , Unless you code、flow Sucks , Otherwise, expect to optimize code To obtain the simulation The acceleration effect is not very obvious .
  • Second, you can use save & store（ Hereinafter referred to as SnR） Method skip common sequence（ chart 1 Is not used SnR And adopt SnR Validation of methods Flow contrast , Figure 2 is SnR Override Sketch Map ）, But traditional SnR The method is only applicable to case、TB Under fixed conditions rerun, And only one time point can be specified SnR. One can specify SnR The number is a little small , both SnR It's hard to grasp the time , Third come Save Later, it will be used to override Of test sequence Can't make any changes , How can I play this ？
  • Besides ,simulation The performance itself is based on 10% Speed up , This heel SoC Size The speed of development is higher than , Not yet. .
• emulation.emulation Soon , but emulator One is Death is expensive , Second, it is generally used in the simulation of large scenes that take a long time , Small case The development process also Not very convenient Frequently put emulator Go for a run , Fear of being run away emulation My classmate called .

chart 1 Proposed Workflows with SnR Methodology

chart 2 Basic Concept of UVM Factory Override Mechanism

The new method

The method proposes

   This paper puts forward a kind of new SnR Method （ chart 3） To reduce simulation turnaround time , The method is still based on UVM factory Of override Mechanism , Use this method You can specify multiple SnR The point of , And stay case、TB It still applies when it is unstable . This method does not require additional hardware investment , Economic material benefit .

   When simulation Run to the designated point and turn DUT、TB Save status , be called snapshot.restore When , First compile your modified test sequence, And then use simulator Of tcl Order the old sequence override fall , Finally, put save Read in your file , From execution save Continue the simulation at the point in time . When the method in this paper is first run , Execute many times save The process ,restore You can specify save point. There are several points to pay attention to ：restore When , designated save point Need to be in test sequence Before the moment of execution .

chart 3 The Dynamic SnR Methodology

Method realization

Save

   In order to save the simulation scene , You need to do the following three steps ：

1. saving_test.sv Insert save point,save point The insertion position is common sequence after 、test sequence Before , Multiple... Can be inserted between them save point. This article is about DPI-C call do_save function , The argument to the function is save point The name of .
2. do_save.c To realize do_save function , Conduct simulation The preservation of state .
3. simulation flow In the open SnR function , Appoint save file Save path, etc .

  Save The schematic code of is shown in Figure 4.

chart 4 Saving Simulation and do_save() function

Restore

   In order to restore the simulation scene , You need to do the following three steps ：

1. test sequence Put it in test_pkg.sv in . Can be in base_sequence discharge common sequence, And from base_sequence Derive from test sequence. In order to achieve sequence Modifiable SnR Mechanism ,test sequence Need to put in SV package in （Top Level）, To share data between multiple environmental components 、 Parameters, etc. .SV package As restore simulation The input file of , stay restore Before re compile、elabrate these test sequence.
2. uvm_factory.tcl of use test_sequence override base_sequence. In order to achieve override_by_type,test_sequence and base_sequence Of type It needs to be the same , This is also the first step test_sequence from base_sequence Derived reasons .
3. simulation flow Specify configuration parameters in , for example sv package、tcl command、start point、datebase name etc. .

  Restore The schematic code of is shown in Figure 5.

chart 5 Restoring Simulation and Tcl command for UVM factory override

experimental result

   This paper sets up two groups of comparative experiments , First, whether there is SnR The simulation time of the method , The second is different common sequence（Block-A with boot,Block-B with boot + PHY init）. The test results are as follows （ chart 6） Shown .

chart 6 Test results

   As you can see from the diagram , have only Boot Of Blocak-A, Using this article SnR Method can TAT promote 88%;Boot+PHY init Of Blocak-B, Using this article SnR Method can TAT promote 97%

Discuss

   The method is very good , The implementation is not complicated , Economic material benefit , Support more save point, stay case It can also be used during debugging , The speed of improvement is also quite obvious , Can greatly speed up case Development iteration speed .

   This method makes some improvements , Use in power-aware sim、gate-sim or regression in , Save time not to say , It can also save a lot memory、licence、server resources .

   Pretty good , recommend .

End

---

Click to enter 【 Chip design verification 】 Community , See more

---

️ Return to the top ️