Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitfc6c727

Browse files
committed
Fix compiler builtin usage in new pg_bitutils.c
Split out these new functions in three parts: one in a new file thatuses the compiler builtin and gets compiled with the -mpopcnt compileroption if it exists; another one that uses the compiler builtin but notthe compiler option; and finally the fallback with open-codedalgorithms.Split out the configure logic: in the original commit, it was selectingto use the -mpopcnt compiler switch together with deciding whether touse the compiler builtin, but those two things are really separate.Split them out. Also, expose whether the builtin exists toMakefile.global, so that src/port's Makefile can decide whether tocompile the hw-optimized file.Remove CPUID test for CTZ/CLZ. Make pg_{right,left}most_ones use eitherthe compiler intrinsic or open-coded algo; trying to use theHW-optimized version is a waste of time. Make them static inlinefunctions.Discussion:https://postgr.es/m/20190213221719.GA15976@alvherre.pgsql
1 parentb060e6c commitfc6c727

File tree

8 files changed

+327
-376
lines changed

8 files changed

+327
-376
lines changed

‎config/c-compiler.m4‎

Lines changed: 8 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -381,22 +381,16 @@ fi])# PGAC_C_BUILTIN_OP_OVERFLOW
381381
# PGAC_C_BUILTIN_POPCOUNT
382382
# -------------------------
383383
AC_DEFUN([PGAC_C_BUILTIN_POPCOUNT],
384-
[define([Ac_cachevar],[AS_TR_SH([pgac_cv_popcount])])dnl
385-
AC_CACHE_CHECK([for __builtin_popcount],[Ac_cachevar],
386-
[pgac_save_CFLAGS=$CFLAGS
387-
CFLAGS="$pgac_save_CFLAGS -mpopcnt"
388-
AC_COMPILE_IFELSE([AC_LANG_SOURCE(
389-
[static int x = __builtin_popcount(255);])],
390-
[Ac_cachevar=yes],
391-
[Ac_cachevar=no])
392-
CFLAGS="$pgac_save_CFLAGS"])
393-
if test x"$Ac_cachevar" = x"yes"; then
394-
CFLAGS_POPCNT="-mpopcnt"
384+
[AC_CACHE_CHECK([for __builtin_popcount],pgac_cv__builtin_popcount,
385+
[AC_COMPILE_IFELSE([AC_LANG_SOURCE(
386+
[static int x = __builtin_popcount(255);]
387+
)],
388+
[pgac_cv__builtin_popcount=yes],
389+
[pgac_cv__builtin_popcount=no])])
390+
if test x"$pgac_cv__builtin_popcount" = x"yes"; then
395391
AC_DEFINE(HAVE__BUILTIN_POPCOUNT,1,
396392
[Define to 1 if your compiler understands __builtin_popcount.])
397-
fi
398-
undefine([Ac_cachevar])dnl
399-
])# PGAC_C_BUILTIN_POPCOUNT
393+
fi])# PGAC_C_BUILTIN_POPCOUNT
400394

401395

402396

‎configure‎

Lines changed: 53 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -651,7 +651,7 @@ CFLAGS_ARMV8_CRC32C
651651
CFLAGS_SSE42
652652
have_win32_dbghelp
653653
LIBOBJS
654-
CFLAGS_POPCNT
654+
have__builtin_popcount
655655
UUID_LIBS
656656
LDAP_LIBS_BE
657657
LDAP_LIBS_FE
@@ -733,6 +733,7 @@ CPP
733733
BITCODE_CXXFLAGS
734734
BITCODE_CFLAGS
735735
CFLAGS_VECTOR
736+
CFLAGS_POPCNT
736737
PERMIT_DECLARATION_AFTER_STATEMENT
737738
LLVM_BINPATH
738739
LLVM_CXXFLAGS
@@ -6581,6 +6582,48 @@ fi
65816582

65826583
fi
65836584

6585+
# Optimization flags and options for bit-twiddling
6586+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether ${CC} supports -mpopcnt, for CFLAGS_POPCNT" >&5
6587+
$as_echo_n "checking whether ${CC} supports -mpopcnt, for CFLAGS_POPCNT... " >&6; }
6588+
if ${pgac_cv_prog_CC_cflags__mpopcnt+:} false; then :
6589+
$as_echo_n "(cached) " >&6
6590+
else
6591+
pgac_save_CFLAGS=$CFLAGS
6592+
pgac_save_CC=$CC
6593+
CC=${CC}
6594+
CFLAGS="${CFLAGS_POPCNT} -mpopcnt"
6595+
ac_save_c_werror_flag=$ac_c_werror_flag
6596+
ac_c_werror_flag=yes
6597+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
6598+
/* end confdefs.h. */
6599+
6600+
int
6601+
main ()
6602+
{
6603+
6604+
;
6605+
return 0;
6606+
}
6607+
_ACEOF
6608+
if ac_fn_c_try_compile "$LINENO"; then :
6609+
pgac_cv_prog_CC_cflags__mpopcnt=yes
6610+
else
6611+
pgac_cv_prog_CC_cflags__mpopcnt=no
6612+
fi
6613+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
6614+
ac_c_werror_flag=$ac_save_c_werror_flag
6615+
CFLAGS="$pgac_save_CFLAGS"
6616+
CC="$pgac_save_CC"
6617+
fi
6618+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_prog_CC_cflags__mpopcnt" >&5
6619+
$as_echo "$pgac_cv_prog_CC_cflags__mpopcnt" >&6; }
6620+
if test x"$pgac_cv_prog_CC_cflags__mpopcnt" = x"yes"; then
6621+
CFLAGS_POPCNT="${CFLAGS_POPCNT} -mpopcnt"
6622+
fi
6623+
6624+
6625+
6626+
65846627
CFLAGS_VECTOR=$CFLAGS_VECTOR
65856628

65866629

@@ -14111,32 +14154,28 @@ $as_echo "#define HAVE__BUILTIN_CTZ 1" >>confdefs.h
1411114154
fi
1411214155
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_popcount" >&5
1411314156
$as_echo_n "checking for __builtin_popcount... " >&6; }
14114-
if ${pgac_cv_popcount+:} false; then :
14157+
if ${pgac_cv__builtin_popcount+:} false; then :
1411514158
$as_echo_n "(cached) " >&6
1411614159
else
14117-
pgac_save_CFLAGS=$CFLAGS
14118-
CFLAGS="$pgac_save_CFLAGS -mpopcnt"
14119-
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14160+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
1412014161
/* end confdefs.h. */
1412114162
static int x = __builtin_popcount(255);
14163+
1412214164
_ACEOF
1412314165
if ac_fn_c_try_compile "$LINENO"; then :
14124-
pgac_cv_popcount=yes
14166+
pgac_cv__builtin_popcount=yes
1412514167
else
14126-
pgac_cv_popcount=no
14168+
pgac_cv__builtin_popcount=no
1412714169
fi
1412814170
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14129-
CFLAGS="$pgac_save_CFLAGS"
1413014171
fi
14131-
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_popcount" >&5
14132-
$as_echo "$pgac_cv_popcount" >&6; }
14133-
if test x"$pgac_cv_popcount" = x"yes"; then
14134-
CFLAGS_POPCNT="-mpopcnt"
14172+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv__builtin_popcount" >&5
14173+
$as_echo "$pgac_cv__builtin_popcount" >&6; }
14174+
if test x"$pgac_cv__builtin_popcount" = x"yes"; then
1413514175

1413614176
$as_echo "#define HAVE__BUILTIN_POPCOUNT 1" >>confdefs.h
1413714177

1413814178
fi
14139-
1414014179
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_unreachable" >&5
1414114180
$as_echo_n "checking for __builtin_unreachable... " >&6; }
1414214181
if ${pgac_cv__builtin_unreachable+:} false; then :
@@ -14654,6 +14693,7 @@ $as_echo "#define LOCALE_T_IN_XLOCALE 1" >>confdefs.h
1465414693

1465514694
fi
1465614695

14696+
have__builtin_popcount=$pgac_cv__builtin_popcount
1465714697

1465814698

1465914699
# MSVC doesn't cope well with defining restrict to __restrict, the

‎configure.in‎

Lines changed: 5 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -547,6 +547,10 @@ elif test "$PORTNAME" = "hpux"; then
547547
PGAC_PROG_CXX_CFLAGS_OPT([+Olibmerrno])
548548
fi
549549

550+
# Optimization flags and options for bit-twiddling
551+
PGAC_PROG_CC_VAR_OPT(CFLAGS_POPCNT, [-mpopcnt])
552+
AC_SUBST(CFLAGS_POPCNT)
553+
550554
AC_SUBST(CFLAGS_VECTOR, $CFLAGS_VECTOR)
551555

552556
# Determine flags used to emit bitcode for JIT inlining. Need to test
@@ -1506,7 +1510,7 @@ AC_TYPE_LONG_LONG_INT
15061510

15071511
PGAC_TYPE_LOCALE_T
15081512

1509-
AC_SUBST(CFLAGS_POPCNT)
1513+
AC_SUBST(have__builtin_popcount, $pgac_cv__builtin_popcount)
15101514

15111515
# MSVC doesn't cope well with defining restrict to __restrict, the
15121516
# spelling it understands, because it conflicts with

‎src/Makefile.global.in‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -517,6 +517,9 @@ WIN32_STACK_RLIMIT=4194304
517517
# Set if we have a working win32 crashdump header
518518
have_win32_dbghelp = @have_win32_dbghelp@
519519

520+
# Set if __builtin_popcount() is supported by $(CC)
521+
have__builtin_popcount = @have__builtin_popcount@
522+
520523
# Pull in platform-specific magic
521524
include$(top_builddir)/src/Makefile.port
522525

‎src/include/port/pg_bitutils.h‎

Lines changed: 168 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -10,17 +10,177 @@
1010
*
1111
*------------------------------------------------------------------------ -
1212
*/
13-
1413
#ifndefPG_BITUTILS_H
1514
#definePG_BITUTILS_H
1615

17-
externint (*pg_popcount32) (uint32word);
18-
externint (*pg_popcount64) (uint64word);
19-
externint (*pg_rightmost_one32) (uint32word);
20-
externint (*pg_rightmost_one64) (uint64word);
21-
externint (*pg_leftmost_one32) (uint32word);
22-
externint (*pg_leftmost_one64) (uint64word);
23-
16+
externint(*pg_popcount32) (uint32word);
17+
externint(*pg_popcount64) (uint64word);
2418
externuint64pg_popcount(constchar*buf,intbytes);
2519

20+
/* in pg_bitutils_hwpopcnt.c */
21+
externintpg_popcount32_hw(uint32word);
22+
externintpg_popcount64_hw(uint64word);
23+
24+
25+
#ifndefHAVE__BUILTIN_CTZ
26+
/*
27+
* Array marking the position of the right-most set bit for each value of
28+
* 1-255. We count the right-most position as the 0th bit, and the
29+
* left-most the 7th bit. The 0th index of the array must not be used.
30+
*/
31+
staticconstuint8rightmost_one_pos[256]= {
32+
0,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
33+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
34+
5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
35+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
36+
6,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
37+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
38+
5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
39+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
40+
7,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
41+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
42+
5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
43+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
44+
6,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
45+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
46+
5,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0,
47+
4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0
48+
};
49+
#endif/* !HAVE__BUILTIN_CTZ */
50+
51+
/*
52+
* pg_rightmost_one32
53+
*Returns the number of trailing 0-bits in word, starting at the least
54+
*significant bit position. word must not be 0.
55+
*/
56+
staticinlineint
57+
pg_rightmost_one32(uint32word)
58+
{
59+
intresult=0;
60+
61+
Assert(word!=0);
62+
63+
#ifdefHAVE__BUILTIN_CTZ
64+
result=__builtin_ctz(word);
65+
#else
66+
while ((word&255)==0)
67+
{
68+
word >>=8;
69+
result+=8;
70+
}
71+
result+=rightmost_one_pos[word&255];
72+
#endif/* HAVE__BUILTIN_CTZ */
73+
74+
returnresult;
75+
}
76+
77+
/*
78+
* pg_rightmost_one64
79+
*Returns the number of trailing 0-bits in word, starting at the least
80+
*significant bit position. word must not be 0.
81+
*/
82+
staticinlineint
83+
pg_rightmost_one64(uint64word)
84+
{
85+
intresult=0;
86+
87+
Assert(word!=0);
88+
#ifdefHAVE__BUILTIN_CTZ
89+
#if defined(HAVE_LONG_INT_64)
90+
return__builtin_ctzl(word);
91+
#elif defined(HAVE_LONG_LONG_INT_64)
92+
return__builtin_ctzll(word);
93+
#else
94+
#error must have a working 64-bit integer datatype
95+
#endif
96+
#else/* HAVE__BUILTIN_CTZ */
97+
while ((word&255)==0)
98+
{
99+
word >>=8;
100+
result+=8;
101+
}
102+
result+=rightmost_one_pos[word&255];
103+
#endif
104+
105+
returnresult;
106+
}
107+
108+
#ifndefHAVE__BUILTIN_CLZ
109+
/*
110+
* Array marking the position of the left-most set bit for each value of
111+
* 1-255. We count the right-most position as the 0th bit, and the
112+
* left-most the 7th bit. The 0th index of the array must not be used.
113+
*/
114+
staticconstuint8leftmost_one_pos[256]= {
115+
0,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,
116+
4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
117+
5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
118+
5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
119+
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
120+
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
121+
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
122+
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
123+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
124+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
125+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
126+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
127+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
128+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
129+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
130+
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7
131+
};
132+
#endif/* !HAVE_BUILTIN_CLZ */
133+
134+
/*
135+
* pg_leftmost_one32
136+
*Returns the 0-based position of the most significant set bit in word
137+
*measured from the least significant bit. word must not be 0.
138+
*/
139+
staticinlineint
140+
pg_leftmost_one32(uint32word)
141+
{
142+
#ifdefHAVE__BUILTIN_CLZ
143+
Assert(word!=0);
144+
145+
return31-__builtin_clz(word);
146+
#else
147+
intshift=32-8;
148+
149+
Assert(word!=0);
150+
151+
while ((word >>shift)==0)
152+
shift-=8;
153+
154+
returnshift+leftmost_one_pos[(word >>shift)&255];
155+
#endif/* HAVE__BUILTIN_CLZ */
156+
}
157+
158+
/*
159+
* pg_leftmost_one64
160+
*Returns the 0-based position of the most significant set bit in word
161+
*measured from the least significant bit. word must not be 0.
162+
*/
163+
staticinlineint
164+
pg_leftmost_one64(uint64word)
165+
{
166+
#ifdefHAVE__BUILTIN_CLZ
167+
Assert(word!=0);
168+
#if defined(HAVE_LONG_INT_64)
169+
return63-__builtin_clzl(word);
170+
#elif defined(HAVE_LONG_LONG_INT_64)
171+
return63-__builtin_clzll(word);
172+
#else
173+
#error must have a working 64-bit integer datatype
174+
#endif
175+
#else/* HAVE__BUILTIN_CLZ */
176+
intshift=64-8;
177+
178+
Assert(word!=0);
179+
while ((word >>shift)==0)
180+
shift-=8;
181+
182+
returnshift+leftmost_one_pos[(word >>shift)&255];
183+
#endif/* !HAVE__BUIILTIN_CLZ */
184+
}
185+
26186
#endif/* PG_BITUTILS_H */

‎src/port/Makefile‎

Lines changed: 12 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -41,6 +41,14 @@ OBJS = $(LIBOBJS) $(PG_CRC32C_OBJS) chklocale.o erand48.o inet_net_ntop.o \
4141
qsort.o qsort_arg.o quotes.o snprintf.o sprompt.o strerror.o\
4242
tar.o thread.o
4343

44+
# If the compiler supports a special flag for the POPCOUNT instruction and it
45+
# has __builtin_popcount, add pg_bitutils_hwpopcnt.o.
46+
ifneq ($(CFLAGS_POPCNT),)
47+
ifeq ($(have__builtin_popcount),yes)
48+
OBJS += pg_bitutils_hwpopcnt.o
49+
endif
50+
endif
51+
4452
# libpgport.a, libpgport_shlib.a, and libpgport_srv.a contain the same files
4553
# foo.o, foo_shlib.o, and foo_srv.o are all built from foo.c
4654
OBJS_SHLIB =$(OBJS:%.o=%_shlib.o)
@@ -78,10 +86,10 @@ pg_crc32c_armv8.o: CFLAGS+=$(CFLAGS_ARMV8_CRC32C)
7886
pg_crc32c_armv8_shlib.o: CFLAGS+=$(CFLAGS_ARMV8_CRC32C)
7987
pg_crc32c_armv8_srv.o: CFLAGS+=$(CFLAGS_ARMV8_CRC32C)
8088

81-
#pg_bitutils.cneeds CFLAGS_POPCNT
82-
pg_bitutils.o: CFLAGS+=$(CFLAGS_POPCNT)
83-
pg_bitutils_shlib.o: CFLAGS+=$(CFLAGS_POPCNT)
84-
pg_bitutils_srv.o: CFLAGS+=$(CFLAGS_POPCNT)
89+
#all versions of pg_bitutils_hwpopcnt.cneed CFLAGS_POPCNT
90+
pg_bitutils_hwpopcnt.o: CFLAGS+=$(CFLAGS_POPCNT)
91+
pg_bitutils_hwpopcnt_shlib.o: CFLAGS+=$(CFLAGS_POPCNT)
92+
pg_bitutils_hwpopcnt_srv.o: CFLAGS+=$(CFLAGS_POPCNT)
8593

8694
#
8795
# Shared library versions of object files

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp