lkml.org 
[lkml]   [2012]   [Feb]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RFC PATCH 13/14] sched: make __update_entity_runnable_avg() fast
From
Date
On Wed, 2012-02-01 at 17:38 -0800, Paul Turner wrote:
> +/* Precomputed fixed inverse multiplies for multiplication by y^n */
> +const static u32 runnable_avg_yN_inv[] = {
> + 0xffffffff,0xfa83b2db,0xf5257d15,0xefe4b99b,0xeac0c6e7,0xe5b906e7,
> + 0xe0ccdeec,0xdbfbb797,0xd744fcca,0xd2a81d91,0xce248c15,0xc9b9bd86,
> + 0xc5672a11,0xc12c4cca,0xbd08a39f,0xb8fbaf47,0xb504f333,0xb123f581,
> + 0xad583eea,0xa9a15ab4,0xa5fed6a9,0xa2704303,0x9ef53260,0x9b8d39b9,
> + 0x9837f051,0x94f4efa8,0x91c3d373,0x8ea4398b,0x8b95c1e3,0x88980e80,
> + 0x85aac367,0x82cd8698,
> +};

I wonder if modern Intel isn't at the point where computing this thing
is cheaper than the cacheline miss. You can compute y^n in O(log n) time
and with n < 32 that's 5 multiplications (see fixed_power_int). Add to
that the division.

Of course there's platforms, ARM?, where reverse is likely true. Bugger
that.

> +/* Precomputed \Sum y^k { 1<=k<=n } */
> +const static u32 runnable_avg_yN_sum[] = {
> + 0, 1002, 1982, 2941, 3880, 4798, 5697, 6576, 7437, 8279, 9103,
> + 9909,10698,11470,12226,12966,13690,14398,15091,15769,16433,17082,
> + 17718,18340,18949,19545,20128,20698,21256,21802,22336,22859,23371,
> +};

Right, can't see a fast way to compute this..

The asymmetry in the tables annoys me though 32 vs 33 entries, hex vs
dec :-)



\
 
 \ /
  Last update: 2012-02-06 21:51    [from the cache]
©2003-2011 Jasper Spaans