ARM Community: Fast Model Evaluation: Helping hardware & software designers sleep better - ARM Community

Virtual Platforms and Fast Models are gaining more advocates: why is this?

It’s probably not the most scientific metric ever. However, if the volume of email, Skype, phone calls and other assorted contacts crossing my desk is anything to go by then the interest in abstract modelling and virtual platforms is escalating rapidly. ARM Partners from all corners of the planet engaged in designs across a broad spectrum of applications are using or evaluating the use of a virtual platform as part of their software and system design process.


When I spent a few minutes - with Outlook switched off - to think about the factors driving this demand it was clear that there are a few recurrent themes and there are a set of conditions in the market place creating (cliché alert) The Perfect Storm.

Each partner will have their own set of criteria that drives the need to adopt the technology, but in the end it comes down to this. The rewards of success are there to be enjoyed but the challenges to overcome drive the need for new weapons in the arsenal. Designs are getting more complex, software content is doubling every ten months, competition is becoming fiercer, end users are demanding more and more from their mobile devices and time to market is more critical than ever. It’s enough to make you lose a few nights’ sleep.

Virtual targets
Virtual targets have been seen for a long time as one of the ways to address these challenges. On paper high performance simulation models of the system under design, offer significant benefits in software development and system analysis. The barriers to adoption have made them inaccessible to many putative users. Are they fast enough? Are the models available? How do I assemble the system? Are they close enough to hardware to make them worthwhile?

These are all valid concerns, but concerns that ARM and the ARM Connected Community are working hard to address. At the heart of the virtual target you need fast, accurate, reliable models of the processors (there are usually a few of those). These models need to be available when they are most needed: before there is silicon, better still, before you start designing the silicon. As the designer of many of those CPU cores, this is what we deliver. We use them internally as part of the IP, and software, design process on new cores. The cores are combined into reference platforms – equivalent to an evaluation board – to enable this early development. These reference platforms are very useful, out of the box, examples to enable software development. But many partners will want to model the complete SoC and for this the processor model, or processor subsystem needs to be integrated into a broader platform. To make this happen the models support standard interfaces: OSCI SystemC TLM 2. This empowers the ARM Connected Community to integrate them into increasingly capable and broad featured tools solutions. These solutions help remove the barriers of the model availability and system assembly as well as offering a cornucopia of analysis and debug views of the software running on the virtual target.

And does it deliver what is needed? All that mail traffic suggests that the answer must be yes.

Recently I have been spending a lot of time running Android on a virtual target of the big.LITTLE^TM architecture that ARM announced last year. Android is very good way of stressing the models in the virtual target and a good demonstrator of capabilities. The big.LITTLE model is capable of booting Android in ~2 minutes on the laptop I use for my everyday work. Not real time, but very usable as a software platform. Linaro have done a lot of work in optimising the Android image for virtual targets and we have been improving the performance of the underlying models. Fancy graphics and sounds during the boot may make a Samsung Galaxy visually appealing but they aren’t that relevant to the software developers. Cutting a few bells and whistles simplifies the software and improves the performance so that designers can focus on what really matters. I set the 0xBenchmark running on the model in an endless loop before I left work the other night. The next morning the model was still happily running – and producing correct results - after executing several trillion instructions: proof of performance and robustness.

Conclusion
The question is still there: “does the return on investment that I need to make in this technology justify the outlay?”. As the cost of ownership comes down, the capabilities grow and the returns increase the answer is a hearty “yes” in many more cases. Virtual targets are not a panacea. They won’t solve every problem for every partner. But they should help a lot more hardware and software design teams get a better night’s sleep.

It’s time to risk another look at Outlook. Now, if only I could develop a virtual persona that would answer all those emails for me...

Rob Kaye, Technical Specialist for Fast Models, ARM, recently celebrated 30 years in and around the semiconductor industry. Rob joined ARM 6 years ago and for the last two years has been focused on the modelling solutions. Prior to joining ARM Rob had lengthy spells at Mentor Graphics and Texas Instruments in a wide variety of roles and locations around the world.