mirror of https://github.com/python/cpython
Issue #13134: optimize finding single-character strings using memchr
This commit is contained in:
parent
60dd7dc23c
commit
2c3b2302ad
|
@ -171,6 +171,15 @@ class UnicodeTest(string_tests.CommonTest,
|
|||
|
||||
def test_find(self):
|
||||
string_tests.CommonTest.test_find(self)
|
||||
# test implementation details of the memchr fast path
|
||||
self.checkequal(100, 'a' * 100 + '\u0102', 'find', '\u0102')
|
||||
self.checkequal(-1, 'a' * 100 + '\u0102', 'find', '\u0201')
|
||||
self.checkequal(-1, 'a' * 100 + '\u0102', 'find', '\u0120')
|
||||
self.checkequal(-1, 'a' * 100 + '\u0102', 'find', '\u0220')
|
||||
self.checkequal(100, 'a' * 100 + '\U00100304', 'find', '\U00100304')
|
||||
self.checkequal(-1, 'a' * 100 + '\U00100304', 'find', '\U00100204')
|
||||
self.checkequal(-1, 'a' * 100 + '\U00100304', 'find', '\U00102004')
|
||||
# check mixed argument types
|
||||
self.checkequalnofix(0, 'abcdefghiabc', 'find', 'abc')
|
||||
self.checkequalnofix(9, 'abcdefghiabc', 'find', 'abc', 1)
|
||||
self.checkequalnofix(-1, 'abcdefghiabc', 'find', 'def', 4)
|
||||
|
@ -180,6 +189,14 @@ class UnicodeTest(string_tests.CommonTest,
|
|||
|
||||
def test_rfind(self):
|
||||
string_tests.CommonTest.test_rfind(self)
|
||||
# test implementation details of the memrchr fast path
|
||||
self.checkequal(0, '\u0102' + 'a' * 100 , 'rfind', '\u0102')
|
||||
self.checkequal(-1, '\u0102' + 'a' * 100 , 'rfind', '\u0201')
|
||||
self.checkequal(-1, '\u0102' + 'a' * 100 , 'rfind', '\u0120')
|
||||
self.checkequal(-1, '\u0102' + 'a' * 100 , 'rfind', '\u0220')
|
||||
self.checkequal(0, '\U00100304' + 'a' * 100, 'rfind', '\U00100304')
|
||||
self.checkequal(-1, '\U00100304' + 'a' * 100, 'rfind', '\U00100204')
|
||||
self.checkequal(-1, '\U00100304' + 'a' * 100, 'rfind', '\U00102004')
|
||||
# check mixed argument types
|
||||
self.checkequalnofix(9, 'abcdefghiabc', 'rfind', 'abc')
|
||||
self.checkequalnofix(12, 'abcdefghiabc', 'rfind', '')
|
||||
|
|
|
@ -32,6 +32,60 @@
|
|||
#define STRINGLIB_BLOOM(mask, ch) \
|
||||
((mask & (1UL << ((ch) & (STRINGLIB_BLOOM_WIDTH -1)))))
|
||||
|
||||
|
||||
Py_LOCAL_INLINE(Py_ssize_t)
|
||||
STRINGLIB(fastsearch_memchr_1char)(const STRINGLIB_CHAR* s, Py_ssize_t n,
|
||||
STRINGLIB_CHAR ch, unsigned char needle,
|
||||
Py_ssize_t maxcount, int mode)
|
||||
{
|
||||
void *candidate;
|
||||
const STRINGLIB_CHAR *found;
|
||||
|
||||
#define DO_MEMCHR(memchr, s, needle, nchars) do { \
|
||||
candidate = memchr((const void *) (s), (needle), (nchars) * sizeof(STRINGLIB_CHAR)); \
|
||||
found = (const STRINGLIB_CHAR *) \
|
||||
((Py_ssize_t) candidate & (~ ((Py_ssize_t) sizeof(STRINGLIB_CHAR) - 1))); \
|
||||
} while (0)
|
||||
|
||||
if (mode == FAST_SEARCH) {
|
||||
const STRINGLIB_CHAR *_s = s;
|
||||
const STRINGLIB_CHAR *e = s + n;
|
||||
while (_s < e) {
|
||||
DO_MEMCHR(memchr, _s, needle, e - _s);
|
||||
if (found == NULL)
|
||||
return -1;
|
||||
if (sizeof(STRINGLIB_CHAR) == 1 || *found == ch)
|
||||
return (found - _s);
|
||||
/* False positive */
|
||||
_s = found + 1;
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
#ifdef HAVE_MEMRCHR
|
||||
/* memrchr() is a GNU extension, available since glibc 2.1.91.
|
||||
it doesn't seem as optimized as memchr(), but is still quite
|
||||
faster than our hand-written loop in FASTSEARCH below */
|
||||
else if (mode == FAST_RSEARCH) {
|
||||
while (n > 0) {
|
||||
DO_MEMCHR(memrchr, s, needle, n);
|
||||
if (found == NULL)
|
||||
return -1;
|
||||
n = found - s;
|
||||
if (sizeof(STRINGLIB_CHAR) == 1 || *found == ch)
|
||||
return n;
|
||||
/* False positive */
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
#endif
|
||||
else {
|
||||
assert(0); /* Should never get here */
|
||||
return 0;
|
||||
}
|
||||
|
||||
#undef DO_MEMCHR
|
||||
}
|
||||
|
||||
Py_LOCAL_INLINE(Py_ssize_t)
|
||||
FASTSEARCH(const STRINGLIB_CHAR* s, Py_ssize_t n,
|
||||
const STRINGLIB_CHAR* p, Py_ssize_t m,
|
||||
|
@ -51,6 +105,25 @@ FASTSEARCH(const STRINGLIB_CHAR* s, Py_ssize_t n,
|
|||
if (m <= 0)
|
||||
return -1;
|
||||
/* use special case for 1-character strings */
|
||||
if (n > 10 && (mode == FAST_SEARCH
|
||||
#ifdef HAVE_MEMRCHR
|
||||
|| mode == FAST_RSEARCH
|
||||
#endif
|
||||
)) {
|
||||
/* use memchr if we can choose a needle without two many likely
|
||||
false positives */
|
||||
unsigned char needle;
|
||||
int use_needle = 1;
|
||||
needle = p[0] & 0xff;
|
||||
if (needle == 0 && sizeof(STRINGLIB_CHAR) > 1) {
|
||||
needle = (p[0] >> 8) & 0xff;
|
||||
if (needle >= 32)
|
||||
use_needle = 0;
|
||||
}
|
||||
if (use_needle)
|
||||
return STRINGLIB(fastsearch_memchr_1char)
|
||||
(s, n, p[0], needle, maxcount, mode);
|
||||
}
|
||||
if (mode == FAST_COUNT) {
|
||||
for (i = 0; i < n; i++)
|
||||
if (s[i] == p[0]) {
|
||||
|
|
|
@ -2566,7 +2566,8 @@ AC_CHECK_FUNCS(alarm accept4 setitimer getitimer bind_textdomain_codeset chown \
|
|||
getgrouplist getgroups getlogin getloadavg getpeername getpgid getpid \
|
||||
getpriority getresuid getresgid getpwent getspnam getspent getsid getwd \
|
||||
if_nameindex \
|
||||
initgroups kill killpg lchmod lchown lockf linkat lstat lutimes mbrtowc mkdirat mkfifo \
|
||||
initgroups kill killpg lchmod lchown lockf linkat lstat lutimes memrchr \
|
||||
mbrtowc mkdirat mkfifo \
|
||||
mkfifoat mknod mknodat mktime mremap nice openat pathconf pause pipe2 plock poll \
|
||||
posix_fallocate posix_fadvise pread \
|
||||
pthread_init pthread_kill putenv pwrite readlink readlinkat readv realpath renameat \
|
||||
|
|
|
@ -515,6 +515,9 @@
|
|||
/* Define to 1 if you have the <memory.h> header file. */
|
||||
#undef HAVE_MEMORY_H
|
||||
|
||||
/* Define to 1 if you have the `memrchr' function. */
|
||||
#undef HAVE_MEMRCHR
|
||||
|
||||
/* Define to 1 if you have the `mkdirat' function. */
|
||||
#undef HAVE_MKDIRAT
|
||||
|
||||
|
|
Loading…
Reference in New Issue