1. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Embedded Computer Architecture 5SAI0 Simulation - chapter 9 - Luc Waeijen 16 Nov 2015

2. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved How to study a computer system Methodologies ➢ Construct a hardware prototype ➢ Mathematical modeling ➢ Simulation

3. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Construct a hardware prototype Advantages ➢ Runs fast Disadvantages ➢ Takes long time to build - RPM (Rapid Prototyping engine for Multiprocessors) Project @ USC; took a few graduate students several years ➢ Expensive ➢ Not flexible

4. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Mathematically model the system Use analytical modeling ➢ Probabilistic ➢ Queuing ➢ Markov ➢ Petri Net Advantages ➢ Very flexible ➢ Very quick to develop ➢ Runs quickly Disadvantages ➢ Can not capture effects of system details ➢ Computer architects are skeptical of models

5. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Simulation Write a program that mimics system behavior Advantages ➢ Very flexible ➢ Relatively quick to develop Disadvantages ➢ Runs slowly (e.g., 30,000 times slower than hardware)

6. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Most popular research method Simulation is chosen by MOST research projects Why? ➢ Mathematical model is NOT accurate ➢ Building prototype is too time-consuming and too expensive for academic researchers

7. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Simulation Bottleneck 1 GHz = 1 Billion Cycles per Second Simulating a second of a future machine execution = Simulate 1B cycles!! Simulation of 1 cycle of a target = 30,000 cycles on a host 1 second of target simulation = 30,000 seconds on host = 8.3 Hours CPU2K run for a few hours natively Speed much worse when simulating CMP targets!! 7

8. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Simulation Bottleneck 1 GHz = 1 Billion Cycles per Second Simulating a second of a future machine execution = Simulate 1B cycles!! Simulation of 1 cycle of a target = 30,000 cycles on a host 1 second of target simulation = 30,000 seconds on host = 8.3 Hours CPU2K run for a few hours natively Speed much worse when simulating CMP targets!! 8

9. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved How to overcome simulation bottleneck Gate level (RTL) Cycle accurate Functional level (ISA) DetailSimulation speed trade accuracy for simulation speed

10. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved How to overcome simulation bottleneck Gate level (RTL) Cycle accurate Functional level (ISA) Model based approximation DetailSimulation speed trade accuracy for simulation speed This trade-off has resulted in a plethora of simulators

12. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Tool classification OS code execution ➢ System-level (complete system) - Does simulate behavior of an entire computer system, including OS and user code - Examples: – Simics – SimOS ➢ User-level - Does NOT simulate OS code - Does emulate system calls - Examples: – SimpleScalar

13. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Tool classification Simulation detail ➢ Instruction set - Does simulate the function of instructions - Does NOT model detailed micro-architectural timing - Examples: – Simics ➢ Micro-architecture - Does clock cycle level simulation - Does speculative, out-of-order multiprocessor timing simulation - May NOT implement functionality of full instruction set or any devices - Examples: – SimpleScalar ➢ RTL - Does logic gate-level simulation - Examples: – Synopsis

14. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Tool classification Simulation input ➢ Trace-driven - Simulator reads a “trace” of inst captured during a previous execution by software/hardware - Easy to implement, no functional component needed - Large trace size; no branch prediction ➢ Execution-driven - Simulator “runs” the program, generating a trace on-the-fly - More difficult to implement, but has many advantages - Interpreter, direct-execution - Examples: – Simics, SimpleScalar…

15. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved

16. Interval Simulation

17. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Multi-Core Simulation Sequential simulation  All target cores are simulated in one thread (on one host core)  Unified memory hierarchy models simulate resource contention Parallel simulation  Each target core is simulated in separate thread

18. © Michel Dubois, Murali Annavaram, Per Strenstrom All rights reserved Multi-Core Simulation Sequential simulation  All target cores are simulated in one thread (on one host core)  Unified memory hierarchy models simulate resource contention Parallel simulation  Each target core is simulated in separate thread There is no relation between the number of target cores and the cores on the host! (except simulation speed)