writer offset as a struct field

[burrbull]

This is retry of one of my old experiments.

It does 2 things:

1. Moves offset from field writer generic to struct field.
2. Always index field array from 0 (slightly confusing but much easier to work).

Generated code is much simpler.

The main advantage is you can now simply use .write(|w| w.fname(3).set_bit() for write/modify field arrays (previously only reading was accepted) same as .write(|w| w.fname3().set_bit().

The main disadvantage is bigger firmware size in debug mode.

cc @kossnikita, @therealprof and anyone who want to help.

I tried several examples from stm32f4xx-hal:

                   before PR vs after PR
blinky-timer-irq  41984/1172 vs 42628/1172
pwm-sinus         49672/8628 vs 50608/8632
rtc-alarm         63704/4680 vs 65648/4680
rtic-usart-shell 88024/17548 vs 88904/17548

So it is 2-3% bigger on debug mode for me.

cargo build --release -j1 --timings. time of stm32f4(f411) compilation on my machine, dependencies not included and peak memory usage during rtic-usart-shell compilation:

0.14.0    4.73s   758860k
0.15.1    3.00s   530916k
before PR 2.67s   476400k
after PR  2.47s   450996k

[kossnikita]

My crate is very similar to yours, and the PAC is also similar. So I don't think there will be a big difference. I don't understand how I can measure the compile time. Just the example build or HAL and PAС?

                   before PR vs after PR
qei               29656/9504 vs 30268/9504
pwm               35004/9524 vs 35612/9524

Sorry, I haven't finished the more complex examples yet.

[burrbull]

I don't understand how I can measure the compile time

You just need to cargo clean and then

cargo build --timings

(+ --release). After finish cargo generates html report and prints a link to it. Doesn't matter what you are building example or HAL. It show compilation time of each dependency.

[kossnikita]

Got it. I mean what time do you need? HAL compile or PAC + HAL?

[burrbull]

I've measured just a PAC compilation time.
Sure a full time is also interesting, as this PR may also affect on linking time. But I don't see good way to estimate difference.

[kossnikita]

cargo build --release -j1 --timings --example qei --features at32f415

PAC compile time:

6.7±0.15s vs 6.5±0.05s

[Emilgardis]

i'm fine with this new interface and zero indexing, but I think we need to signal the break a little better, not sure how though

[Emilgardis]

lgtm! I'd like it if someone else on @rust-embedded/tools could also review

[therealprof]

I don't quite understand why we want to do this but I'm fine in general.