From patchwork Fri Nov 18 15:40:54 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Wilco Dijkstra X-Patchwork-Id: 17565 Received: (qmail 74348 invoked by alias); 18 Nov 2016 15:41:08 -0000 Mailing-List: contact libc-alpha-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-alpha-owner@sourceware.org Delivered-To: mailing list libc-alpha@sourceware.org Received: (qmail 74315 invoked by uid 89); 18 Nov 2016 15:41:07 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-1.6 required=5.0 tests=AWL, BAYES_00, KAM_LOTSOFHASH, RCVD_IN_DNSWL_NONE, SPF_HELO_PASS autolearn=no version=3.3.2 spammy=4956, 1257, 1277 X-HELO: EUR01-VE1-obe.outbound.protection.outlook.com From: Wilco Dijkstra To: "libc-alpha@sourceware.org" CC: nd Subject: [PATCH] Improve strsep Date: Fri, 18 Nov 2016 15:40:54 +0000 Message-ID: authentication-results: spf=none (sender IP is ) smtp.mailfrom=Wilco.Dijkstra@arm.com; x-microsoft-exchange-diagnostics: 1; AM5PR0802MB2611; 7:PexGxSiW0+c0vXr6UVm3an+Bn6Quzkc/KcbvSaz/xzV/yj7fQKW2leEU8uc0hZZ1EaZkDexlBUX62xKBzdCybFBgTh1RijAJG9Xkqhm2eTekc/i5plj2m0PJ9Kf6YDlrNMsIOVPJ5tnwVV/+1Lbrn2Q3nfdRsri+JS74Zhs0xoh8KgutJlja9lrmh7BiIEa0YLjLSSc+aDr3TD+f8U1XBieiit7GKJa0HLj1dPPSNjWZ3ZRydsev2w32donmlFJYC0hTnz9kAp6EM/2d/LE57qXmWJU9OtDS5/2BWnG/vI1Ok+FGlOjh3bvtCAIVBbkeMJE91wGJL/x2qEOoHMZG60ij5aum+ad/D24j927UBKM= x-ms-office365-filtering-correlation-id: de7fa169-d559-4283-8011-08d40fc9488b x-microsoft-antispam: UriScan:; BCL:0; PCL:0; RULEID:(22001); SRVR:AM5PR0802MB2611; nodisclaimer: True x-microsoft-antispam-prvs: x-exchange-antispam-report-test: UriScan:(180628864354917); x-exchange-antispam-report-cfa-test: BCL:0; PCL:0; RULEID:(6040281)(6060326)(601004)(2401047)(8121501046)(5005006)(3002001)(10201501046)(6055026)(6041223)(6061324); SRVR:AM5PR0802MB2611; BCL:0; PCL:0; RULEID:; SRVR:AM5PR0802MB2611; x-forefront-prvs: 01304918F3 x-forefront-antispam-report: SFV:NSPM; SFS:(10009020)(6009001)(7916002)(54534003)(377424004)(189002)(199003)(2501003)(5640700001)(54356999)(76576001)(74316002)(86362001)(7736002)(575784001)(3280700002)(105586002)(305945005)(106356001)(7846002)(50986999)(6506003)(101416001)(5660300001)(106116001)(8676002)(3660700001)(3846002)(2351001)(102836003)(6116002)(450100001)(92566002)(6916009)(4001150100001)(122556002)(33656002)(4326007)(68736007)(66066001)(7696004)(2906002)(110136003)(189998001)(87936001)(81156014)(97736004)(2900100001)(81166006)(9686002)(8936002)(77096005)(38730400001); DIR:OUT; SFP:1101; SCL:1; SRVR:AM5PR0802MB2611; H:AM5PR0802MB2610.eurprd08.prod.outlook.com; FPR:; SPF:None; PTR:InfoNoRecords; MX:1; A:1; LANG:en; received-spf: None (protection.outlook.com: arm.com does not designate permitted sender hosts) spamdiagnosticoutput: 1:99 spamdiagnosticmetadata: NSPM MIME-Version: 1.0 X-OriginatorOrg: arm.com X-MS-Exchange-CrossTenant-originalarrivaltime: 18 Nov 2016 15:40:54.4706 (UTC) X-MS-Exchange-CrossTenant-fromentityheader: Hosted X-MS-Exchange-CrossTenant-id: f34e5979-57d9-4aaa-ad4d-b122a662184d X-MS-Exchange-Transport-CrossTenantHeadersStamped: AM5PR0802MB2611 This patch cleans up the strsep implementation and improves performance. Currently strsep calls strpbrk which is is now a veneer to strcspn. Calling strcspn directly is faster. Since it handles a delimiter string of size 1 as a special case, this is not needed in strsep itself. Although this means there is a slightly higher overhead if the delimiter size is 1, all other cases are slightly faster. The overall performance gain is 5-10% on AArch64. The string/bits/string2.h header contains optimizations for constant delimiters of size 1-3. Benchmarking these showed similar performance for size 1 (since in all cases strchr/strchrnul is used), while size 2 and 3 can give up to 2x speedup for small input strings. However if these cases are common it seems much better to add this optimization to strcspn. So move these header optimizations to string-inlines.c. Improve the strsep benchmark so that it actually benchmarks something. The current version contains a delimiter character at every position in the input string, so there is very little work to do, and the extremely inefficent simple_strsep implementation appears fastest in every case. The new version has either no match in the input for the fail case and a match halfway in the input for the success case. The input is then restored so that each iteration does exactly the same amount of work. Reduce the number of testcases since simple_strsep takes a lot of time now. Passes GLIBC tests, OK for commit? ChangeLog: 2015-11-18 Wilco Dijkstra * benchtests/bench-strsep.c (oldstrsep): Add old implementation. (do_one_test) Restore original string so iteration works. * string/string-inlines.c (do_test): Create better input strings. (test_main) Reduce number of testruns. * string/string-inlines.c (__old_strsep_1c): New function. (__old_strsep_2c): Likewise. (__old_strsep_3c): Likewise. * string/strsep.c (__strsep): Remove case of small delim string. Call strcspn directly rather than strpbrk. * string/bits/string2.h (__strsep): Remove define. (__strsep_1c): Remove. (__strsep_2c): Remove. (__strsep_3c): Remove. diff --git a/benchtests/bench-strsep.c b/benchtests/bench-strsep.c index 70dbb377563232a43895abc940d396f2c65237c6..9e0c00d9f198776a3de4857446b82125566ba4b4 100644 --- a/benchtests/bench-strsep.c +++ b/benchtests/bench-strsep.c @@ -49,10 +49,56 @@ simple_strsep (char **s1, char *s2) return begin; } +char * +oldstrsep (char **stringp, const char *delim) +{ + char *begin, *end; + + begin = *stringp; + if (begin == NULL) + return NULL; + + /* A frequent case is when the delimiter string contains only one + character. Here we don't need to call the expensive `strpbrk' + function and instead work using `strchr'. */ + if (delim[0] == '\0' || delim[1] == '\0') + { + char ch = delim[0]; + + if (ch == '\0') + end = NULL; + else + { + if (*begin == ch) + end = begin; + else if (*begin == '\0') + end = NULL; + else + end = strchr (begin + 1, ch); + } + } + else + /* Find the end of the token. */ + end = strpbrk (begin, delim); + + if (end) + { + /* Terminate the token and set *STRINGP past NUL character. */ + *end++ = '\0'; + *stringp = end; + } + else + /* No more delimiters; this is the last token. */ + *stringp = NULL; + + return begin; +} + typedef char *(*proto_t) (const char **, const char *); IMPL (simple_strsep, 0) IMPL (strsep, 1) +IMPL (oldstrsep, 2) static void do_one_test (impl_t * impl, const char *s1, const char *s2) @@ -63,7 +109,10 @@ do_one_test (impl_t * impl, const char *s1, const char *s2) TIMING_NOW (start); for (i = 0; i < iters; ++i) { - CALL (impl, &s1, s2); + const char *s1a = s1; + CALL (impl, &s1a, s2); + if (s1a != NULL) + ((char*)s1a)[-1] = '1'; } TIMING_NOW (stop); @@ -76,7 +125,7 @@ static void do_test (size_t align1, size_t align2, size_t len1, size_t len2, int fail) { char *s2 = (char *) (buf2 + align2); - static const char d[] = "1234567890abcdef"; + static const char d[] = "1234567891abcdef"; #define dl (sizeof (d) - 1) char *ss2 = s2; for (size_t l = len2; l > 0; l = l > dl ? l - dl : 0) @@ -92,24 +141,9 @@ do_test (size_t align1, size_t align2, size_t len1, size_t len2, int fail) FOR_EACH_IMPL (impl, 0) { char *s1 = (char *) (buf1 + align1); - if (fail) - { - char *ss1 = s1; - for (size_t l = len1; l > 0; l = l > dl ? l - dl : 0) - { - size_t t = l > dl ? dl : l; - memcpy (ss1, d, t); - ++ss1[len2 > 7 ? 7 : len2 - 1]; - ss1 += t; - } - } - else - { - memset (s1, '0', len1); - memcpy (s1 + (len1 - len2) - 2, s2, len2); - if ((len1 / len2) > 4) - memcpy (s1 + (len1 - len2) - (3 * len2), s2, len2); - } + memset (s1, '0', len1); + if (!fail) + s1[len1 / 2] = '1'; s1[len1] = '\0'; do_one_test (impl, s1, s2); } @@ -127,7 +161,7 @@ test_main (void) putchar ('\n'); for (size_t klen = 2; klen < 32; ++klen) - for (size_t hlen = 2 * klen; hlen < 16 * klen; hlen += klen) + for (size_t hlen = 4 * klen; hlen < 8 * klen; hlen += klen) { do_test (0, 0, hlen, klen, 0); do_test (0, 0, hlen, klen, 1); diff --git a/string/bits/string2.h b/string/bits/string2.h index 0370b9777b68ce8a152e6c49985503a0f6ff8bbd..8fa35d52e7c8e3ff592573fa64472da526e8616d 100644 --- a/string/bits/string2.h +++ b/string/bits/string2.h @@ -175,91 +175,7 @@ #endif -#if !defined _HAVE_STRING_ARCH_strsep || defined _FORCE_INLINES -# ifndef _HAVE_STRING_ARCH_strsep - -extern char *__strsep_g (char **__stringp, const char *__delim); -# define __strsep(s, reject) \ - __extension__ \ - ({ char __r0, __r1, __r2; \ - (__builtin_constant_p (reject) && __string2_1bptr_p (reject) \ - && (__r0 = ((const char *) (reject))[0], \ - ((const char *) (reject))[0] != '\0') \ - ? ((__r1 = ((const char *) (reject))[1], \ - ((const char *) (reject))[1] == '\0') \ - ? __strsep_1c (s, __r0) \ - : ((__r2 = ((const char *) (reject))[2], __r2 == '\0') \ - ? __strsep_2c (s, __r0, __r1) \ - : (((const char *) (reject))[3] == '\0' \ - ? __strsep_3c (s, __r0, __r1, __r2) \ - : __strsep_g (s, reject)))) \ - : __strsep_g (s, reject)); }) -# endif - -__STRING_INLINE char *__strsep_1c (char **__s, char __reject); -__STRING_INLINE char * -__strsep_1c (char **__s, char __reject) -{ - char *__retval = *__s; - if (__retval != NULL && (*__s = strchr (__retval, __reject)) != NULL) - *(*__s)++ = '\0'; - return __retval; -} - -__STRING_INLINE char *__strsep_2c (char **__s, char __reject1, char __reject2); -__STRING_INLINE char * -__strsep_2c (char **__s, char __reject1, char __reject2) -{ - char *__retval = *__s; - if (__retval != NULL) - { - char *__cp = __retval; - while (1) - { - if (*__cp == '\0') - { - __cp = NULL; - break; - } - if (*__cp == __reject1 || *__cp == __reject2) - { - *__cp++ = '\0'; - break; - } - ++__cp; - } - *__s = __cp; - } - return __retval; -} - -__STRING_INLINE char *__strsep_3c (char **__s, char __reject1, char __reject2, - char __reject3); -__STRING_INLINE char * -__strsep_3c (char **__s, char __reject1, char __reject2, char __reject3) -{ - char *__retval = *__s; - if (__retval != NULL) - { - char *__cp = __retval; - while (1) - { - if (*__cp == '\0') - { - __cp = NULL; - break; - } - if (*__cp == __reject1 || *__cp == __reject2 || *__cp == __reject3) - { - *__cp++ = '\0'; - break; - } - ++__cp; - } - *__s = __cp; - } - return __retval; -} +#if !defined _HAVE_STRING_ARCH_strsep # ifdef __USE_MISC # define strsep(s, reject) __strsep (s, reject) # endif diff --git a/string/string-inlines.c b/string/string-inlines.c index d43e5897c37430e5f97940469d65e7ddbdcbd09c..7b1de7180f8be3dca1515e23553a9ecbdc76f5e2 100644 --- a/string/string-inlines.c +++ b/string/string-inlines.c @@ -63,6 +63,70 @@ __old_strtok_r_1c (char *__s, char __sep, char **__nextp) return __result; } compat_symbol (libc, __old_strtok_r_1c, __strtok_r_1c, GLIBC_2_1_1); + +char * +__old_strsep_1c (char **__s, char __reject) +{ + char *__retval = *__s; + if (__retval != NULL && (*__s = strchr (__retval, __reject)) != NULL) + *(*__s)++ = '\0'; + return __retval; +} +compat_symbol (libc, __old_strsep_1c, __strsep_1c, GLIBC_2_1_1); + +char * +__old_strsep_2c (char **__s, char __reject1, char __reject2) +{ + char *__retval = *__s; + if (__retval != NULL) + { + char *__cp = __retval; + while (1) + { + if (*__cp == '\0') + { + __cp = NULL; + break; + } + if (*__cp == __reject1 || *__cp == __reject2) + { + *__cp++ = '\0'; + break; + } + ++__cp; + } + *__s = __cp; + } + return __retval; +} +compat_symbol (libc, __old_strsep_2c, __strsep_2c, GLIBC_2_1_1); + +char * +__old_strsep_3c (char **__s, char __reject1, char __reject2, char __reject3) +{ + char *__retval = *__s; + if (__retval != NULL) + { + char *__cp = __retval; + while (1) + { + if (*__cp == '\0') + { + __cp = NULL; + break; + } + if (*__cp == __reject1 || *__cp == __reject2 || *__cp == __reject3) + { + *__cp++ = '\0'; + break; + } + ++__cp; + } + *__s = __cp; + } + return __retval; +} +compat_symbol (libc, __old_strsep_3c, __strsep_3c, GLIBC_2_1_1); #endif #if SHLIB_COMPAT (libc, GLIBC_2_1_1, GLIBC_2_24) diff --git a/string/strsep.c b/string/strsep.c index 10547740481bed8f7a065f76eccfe7b320c3c49c..68581c863980c1e86ae8559400ca0d1cec80d95e 100644 --- a/string/strsep.c +++ b/string/strsep.c @@ -29,30 +29,10 @@ __strsep (char **stringp, const char *delim) if (begin == NULL) return NULL; - /* A frequent case is when the delimiter string contains only one - character. Here we don't need to call the expensive `strpbrk' - function and instead work using `strchr'. */ - if (delim[0] == '\0' || delim[1] == '\0') - { - char ch = delim[0]; - - if (ch == '\0') - end = NULL; - else - { - if (*begin == ch) - end = begin; - else if (*begin == '\0') - end = NULL; - else - end = strchr (begin + 1, ch); - } - } - else - /* Find the end of the token. */ - end = strpbrk (begin, delim); + /* Find the end of the token. */ + end = begin + strcspn (begin, delim); - if (end) + if (*end) { /* Terminate the token and set *STRINGP past NUL character. */ *end++ = '\0';