Summary

We hope this primer will serve as a useful starting point for some future content in the coming weeks and months. Furthermore, the trade-offs between specialisation and unification (and how that affects programmability and vice-versa) are certainly at least as important as those involving parallelism. So we certainly won’t just look at serial performance and restrictive forms of parallelism, but also original ways to specialise programmable hardware and many other interesting ideas and trade-offs.

And most importantly, there are plenty of aspects that don’t fit nicely into these little boxes - ideas that aren’t just more specialised or restrictive, but definitely very smart and effective. They are still necessarily affected by the trade-offs explained in this primer, though, and so should still be useful to set the terms of the debate. If you talk about programmability, you need to consider specialisation/unification trade-offs at a very deep level. If you talk about parallelism, you need to think about data dependencies. They’re not the only factors, but it is hard to imagine an analysis without that to be of much value.

Conclusion

Next up, we’ll be looking at the 3G/4G baseband architecture from Icera. It is a fully programmable solution, unlike the incumbent products it fights against, and provides a very interesting case study when it comes to specialization and exotic forms of explicit parallelism. After that, we’ll look at picoChip; they have a very interesting approach to programmer-friendly parallelism, and their architecture is also designed for the 3G/4G market, although aimed at the infrastructure and femtocell markets.

Afterwards, we’ll hopefully examine plenty more examples in every field, although often still with an emphasis on wireless technology; this is obviously far from being the only place where all of this matters, but it is definitely where the most venture capital (and/or high-risk capital inside a company) has gone in recent years because it has been the largest growth sector in the semiconductor industry. So whenever someone has had an interesting architectural idea, that’s often where they decided to apply it first.

No promise on how deeply each architecture can be analysed, but here’s hoping you’ll enjoy all this in the coming weeks and probably months. At the very least, we expect an article on Icera to be online early next week.

Feedback? Feel free to post it in our forum thread!