Optimize stripos/stristr

Closes GH-7847

Previously stripos/stristr would lowercase both the haystack and the
needle to reuse strpos. The approach in this PR is similar to strpos.
memchr is highly optimized so we're using it to search for the first
character of the needle in the haystack. If we find it we compare the
remaining characters of the needle manually.

The new implementation seems to perform about half as well as strpos (as
two memchr calls are necessary to find the next candidate).

Author: iluuu1994

Zend/zend_operators.h

@@ -919,6 +919,63 @@ static zend_always_inline void zend_unwrap_reference(zval *op) /* {{{ */
 }
 /* }}} */
 
+static zend_always_inline bool zend_strnieq(const char *ptr1, const char *ptr2, size_t num)
+{
+	const char *end = ptr1 + num;
+	while (ptr1 < end) {
+		if (zend_tolower_ascii(*ptr1++) != zend_tolower_ascii(*ptr2++)) {
+			return 0;
+		}
+	}
+	return 1;
+}
+
+static zend_always_inline const char *
+zend_memnistr(const char *haystack, const char *needle, size_t needle_len, const char *end)
+{
+	ZEND_ASSERT(end >= haystack);
+
+	if (UNEXPECTED(needle_len == 0)) {
+		return haystack;
+	}
+
+	const char first_lower = zend_tolower_ascii(*needle);
+	const char first_upper = zend_toupper_ascii(*needle);
+	const char *p_lower = (const char *)memchr(haystack, first_lower, end - haystack);
+	const char *p_upper = (first_lower != first_upper) ? (const char *)memchr(haystack, first_upper, end - haystack) : NULL;
+	const char *p = !p_upper || (p_lower && p_lower < p_upper) ? p_lower : p_upper;
+
+	if (needle_len == 1) {
+		return p;
+	}
+
+	size_t off_s = (size_t)(end - p);
+	if (needle_len > off_s) {
+		return NULL;
+	}
+
+	const char needle_end_lower = zend_tolower_ascii(needle[needle_len - 1]);
+	const char needle_end_upper = zend_toupper_ascii(needle[needle_len - 1]);
+	end -= needle_len;
+
+	while (p && p <= end) {
+		if (needle_end_lower == p[needle_len - 1] || needle_end_upper == p[needle_len - 1]) {
+			if(zend_strnieq(needle + 1, p + 1, needle_len - 2)){
+				return p;
+			}
+		}
+		if (p_lower == p) {
+			p_lower = (const char *)memchr(p_lower + 1, first_lower, end - p_lower);
+		}
+		if (p_upper == p) {
+			p_upper = (const char *)memchr(p_upper + 1, first_upper, end - p_upper);
+		}
+		p = !p_upper || (p_lower && p_lower < p_upper) ? p_lower : p_upper;
+	}
+
+	return NULL;
+}
+
 
 END_EXTERN_C()
 

ext/standard/php_string.h

@@ -48,6 +48,7 @@ PHPAPI void php_stripcslashes(zend_string *str);
 PHPAPI zend_string *php_basename(const char *s, size_t len, const char *suffix, size_t sufflen);
 PHPAPI size_t php_dirname(char *str, size_t len);
 PHPAPI char *php_stristr(char *s, char *t, size_t s_len, size_t t_len);
+PHPAPI const char *php_stristr_no_tolower(const char *s, const char *t, size_t s_len, size_t t_len);
 PHPAPI zend_string *php_str_to_str(const char *haystack, size_t length, const char *needle,
 		size_t needle_len, const char *str, size_t str_len);
 PHPAPI zend_string *php_trim(zend_string *str, const char *what, size_t what_len, int mode);

ext/standard/string.c

@@ -1693,6 +1693,11 @@ PHPAPI char *php_stristr(char *s, char *t, size_t s_len, size_t t_len)
 }
 /* }}} */
 
+PHPAPI const char *php_stristr_no_tolower(const char *s, const char *t, size_t s_len, size_t t_len)
+{
+	return (const char*)php_memnistr(s, t, t_len, s + s_len);
+}
+
 /* {{{ php_strspn */
 PHPAPI size_t php_strspn(const char *s1, const char *s2, const char *s1_end, const char *s2_end)
 {
@@ -1735,8 +1740,6 @@ PHP_FUNCTION(stristr)
 	zend_string *haystack, *needle;
 	const char *found = NULL;
 	size_t  found_offset;
-	char *haystack_dup;
-	char *orig_needle;
 	bool part = 0;
 
 	ZEND_PARSE_PARAMETERS_START(2, 3)
@@ -1746,13 +1749,10 @@ PHP_FUNCTION(stristr)
 		Z_PARAM_BOOL(part)
 	ZEND_PARSE_PARAMETERS_END();
 
-	haystack_dup = estrndup(ZSTR_VAL(haystack), ZSTR_LEN(haystack));
-	orig_needle = estrndup(ZSTR_VAL(needle), ZSTR_LEN(needle));
-	found = php_stristr(haystack_dup, orig_needle, ZSTR_LEN(haystack), ZSTR_LEN(needle));
-	efree(orig_needle);
+	found = php_stristr_no_tolower(ZSTR_VAL(haystack), ZSTR_VAL(needle), ZSTR_LEN(haystack), ZSTR_LEN(needle));
 
 	if (found) {
-		found_offset = found - haystack_dup;
+		found_offset = found - ZSTR_VAL(haystack);
 		if (part) {
 			RETVAL_STRINGL(ZSTR_VAL(haystack), found_offset);
 		} else {
@@ -1761,8 +1761,6 @@ PHP_FUNCTION(stristr)
 	} else {
 		RETVAL_FALSE;
 	}
-
-	efree(haystack_dup);
 }
 /* }}} */
 
@@ -1890,7 +1888,6 @@ PHP_FUNCTION(stripos)
 	const char *found = NULL;
 	zend_string *haystack, *needle;
 	zend_long offset = 0;
-	zend_string *needle_dup = NULL, *haystack_dup;
 
 	ZEND_PARSE_PARAMETERS_START(2, 3)
 		Z_PARAM_STR(haystack)
@@ -1911,19 +1908,14 @@ PHP_FUNCTION(stripos)
 		RETURN_FALSE;
 	}
 
-	haystack_dup = zend_string_tolower(haystack);
-	needle_dup = zend_string_tolower(needle);
-	found = (char*)php_memnstr(ZSTR_VAL(haystack_dup) + offset,
-			ZSTR_VAL(needle_dup), ZSTR_LEN(needle_dup), ZSTR_VAL(haystack_dup) + ZSTR_LEN(haystack));
+	found = (char*)php_memnistr(ZSTR_VAL(haystack) + offset,
+			ZSTR_VAL(needle), ZSTR_LEN(needle), ZSTR_VAL(haystack) + ZSTR_LEN(haystack));
 
 	if (found) {
-		RETVAL_LONG(found - ZSTR_VAL(haystack_dup));
+		RETVAL_LONG(found - ZSTR_VAL(haystack));
 	} else {
 		RETVAL_FALSE;
 	}
-
-	zend_string_release_ex(haystack_dup, 0);
-	zend_string_release_ex(needle_dup, 0);
 }
 /* }}} */
 

main/php.h

@@ -344,6 +344,7 @@ END_EXTERN_C()
 #define phpin zendin
 
 #define php_memnstr zend_memnstr
+#define php_memnistr zend_memnistr
 
 /* functions */
 BEGIN_EXTERN_C()