[FEATURE] Chip adaptation project

[zyy37]

Is your feature request related to a problem? Please describe.

Can you establish a continuous plan for adapting chips? Like the Windows operating system adapting to various chips, the code is either from their own people or from chip manufacturers. Similarly, the CPU acceleration library mkldnn code in TensorFlow, such as tensorflow/core/kernels/mkl, is also developed by the project's own team or people from Intel Corporation. Can your well-known Apache association send cooperation invitations to various chip manufacturers to adapt chips such as gigadevice GD32W515PIQ6, xhsc HC32F451JEUB, or can you also participate in this continuous cycle chip adaptation work? The other company usually sends employees to submit, just like the above, which is also beneficial for them. This is indeed a tough nut for ordinary users. This chip adaptation project is definitely unavoidable in the long run. Thank you.

Describe the solution you'd like

na

Describe alternatives you've considered

No response

Verification

• I have verified before submitting the report.

[zyy37]

Hello, can you support those two chips? They are all very representative. I asked the AI about the workload and duration for porting chips for several RTOS systems such as FreeRTOS. It seems that FreeRTOS is easier, requiring less than 3 days, while your NuttX requires more than 2 weeks? I hope you can provide feedback and improve this. Actually, developers just want it to be easier to get started, rather than being blocked by these issues. You can compare it to the market share of desktop and mobile systems. Thank you. @raiden00pl @acassis

---

[acassis]

Hi @zyy37 actually recently @JorgeGzm ported NuttX to GigaDevice gd32vw55x and the port was done in a few days (even with WiFi and BLE working), not 2 weeks as that AI told you.

Unfortunately Apache will not send this invitation to companies to port NuttX to their chip, because they even have other RTOS under their umbrella: mynewt, it is different from Linux Foundation which has only one.

I think a way to get NuttX ported to more devices is: 1) the own community does it, like Jorge did; 2) NuttX get critical mass, more users, and more customers requiring semiconductor companies to do the port; 3) we can create a NuttX International (like RISC-V International) to get more companies associated and get support to the project. Actually I and @cederom already discussed this ideas some days ago.

So, if you want to help, please start with 1.

[cederom]

NuttX International or NuttX Foundation is a good idea to find funding for core developers, this is something we lack badly at this point. Everyone just takes results for free and does not really want to invest in development. We need to find a way to fix this :-)

Regarding using AI for porting this can be shoot in the feet because companies would then use AI to port instead hire core team developers :-(

[cederom]

@zyy37: Can you establish a continuous plan for adapting chips? (..)

There is no such plan and it will be hard to create one without vendors being interested in the first place, then providing resources (boards, funding) for the development to happen.

I have contacted numerous vendors and none of them was interested :-(

Therefore we only support chips that someone uses on their own and shares back to the upstream which is also not very common thing.

[zyy37]

Understood. Thank you for your efforts and contributions. I wish you all the best in the future.

It's really difficult to plan, truly. Perhaps you can refer to the organizational and cooperative approaches for chip adaptation used by desktop OS like Windows and mobile OS like Android, or abstract generic chip adaptation instead of specific models. Otherwise, the models and workload would be enormous.

[zyy37]

Hi @zyy37 actually recently @JorgeGzm ported NuttX to GigaDevice gd32vw55x and the port was done in a few days (even with WiFi and BLE working), not 2 weeks as that AI told you.

Unfortunately Apache will not send this invitation to companies to port NuttX to their chip, because they even have other RTOS under their umbrella: mynewt, it is different from Linux Foundation which has only one.

I think a way to get NuttX ported to more devices is: 1) the own community does it, like Jorge did; 2) NuttX get critical mass, more users, and more customers requiring semiconductor companies to do the port; 3) we can create a NuttX International (like RISC-V International) to get more companies associated and get support to the project. Actually I and @cederom already discussed this ideas some days ago.

So, if you want to help, please start with 1.

Thank you for your reply. Perhaps you can clarify the step-by-step documentation for chip porting and simplify the porting architecture and workload, as mentioned in AI. It would be best not to use pseudo-code, and it should be something that actually works. By clarifying the documentation and teaching others, others can contribute and help you with these workloads? Currently, only those who are very experienced and familiar with embedded systems can easily get started. If it becomes easy to port, maybe many developers will be willing to contribute, just like in approach 1.

[acassis]

Hi @zyy37 actually recently @JorgeGzm ported NuttX to GigaDevice gd32vw55x and the port was done in a few days (even with WiFi and BLE working), not 2 weeks as that AI told you.
Unfortunately Apache will not send this invitation to companies to port NuttX to their chip, because they even have other RTOS under their umbrella: mynewt, it is different from Linux Foundation which has only one.
I think a way to get NuttX ported to more devices is: 1) the own community does it, like Jorge did; 2) NuttX get critical mass, more users, and more customers requiring semiconductor companies to do the port; 3) we can create a NuttX International (like RISC-V International) to get more companies associated and get support to the project. Actually I and @cederom already discussed this ideas some days ago.
So, if you want to help, please start with 1.

Thank you for your reply. Perhaps you can clarify the step-by-step documentation for chip porting and simplify the porting architecture and workload, as mentioned in AI. It would be best not to use pseudo-code, and it should be something that actually works. By clarifying the documentation and teaching others, others can contribute and help you with these workloads? Currently, only those who are very experienced and familiar with embedded systems can easily get started. If it becomes easy to port, maybe many developers will be willing to contribute, just like in approach 1.

There are two guides already based on real port:
https://nuttx.apache.org/docs/latest/guides/port.html

If you want to learn, I suggest you try to port without using AI, just to understand all the details.

[zyy37]

@zyy37: Can you establish a continuous plan for adapting chips? (..)

There is no such plan and it will be hard to create one without vendors being interested in the first place, then providing resources (boards, funding) for the development to happen.

I have contacted numerous vendors and none of them was interested :-(

Therefore we only support chips that someone uses on their own and shares back to the upstream which is also not very common thing.

Thank you. It's really difficult to coordinate. I think your operating system and chip manufacturers are upstream and downstream partners, so there should be interfaces that others can use or be used for. Actually, many manufacturers provide ready-made chip adaptation code, but they may not know the interface of your operating system and cannot use it. The force did not move in one direction. You can also refer to the manufacturer's ready-made code when porting a chip, which can help reduce your workload.

---

[cederom]

@acassis: There are two guides already based on real port: https://nuttx.apache.org/docs/latest/guides/port.html

If you want to learn, I suggest you try to port without using AI, just to understand all the details.

I am playing around with CCNV2 from ChipCraft-IC from Poland, if the port works out I will add next case study.. this is also so called "free time" effort :-)

Yes do not use AI to learn basics, do your own fun, play around, experiment, make mistakes, make fixes, that way you learn yourself and on your own. After you know basics that you could do yourself on your own you can use AI tools to assist you but not do the work for you.. you need to assess the results, verify, and sign your own name under the results :-)

[cederom]

@zyy37: Can you establish a continuous plan for adapting chips? (..)

@cederom: There is no such plan and it will be hard to create one without vendors being interested in the first place, then providing resources (boards, funding) for the development to happen.
I have contacted numerous vendors and none of them was interested :-(
Therefore we only support chips that someone uses on their own and shares back to the upstream which is also not very common thing.

@zyy37: Thank you. It's really difficult to coordinate. I think your operating system and chip manufacturers are upstream and downstream partners, so there should be interfaces that others can use or be used for. Actually, many manufacturers provide ready-made chip adaptation code, but they may not know the interface of your operating system and cannot use it. The force did not move in one direction. You can also refer to the manufacturer's ready-made code when porting a chip, which can help reduce your workload.

It's a theory. If you work with different chips, you will notice there is no one standard, every vendor has their own preferences, and each OS/RTOS use different approach under the hood (that approach changes in time faster or slower). Each vendor may provide the SDK of some sort, we may adopt it (the interface layer slows things down and makes things bigger), but problem arises when that SDK changes, its a never ending story, like chasing a rabbit (in our free time) and this is why avoid SDKs here in NuttX in favor of in-tree drivers that are integral part of our project and do not depend on external politics.

The problem here is clear: no vendor is willing to pay NuttX developers to port their devices, everyone is only happy to use results for free :-(