I can't envision why the below would give us any improvement in
sysbench, since it only changes the symbol compilation for the
weight_string() function, which I am 100% positive does not exist in any
workload used by sysbench :)
What is the other user of it, Stewart?
-jay
Stewart Smith wrote:
On Thu, May 21, 2009 at 09:46:13PM -0700, Brian Aker wrote:
Drizzle:
==14109== L2d misses: 12,816,504 ( 10,170,854 rd +
2,645,650 wr)
MySQL:
==26463== L2d misses: 1,371,822 ( 591,124 rd +
780,698 wr)
So I put OProfile to work for a quick looksy.... found that this is
giving me at least a 5% speed boost on small read only sysbench
(innodb) as a lot of l2 cache misses were in my_strxfrm_utf8mb4.
The other main user of it... i didn't manage to "fix" so easily... but
this colud be an opportunity for improvement.
=== modified file 'drizzled/sql_yacc.yy'
--- drizzled/sql_yacc.yy 2009-05-21 19:15:01 +0000
+++ drizzled/sql_yacc.yy 2009-05-23 07:22:32 +0000
@@ -3203,13 +3203,13 @@ function_call_conflict:
| WEIGHT_STRING_SYM '(' expr AS CHAR_SYM ws_nweights opt_ws_levels ')'
{
$$= new (YYSession->mem_root)
- Item_func_weight_string($3, $6, $7|MY_STRXFRM_PAD_WITH_SPACE);
+ Item_func_weight_string($3, $6, $7);
}
| WEIGHT_STRING_SYM '(' expr AS BINARY ws_nweights ')'
{
$3= create_func_char_cast(YYSession, $3, $6, &my_charset_bin);
$$= new (YYSession->mem_root)
- Item_func_weight_string($3, $6, MY_STRXFRM_PAD_WITH_SPACE);
+ Item_func_weight_string($3, $6, 0);
}
;
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help : https://help.launchpad.net/ListHelp
