lkml.org 
[lkml]   [2017]   [Jun]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH] bsearch: micro-optimize pivot position calculation
Date
There is a slightly faster way (in terms of the number of
instructions being used) to calculate the position of a
middle element, preserving integer overflow safeness.

./scripts/bloat-o-meter lib/bsearch.o.old lib/bsearch.o.new
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-24 (-24)
function old new delta
bsearch 122 98 -24

TEST

INT array of size 100001, elements [0..100000]. gcc 7.1, Os, x86_64.

a) bsearch() of existing key "100001 - 2":

BASE
====

$ perf stat ./a.out

Performance counter stats for './a.out':

619.445196 task-clock:u (msec) # 0.999 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
133 page-faults:u # 0.215 K/sec
1,949,517,279 cycles:u # 3.147 GHz (83.06%)
181,017,938 stalled-cycles-frontend:u # 9.29% frontend cycles idle (83.05%)
82,959,265 stalled-cycles-backend:u # 4.26% backend cycles idle (67.02%)
4,355,706,383 instructions:u # 2.23 insn per cycle
# 0.04 stalled cycles per insn (83.54%)
1,051,539,242 branches:u # 1697.550 M/sec (83.54%)
15,263,381 branch-misses:u # 1.45% of all branches (83.43%)

0.620082548 seconds time elapsed

PATCHED
=======

$ perf stat ./a.out

Performance counter stats for './a.out':

475.097316 task-clock:u (msec) # 0.999 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
135 page-faults:u # 0.284 K/sec
1,487,467,717 cycles:u # 3.131 GHz (82.95%)
186,537,162 stalled-cycles-frontend:u # 12.54% frontend cycles idle (82.93%)
28,797,869 stalled-cycles-backend:u # 1.94% backend cycles idle (67.10%)
3,807,564,203 instructions:u # 2.56 insn per cycle
# 0.05 stalled cycles per insn (83.57%)
1,049,344,291 branches:u # 2208.693 M/sec (83.60%)
5,485 branch-misses:u # 0.00% of all branches (83.58%)

0.475760235 seconds time elapsed

b) bsearch() of un-existing key "100001 + 2":

BASE
====

$ perf stat ./a.out

Performance counter stats for './a.out':

499.244480 task-clock:u (msec) # 0.999 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
132 page-faults:u # 0.264 K/sec
1,571,194,855 cycles:u # 3.147 GHz (83.18%)
13,450,980 stalled-cycles-frontend:u # 0.86% frontend cycles idle (83.18%)
21,256,072 stalled-cycles-backend:u # 1.35% backend cycles idle (66.78%)
4,171,197,909 instructions:u # 2.65 insn per cycle
# 0.01 stalled cycles per insn (83.68%)
1,009,175,281 branches:u # 2021.405 M/sec (83.79%)
3,122 branch-misses:u # 0.00% of all branches (83.37%)

0.499871144 seconds time elapsed

PATCHED
=======

$ perf stat ./a.out

Performance counter stats for './a.out':

399.023499 task-clock:u (msec) # 0.998 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
134 page-faults:u # 0.336 K/sec
1,245,793,991 cycles:u # 3.122 GHz (83.39%)
11,529,273 stalled-cycles-frontend:u # 0.93% frontend cycles idle (83.46%)
12,116,311 stalled-cycles-backend:u # 0.97% backend cycles idle (66.92%)
3,679,710,005 instructions:u # 2.95 insn per cycle
# 0.00 stalled cycles per insn (83.47%)
1,009,792,625 branches:u # 2530.660 M/sec (83.46%)
2,590 branch-misses:u # 0.00% of all branches (83.12%)

0.399733539 seconds time elapsed

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
---
lib/bsearch.c | 22 ++++++++++++----------
1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/lib/bsearch.c b/lib/bsearch.c
index e33c179089db..18b445b010c3 100644
--- a/lib/bsearch.c
+++ b/lib/bsearch.c
@@ -33,19 +33,21 @@
void *bsearch(const void *key, const void *base, size_t num, size_t size,
int (*cmp)(const void *key, const void *elt))
{
- size_t start = 0, end = num;
+ const char *pivot;
int result;

- while (start < end) {
- size_t mid = start + (end - start) / 2;
+ while (num > 0) {
+ pivot = base + (num >> 1) * size;
+ result = cmp(key, pivot);

- result = cmp(key, base + mid * size);
- if (result < 0)
- end = mid;
- else if (result > 0)
- start = mid + 1;
- else
- return (void *)base + mid * size;
+ if (result == 0)
+ return (void *)pivot;
+
+ if (result > 0) {
+ base = pivot + size;
+ num--;
+ }
+ num >>= 1;
}

return NULL;
--
2.13.1
\
 
 \ /
  Last update: 2017-06-12 01:07    [W:1.254 / U:0.020 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site