[Mesa-dev] [PATCH 1/5] util: import public domain code for integer division by a constant

[Jason Ekstrand]

On Tue, Oct 2, 2018 at 3:20 PM Marek Olšák <maraeo at gmail.com> wrote:

>
> On Tue, Oct 2, 2018, 1:15 PM Jason Ekstrand <jason at jlekstrand.net> wrote:
>
>> Reading through things in a bit more detail, I do believe that importing
>> this version in some form would be better than using mine for a number of
>> reasons:
>>
>>  * It is better optimized for signed integers
>>  * The struct of division factors is much better than what I did.  (I did
>> consider a struct and discarded the idea; I was wrong).
>>  * Computation of the division factors doesn't involve N*2-bit
>> multiplication
>>  * The round-up algorithm here results in significantly better code than
>> the N+1-bit round-down.
>>  * I trust ridiculousfish to get this right more than I trust myself
>>
>> That said, I have a few caveats on merging this as-is:
>>
>>  * I would like to see some unit tests.  I already spent the time to
>> write some; they just have to be ported.
>>  * It needs to be adjusted to handle 64-bit integers (right now, it
>> appears to only work for num_bits <= 32)
>>  * We shouldn't define uint_t and sint_t in a header
>>
>> How do you want to proceed?
>>
>
> I don't have a plan. Anything that works for you would be OK with me, so
> if you wanna just rework it according to you, that's fine.
>

I'm not in too much of a hurry but I can probably rework it if you don't
get to it first.


> Changing the types is tricky. Template code in a C header included several
> times would work. C++ templates would be ideal.
>

We may be able to just s/uint_t/uint64_t/ for most of it.  I *think* the
functions to create the division parameters should fairly nicely
generalize.  It's the functions which actually do divisions that we'll have
to re-type for each size.


> What's your timeframe for this? Mine is certainly more than a month.
>

Unsure.  I've got other (not the NIR pass) code pending that requires this
and will hopefully be landing in a month or so.  Sadly, I can't send it to
the ML yet.  In any case, I'd be looking at one to two months, probably.
I'm happy for it to be a matter of whoever gets to it first.

--Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20181002/35681a31/attachment.html>