Micro-GMP

[Marco Bodrato]

Ciao,

Il Mar, 11 Dicembre 2018 9:57 am, Niels Möller ha scritto:
> "Marco Bodrato" <bodrato at mail.dm.unipi.it> writes:
>
>> The test suite is a test suite and not a benchmark, I know... but,
>> before
>> the patch you propose
>>  make check-mini-gmp;time make check-mini-gmp
>> reports (on a slow machine)
>>  user	0m24.610s
>>
>> after the patch (on the same machine):
>>  user	0m45.160s
>
> That's a big difference. I wonder which functions are responsible. And

I believe I've found the problem, and healed it. In mini-, not in your
proposed patch: a couple of mpz_divexact_ui ( , , 2)
https://gmplib.org/repo/gmp/rev/afae8fd2739d#l1.39

Before my change, lcov said that mpz_divexact_ui was called 11852062
times, now it says that the function is not covered by the tests, and the
impact of your proposed patch seems negligible now, on the total running
time of the test-suite.

The main impact of your patch (probably) is then due to the removal of the
function mpn_div_qr_1(), with its /* Special case for powers of two. */.

Ĝis,
m

-- 
http://bodrato.it/papers/