On 12/20/24 8:24 PM, Jeff Davis wrote:
On Fri, 2024-12-20 at 06:20 +0100, Andreas Karlsson wrote:
SELECT count(upper) FROM (SELECT upper(('Kålhuvud ' || i) COLLATE
"sv-SE-x-icu") FROM generate_series(1, 1000000) i);
master: ~540 ms
Patched: ~460 ms
glibc: ~410 ms
It looks like you are opening and closing the UCaseMap object each
time. Why not save it in pg_locale_t? That should speed it up even more
and hopefully beat libc.
Fixed. New benchmarks are:
SELECT count(upper) FROM (SELECT upper(('Kålhuvud ' || i) COLLATE
"sv-SE-x-icu") FROM generate_series(1, 1000000) i);
master: ~570 ms
Patched: ~340 ms
glibc: ~400 ms
So it does indeed seem like we got a further speedup and now are faster
than glibc.
Also, to support older ICU versions consistently, we need to fix up the
locale name to support "und"; cf. pg_ucol_open(). Perhaps factor out
that logic?
Fixed.
Andreas
From 138ecc65c85aeec7a1c0459f69642fd1ea3103db Mon Sep 17 00:00:00 2001
From: Andreas Karlsson <[email protected]>
Date: Tue, 17 Dec 2024 22:47:00 +0100
Subject: [PATCH v2] Use optimized versions of ICU case conversion for UTF-8
Instead of converting to and from UChar when doing case conversions we
use the UTF-8 versions of the functions. This can give a signficant
speedup, 30-40%, on short to medium length strings.
The only cost we incur is that we have to allocate a casemap object on
locale initialization for UTF-8 databases but the object is realtively
small and the assumption is that most users will at some point want to
run case conversion functions.
While at it we also remove some duplication in the non-UTF-8 code.
---
src/backend/utils/adt/pg_locale_icu.c | 253 +++++++++++++++++---------
src/include/utils/pg_locale.h | 2 +
2 files changed, 164 insertions(+), 91 deletions(-)
diff --git a/src/backend/utils/adt/pg_locale_icu.c b/src/backend/utils/adt/pg_locale_icu.c
index 43d44fe43bd..02d9efd0d64 100644
--- a/src/backend/utils/adt/pg_locale_icu.c
+++ b/src/backend/utils/adt/pg_locale_icu.c
@@ -52,6 +52,7 @@ extern pg_locale_t create_pg_locale_icu(Oid collid, MemoryContext context);
#ifdef USE_ICU
extern UCollator *pg_ucol_open(const char *loc_str);
+static UCaseMap *pg_ucasemap_open(const char *loc_str);
static size_t strlower_icu(char *dest, size_t destsize, const char *src,
ssize_t srclen, pg_locale_t locale);
@@ -111,9 +112,12 @@ static size_t icu_from_uchar(char *dest, size_t destsize,
const UChar *buff_uchar, int32_t len_uchar);
static void icu_set_collation_attributes(UCollator *collator, const char *loc,
UErrorCode *status);
-static int32_t icu_convert_case(ICU_Convert_Func func, pg_locale_t mylocale,
- UChar **buff_dest, UChar *buff_source,
- int32_t len_source);
+static int32_t icu_convert_case_no_utf8(ICU_Convert_Func func, char *dest,
+ size_t destsize, const char *src,
+ ssize_t srclen, pg_locale_t locale);
+static int32_t icu_convert_case_uchar(ICU_Convert_Func func, pg_locale_t mylocale,
+ UChar **buff_dest, UChar *buff_source,
+ int32_t len_source);
static int32_t u_strToTitle_default_BI(UChar *dest, int32_t destCapacity,
const UChar *src, int32_t srcLength,
const char *locale,
@@ -140,6 +144,8 @@ tolower_icu(pg_wchar wc, pg_locale_t locale)
return u_tolower(wc);
}
+static int32_t icu_foldcase_options(const char *locale);
+
static const struct collate_methods collate_methods_icu = {
.strncoll = strncoll_icu,
.strnxfrm = strnxfrm_icu,
@@ -278,6 +284,7 @@ create_pg_locale_icu(Oid collid, MemoryContext context)
const char *icurules = NULL;
UCollator *collator;
locale_t loc = (locale_t) 0;
+ UCaseMap *casemap = NULL;
pg_locale_t result;
if (collid == DEFAULT_COLLATION_OID)
@@ -339,10 +346,14 @@ create_pg_locale_icu(Oid collid, MemoryContext context)
collator = make_icu_collator(iculocstr, icurules);
+ if (GetDatabaseEncoding() == PG_UTF8)
+ casemap = pg_ucasemap_open(iculocstr);
+
result = MemoryContextAllocZero(context, sizeof(struct pg_locale_struct));
result->icu.locale = MemoryContextStrdup(context, iculocstr);
result->icu.ucol = collator;
result->icu.lt = loc;
+ result->icu.ucasemap = casemap;
result->deterministic = deterministic;
result->collate_is_c = false;
result->ctype_is_c = false;
@@ -366,41 +377,18 @@ create_pg_locale_icu(Oid collid, MemoryContext context)
#ifdef USE_ICU
/*
- * Wrapper around ucol_open() to handle API differences for older ICU
- * versions.
- *
- * Ensure that no path leaks a UCollator.
+ * In ICU versions 54 and earlier, "und" is not a recognized spelling of the
+ * root locale. If the first component of the locale is "und", replace with
+ * "root" before opening.
*/
-UCollator *
-pg_ucol_open(const char *loc_str)
+static char *
+fix_icu_locale_str(const char *loc_str)
{
- UCollator *collator;
- UErrorCode status;
- const char *orig_str = loc_str;
- char *fixed_str = NULL;
-
- /*
- * Must never open default collator, because it depends on the environment
- * and may change at any time. Should not happen, but check here to catch
- * bugs that might be hard to catch otherwise.
- *
- * NB: the default collator is not the same as the collator for the root
- * locale. The root locale may be specified as the empty string, "und", or
- * "root". The default collator is opened by passing NULL to ucol_open().
- */
- if (loc_str == NULL)
- elog(ERROR, "opening default collator is not supported");
-
- /*
- * In ICU versions 54 and earlier, "und" is not a recognized spelling of
- * the root locale. If the first component of the locale is "und", replace
- * with "root" before opening.
- */
if (U_ICU_VERSION_MAJOR_NUM < 55)
{
char lang[ULOC_LANG_CAPACITY];
+ UErrorCode status = U_ZERO_ERROR;
- status = U_ZERO_ERROR;
uloc_getLanguage(loc_str, lang, ULOC_LANG_CAPACITY, &status);
if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
{
@@ -413,15 +401,49 @@ pg_ucol_open(const char *loc_str)
if (strcmp(lang, "und") == 0)
{
const char *remainder = loc_str + strlen("und");
+ char *fixed_str;
fixed_str = palloc(strlen("root") + strlen(remainder) + 1);
strcpy(fixed_str, "root");
strcat(fixed_str, remainder);
- loc_str = fixed_str;
+ return fixed_str;
}
}
+ return NULL;
+}
+
+/*
+ * Wrapper around ucol_open() to handle API differences for older ICU
+ * versions.
+ *
+ * Ensure that no path leaks a UCollator.
+ */
+UCollator *
+pg_ucol_open(const char *loc_str)
+{
+ UCollator *collator;
+ UErrorCode status;
+ const char *orig_str = loc_str;
+ char *fixed_str;
+
+ /*
+ * Must never open default collator, because it depends on the environment
+ * and may change at any time. Should not happen, but check here to catch
+ * bugs that might be hard to catch otherwise.
+ *
+ * NB: the default collator is not the same as the collator for the root
+ * locale. The root locale may be specified as the empty string, "und", or
+ * "root". The default collator is opened by passing NULL to ucol_open().
+ */
+ if (loc_str == NULL)
+ elog(ERROR, "opening default collator is not supported");
+
+ fixed_str = fix_icu_locale_str(loc_str);
+ if (fixed_str)
+ loc_str = fixed_str;
+
status = U_ZERO_ERROR;
collator = ucol_open(loc_str, &status);
if (U_FAILURE(status))
@@ -456,6 +478,34 @@ pg_ucol_open(const char *loc_str)
return collator;
}
+/*
+ * Wrapper around ucasemap_open() to handle API differences for older ICU
+ * versions.
+ *
+ * Additional makes sure we get the right options for case folding.
+ */
+static UCaseMap *
+pg_ucasemap_open(const char *loc_str)
+{
+ UErrorCode status = U_ZERO_ERROR;
+ UCaseMap *casemap;
+ char *fixed_str;
+
+ fixed_str = fix_icu_locale_str(loc_str);
+ if (fixed_str)
+ loc_str = fixed_str;
+
+ casemap = ucasemap_open(loc_str, icu_foldcase_options(loc_str), &status);
+ if (U_FAILURE(status))
+ ereport(ERROR,
+ (errmsg("casemap lookup failed: %s", u_errorName(status))));
+
+ if (fixed_str != NULL)
+ pfree(fixed_str);
+
+ return casemap;
+}
+
/*
* Create a UCollator with the given locale string and rules.
*
@@ -528,80 +578,76 @@ static size_t
strlower_icu(char *dest, size_t destsize, const char *src, ssize_t srclen,
pg_locale_t locale)
{
- int32_t len_uchar;
- int32_t len_conv;
- UChar *buff_uchar;
- UChar *buff_conv;
- size_t result_len;
-
- len_uchar = icu_to_uchar(&buff_uchar, src, srclen);
- len_conv = icu_convert_case(u_strToLower, locale,
- &buff_conv, buff_uchar, len_uchar);
- result_len = icu_from_uchar(dest, destsize, buff_conv, len_conv);
- pfree(buff_uchar);
- pfree(buff_conv);
+ if (GetDatabaseEncoding() == PG_UTF8)
+ {
+ UErrorCode status = U_ZERO_ERROR;
+ int32_t needed;
- return result_len;
+ needed = ucasemap_utf8ToLower(locale->icu.ucasemap, dest, destsize, src, srclen, &status);
+ if (status != U_BUFFER_OVERFLOW_ERROR && U_FAILURE(status))
+ ereport(ERROR,
+ (errmsg("case conversion failed: %s", u_errorName(status))));
+ return needed;
+ }
+ else
+ return icu_convert_case_no_utf8(u_strToLower, dest, destsize, src, srclen, locale);
}
static size_t
strtitle_icu(char *dest, size_t destsize, const char *src, ssize_t srclen,
pg_locale_t locale)
{
- int32_t len_uchar;
- int32_t len_conv;
- UChar *buff_uchar;
- UChar *buff_conv;
- size_t result_len;
-
- len_uchar = icu_to_uchar(&buff_uchar, src, srclen);
- len_conv = icu_convert_case(u_strToTitle_default_BI, locale,
- &buff_conv, buff_uchar, len_uchar);
- result_len = icu_from_uchar(dest, destsize, buff_conv, len_conv);
- pfree(buff_uchar);
- pfree(buff_conv);
+ if (GetDatabaseEncoding() == PG_UTF8)
+ {
+ UErrorCode status = U_ZERO_ERROR;
+ int32_t needed;
- return result_len;
+ needed = ucasemap_utf8ToTitle(locale->icu.ucasemap, dest, destsize, src, srclen, &status);
+ if (status != U_BUFFER_OVERFLOW_ERROR && U_FAILURE(status))
+ ereport(ERROR,
+ (errmsg("case conversion failed: %s", u_errorName(status))));
+ return needed;
+ }
+ else
+ return icu_convert_case_no_utf8(u_strToTitle_default_BI, dest, destsize, src, srclen, locale);
}
static size_t
strupper_icu(char *dest, size_t destsize, const char *src, ssize_t srclen,
pg_locale_t locale)
{
- int32_t len_uchar;
- int32_t len_conv;
- UChar *buff_uchar;
- UChar *buff_conv;
- size_t result_len;
-
- len_uchar = icu_to_uchar(&buff_uchar, src, srclen);
- len_conv = icu_convert_case(u_strToUpper, locale,
- &buff_conv, buff_uchar, len_uchar);
- result_len = icu_from_uchar(dest, destsize, buff_conv, len_conv);
- pfree(buff_uchar);
- pfree(buff_conv);
+ if (GetDatabaseEncoding() == PG_UTF8)
+ {
+ UErrorCode status = U_ZERO_ERROR;
+ int32_t needed;
- return result_len;
+ needed = ucasemap_utf8ToUpper(locale->icu.ucasemap, dest, destsize, src, srclen, &status);
+ if (status != U_BUFFER_OVERFLOW_ERROR && U_FAILURE(status))
+ ereport(ERROR,
+ (errmsg("case conversion failed: %s", u_errorName(status))));
+ return needed;
+ }
+ else
+ return icu_convert_case_no_utf8(u_strToUpper, dest, destsize, src, srclen, locale);
}
static size_t
strfold_icu(char *dest, size_t destsize, const char *src, ssize_t srclen,
pg_locale_t locale)
{
- int32_t len_uchar;
- int32_t len_conv;
- UChar *buff_uchar;
- UChar *buff_conv;
- size_t result_len;
-
- len_uchar = icu_to_uchar(&buff_uchar, src, srclen);
- len_conv = icu_convert_case(u_strFoldCase_default, locale,
- &buff_conv, buff_uchar, len_uchar);
- result_len = icu_from_uchar(dest, destsize, buff_conv, len_conv);
- pfree(buff_uchar);
- pfree(buff_conv);
+ if (GetDatabaseEncoding() == PG_UTF8)
+ {
+ UErrorCode status = U_ZERO_ERROR;
+ int32_t needed;
- return result_len;
+ needed = ucasemap_utf8FoldCase(locale->icu.ucasemap, dest, destsize, src, srclen, &status);
+ if (status != U_BUFFER_OVERFLOW_ERROR && U_FAILURE(status))
+ ereport(ERROR,
+ (errmsg("case conversion failed: %s", u_errorName(status))));
+ return needed;
+ }
+ else
+ return icu_convert_case_no_utf8(u_strFoldCase_default, dest, destsize, src, srclen, locale);
}
/*
@@ -829,8 +875,28 @@ icu_from_uchar(char *dest, size_t destsize, const UChar *buff_uchar, int32_t len
}
static int32_t
-icu_convert_case(ICU_Convert_Func func, pg_locale_t mylocale,
- UChar **buff_dest, UChar *buff_source, int32_t len_source)
+icu_convert_case_no_utf8(ICU_Convert_Func func, char *dest, size_t destsize,
+ const char *src, ssize_t srclen, pg_locale_t locale)
+{
+ int32_t len_uchar;
+ int32_t len_conv;
+ UChar *buff_uchar;
+ UChar *buff_conv;
+ size_t result_len;
+
+ len_uchar = icu_to_uchar(&buff_uchar, src, srclen);
+ len_conv = icu_convert_case_uchar(func, locale, &buff_conv,
+ buff_uchar, len_uchar);
+ result_len = icu_from_uchar(dest, destsize, buff_conv, len_conv);
+ pfree(buff_uchar);
+ pfree(buff_conv);
+
+ return result_len;
+}
+
+static int32_t
+icu_convert_case_uchar(ICU_Convert_Func func, pg_locale_t mylocale,
+ UChar **buff_dest, UChar *buff_source, int32_t len_source)
{
UErrorCode status;
int32_t len_dest;
@@ -870,10 +936,17 @@ u_strFoldCase_default(UChar *dest, int32_t destCapacity,
const UChar *src, int32_t srcLength,
const char *locale,
UErrorCode *pErrorCode)
+{
+ return u_strFoldCase(dest, destCapacity, src, srcLength,
+ icu_foldcase_options(locale), pErrorCode);
+}
+
+static int32_t
+icu_foldcase_options(const char *locale)
{
uint32 options = U_FOLD_CASE_DEFAULT;
char lang[3];
- UErrorCode status;
+ UErrorCode status = U_ZERO_ERROR;
/*
* Unlike the ICU APIs for lowercasing, titlecasing, and uppercasing, case
@@ -881,7 +954,6 @@ u_strFoldCase_default(UChar *dest, int32_t destCapacity,
* option relevant to Turkic languages 'az' and 'tr'; check for those
* languages to enable the option.
*/
- status = U_ZERO_ERROR;
uloc_getLanguage(locale, lang, 3, &status);
if (U_SUCCESS(status))
{
@@ -893,8 +965,7 @@ u_strFoldCase_default(UChar *dest, int32_t destCapacity,
options = U_FOLD_CASE_EXCLUDE_SPECIAL_I;
}
- return u_strFoldCase(dest, destCapacity, src, srcLength,
- options, pErrorCode);
+ return options;
}
/*
diff --git a/src/include/utils/pg_locale.h b/src/include/utils/pg_locale.h
index 86016b9344e..a4995e046aa 100644
--- a/src/include/utils/pg_locale.h
+++ b/src/include/utils/pg_locale.h
@@ -21,6 +21,7 @@
#undef U_SHOW_CPLUSPLUS_HEADER_API
#define U_SHOW_CPLUSPLUS_HEADER_API 0
#include <unicode/ucol.h>
+#include <unicode/ucasemap.h>
#endif
/* use for libc locale names */
@@ -168,6 +169,7 @@ struct pg_locale_struct
const char *locale;
UCollator *ucol;
locale_t lt;
+ UCaseMap *ucasemap;
} icu;
#endif
};
--
2.47.3
