Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20170530122324.23733-1-ncopa@alpinelinux.org>
Date: Tue, 30 May 2017 14:23:24 +0200
From: Natanael Copa <ncopa@...inelinux.org>
To: musl@...ts.openwall.com
Cc: Natanael Copa <ncopa@...inelinux.org>
Subject: [PATCH] towupper/towlower: fast path for ascii chars

Make a fast path for ascii chars which is assumed to be the most common
case. This has significant performance benefit on xml json and similar

---

This gives a performance boost for the given testcase:
https://gist.github.com/maksis/92ad04f525d69043283350675d04f160

Before:
Completed in 8.302969s, compare count 54136421

After:
Completed in 2.745886s, compare count 54136421

I don't consider this the final solution but it is atleast a significant
improvement.

 src/ctype/towctrans.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/ctype/towctrans.c b/src/ctype/towctrans.c
index 6af61875..cf13a862 100644
--- a/src/ctype/towctrans.c
+++ b/src/ctype/towctrans.c
@@ -1,3 +1,4 @@
+#include <ctype.h>
 #include <wctype.h>
 #include "libc.h"
 
@@ -9,7 +10,6 @@ static const struct {
 	signed char lower;
 	unsigned char len;
 } casemaps[] = {
-	CASEMAP('A','Z','a'),
 	CASEMAP(0xc0,0xde,0xe0),
 
 	CASELACE(0x0100,0x012e),
@@ -257,12 +257,12 @@ static wchar_t __towcase(wchar_t wc, int lower)
 
 wint_t towupper(wint_t wc)
 {
-	return __towcase(wc, 0);
+	return (unsigned)wc < 128 ? toupper(wc) : __towcase(wc, 0);
 }
 
 wint_t towlower(wint_t wc)
 {
-	return __towcase(wc, 1);
+	return (unsigned)wc < 128 ? tolower(wc) : __towcase(wc, 1);
 }
 
 wint_t __towupper_l(wint_t c, locale_t l)
-- 
2.13.0

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.