ni...@lysator.liu.se (Niels Möller) writes:
> My guess is that special q=1 is often beneficial for the double-limb
> loop, and rarely beneficial in the single-limb loop. But needs some
> measurements.
I've made a quick try deleting it from the single-limb loop. See patch
below. Measurements are
t...@gmplib.org (Torbjörn Granlund) writes:
> ni...@lysator.liu.se (Niels Möller) writes:
>
> Pushed now, with minor changes and deletion of the "current" code
> (method 2 above).
>
> It is not in the main repo yet.
Ooops. Now it should be in.
Regards,
/Niels
--
Niels Möller.
ni...@lysator.liu.se (Niels Möller) writes:
Pushed now, with minor changes and deletion of the "current" code
(method 2 above).
It is not in the main repo yet.
--
Torbjörn
Please encrypt, key id 0xC8601622
___
gmp-devel mailing list
ni...@lysator.liu.se (Niels Möller) writes:
>> I'm appending another iteration of the patch to add div2 function based
>> on div1 on the high limbs. Selected via HGCD2_DIV2_METHOD. Benchmarks:
>>
>> HGCD2_DIV2_METHOD mpn_hgcd2_1 mpn_hgcd2_2 mpn_hgcd2_3
>> 1#1504.47
Hello GMP developers,
I am chair of the 2019 Jenks Prize Committee.
If the person in charge of GMP would like to nominate the GMP team for the 2019
Jenks Prize for Software Excellence in Computer Algebra, please could they
complete the nomination form or contact me.
Information about the
ni...@lysator.liu.se (Niels Möller) writes:
> I've written several div1 in asm (arm v5 method 2, 64-bit arm v8 method
> 2 and 3, 64-bit x86 method 2).
Nice.
x64-div_11-m2.asm
Description: Binary data
arm32-div_11-m2.asm
Description: Binary data
arm64-div_11-m2.asm
Description:
t...@gmplib.org (Torbjörn Granlund) writes:
> Not exactly. I think running div1 like hgcd2 but without any of hgcd2
> bookkeeping would make some sense. I.e., feed div1 with the input of
> Euclid's algorithm. To avoid skew from particular operands, perhaps
> table 10 uniformly distrinuted