[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: 70x speed-up for seq's common cases
From: |
Jim Meyering |
Subject: |
Re: 70x speed-up for seq's common cases |
Date: |
Sat, 15 Sep 2012 20:18:12 +0200 |
Torbjorn Granlund wrote:
> For general seq use, I suppose we should do away with `incr' and instead
> use a general `add'. Below is a suggested implementation, and as bonus
> also a `sub'.
>
> Like `incr', `add' needs an extra byte if space, since if the end value
> is within the increment from a power of 10, a number with one more digit
> than the end value will be created.
>
> Implementing negative values is a bit tricky. If a-b is wanted and a <
> b, one must of course compute b-a and separately store a minus sign. I
> just write a `sub' function, not any code actually handling negative
> numbers.
>
> The function `add' is reasonably well-tested. The function `sub' is
> minimally tested (but as a trivial edit of `add' it might b correct
> anyway).
Thanks. I'm sure someone will find time to experiment with those.
Looking at factor.c, I saw this bit of code,
and confirm that the current tests do not exercise it.
{
/* The found factor is two words. This is highly unlikely, thus hard
to trigger. Please be careful before you change this code! */
uintmax_t ginv;
binv (ginv, g0); /* Compute n = n / g. Since the result will */
n0 = ginv * n0; /* fit one word, we can compute the quotient */
n1 = 0; /* modulo B, ignoring the high divisor word. */
if (!prime2_p (g1, g0))
factor_using_pollard_rho2 (g1, g0, a + 1, factors);
else
factor_insert_large (factors, g1, g0);
}
Can you provide a test case that exercises that code?
- Re: 70x speed-up for seq's common cases, (continued)