[libre-riscv-dev] getting important: need to create the project plan, involves estimating times for tasks (and creating them)

[Luke Kenneth Casson Leighton]

On Thu, Apr 25, 2019 at 4:59 AM Aleksandar Kostovic
<alexandar.kostovic at gmail.com> wrote:
>
> "So would you be happy with say USD 1500 to do the sqrt FP? Yes you get the
> choice to be included. Will need you on the MoU to do so, ideally."
>
> I would be more than happy with that! :)

awesome, ok, so i'll introduce you to michiel, you'll need to send him
your address.  otherwise, the money has to come to me, and i then need
to send it on to you (and pay tax on money i *didn't receive*) in the
process.

on a practical note, i cannot recommend creating a milestone that says
"do the unit tests later".  fulfilling that in-order, in order to
receive the funds on sequential completion of milestones, will
massively interfere with the ability to debug and complete the entire
unit, as you would be tempted to develop the code without *ANY* unit
tests... because you won't get paid until the unit tests are
COMPLETED.

in full.

what i recommend instead is to create a milestone "preliminary
exploratory software emulation of FP SQRT" milestone.

this could be to convert that c code you found (from the academic
paper) into python, which we will *need anyway as part of the unit
tests*.

no i am not talking about the "high level" unit tests (using softfloat
/ sfpy), i'm talking about the "sub-unit" (modules) - the individual
pipeline stages.

if we have a python function that creates one iteration, for example,
we can write a unit test in future for the hardware that tests *one
iteration of the pipeline*, calculating *one* iteration of the
mantissa.

doing that with the c-based jdawson softtfloat library will *NOT* be
possible... not without a hell of a lot of work, at least.

what do you think?

l.