Improve strstr performance of short needles

Improve strstr performance for the common case of short needles.  For a single
character strchr is best, for 2-4 characters a small loop is fastest.  For these
the speedup over the Two-Way algorithm is ~10 times on large strings.

Newlib builds, the new code passes GLIBC testsuite. OK for commit?
This commit is contained in:
Wilco Dijkstra 2018-09-04 17:54:21 +00:00 committed by Corinna Vinschen
parent 86c31ae47b
commit 6dbb20dfc7
1 changed files with 65 additions and 10 deletions

View File

@ -30,20 +30,11 @@ QUICKREF
#include <string.h> #include <string.h>
#if !defined(PREFER_SIZE_OVER_SPEED) && !defined(__OPTIMIZE_SIZE__) #if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
# define RETURN_TYPE char *
# define AVAILABLE(h, h_l, j, n_l) \
(!memchr ((h) + (h_l), '\0', (j) + (n_l) - (h_l)) \
&& ((h_l) = (j) + (n_l)))
# include "str-two-way.h"
#endif
char * char *
strstr (const char *searchee, strstr (const char *searchee,
const char *lookfor) const char *lookfor)
{ {
#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
/* Less code size, but quadratic performance in the worst case. */ /* Less code size, but quadratic performance in the worst case. */
if (*searchee == 0) if (*searchee == 0)
{ {
@ -77,6 +68,58 @@ strstr (const char *searchee,
#else /* compilation for speed */ #else /* compilation for speed */
# define RETURN_TYPE char *
# define AVAILABLE(h, h_l, j, n_l) \
(!memchr ((h) + (h_l), '\0', (j) + (n_l) - (h_l)) \
&& ((h_l) = (j) + (n_l)))
# include "str-two-way.h"
static inline char *
strstr2 (const char *hs, const char *ne)
{
uint32_t h1 = (ne[0] << 16) | ne[1];
uint32_t h2 = 0;
int c = hs[0];
while (h1 != h2 && c != 0)
{
h2 = (h2 << 16) | c;
c = *++hs;
}
return h1 == h2 ? (char *)hs - 2 : NULL;
}
static inline char *
strstr3 (const char *hs, const char *ne)
{
uint32_t h1 = (ne[0] << 24) | (ne[1] << 16) | (ne[2] << 8);
uint32_t h2 = 0;
int c = hs[0];
while (h1 != h2 && c != 0)
{
h2 = (h2 | c) << 8;
c = *++hs;
}
return h1 == h2 ? (char *)hs - 3 : NULL;
}
static inline char *
strstr4 (const char *hs, const char *ne)
{
uint32_t h1 = (ne[0] << 24) | (ne[1] << 16) | (ne[2] << 8) | ne[3];
uint32_t h2 = 0;
int c = hs[0];
while (h1 != h2 && c != 0)
{
h2 = (h2 << 8) | c;
c = *++hs;
}
return h1 == h2 ? (char *)hs - 4 : NULL;
}
char *
strstr (const char *searchee,
const char *lookfor)
{
/* Larger code size, but guaranteed linear performance. */ /* Larger code size, but guaranteed linear performance. */
const char *haystack = searchee; const char *haystack = searchee;
const char *needle = lookfor; const char *needle = lookfor;
@ -84,6 +127,18 @@ strstr (const char *searchee,
size_t haystack_len; /* Known minimum length of HAYSTACK. */ size_t haystack_len; /* Known minimum length of HAYSTACK. */
int ok = 1; /* True if NEEDLE is prefix of HAYSTACK. */ int ok = 1; /* True if NEEDLE is prefix of HAYSTACK. */
/* Handle short needle special cases first. */
if (needle[0] == '\0')
return (char *) haystack;
if (needle[1] == '\0')
return strchr (haystack, needle[0]);
if (needle[2] == '\0')
return strstr2 (haystack, needle);
if (needle[3] == '\0')
return strstr3 (haystack, needle);
if (needle[4] == '\0')
return strstr4 (haystack, needle);
/* Determine length of NEEDLE, and in the process, make sure /* Determine length of NEEDLE, and in the process, make sure
HAYSTACK is at least as long (no point processing all of a long HAYSTACK is at least as long (no point processing all of a long
NEEDLE if HAYSTACK is too short). */ NEEDLE if HAYSTACK is too short). */