[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: speed up u8_strstr
From: |
Bruno Haible |
Subject: |
Re: speed up u8_strstr |
Date: |
Fri, 21 Jan 2011 13:41:11 +0100 |
User-agent: |
KMail/1.9.9 |
Hi Pádraig,
> > But at the same time, can you also try to apply the O(n) algorithm to
> > u16_strstr and u32_strstr?
>
> Attached is a first attempt at this.
Thanks for working on this.
> Questions I have are:
>
> Is it OK to include str-mp.h and malloca
> in unistr/ modules like I do?
Yes, sure. Why not?
Possible namespacing problems are solved by libunistring's Makefile.
Let me go through your patch.
1) It's a good choice to use the Knuth-Morris-Pratt algorithm for u16_strstr
and u32_strstr. It's O(n). Eric's two-way algorithm is O(n) as well, but is
more complicated code and requires to set up a table whose size is 1 <<
CHAR_BIT,
which would be 1 << 16 for uint16_t or 1 << 32 for uint32_t - infeasible.
2) It's a good choice to parameterize str-kmp.h, rather than to duplicate code.
But while doing so, please try to avoid gcc warnings. Always test your patches
with CPPFLAGS="-Wall" before submitting them:
mbscasestr.c:376: warning: pointer targets in passing argument 1 of
‘knuth_morris_pratt’ differ in signedness
mbscasestr.c:376: warning: pointer targets in passing argument 2 of
‘knuth_morris_pratt’ differ in signedness
mbscasestr.c:376: warning: passing argument 4 of ‘knuth_morris_pratt’ from
incompatible pointer type
mbsstr.c:347: warning: pointer targets in passing argument 1 of
‘knuth_morris_pratt’ differ in signedness
mbsstr.c:347: warning: pointer targets in passing argument 2 of
‘knuth_morris_pratt’ differ in signedness
And there's no reason to remove part of the specification of CANON_ELEMENT in
lib/str-kmp.h. Removing reasonable comments is a good way to have your patches
rejected.
3) About the module descriptions:
Makefile.am:
if LIBUNISTRING_COMPILE_UNISTR_U16_STRSTR
-lib_SOURCES += unistr/u16-strstr.c
+lib_SOURCES += unistr/u16-strstr.c str-kmp.h
endif
You don't need to add .h files to lib_SOURCES.
4) The line gl_LIBUNISTRING_MODULE([0.9], [unistr/u32-strstr])
needs to be updated, because with a libunistring from versions 0.9 .. 0.9.3
the new unit tests will fail. So when a libunistring version < 0.9.4 is found,
you need to enforce compilation of the module from source. This is done by
changing
the line to
gl_LIBUNISTRING_MODULE([0.9.4], [unistr/u32-strstr])
5) The unit test tests/unistr/test-u-strstr.h: It is good to share code among
tests.
rather than write down the same code 3 times. But please follow GNU coding
style:
indentation by 2 spaces, not 4, and put spaces around arithmetic operators.
More importantly, though, the test code that you put there is part of
tests/test-strstr.c and tests/test-mbsstr1.c. Why only a part? u16_strstr and
u32_strstr should have all the properties that strstr and mbsstr have. So the
right thing to do is to adapt all of tests/test-strstr.c into
tests/unistr/test-u-strstr.h. I've done this, and now I get test failures:
/bin/sh: line 5: 24308 Alarm clock EXEEXT='' srcdir='.' MAKE='make'
${dir}$tst
FAIL: test-u16-strstr
/bin/sh: line 5: 24325 Alarm clock EXEEXT='' srcdir='.' MAKE='make'
${dir}$tst
FAIL: test-u32-strstr
6) In unit tests, the order of the #includes is the following:
- First comes <config.h>.
- Then comes the header file supposed to declare the particular functionality.
This comes immediately after <config.h> in order to check that the header is
self-contained.
- Then come system includes <...>.
- Then come gnulib includes "...".
7) Also please go into more details when writing a ChangeLog entry. Always
mention
the function names, especially when you're renaming a function.
So, the first part that can go in is this:
2011-01-21 Pádraig Brady <address@hidden>
Bruno Haible <address@hidden>
Prepare for faster uN_strstr functions.
* lib/str-kmp.h: Support definable UNITs.
(knuth_morris_pratt): Renamed from knuth_morris_pratt_unibyte. Add
needle_len argument.
* lib/mbsstr.c (mbsstr): Adjust for the changed str-kmp.h.
* lib/mbscasestr.c (mbscasestr): Likewise.
--- lib/mbscasestr.c.orig Fri Jan 21 13:29:32 2011
+++ lib/mbscasestr.c Fri Jan 21 13:05:06 2011
@@ -30,6 +30,7 @@
#define TOLOWER(Ch) (isupper (Ch) ? tolower (Ch) : (Ch))
/* Knuth-Morris-Pratt algorithm. */
+#define UNIT unsigned char
#define CANON_ELEMENT(c) TOLOWER (c)
#include "str-kmp.h"
@@ -368,10 +369,12 @@
if (needle_last_ccount == NULL)
{
/* Try the Knuth-Morris-Pratt algorithm. */
- const char *result;
+ const unsigned char *result;
bool success =
- knuth_morris_pratt_unibyte (haystack, needle - 1,
- &result);
+ knuth_morris_pratt ((const unsigned char *) haystack,
+ (const unsigned char *) (needle -
1),
+ strlen (needle - 1),
+ &result);
if (success)
return (char *) result;
try_kmp = false;
--- lib/mbsstr.c.orig Fri Jan 21 13:29:32 2011
+++ lib/mbsstr.c Fri Jan 21 13:05:06 2011
@@ -27,6 +27,7 @@
#include "mbuiter.h"
/* Knuth-Morris-Pratt algorithm. */
+#define UNIT unsigned char
#define CANON_ELEMENT(c) c
#include "str-kmp.h"
@@ -339,10 +340,12 @@
if (needle_last_ccount == NULL)
{
/* Try the Knuth-Morris-Pratt algorithm. */
- const char *result;
+ const unsigned char *result;
bool success =
- knuth_morris_pratt_unibyte (haystack, needle - 1,
- &result);
+ knuth_morris_pratt ((const unsigned char *) haystack,
+ (const unsigned char *) (needle -
1),
+ strlen (needle - 1),
+ &result);
if (success)
return (char *) result;
try_kmp = false;
--- lib/str-kmp.h.orig Fri Jan 21 13:29:32 2011
+++ lib/str-kmp.h Fri Jan 21 13:05:06 2011
@@ -1,4 +1,4 @@
-/* Substring search in a NUL terminated string of 'char' elements,
+/* Substring search in a NUL terminated string of UNIT elements,
using the Knuth-Morris-Pratt algorithm.
Copyright (C) 2005-2011 Free Software Foundation, Inc.
Written by Bruno Haible <address@hidden>, 2005.
@@ -18,21 +18,26 @@
Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */
/* Before including this file, you need to define:
+ UNIT The element type of the needle and haystack.
CANON_ELEMENT(c) A macro that canonicalizes an element right after
- it has been fetched from one of the two strings.
- The argument is an 'unsigned char'; the result
- must be an 'unsigned char' as well. */
+ it has been fetched from needle or haystack.
+ The argument is of type UNIT; the result must be
+ of type UNIT as well. */
/* Knuth-Morris-Pratt algorithm.
See http://en.wikipedia.org/wiki/Knuth-Morris-Pratt_algorithm
+ HAYSTACK is the NUL terminated string in which to search for.
+ NEEDLE is the string to search for in HAYSTACK, consisting of NEEDLE_LEN
+ units.
Return a boolean indicating success:
Return true and set *RESULTP if the search was completed.
Return false if it was aborted because not enough memory was available. */
static bool
-knuth_morris_pratt_unibyte (const char *haystack, const char *needle,
- const char **resultp)
+knuth_morris_pratt (const UNIT *haystack,
+ const UNIT *needle, size_t needle_len,
+ const UNIT **resultp)
{
- size_t m = strlen (needle);
+ size_t m = needle_len;
/* Allocate the table. */
size_t *table = (size_t *) nmalloca (m, sizeof (size_t));
@@ -66,14 +71,14 @@
The inequality needle[x..i-1] != needle[0..i-1-x] is known to hold
for x < table[i-1], by induction.
Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
- unsigned char b = CANON_ELEMENT ((unsigned char) needle[i - 1]);
+ UNIT b = CANON_ELEMENT (needle[i - 1]);
for (;;)
{
/* Invariants: The inequality needle[x..i-1] != needle[0..i-1-x]
is known to hold for x < i-1-j.
Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
- if (b == CANON_ELEMENT ((unsigned char) needle[j]))
+ if (b == CANON_ELEMENT (needle[j]))
{
/* Set table[i] := i-1-j. */
table[i] = i - ++j;
@@ -108,17 +113,16 @@
/* Search, using the table to accelerate the processing. */
{
size_t j;
- const char *rhaystack;
- const char *phaystack;
+ const UNIT *rhaystack;
+ const UNIT *phaystack;
*resultp = NULL;
j = 0;
rhaystack = haystack;
phaystack = haystack;
/* Invariant: phaystack = rhaystack + j. */
- while (*phaystack != '\0')
- if (CANON_ELEMENT ((unsigned char) needle[j])
- == CANON_ELEMENT ((unsigned char) *phaystack))
+ while (*phaystack != 0)
+ if (CANON_ELEMENT (needle[j]) == CANON_ELEMENT (*phaystack))
{
j++;
phaystack++;