Skip to content

Commit 711bab1

Browse files
alvherrealvherre
committed
Add basic support for using the POPCNT and SSE4.2s LZCNT opcodes
These opcodes have been around in the AMD world since 2007, and 2008 in the case of intel. They're supported in GCC and Clang via some __builtin macros. The opcodes may be unavailable during runtime, in which case we fall back on a C-based implementation of the code. In order to get the POPCNT instruction we must pass the -mpopcnt option to the compiler. We do this only for the pg_bitutils.c file. David Rowley (with fragments taken from a patch by Thomas Munro) Discussion: https://postgr.es/m/CAKJS1f9WTAGG1tPeJnD18hiQW5gAk59fQ6WK-vfdAKEHyRg2RA@mail.gmail.com
1 parent 754ca99 commit 711bab1

File tree

13 files changed

+914
-169
lines changed

13 files changed

+914
-169
lines changed

‎config/c-compiler.m4‎

Lines changed: 116 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -378,6 +378,122 @@ fi])# PGAC_C_BUILTIN_OP_OVERFLOW
378378

379379

380380

381+
# PGAC_C_BUILTIN_POPCOUNT
382+
# -------------------------
383+
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"
395+
AC_DEFINE(HAVE__BUILTIN_POPCOUNT, 1,
396+
[Define to 1 if your compiler understands __builtin_popcount.])
397+
fi
398+
undefine([Ac_cachevar])dnl
399+
])# PGAC_C_BUILTIN_POPCOUNT
400+
401+
402+
403+
# PGAC_C_BUILTIN_POPCOUNTL
404+
# -------------------------
405+
AC_DEFUN([PGAC_C_BUILTIN_POPCOUNTL],
406+
[define([Ac_cachevar], [AS_TR_SH([pgac_cv_popcountl])])dnl
407+
AC_CACHE_CHECK([for __builtin_popcountl], [Ac_cachevar],
408+
[pgac_save_CFLAGS=$CFLAGS
409+
CFLAGS="$pgac_save_CFLAGS -mpopcnt"
410+
AC_COMPILE_IFELSE([AC_LANG_SOURCE(
411+
[static int x = __builtin_popcountl(255);])],
412+
[Ac_cachevar=yes],
413+
[Ac_cachevar=no])
414+
CFLAGS="$pgac_save_CFLAGS"])
415+
if test x"$Ac_cachevar" = x"yes"; then
416+
CFLAGS_POPCNT="-mpopcnt"
417+
AC_DEFINE(HAVE__BUILTIN_POPCOUNTL, 1,
418+
[Define to 1 if your compiler understands __builtin_popcountl.])
419+
fi
420+
undefine([Ac_cachevar])dnl
421+
])# PGAC_C_BUILTIN_POPCOUNTL
422+
423+
424+
425+
# PGAC_C_BUILTIN_CTZ
426+
# -------------------------
427+
# Check if the C compiler understands __builtin_ctz(),
428+
# and define HAVE__BUILTIN_CTZ if so.
429+
AC_DEFUN([PGAC_C_BUILTIN_CTZ],
430+
[AC_CACHE_CHECK(for __builtin_ctz, pgac_cv__builtin_ctz,
431+
[AC_COMPILE_IFELSE([AC_LANG_SOURCE(
432+
[static int x = __builtin_ctz(256);]
433+
)],
434+
[pgac_cv__builtin_ctz=yes],
435+
[pgac_cv__builtin_ctz=no])])
436+
if test x"$pgac_cv__builtin_ctz" = xyes ; then
437+
AC_DEFINE(HAVE__BUILTIN_CTZ, 1,
438+
[Define to 1 if your compiler understands __builtin_ctz.])
439+
fi])# PGAC_C_BUILTIN_CTZ
440+
441+
442+
443+
# PGAC_C_BUILTIN_CTZL
444+
# -------------------------
445+
# Check if the C compiler understands __builtin_ctzl(),
446+
# and define HAVE__BUILTIN_CTZL if so.
447+
AC_DEFUN([PGAC_C_BUILTIN_CTZL],
448+
[AC_CACHE_CHECK(for __builtin_ctzl, pgac_cv__builtin_ctzl,
449+
[AC_COMPILE_IFELSE([AC_LANG_SOURCE(
450+
[static int x = __builtin_ctzl(256);]
451+
)],
452+
[pgac_cv__builtin_ctzl=yes],
453+
[pgac_cv__builtin_ctzl=no])])
454+
if test x"$pgac_cv__builtin_ctzl" = xyes ; then
455+
AC_DEFINE(HAVE__BUILTIN_CTZL, 1,
456+
[Define to 1 if your compiler understands __builtin_ctzl.])
457+
fi])# PGAC_C_BUILTIN_CTZL
458+
459+
460+
461+
# PGAC_C_BUILTIN_CLZ
462+
# -------------------------
463+
# Check if the C compiler understands __builtin_clz(),
464+
# and define HAVE__BUILTIN_CLZ if so.
465+
AC_DEFUN([PGAC_C_BUILTIN_CLZ],
466+
[AC_CACHE_CHECK(for __builtin_clz, pgac_cv__builtin_clz,
467+
[AC_COMPILE_IFELSE([AC_LANG_SOURCE(
468+
[static int x = __builtin_clz(256);]
469+
)],
470+
[pgac_cv__builtin_clz=yes],
471+
[pgac_cv__builtin_clz=no])])
472+
if test x"$pgac_cv__builtin_clz" = xyes ; then
473+
AC_DEFINE(HAVE__BUILTIN_CLZ, 1,
474+
[Define to 1 if your compiler understands __builtin_clz.])
475+
fi])# PGAC_C_BUILTIN_CLZ
476+
477+
478+
479+
# PGAC_C_BUILTIN_CLZL
480+
# -------------------------
481+
# Check if the C compiler understands __builtin_clzl(),
482+
# and define HAVE__BUILTIN_CLZL if so.
483+
AC_DEFUN([PGAC_C_BUILTIN_CLZL],
484+
[AC_CACHE_CHECK(for __builtin_clzl, pgac_cv__builtin_clzl,
485+
[AC_COMPILE_IFELSE([AC_LANG_SOURCE(
486+
[static int x = __builtin_clzl(256);]
487+
)],
488+
[pgac_cv__builtin_clzl=yes],
489+
[pgac_cv__builtin_clzl=no])])
490+
if test x"$pgac_cv__builtin_clzl" = xyes ; then
491+
AC_DEFINE(HAVE__BUILTIN_CLZL, 1,
492+
[Define to 1 if your compiler understands __builtin_clzl.])
493+
fi])# PGAC_C_BUILTIN_CLZL
494+
495+
496+
381497
# PGAC_C_BUILTIN_UNREACHABLE
382498
# --------------------------
383499
# Check if the C compiler understands __builtin_unreachable(),

‎configure‎

Lines changed: 155 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -651,6 +651,7 @@ CFLAGS_ARMV8_CRC32C
651651
CFLAGS_SSE42
652652
have_win32_dbghelp
653653
LIBOBJS
654+
CFLAGS_POPCNT
654655
UUID_LIBS
655656
LDAP_LIBS_BE
656657
LDAP_LIBS_FE
@@ -14059,6 +14060,158 @@ if test x"$pgac_cv__builtin_constant_p" = xyes ; then
1405914060

1406014061
$as_echo "#define HAVE__BUILTIN_CONSTANT_P 1" >>confdefs.h
1406114062

14063+
fi
14064+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_popcount" >&5
14065+
$as_echo_n "checking for __builtin_popcount... " >&6; }
14066+
if ${pgac_cv_popcount+:} false; then :
14067+
$as_echo_n "(cached) " >&6
14068+
else
14069+
pgac_save_CFLAGS=$CFLAGS
14070+
CFLAGS="$pgac_save_CFLAGS -mpopcnt"
14071+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14072+
/* end confdefs.h. */
14073+
static int x = __builtin_popcount(255);
14074+
_ACEOF
14075+
if ac_fn_c_try_compile "$LINENO"; then :
14076+
pgac_cv_popcount=yes
14077+
else
14078+
pgac_cv_popcount=no
14079+
fi
14080+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14081+
CFLAGS="$pgac_save_CFLAGS"
14082+
fi
14083+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_popcount" >&5
14084+
$as_echo "$pgac_cv_popcount" >&6; }
14085+
if test x"$pgac_cv_popcount" = x"yes"; then
14086+
CFLAGS_POPCNT="-mpopcnt"
14087+
14088+
$as_echo "#define HAVE__BUILTIN_POPCOUNT 1" >>confdefs.h
14089+
14090+
fi
14091+
14092+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_popcountl" >&5
14093+
$as_echo_n "checking for __builtin_popcountl... " >&6; }
14094+
if ${pgac_cv_popcountl+:} false; then :
14095+
$as_echo_n "(cached) " >&6
14096+
else
14097+
pgac_save_CFLAGS=$CFLAGS
14098+
CFLAGS="$pgac_save_CFLAGS -mpopcnt"
14099+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14100+
/* end confdefs.h. */
14101+
static int x = __builtin_popcountl(255);
14102+
_ACEOF
14103+
if ac_fn_c_try_compile "$LINENO"; then :
14104+
pgac_cv_popcountl=yes
14105+
else
14106+
pgac_cv_popcountl=no
14107+
fi
14108+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14109+
CFLAGS="$pgac_save_CFLAGS"
14110+
fi
14111+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_popcountl" >&5
14112+
$as_echo "$pgac_cv_popcountl" >&6; }
14113+
if test x"$pgac_cv_popcountl" = x"yes"; then
14114+
CFLAGS_POPCNT="-mpopcnt"
14115+
14116+
$as_echo "#define HAVE__BUILTIN_POPCOUNTL 1" >>confdefs.h
14117+
14118+
fi
14119+
14120+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_ctz" >&5
14121+
$as_echo_n "checking for __builtin_ctz... " >&6; }
14122+
if ${pgac_cv__builtin_ctz+:} false; then :
14123+
$as_echo_n "(cached) " >&6
14124+
else
14125+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14126+
/* end confdefs.h. */
14127+
static int x = __builtin_ctz(256);
14128+
14129+
_ACEOF
14130+
if ac_fn_c_try_compile "$LINENO"; then :
14131+
pgac_cv__builtin_ctz=yes
14132+
else
14133+
pgac_cv__builtin_ctz=no
14134+
fi
14135+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14136+
fi
14137+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv__builtin_ctz" >&5
14138+
$as_echo "$pgac_cv__builtin_ctz" >&6; }
14139+
if test x"$pgac_cv__builtin_ctz" = xyes ; then
14140+
14141+
$as_echo "#define HAVE__BUILTIN_CTZ 1" >>confdefs.h
14142+
14143+
fi
14144+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_ctzl" >&5
14145+
$as_echo_n "checking for __builtin_ctzl... " >&6; }
14146+
if ${pgac_cv__builtin_ctzl+:} false; then :
14147+
$as_echo_n "(cached) " >&6
14148+
else
14149+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14150+
/* end confdefs.h. */
14151+
static int x = __builtin_ctzl(256);
14152+
14153+
_ACEOF
14154+
if ac_fn_c_try_compile "$LINENO"; then :
14155+
pgac_cv__builtin_ctzl=yes
14156+
else
14157+
pgac_cv__builtin_ctzl=no
14158+
fi
14159+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14160+
fi
14161+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv__builtin_ctzl" >&5
14162+
$as_echo "$pgac_cv__builtin_ctzl" >&6; }
14163+
if test x"$pgac_cv__builtin_ctzl" = xyes ; then
14164+
14165+
$as_echo "#define HAVE__BUILTIN_CTZL 1" >>confdefs.h
14166+
14167+
fi
14168+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_clz" >&5
14169+
$as_echo_n "checking for __builtin_clz... " >&6; }
14170+
if ${pgac_cv__builtin_clz+:} false; then :
14171+
$as_echo_n "(cached) " >&6
14172+
else
14173+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14174+
/* end confdefs.h. */
14175+
static int x = __builtin_clz(256);
14176+
14177+
_ACEOF
14178+
if ac_fn_c_try_compile "$LINENO"; then :
14179+
pgac_cv__builtin_clz=yes
14180+
else
14181+
pgac_cv__builtin_clz=no
14182+
fi
14183+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14184+
fi
14185+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv__builtin_clz" >&5
14186+
$as_echo "$pgac_cv__builtin_clz" >&6; }
14187+
if test x"$pgac_cv__builtin_clz" = xyes ; then
14188+
14189+
$as_echo "#define HAVE__BUILTIN_CLZ 1" >>confdefs.h
14190+
14191+
fi
14192+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_clzl" >&5
14193+
$as_echo_n "checking for __builtin_clzl... " >&6; }
14194+
if ${pgac_cv__builtin_clzl+:} false; then :
14195+
$as_echo_n "(cached) " >&6
14196+
else
14197+
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
14198+
/* end confdefs.h. */
14199+
static int x = __builtin_clzl(256);
14200+
14201+
_ACEOF
14202+
if ac_fn_c_try_compile "$LINENO"; then :
14203+
pgac_cv__builtin_clzl=yes
14204+
else
14205+
pgac_cv__builtin_clzl=no
14206+
fi
14207+
rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
14208+
fi
14209+
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv__builtin_clzl" >&5
14210+
$as_echo "$pgac_cv__builtin_clzl" >&6; }
14211+
if test x"$pgac_cv__builtin_clzl" = xyes ; then
14212+
14213+
$as_echo "#define HAVE__BUILTIN_CLZL 1" >>confdefs.h
14214+
1406214215
fi
1406314216
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_unreachable" >&5
1406414217
$as_echo_n "checking for __builtin_unreachable... " >&6; }
@@ -14577,6 +14730,8 @@ $as_echo "#define LOCALE_T_IN_XLOCALE 1" >>confdefs.h
1457714730

1457814731
fi
1457914732

14733+
14734+
1458014735
# MSVC doesn't cope well with defining restrict to __restrict, the
1458114736
# spelling it understands, because it conflicts with
1458214737
# __declspec(restrict). Therefore we define pg_restrict to the

‎configure.in‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1489,6 +1489,12 @@ PGAC_C_BUILTIN_BSWAP16
14891489
PGAC_C_BUILTIN_BSWAP32
14901490
PGAC_C_BUILTIN_BSWAP64
14911491
PGAC_C_BUILTIN_CONSTANT_P
1492+
PGAC_C_BUILTIN_POPCOUNT
1493+
PGAC_C_BUILTIN_POPCOUNTL
1494+
PGAC_C_BUILTIN_CTZ
1495+
PGAC_C_BUILTIN_CTZL
1496+
PGAC_C_BUILTIN_CLZ
1497+
PGAC_C_BUILTIN_CLZL
14921498
PGAC_C_BUILTIN_UNREACHABLE
14931499
PGAC_C_COMPUTED_GOTO
14941500
PGAC_STRUCT_TIMEZONE
@@ -1503,6 +1509,8 @@ AC_TYPE_LONG_LONG_INT
15031509

15041510
PGAC_TYPE_LOCALE_T
15051511

1512+
AC_SUBST(CFLAGS_POPCNT)
1513+
15061514
# MSVC doesn't cope well with defining restrict to __restrict, the
15071515
# spelling it understands, because it conflicts with
15081516
# __declspec(restrict). Therefore we define pg_restrict to the

‎src/Makefile.global.in‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -260,6 +260,7 @@ CXX = @CXX@
260260
CFLAGS = @CFLAGS@
261261
CFLAGS_VECTOR = @CFLAGS_VECTOR@
262262
CFLAGS_SSE42 = @CFLAGS_SSE42@
263+
CFLAGS_POPCNT = @CFLAGS_POPCNT@
263264
CFLAGS_ARMV8_CRC32C = @CFLAGS_ARMV8_CRC32C@
264265
PERMIT_DECLARATION_AFTER_STATEMENT = @PERMIT_DECLARATION_AFTER_STATEMENT@
265266
CXXFLAGS = @CXXFLAGS@

0 commit comments

Comments
 (0)