Video: QuantumLeap Parallel Simulation using multi-core host PC gaining significant simulation speed.


A demonstration video showing the Imperas QuantumLeap accelerator on several examples on a 4 core i7-3770 host PC to gain up to 16 billion Instructions a second.

The video show gains from using QuantumLeap on peripheral accelerators.

It then covers booting SMP Linux on an ARM VersatileExpress virtual platform that incorporates a Cortex-A9MPx4 and shows running 4 applications in parallel under the simulated Linux and shows the QuantumLeap performance increase.

The video concludes showing AMP peak performance of over 16 billion instructions per second on a Cortex-A9:

This video has several sections. Its contents are listed after the image below. Click on the section links to view the video from the start of that section.

OVP QuantumLeap1 Video

Introduction
  • [00:00] Start & Introduction
  • [00:22] Faster Simulators Find More Bugs

  • Imperas Simulators are Fast
  • [01:06] Table showing performance across many processor families
  • [01:53] ARM Cortex-A9UP running on average 6x faster than other simulators - single threaded
  • [02:16] Modern SoCs have many processors running concurrently SMP & AMP within them
  • [02:50] Single threaded simulations become challenged for many core platforms

  • The big challenge is getting simulators to scale
  • [03:53] The solution is to make use of the parallel desktop hardware: QuantumLeap

  • QuantumLeap features
  • [04:33] A fully deterministic solution with no changes needed to the models
  • [05:06] The QuantumLeap Synchronization Algorithmn

  • Demonstration 1 - Accelerating Peripheral and Hardware Algorithms
  • [05:38] Platform example walkthrough with 6 hardware accelerators as peripherals
  • [06:27] Running in a single thread - a lot of time simulating peripherals
  • [07:20] Analyzing the run performance
  • [07:45] Running QuantumLeap with MPonMP to accelerate the peripherals
  • [08:16] Analyzing the parallel simulation run performance of QuantumLeap

  • Demonstration 2 - Accelerating Symmetric Multicore Processor (SMP) designs
  • [08:35] ARM VersatileExpress platform with quad core Cortex-A9MPx4 processor
  • [08:35] The application runs booting Linux and then 4 x linpack benchmarks in a single host thread
  • [09:38] And when we run with QuantumLeap we get significant speed up
  • [10:20] Comparing single thread simulations to QuantumLeap parallel runs

  • Demonstration 3 - Accelerating Asymmetric Multicore Processor (AMP) designs
  • [11:13] 4 x Cortex-A9UP bare metal processors individually running dhrystones benchmark
  • [11:44] Running MPonMP in parallel with QuantumLeap significantly speeds up simulation
  • [12:08] Running peakspeed with QuantumLeap to see maximum speed of 13,000 MIPS

  • Summary
  • [13:30] Making use of your host parallelism using QuantumLeap
  • [14:43] More information on QuantumLeap