Amit Rohatgi, vice president of segment marketing, at Imagination, acknowledged that while 80 to 85 percent of Android apps run well on MIPS, 15 to 20 percent of apps are natively compiled specific to its rival architectures [such as ARM].
This is a reality that could spell trouble for MIPS customers -- at least for the time being.
But one needs to look at a few long-term industry trends, Rohatgi explained.
One is Google’s latest view that “going native is generally not worth it,” as Rohatgi put it. The performance delta between native and non-native apps is fast approaching insignificance.
Most apps that are architecture-specific are games, sensitive to performance. Yet, in an example like ‘Temple Run,’ a graphically intensive video game, Rohatgi noted that the total code written to the system (and to run on CPU) is small, compared to code written to the Open GL, running on GPU. Most developers feel that it’s not worth developing games to a native platform, he said. In Rohatgi’s view, Native Development Kit -- libraries written in C and other languages to be compiled to ARM, MIPS or x86 native code and installed -- is passé. “The whole industry doesn’t want that.”
Another big industry trend to consider is a proliferation of portability tools. One good example is Renderscript, according to Rohatgi. It offers a platform-independent computation engine operating at the native level. Developers can use it to accelerate apps that require extensive computational horsepower.
There is also a new complier called ‘Clang’ -- designed under the LLVM umbrella. LLVM was developed to provide the middle layers of a complete compiler system, taking intermediate form code from a compiler and emitting an optimized intermediate form.
The combination of Clang and LLVM reportedly provides the lion’s share of a tool chain. Since Clang is built with a library-based design, like the rest of LLVM, Clang is said to be easy to embed into other applications.
That Imagination decided to cut the MIPS sales team is not surprising. Marketing and Sales are usually the first to be let go from the acquired company.
The challenge that Imagination will face is that eventhough they may be right in predicting that in due time most (still not all) applications will be not be platform specific, there will not be too many takers. Who would want to risk their android phones from not running a few apps. I cannot imagine (pun un-intended) the magic-code being as optimal as the ARM code.
The other challenge will be the variety of battle-hardended silicon available from the likes of qualcomm, samsung, apple, nVidia with ARM. Can imagination really imagine (pun intended) developing silicon that can out-perform and compete with those companies. All this while ARM continues to deliver on its roadmap. I feel nostalgic for MiPs but I still don't see them winning in this space.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.