Rapidly developing RISC-V software ecosystem

The TARSIER team's vision is to make RISC-V the Tier-1 platform for all mainstream open source software. We hope to ensure that all popular Linux distributions run smoothly and smoothly on RISC-V platform, and the software ecological richness, availability and user experience reach and exceed the X86 and Arm64 platforms.

It is expected that in 2025, we can promote the mainstream Linux distribution to upgrade RISC-V to the default support architecture. The software on RISC-V notebooks can meet the daily office needs and support all open source software stacks required by RISC-V to enter the supercomputing field.

The bottom layer in the figure above is the infrastructure layer. The software has purchased almost all the available RISC-V devices on the market, providing open source CI Farms for all open source communities around the world, and has also built a large number of cross-compilation environments through Alibaba Cloud, with a hardware environment of more than 2000 nodes (including x86 computing nodes).

The second layer is the basic service layer for developers and operating systems. At present, it provides CI Farm services to more than 50 communities around the world. It can log in and access all RISC-V devices remotely through SSH.

The third layer is the language and execution environment layer. All compilers, virtual machines and simulators can be executed in interpreter mode. In addition, we are also doing JIT work such as SpiderMonkey.

The top layer is the Linux distribution layer. Almost all distributions, including the Dragon monitor operating system, have supported the RISC-V architecture.

Moore's law has stagnated in 2003, but the development of computational power and the increase of data scale are infinite. There are endogenous contradictions among them. The revolution in the field of hardware has been carried out for more than ten years, and the top advantages have been gradually eliminated. With the reduction of chip design costs and manufacturing costs, the threshold of technology has also been gradually reduced, and more and more manufacturers have begun to try to make specific chips in specific fields.

The complexity of the software system has a super-linear growth (which may be a square or exponential growth). For example, a software in a mobile phone may have millions of lines of code running, and each update may add hundreds of thousands of lines of code. This scale has exceeded the level that any company or country can maintain.

Therefore, open source in 2022 has a different meaning from that in the 1980s. All companies have to use open source, otherwise the product cost will be very high.

Open source software has become a community of human knowledge, especially information industry knowledge.

Globally, the number of developers who have the ability to control software complexity is also very limited. If we do not realize this, we will face great difficulties in product promotion and competition. At present, only one or two open source communities in a segment are finally active. The code that is not maintained by upstream is like living in ICU, which is expensive and has a high mortality rate.

To sum up, we can draw two inferences:

First, open source software will devour everything: after the emergence of an active community of open source software in the market or market segments, it will likely become the final top winner, and will suppress other non-open source or emerging products.

Second, a free and open instruction set is bound to emerge. The instruction set is open to software developers, but closed to hardware manufacturers. The open instruction set enables the hardware design members of the domain specific architecture to design their own instructions and chips. In addition, we believe that there are only 1-2 free and open instruction sets that can survive.

All technical fields will have open source standards.

In a narrow sense, RISC-V only contains specifications, which are open. For hardware developers, as long as they meet the standards, they can compile and run the software through the compiler immediately after development, which is a huge change.

RISC-V was successful. It was proposed at a suitable time (when Moore's law has expired and DSA is needed). It uses modular instructions. The minimum instruction set is only 47 instructions. It can be directly used in small chip control and IoT MCU. It can also add floating point calculation instructions and DSP instructions to become watch chips. After adding 64 bits, it can become mobile phone or laptop chips.

RISC-V adopted open standards when it was proposed, and hardware manufacturers can use it directly without considering patents and other issues. RISC-V decided to change from architecture orientation to software orientation in 2020. The biggest feature of software orientation is that it has full respect for open source software and developers, and can produce a unified and complete community in a collaborative way. This has also triggered a new business model, that is, select RISC-V architecture first, and then select manufacturers.

The figure above shows that RISC-V will turn in 2020. We also expect domestic manufacturers to think more from the perspective of software.

The figure above shows the statistics of RISC-V Foundation in 2021. In fact, 10 billion RISC-V chips have been shipped, and it is likely to exceed 80 billion in 2025.

Taking Debian as an example, the basic support of RISC-V open source software ecosystem for the operating system has been completed, including web browsing, office software, graphic editing, etc., and less than 5% of the unfinished projects are mainly JIT compilers, which are currently being solved in succession.

China is a region with rapid development of RISC-V. The second RISC-V China Summit is the only summit-level summit in the world except North America.

All speech videos have been online: https://space.bilibili.com/1121469705 。

The main applications of RISC-V currently include compiler, virtual machine, simulator, application and RISC-V distribution.

There are a lot of new opportunities for the ecosystem of RISC-V open source software, which is a carnival in the field of basic software. We hope that more people will participate in the ecological construction of RISC-V, especially the ecological construction of software.

RISC-V opens up new possibilities for the security field, and can explore more possibilities through FPGA and open source RISC-V. We believe that a large number of RISC-V related security products may emerge on the market in five years.

We are implementing RISC-V to build a cluster with more than 1000 cores, hoping to verify the support of open source software for RISC-V in the HPC field. The trainee team of PLCT laboratory will fully tap the D1 computing power potential, including the expansion of Vector v0.7.1, and will form a set of Linux distribution for RISC-V supercomputing field: RobinOS, which is based on the implementation of Dragon lizard RISC-V.

For this, we plan to publish the project before December 31, 2022. You can log in https://mirror.iscas.ac.cn/RobinOS/ For download. Users can submit directly through github's pull request, or even dismantle the machine and rebuild according to their own topology.

In addition, PLCT laboratory began to prepare to build a RISC-V cluster with ≥ 1024 nodes using cheap switches.

The PLCT wishing pool plan is a unique community cooperation model of PLCT laboratory (and TARSIER team). We collect information from global developers about which software we want to see, use, or have in the RISC-V software ecosystem, and select some of them to be included in the new year's roadmap.