1 / 32

Simulation Verification of Different Constraints in System Level Design in SystemC

Simulation Verification of Different Constraints in System Level Design in SystemC. Piyush Ranjan Satapathy CS220 Class Project Presentation. The Problem Statement. How can we detect deadlock in the system level design in systemC ?

blanca
Download Presentation

Simulation Verification of Different Constraints in System Level Design in SystemC

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Simulation Verification of Different Constraints in System Level Design in SystemC Piyush Ranjan Satapathy CS220 Class Project Presentation

  2. The Problem Statement • How can we detect deadlock in the system level design in systemC ? • What are the other possibilities of design constraints in systemC ? (E.g. Live lock, Starvation) • We know that system level designs are highly complex, heterogeneous and concurrent. so how can we detect the above constraints in the complex model of designing ?

  3. What is the solution ? • Analyse the synchronization dependencies of the systems designed in systemC • Capture the run time dependencies of various blocks in the system and maintain a data structure named Dynamic Synchronization Dependency Graph • Use a loop detection algorithm to detect the deadlocks. • Similar to Metropolis Approach but varies with the Environment and synchronization lists.

  4. Motivation • A System level designer deals with  Function Vs Architecture  Computation Vs Communication  Data Path Vs Control • Possible to introduce unintended and undesirable behaviors into function specifications, high level architectural models or functional mappings. • Deadlock, Livelock and Starvation are such undesired behaviors which designer may face at the runtime. • So detecting such a scenario by simulation helps us knowing the design faults and also helps to figure out the exact portion of the design/code to change.

  5. Road Map • Some back Grounds in SystemC • Synchronizations and Dependencies in SystemC • Constraints scenario in systemC • Deadlock Detection Data Structure • Loop Detection Algorithm • Algorithm in Action in 1 systemC Model • A Comparison with Related Work (MMM Environment) • Conclusion • Future Work

  6. Some back Grounds in SystemC Slide From http://www-cad.eecs.berkeley.edu/~polis/class/ee249/lectures/l10-systemC.pdf

  7. Slide From http://www-cad.eecs.berkeley.edu/~polis/class/ee249/lectures/l10-systemC.pdf

  8. Slide From http://www-cad.eecs.berkeley.edu/~polis/class/ee249/lectures/l10-SystemC.pdf

  9. Slide From http://www-cad.eecs.berkeley.edu/~polis/class/ee249/lectures/l10-SystemC.pdf

  10. SystemC Highlights • It supports Hardware-Software Co-Design. All Constructs are in a C++ Environment. It introduces • 3 kinds of processes such as Method, Thread, and Cthread • Ports and Modules • It introduces • Events (Flexible and Low Level synchronization Primitive) • Channels (A Container class for communication and Synchronization) • Interfaces (Specify a set of access methods to channel) • It supports • Clocks (time Keeper of Simulation) • Event Driven Simulation by Dynamic sensitivity of Events E.g. (Wait(), next_trigger(),wait_until)

  11. Slide From http://www-cad.eecs.berkeley.edu/~polis/class/ee249/lectures/l10-SystemC.pdf

  12. Synchronizations and Dependencies in SystemC Transferring Control of Execution From One process to Another Process 1. For Thread Process: • Wait() • Wait(e1) • Wait(e1 | e2 | e3) • Wait(e1 & e2 & e3) • For Method Process: • Next_trigger() : Returns immediately so no constraints here. • For Clocked Thread Process: • Wait_until() • Watching() 4. Access to shared data using Mutex through proper synchronization

  13. Synchronizations and Dependencies in SystemC Resuming the Execution of the Dependent Process • For Thread Process: • Notify() //Notify Immediately • Notify(SC_ZERO_TIME) //Notify Next Delta Cycle • Notify(t) //Notify at time t • Cancel() //Cancels a delayed notification • For Method Process: • Next_trigger method() returns immediately without passing the control to another Process • For Clocked Thread Process: • Boolean true value to wait_until() or watching() methods.

  14. SystemC Event Scheduler INIT Evaluate YES Immediate Notification ? Update Delta Event Notification ? YES Next Simulation Time Pending Events ? YES DONE

  15. P1 e P2 || P0 e e P3 p4 e

  16. P1 e P0 && P2 e P3 e

  17. Live Lock in SystemC Live lock is defined as a situation where the system falls into dead loop and responds to no further interrupts. Its defined as infinite cyclic executions of any events. Case1: Waiting events of the same process.. (But I have not tried) Case2: Data Watch in Cthread Process..What if Wait_until() and watching not get satisfied…It will loop infinitely. Because that is the property of the clocked thread process.

  18. Starvation in SystemC Starvation is defined as a situation where a process will be blocked infinitely. Live lock of some processes may cause starvation for the dependent processes. Case1: Notify() at a circular basis at immediate stage or at Next Delta cycle stage Case2: Data path blocking due to watch() or wait_until() or due to mutex sharing.

  19. Dead Lock Analysis Update DSDG DSDG Deadlock Detection Deadlock ? System level Design YES Deadlock Warnings? Compilation Output Dependencies Simulation Model Simulation vectors Simulation Deadlock Analysis Simulation trace Simulation Dependencies Analysis Report Revise the design and/or simulation vectors

  20. Data Structure (Dynamic Synchronization Dependency Graph) • Let’s represent the synchronization dependency as a directed graph S=(V,E) where V is a set of 4 categories of vertices representing processes in the system, one dependency, set of and dependency, set of or dependency. E is a set of directed edges between vertices indicating dynamic synchronization dependencies. P Process vertex && And vertex || Or vertex Single Dependency vertex e

  21. DSDG Examples

  22. Updating DSDG Dynamically Algorithm(1): Checking When to update and what to update For each process Pi in the system If Pi is in Evaluation Phase if Pi is unblocked by one or more synch constraints then Remove all the dependency vertices and edges from Pi end if if Pi is blocked by one or more synch constraints then UPDATE_PROCESS(Pi) end if end If End For

  23. Updating DSDG Dynamically Algorithm(2): How to update UPDATE_PROCESS(Pi) For each synch construct that blocks Pi do if Pi is blocked by any one of P1, P2,…Pn then add an OR dependency vertex Oi CONNECT(Pi, Oi, pj:j Є [1,n]) else if Pi is blocked by any all of P1, P2,…Pn then add an AND dependency vertex ai CONNECT(Pi, ai, pj:j Є [1,n]) else if Pi is blocked by one process PJ then add an single dependency vertes si CONNECT(Pi, si, pJ) end if End for End procedure

  24. Updating DSDG Dynamically Algorithm(3): Connecting… CONNECT (source,Mid, Desti: i Є [1,n]) add an edge from source to Mid if (i = = 1) then add an edge from Mid to Dest1 end if for i := 1 to n do add an edge from Mid to Desti end for

  25. Loop Detection Algorithm 1 DEADLOCK_DETECTION(S,P) 2 Search for simple cycles in S starting from process P 3 Let L = {Vi, Si} be the set of all these cycles 4 If L = Ф then 5 Return “No Deadlock” 6 End if 7 For each Li in L do 8 if the cycle is marked then 9 continue; 10 end if 11 Mark the cycle Li 12 if each vertex in the cycle Li are either process or and dependency or single vertex dependency then the process in Li are deadlocked, return; 13 else 14 D:= OR dependency vertices that have two or more outgoing edges 15 L’ = {Li} 16 repeat 17 Find unmarked cycles in L that contains vertices in Di 18 mark all these cycles 19 L’ = L’ U {these cycles} 20 until L’ becomes stable 21 If vertex in D that has an outgoing edge not belongs to L’ then 22 continue 23 Endif 24 The process in L’ are deadlocked , return 25 Endif 26 End for 27 Return No_Deadlock; 28 End procedure

  26. Algorithm in Action P3 e2 e3 P4 Example: Dining philosopher problem P2 e4 e1 e0 P0 P1 One Instance of Simulation: Got e4 Got e3 Got e2 Got e1 Got e0 e4 e3 e2 e1 P0 P4 P3 P2 P1 e0 Satisfied by Line 12 of algorithm..so deadlock….

  27. Wait(), notify(), wait_until(), watching() Based on systemC kernel scheduler Based on the top of the C++. So systemC is more approachable for system level design More or less C++ Library Loop Detection algorithm much easier compared to MMM Await(), synch, eval Based on metro Quantity manager Based on the Java Platform New keywords and syntaxes Comparison With MMM

  28. Conclusion • Detecting deadlock at early time of the design saves both in time and money. • In this project, I have studied the deadlock and other constraints in system level design in systemC • Implemented a prototype of deadlock and verified the modified algorithms

  29. Future work • More research on Live lock and Starvation • How to resolve the deadlock dynamically • Possible modify the simulator of systemC-2.0.1 to hold this deadlock detection feautre

  30. Acknowledgement • “Simulation Based Deadlock Analysis for system Level Designs” by Xi Chen, et al. DAC’0 • http://www-cad.eecs.berkeley.edu/~polis/class/ee249/lectures/l10-systemC.pdf • SystemC Func Guide and users guide (SystemC-2.0.1)

  31. Thank You !!

  32. P Process vertex And vertex && Or vertex || e Single dependency vertex A B

More Related