nRF52 development using clang

RE: nRF52 development using clang

Håkon Alseth — Mon, 12 Aug 2019 11:55:21 GMT

Sorry, but there are no current plans to add LLVM/clang to the nRF5 SDK.

RE: nRF52 development using clang

eyalasko — Mon, 12 Aug 2019 11:39:28 GMT

[quote userid="2115" url="~/f/nordic-q-a/50605/nrf52-development-using-clang/203490"]you can see that most compilers perform equally when optimizing for size[/quote]

actually when it comes to RAM consumption, I come the opposite conclusion...

RE: nRF52 development using clang

eyalasko — Mon, 12 Aug 2019 11:36:32 GMT

Thanks.

We are aware of the git repo you've mentioned. We have tried to follow the instructions with SDK14.2 with limited success.

Is there any clang support on Nordic's roadmap on the foreseeable future?

Thanks

RE: nRF52 development using clang

Håkon Alseth — Mon, 12 Aug 2019 07:36:45 GMT

Hi,

I just want to initially point out that armclang / llvm is not supported in nRF5 SDK, so please be aware that there might be unwanted side effects by using this toolchain.

[quote user=""]I've seen some pretty old posts (e.g. RK's comment here, and here) that the app breaks/softdevice calls not supported when using clang compiler. Is this still the case?[/quote]

Yes, that would still be the case. llvm/clang is close to gcc compatible, but not 100 % on the asm parts, and other ARM specific attributes (SVC for instance).

[quote user=""]Is there a (reasonably) easy way to make the project clang/Keil V6 compatible?[/quote]

I haven't tried since SDK 13/14, but other forum users have. From the first thread you linked to, you can find a link to this github repo:

https://github.com/eblot/nrf5-llvm

You can try it out and see if that works for you.

[quote user=""]Are there any Code/RAM/Performance deltas when comparing Keil V5 and clang? (I've found this old post , but it compares gcc only and its outdated)[/quote]

I am not sure if moving to llvm/armclang would be the best option if you're looking to tune your code, but for research purposes, it makes sense to look at it.

From the link you posted, you can see that most compilers perform equally when optimizing for size. Without having done any testing with llvm/armclang in a while, I do suspect that the performance would be similar to other ARM compilers out there (IAR, gcc, armcc).

Kind regards,

Håkon

RE: nRF52 development using clang

Dmitry — Wed, 07 Aug 2019 08:38:50 GMT

Hello,

I can confirm that the problem exists in SDK 15.3.0 (SVCALL is implemented in the same way). We're also considering to use 6.x compiler, so looking forward..

Concerning code/RAM results, maybe statistics from our project will be interesting for you.

v5.06 (-O2): Full build time 0:43, Program Size: Code=27962 RO-data=13030 RW-data=308 ZI-data=10024
v6.9 (-Os balanced): Full build time 2:06, Program Size: Code=32600 RO-data=8392 RW-data=92 ZI-data=10236

A noticeable difference in RO-data is because LLVM removes unused constant sections (for example, we have sha256/512 common source where sha512 functions are not called, v5.06 leaves sha512 transform tables in resulting image).

RE: nRF52 development using clang

Susheel Nuguru — Wed, 07 Aug 2019 06:43:47 GMT

Hi eyalasko,

The clang expert on our team is on holiday and will be back on Monday. He will answer to this post. I know the posts you linked in but I am worried that this information might be outdated.