mirror/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2024-11-21 19:46:16 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

sch_cake: constify inverse square root cache

Browse Source 
sch_cake uses a cache of the first 16 values of the inverse square root
calculation for the Cobalt AQM to save some cycles on the fast path.
This cache is populated when the qdisc is first loaded, but there's
really no reason why it can't just be pre-populated. So change it to be
pre-populated with constants, which also makes it possible to constify
it.

This gives a modest space saving for the module (not counting debug data):
.text:  -224 bytes
.rodata: +80 bytes
.bss:    -64 bytes
Total:  -192 bytes

Signed-off-by: Dave Taht <dave.taht@gmail.com>
[ fixed up comment, rewrote commit message ]
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://patch.msgid.link/20240909091630.22177-1-toke@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit is contained in:
Dave Taht 2024-09-09 11:16:28 +02:00 committed by Jakub Kicinski
parent 3f464b193d
commit c48994baef
1 changed files with 18 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

53
net/sched/sch_cake.c
View File

@ -361,8 +361,24 @@ static const u8 besteffort[] = {
static const u8 normal_order[] = {0, 1, 2, 3, 4, 5, 6, 7};
static const u8 bulk_order[] = {1, 0, 2, 3};
/* There is a big difference in timing between the accurate values placed in the
* cache and the approximations given by a single Newton step for small count
* values, particularly when stepping from count 1 to 2 or vice versa. Hence,
* these values are calculated using eight Newton steps, using the
* implementation below. Above 16, a single Newton step gives sufficient
* accuracy in either direction, given the precision stored.
*
* The magnitude of the error when stepping up to count 2 is such as to give the
* value that *should* have been produced at count 4.
*/
#define REC_INV_SQRT_CACHE (16)
static u32 cobalt_rec_inv_sqrt_cache[REC_INV_SQRT_CACHE] = {0};
static const u32 inv_sqrt_cache[REC_INV_SQRT_CACHE] = {
~0, ~0, 3037000500, 2479700525,
2147483647, 1920767767, 1753413056, 1623345051,
1518500250, 1431655765, 1358187914, 1294981364,
1239850263, 1191209601, 1147878294, 1108955788
};
/* http://en.wikipedia.org/wiki/Methods_of_computing_square_roots
* new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2)
@ -388,47 +404,14 @@ static void cobalt_newton_step(struct cobalt_vars *vars)
static void cobalt_invsqrt(struct cobalt_vars *vars)
{
if (vars->count < REC_INV_SQRT_CACHE)
vars->rec_inv_sqrt = cobalt_rec_inv_sqrt_cache[vars->count];
vars->rec_inv_sqrt = inv_sqrt_cache[vars->count];
else
cobalt_newton_step(vars);
}
/* There is a big difference in timing between the accurate values placed in
* the cache and the approximations given by a single Newton step for small
* count values, particularly when stepping from count 1 to 2 or vice versa.
* Above 16, a single Newton step gives sufficient accuracy in either
* direction, given the precision stored.
*
* The magnitude of the error when stepping up to count 2 is such as to give
* the value that *should* have been produced at count 4.
*/
static void cobalt_cache_init(void)
{
struct cobalt_vars v;
memset(&v, 0, sizeof(v));
v.rec_inv_sqrt = ~0U;
cobalt_rec_inv_sqrt_cache[0] = v.rec_inv_sqrt;
for (v.count = 1; v.count < REC_INV_SQRT_CACHE; v.count++) {
cobalt_newton_step(&v);
cobalt_newton_step(&v);
cobalt_newton_step(&v);
cobalt_newton_step(&v);
cobalt_rec_inv_sqrt_cache[v.count] = v.rec_inv_sqrt;
}
}
static void cobalt_vars_init(struct cobalt_vars *vars)
{
memset(vars, 0, sizeof(*vars));
if (!cobalt_rec_inv_sqrt_cache[0]) {
cobalt_cache_init();
cobalt_rec_inv_sqrt_cache[0] = ~0;
}
}
/* CoDel control_law is t + interval/sqrt(count)