[Libre-soc-dev] funding & go-ahead requested for #1134

[Jacob Lifshay]

On Fri, Aug 18, 2023, 20:32 Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

> https://bugs.libre-soc.org/show_bug.cgi?id=1025#c0
>
>      * fptrans
>        * huge, needs to be split into much smaller subtasks.
>          it's enough work that, except for the tasks already split out
>          (recip/[r]sqrt/mod/rem and fminmax) we should probably eliminate
>          it entirely from this grant due to being lower priority and
> difficult.
>
> now you're starting to get it, ehn? it's a frickin lot of
> work, to think like this, doing advance planning, isn't it?
>

Luke, I already knew it is a lot of work. please don't treat me as if I'm
clueless.

>
> i.e. *when* you have the estimates, you *can* start to go
> "oh shite that's way too much work for the money" rather
>

I already did that several weeks ago. edit i made on 2023-08-02 to
https://bugs.libre-soc.org/show_bug.cgi?id=1026#c0
> Because there are too many instructions to fit within
> this budget, this task only covers some of the instructions,
> to be determined later since they may be easier/harder than
> expected.

than rush ahead and do 5 months work for 2 weeks money
> (or zero because the actual 2 week task was never done).
>

that happened because I was more interested in solving the (more complex
than expected) interesting problem of register allocation for SVP64 than in
how much money I was making.

>
>      * FPSCR/FP regs and dependency stuff, and loads/stores.
>        jacob: I have no experience working on that part of the code,
>        so I'd like someone else to estimate that.
>
> nnope.  it will be good practice for you to do some estimates
> on things you *don't* know anything about, because if *you*
> are the one to implement it, *you* are the one that needs
> the direct feedback - to yourself - to get better at doing the
> estimates, aren't you?
>

ok, i'm estimating 500 lines plus 500 lines of tests for that one, based on
grepping for MSR in soc.git.

I'll fill in the rest of them, except for the rest of fptrans which I'm
pretty sure we'll be out of budget before getting to.

>
> plus, you need to know in advance clearly what you are going
> to be doing, don't you?


I see that more as part of actually doing the task.

now would be a good time to find out,
> wouldn't it? (hint: i already described it to you, last week).
>
> plus, what if you find that the estimates come out at 4 weeks,
> or 8 weeks, or 4 months?
>

I know it's not an 8 week task because I was already vaguely familiar with
the code and expected it to be a few days at most for someone familiar with
that part of the codebase.

Jacob