mirror/glibc
2
0
Fork 0
mirror of git://sourceware.org/git/glibc.git synced 2024-11-27 03:41:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

Remove configure tests for AVX support.

Browse Source 
GCC added support for -mavx and -msse2avx in version 4.4.  Thus the
configure tests for this support are obsolete, and this patch removes
them.

Tested for x86_64 and x86 (testsuite, and that installed stripped
shared libraries are unchanged by this patch).

	* sysdeps/i386/configure.ac (libc_cv_cc_avx): Remove configure
	test.
	(libc_cv_cc_sse2avx): Likewise.
	* sysdeps/i386/configure: Regenerated.
	* sysdeps/i386/i686/multiarch/Makefile
	[$(subdir)$(config-cflags-avx) = mathyes]: Change conditional to
	[$(subdir) = math].
	* sysdeps/i386/i686/multiarch/s_fma-fma.c [HAVE_AVX_SUPPORT]: Make
	code unconditional.
	* sysdeps/i386/i686/multiarch/s_fma.c [HAVE_AVX_SUPPORT]:
	Likewise.
	* sysdeps/i386/i686/multiarch/s_fmaf-fma.c [HAVE_AVX_SUPPORT]:
	Likewise.
	* sysdeps/i386/i686/multiarch/s_fmaf.c [HAVE_AVX_SUPPORT]:
	Likewise.
	* sysdeps/x86_64/configure.ac (libc_cv_cc_avx): Remove configure
	test.
	(libc_cv_cc_sse2avx): Likewise.
	* sysdeps/x86_64/configure: Regenerated.
	* sysdeps/x86_64/Makefile [$(config-cflags-avx) = yes]: Make code
	unconditional.
	* sysdeps/x86_64/dl-trampoline.h (_dl_runtime_profile)
	[HAVE_AVX_SUPPORT || HAVE_AVX512_ASM_SUPPORT]: Make code
	unconditional.
	(_dl_runtime_profile)
	[!(HAVE_AVX_SUPPORT || HAVE_AVX512_ASM_SUPPORT)]: Remove
	conditional code.
	* sysdeps/x86_64/fpu/multiarch/Makefile
	[$(config-cflags-sse2avx) = yes]: Make code unconditional.
	* sysdeps/x86_64/fpu/multiarch/e_atan2.c
	[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
	* sysdeps/x86_64/fpu/multiarch/e_exp.c
	[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
	* sysdeps/x86_64/fpu/multiarch/e_log.c
	[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
	* sysdeps/x86_64/fpu/multiarch/s_atan.c
	[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
	* sysdeps/x86_64/fpu/multiarch/s_fma.c [HAVE_AVX_SUPPORT]:
	Likewise.
	* sysdeps/x86_64/fpu/multiarch/s_fmaf.c [HAVE_AVX_SUPPORT]:
	Likewise.
	* sysdeps/x86_64/fpu/multiarch/s_sin.c
	[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
	* sysdeps/x86_64/fpu/multiarch/s_tan.c
	[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
	* sysdeps/x86_64/multiarch/strcmp.S [HAVE_AVX_SUPPORT]: Likewise.
	* config.h.in (HAVE_AVX_SUPPORT): Remove #undef.
	(HAVE_SSE2AVX_SUPPORT): Likewise.
This commit is contained in:
Joseph Myers 2015-10-08 15:59:32 +00:00
parent 104ff17183
commit 1b12cd7f4d
23 changed files with 137 additions and 279 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
ChangeLog
View File

@ -1,3 +1,54 @@
2015-10-08 Joseph Myers <joseph@codesourcery.com>
* sysdeps/i386/configure.ac (libc_cv_cc_avx): Remove configure
test.
(libc_cv_cc_sse2avx): Likewise.
* sysdeps/i386/configure: Regenerated.
* sysdeps/i386/i686/multiarch/Makefile
[$(subdir)$(config-cflags-avx) = mathyes]: Change conditional to
[$(subdir) = math].
* sysdeps/i386/i686/multiarch/s_fma-fma.c [HAVE_AVX_SUPPORT]: Make
code unconditional.
* sysdeps/i386/i686/multiarch/s_fma.c [HAVE_AVX_SUPPORT]:
Likewise.
* sysdeps/i386/i686/multiarch/s_fmaf-fma.c [HAVE_AVX_SUPPORT]:
Likewise.
* sysdeps/i386/i686/multiarch/s_fmaf.c [HAVE_AVX_SUPPORT]:
Likewise.
* sysdeps/x86_64/configure.ac (libc_cv_cc_avx): Remove configure
test.
(libc_cv_cc_sse2avx): Likewise.
* sysdeps/x86_64/configure: Regenerated.
* sysdeps/x86_64/Makefile [$(config-cflags-avx) = yes]: Make code
unconditional.
* sysdeps/x86_64/dl-trampoline.h (_dl_runtime_profile)
[HAVE_AVX_SUPPORT || HAVE_AVX512_ASM_SUPPORT]: Make code
unconditional.
(_dl_runtime_profile)
[!(HAVE_AVX_SUPPORT || HAVE_AVX512_ASM_SUPPORT)]: Remove
conditional code.
* sysdeps/x86_64/fpu/multiarch/Makefile
[$(config-cflags-sse2avx) = yes]: Make code unconditional.
* sysdeps/x86_64/fpu/multiarch/e_atan2.c
[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_exp.c
[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
* sysdeps/x86_64/fpu/multiarch/e_log.c
[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_atan.c
[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_fma.c [HAVE_AVX_SUPPORT]:
Likewise.
* sysdeps/x86_64/fpu/multiarch/s_fmaf.c [HAVE_AVX_SUPPORT]:
Likewise.
* sysdeps/x86_64/fpu/multiarch/s_sin.c
[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
* sysdeps/x86_64/fpu/multiarch/s_tan.c
[HAVE_FMA4_SUPPORT || HAVE_AVX_SUPPORT]: Likewise.
* sysdeps/x86_64/multiarch/strcmp.S [HAVE_AVX_SUPPORT]: Likewise.
* config.h.in (HAVE_AVX_SUPPORT): Remove #undef.
(HAVE_SSE2AVX_SUPPORT): Likewise.
2015-10-07 Paul Pluzhnikov <ppluzhnikov@google.com> 2015-10-07 Paul Pluzhnikov <ppluzhnikov@google.com>
* stdio-common/tst-printf-bz18872.sh: Use attribute optimize * stdio-common/tst-printf-bz18872.sh: Use attribute optimize

6
config.h.in
View File

@ -76,12 +76,6 @@
/* Define if _rtld_local structure should be forced into .sdata section. */ /* Define if _rtld_local structure should be forced into .sdata section. */
#undef HAVE_SDATA_SECTION #undef HAVE_SDATA_SECTION
/* Define if gcc supports AVX. */
#undef HAVE_AVX_SUPPORT
/* Define if gcc supports VEX encoding. */
#undef HAVE_SSE2AVX_SUPPORT
/* Define if compiler supports AVX512. */ /* Define if compiler supports AVX512. */
#undef HAVE_AVX512_SUPPORT #undef HAVE_AVX512_SUPPORT

52
sysdeps/i386/configure vendored
View File

@ -114,58 +114,6 @@ $as_echo "$libc_cv_as_i686" >&6; }
config_vars="$config_vars config_vars="$config_vars
config-asflags-i686 = $libc_cv_as_i686" config-asflags-i686 = $libc_cv_as_i686"
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for AVX support" >&5
$as_echo_n "checking for AVX support... " >&6; }
if ${libc_cv_cc_avx+:} false; then :
$as_echo_n "(cached) " >&6
else
if { ac_try='${CC-cc} -mavx -xc /dev/null -S -o /dev/null'
{ { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5
(eval $ac_try) 2>&5
ac_status=$?
$as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
test $ac_status = 0; }; }; then :
libc_cv_cc_avx=yes
else
libc_cv_cc_avx=no
fi
fi
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_cc_avx" >&5
$as_echo "$libc_cv_cc_avx" >&6; }
if test $libc_cv_cc_avx = yes; then
$as_echo "#define HAVE_AVX_SUPPORT 1" >>confdefs.h
fi
config_vars="$config_vars
config-cflags-avx = $libc_cv_cc_avx"
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for AVX encoding of SSE instructions" >&5
$as_echo_n "checking for AVX encoding of SSE instructions... " >&6; }
if ${libc_cv_cc_sse2avx+:} false; then :
$as_echo_n "(cached) " >&6
else
if { ac_try='${CC-cc} -msse2avx -xc /dev/null -S -o /dev/null'
{ { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5
(eval $ac_try) 2>&5
ac_status=$?
$as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
test $ac_status = 0; }; }; then :
libc_cv_cc_sse2avx=yes
else
libc_cv_cc_sse2avx=no
fi
fi
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_cc_sse2avx" >&5
$as_echo "$libc_cv_cc_sse2avx" >&6; }
if test $libc_cv_cc_sse2avx = yes; then
$as_echo "#define HAVE_SSE2AVX_SUPPORT 1" >>confdefs.h
fi
config_vars="$config_vars
config-cflags-sse2avx = $libc_cv_cc_sse2avx"
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for FMA4 support" >&5 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for FMA4 support" >&5
$as_echo_n "checking for FMA4 support... " >&6; } $as_echo_n "checking for FMA4 support... " >&6; }
if ${libc_cv_cc_fma4+:} false; then : if ${libc_cv_cc_fma4+:} false; then :

20
sysdeps/i386/configure.ac
View File

@ -42,26 +42,6 @@ LIBC_TRY_CC_OPTION([-Wa,-mtune=i686],
]) ])
LIBC_CONFIG_VAR([config-asflags-i686], [$libc_cv_as_i686]) LIBC_CONFIG_VAR([config-asflags-i686], [$libc_cv_as_i686])
dnl Check if -mavx works.
AC_CACHE_CHECK(for AVX support, libc_cv_cc_avx, [dnl
LIBC_TRY_CC_OPTION([-mavx], [libc_cv_cc_avx=yes], [libc_cv_cc_avx=no])
])
if test $libc_cv_cc_avx = yes; then
AC_DEFINE(HAVE_AVX_SUPPORT)
fi
LIBC_CONFIG_VAR([config-cflags-avx], [$libc_cv_cc_avx])
dnl Check if -msse2avx works.
AC_CACHE_CHECK(for AVX encoding of SSE instructions, libc_cv_cc_sse2avx, [dnl
LIBC_TRY_CC_OPTION([-msse2avx],
[libc_cv_cc_sse2avx=yes],
[libc_cv_cc_sse2avx=no])
])
if test $libc_cv_cc_sse2avx = yes; then
AC_DEFINE(HAVE_SSE2AVX_SUPPORT)
fi
LIBC_CONFIG_VAR([config-cflags-sse2avx], [$libc_cv_cc_sse2avx])
dnl Check if -mfma4 works. dnl Check if -mfma4 works.
AC_CACHE_CHECK(for FMA4 support, libc_cv_cc_fma4, [dnl AC_CACHE_CHECK(for FMA4 support, libc_cv_cc_fma4, [dnl
LIBC_TRY_CC_OPTION([-mfma4], [libc_cv_cc_fma4=yes], [libc_cv_cc_fma4=no]) LIBC_TRY_CC_OPTION([-mfma4], [libc_cv_cc_fma4=yes], [libc_cv_cc_fma4=no])

2
sysdeps/i386/i686/multiarch/Makefile
View File

@ -38,7 +38,7 @@ sysdep_routines += wcscmp-sse2 wcscmp-c wcslen-sse2 wcslen-c \
wcschr-c wcsrchr-sse2 wcsrchr-c wcscpy-ssse3 wcscpy-c wcschr-c wcsrchr-sse2 wcsrchr-c wcscpy-ssse3 wcscpy-c
endif endif
ifeq (mathyes,$(subdir)$(config-cflags-avx)) ifeq ($(subdir),math)
libm-sysdep_routines += s_fma-fma s_fmaf-fma libm-sysdep_routines += s_fma-fma s_fmaf-fma
CFLAGS-s_fma-fma.c += -mavx -mfpmath=sse CFLAGS-s_fma-fma.c += -mavx -mfpmath=sse
CFLAGS-s_fmaf-fma.c += -mavx -mfpmath=sse CFLAGS-s_fmaf-fma.c += -mavx -mfpmath=sse

2
sysdeps/i386/i686/multiarch/s_fma-fma.c
View File

@ -19,11 +19,9 @@
#include <config.h> #include <config.h>
#ifdef HAVE_AVX_SUPPORT
double double
__fma_fma (double x, double y, double z) __fma_fma (double x, double y, double z)
{ {
asm ("vfmadd213sd %3, %2, %0" : "=x" (x) : "0" (x), "x" (y), "xm" (z)); asm ("vfmadd213sd %3, %2, %0" : "=x" (x) : "0" (x), "x" (y), "xm" (z));
return x; return x;
} }
#endif

4
sysdeps/i386/i686/multiarch/s_fma.c
View File

@ -19,7 +19,6 @@
#include <config.h> #include <config.h>
#ifdef HAVE_AVX_SUPPORT
#include <math.h> #include <math.h>
#include <init-arch.h> #include <init-arch.h>
@ -30,7 +29,6 @@ libm_ifunc (__fma,
HAS_ARCH_FEATURE (FMA_Usable) ? __fma_fma : __fma_ia32); HAS_ARCH_FEATURE (FMA_Usable) ? __fma_fma : __fma_ia32);
weak_alias (__fma, fma) weak_alias (__fma, fma)
# define __fma __fma_ia32 #define __fma __fma_ia32
#endif
#include <sysdeps/ieee754/ldbl-96/s_fma.c> #include <sysdeps/ieee754/ldbl-96/s_fma.c>

2
sysdeps/i386/i686/multiarch/s_fmaf-fma.c
View File

@ -19,11 +19,9 @@
#include <config.h> #include <config.h>
#ifdef HAVE_AVX_SUPPORT
float float
__fmaf_fma (float x, float y, float z) __fmaf_fma (float x, float y, float z)
{ {
asm ("vfmadd213ss %3, %2, %0" : "=x" (x) : "0" (x), "x" (y), "xm" (z)); asm ("vfmadd213ss %3, %2, %0" : "=x" (x) : "0" (x), "x" (y), "xm" (z));
return x; return x;
} }
#endif

4
sysdeps/i386/i686/multiarch/s_fmaf.c
View File

@ -19,7 +19,6 @@
#include <config.h> #include <config.h>
#ifdef HAVE_AVX_SUPPORT
#include <math.h> #include <math.h>
#include <init-arch.h> #include <init-arch.h>
@ -30,7 +29,6 @@ libm_ifunc (__fmaf,
HAS_ARCH_FEATURE (FMA_Usable) ? __fmaf_fma : __fmaf_ia32); HAS_ARCH_FEATURE (FMA_Usable) ? __fmaf_fma : __fmaf_ia32);
weak_alias (__fmaf, fmaf) weak_alias (__fmaf, fmaf)
# define __fmaf __fmaf_ia32 #define __fmaf __fmaf_ia32
#endif
#include <sysdeps/ieee754/dbl-64/s_fmaf.c> #include <sysdeps/ieee754/dbl-64/s_fmaf.c>

7
sysdeps/x86_64/Makefile
View File

@ -43,10 +43,7 @@ tests-pie += $(quad-pie-test)
$(objpfx)tst-quad1pie: $(objpfx)tst-quadmod1pie.o $(objpfx)tst-quad1pie: $(objpfx)tst-quadmod1pie.o
$(objpfx)tst-quad2pie: $(objpfx)tst-quadmod2pie.o $(objpfx)tst-quad2pie: $(objpfx)tst-quadmod2pie.o
tests += tst-audit3 tst-audit4 tst-audit5 tst-audit10 tests += tst-audit3 tst-audit4 tst-audit5 tst-audit6 tst-audit7 tst-audit10
ifeq (yes,$(config-cflags-avx))
tests += tst-audit6 tst-audit7
endif
tests += tst-split-dynreloc tests += tst-split-dynreloc
LDFLAGS-tst-split-dynreloc = -Wl,-T,$(..)sysdeps/x86_64/tst-split-dynreloc.lds LDFLAGS-tst-split-dynreloc = -Wl,-T,$(..)sysdeps/x86_64/tst-split-dynreloc.lds
@ -84,7 +81,6 @@ $(objpfx)tst-audit10: $(objpfx)tst-auditmod10a.so
$(objpfx)tst-audit10.out: $(objpfx)tst-auditmod10b.so $(objpfx)tst-audit10.out: $(objpfx)tst-auditmod10b.so
tst-audit10-ENV = LD_AUDIT=$(objpfx)tst-auditmod10b.so tst-audit10-ENV = LD_AUDIT=$(objpfx)tst-auditmod10b.so
ifeq (yes,$(config-cflags-avx))
AVX-CFLAGS=-mavx AVX-CFLAGS=-mavx
ifeq (yes,$(config-cflags-novzeroupper)) ifeq (yes,$(config-cflags-novzeroupper))
AVX-CFLAGS+=-mno-vzeroupper AVX-CFLAGS+=-mno-vzeroupper
@ -95,7 +91,6 @@ CFLAGS-tst-auditmod4b.c += $(AVX-CFLAGS)
CFLAGS-tst-auditmod6b.c += $(AVX-CFLAGS) CFLAGS-tst-auditmod6b.c += $(AVX-CFLAGS)
CFLAGS-tst-auditmod6c.c += $(AVX-CFLAGS) CFLAGS-tst-auditmod6c.c += $(AVX-CFLAGS)
CFLAGS-tst-auditmod7b.c += $(AVX-CFLAGS) CFLAGS-tst-auditmod7b.c += $(AVX-CFLAGS)
endif
ifeq (yes,$(config-cflags-avx512)) ifeq (yes,$(config-cflags-avx512))
AVX512-CFLAGS = -mavx512f AVX512-CFLAGS = -mavx512f
CFLAGS-tst-audit10.c += $(AVX512-CFLAGS) CFLAGS-tst-audit10.c += $(AVX512-CFLAGS)

52
sysdeps/x86_64/configure vendored
View File

@ -43,32 +43,6 @@ fi
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for AVX support" >&5
$as_echo_n "checking for AVX support... " >&6; }
if ${libc_cv_cc_avx+:} false; then :
$as_echo_n "(cached) " >&6
else
if { ac_try='${CC-cc} -mavx -xc /dev/null -S -o /dev/null'
{ { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5
(eval $ac_try) 2>&5
ac_status=$?
$as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
test $ac_status = 0; }; }; then :
libc_cv_cc_avx=yes
else
libc_cv_cc_avx=no
fi
fi
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_cc_avx" >&5
$as_echo "$libc_cv_cc_avx" >&6; }
if test $libc_cv_cc_avx = yes; then
$as_echo "#define HAVE_AVX_SUPPORT 1" >>confdefs.h
fi
config_vars="$config_vars
config-cflags-avx = $libc_cv_cc_avx"
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for AVX512 support in assembler" >&5 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for AVX512 support in assembler" >&5
$as_echo_n "checking for AVX512 support in assembler... " >&6; } $as_echo_n "checking for AVX512 support in assembler... " >&6; }
if ${libc_cv_asm_avx512+:} false; then : if ${libc_cv_asm_avx512+:} false; then :
@ -123,32 +97,6 @@ fi
config_vars="$config_vars config_vars="$config_vars
config-cflags-avx512 = $libc_cv_cc_avx512" config-cflags-avx512 = $libc_cv_cc_avx512"
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for AVX encoding of SSE instructions" >&5
$as_echo_n "checking for AVX encoding of SSE instructions... " >&6; }
if ${libc_cv_cc_sse2avx+:} false; then :
$as_echo_n "(cached) " >&6
else
if { ac_try='${CC-cc} -msse2avx -xc /dev/null -S -o /dev/null'
{ { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5
(eval $ac_try) 2>&5
ac_status=$?
$as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
test $ac_status = 0; }; }; then :
libc_cv_cc_sse2avx=yes
else
libc_cv_cc_sse2avx=no
fi
fi
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_cc_sse2avx" >&5
$as_echo "$libc_cv_cc_sse2avx" >&6; }
if test $libc_cv_cc_sse2avx = yes; then
$as_echo "#define HAVE_SSE2AVX_SUPPORT 1" >>confdefs.h
fi
config_vars="$config_vars
config-cflags-sse2avx = $libc_cv_cc_sse2avx"
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for FMA4 support" >&5 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for FMA4 support" >&5
$as_echo_n "checking for FMA4 support... " >&6; } $as_echo_n "checking for FMA4 support... " >&6; }
if ${libc_cv_cc_fma4+:} false; then : if ${libc_cv_cc_fma4+:} false; then :

20
sysdeps/x86_64/configure.ac
View File

@ -5,15 +5,6 @@ AC_CHECK_HEADER([cpuid.h], ,
[AC_MSG_ERROR([gcc must provide the <cpuid.h> header])], [AC_MSG_ERROR([gcc must provide the <cpuid.h> header])],
[/* No default includes. */]) [/* No default includes. */])
dnl Check if -mavx works.
AC_CACHE_CHECK(for AVX support, libc_cv_cc_avx, [dnl
LIBC_TRY_CC_OPTION([-mavx], [libc_cv_cc_avx=yes], [libc_cv_cc_avx=no])
])
if test $libc_cv_cc_avx = yes; then
AC_DEFINE(HAVE_AVX_SUPPORT)
fi
LIBC_CONFIG_VAR([config-cflags-avx], [$libc_cv_cc_avx])
dnl Check if asm supports AVX512. dnl Check if asm supports AVX512.
AC_CACHE_CHECK(for AVX512 support in assembler, libc_cv_asm_avx512, [dnl AC_CACHE_CHECK(for AVX512 support in assembler, libc_cv_asm_avx512, [dnl
cat > conftest.s <<\EOF cat > conftest.s <<\EOF
@ -39,17 +30,6 @@ if test $libc_cv_cc_avx512 = yes; then
fi fi
LIBC_CONFIG_VAR([config-cflags-avx512], [$libc_cv_cc_avx512]) LIBC_CONFIG_VAR([config-cflags-avx512], [$libc_cv_cc_avx512])
dnl Check if -msse2avx works.
AC_CACHE_CHECK(for AVX encoding of SSE instructions, libc_cv_cc_sse2avx, [dnl
LIBC_TRY_CC_OPTION([-msse2avx],
[libc_cv_cc_sse2avx=yes],
[libc_cv_cc_sse2avx=no])
])
if test $libc_cv_cc_sse2avx = yes; then
AC_DEFINE(HAVE_SSE2AVX_SUPPORT)
fi
LIBC_CONFIG_VAR([config-cflags-sse2avx], [$libc_cv_cc_sse2avx])
dnl Check if -mfma4 works. dnl Check if -mfma4 works.
AC_CACHE_CHECK(for FMA4 support, libc_cv_cc_fma4, [dnl AC_CACHE_CHECK(for FMA4 support, libc_cv_cc_fma4, [dnl
LIBC_TRY_CC_OPTION([-mfma4], [libc_cv_cc_fma4=yes], [libc_cv_cc_fma4=no]) LIBC_TRY_CC_OPTION([-mfma4], [libc_cv_cc_fma4=yes], [libc_cv_cc_fma4=no])

4
sysdeps/x86_64/dl-trampoline.h
View File

@ -204,14 +204,10 @@ _dl_runtime_profile:
/* Actively align the La_x86_64_regs structure. */ /* Actively align the La_x86_64_regs structure. */
and $-VEC_SIZE, %RSP_LP and $-VEC_SIZE, %RSP_LP
# if defined HAVE_AVX_SUPPORT || defined HAVE_AVX512_ASM_SUPPORT
/* sizeof(La_x86_64_regs). Need extra space for 8 SSE registers /* sizeof(La_x86_64_regs). Need extra space for 8 SSE registers
to detect if any xmm0-xmm7 registers are changed by audit to detect if any xmm0-xmm7 registers are changed by audit
module. */ module. */
sub $(LR_SIZE + XMM_SIZE*8), %RSP_LP sub $(LR_SIZE + XMM_SIZE*8), %RSP_LP
# else
sub $LR_SIZE, %RSP_LP # sizeof(La_x86_64_regs)
# endif
movq %rsp, 24(%rbx) movq %rsp, 24(%rbx)
/* Fill the La_x86_64_regs structure. */ /* Fill the La_x86_64_regs structure. */

2
sysdeps/x86_64/fpu/multiarch/Makefile
View File

@ -33,7 +33,6 @@ CFLAGS-s_sin-fma4.c = -mfma4
CFLAGS-s_tan-fma4.c = -mfma4 CFLAGS-s_tan-fma4.c = -mfma4
endif endif
ifeq ($(config-cflags-sse2avx),yes)
libm-sysdep_routines += e_exp-avx e_log-avx s_atan-avx \ libm-sysdep_routines += e_exp-avx e_log-avx s_atan-avx \
e_atan2-avx s_sin-avx s_tan-avx \ e_atan2-avx s_sin-avx s_tan-avx \
mplog-avx mpa-avx slowexp-avx \ mplog-avx mpa-avx slowexp-avx \
@ -50,7 +49,6 @@ CFLAGS-s_sin-avx.c = -msse2avx -DSSE2AVX
CFLAGS-slowexp-avx.c = -msse2avx -DSSE2AVX CFLAGS-slowexp-avx.c = -msse2avx -DSSE2AVX
CFLAGS-s_tan-avx.c = -msse2avx -DSSE2AVX CFLAGS-s_tan-avx.c = -msse2avx -DSSE2AVX
endif endif
endif
ifeq ($(subdir),mathvec) ifeq ($(subdir),mathvec)
libmvec-sysdep_routines += svml_d_cos2_core_sse4 svml_d_cos4_core_avx2 \ libmvec-sysdep_routines += svml_d_cos2_core_sse4 svml_d_cos4_core_avx2 \

22
sysdeps/x86_64/fpu/multiarch/e_atan2.c
View File

@ -1,17 +1,16 @@
#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT #include <init-arch.h>
# include <init-arch.h> #include <math.h>
# include <math.h> #include <math_private.h>
# include <math_private.h>
extern double __ieee754_atan2_sse2 (double, double); extern double __ieee754_atan2_sse2 (double, double);
extern double __ieee754_atan2_avx (double, double); extern double __ieee754_atan2_avx (double, double);
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_atan2_fma4 (double, double); extern double __ieee754_atan2_fma4 (double, double);
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __ieee754_atan2_fma4 ((void *) 0) # define __ieee754_atan2_fma4 ((void *) 0)
# endif #endif
libm_ifunc (__ieee754_atan2, libm_ifunc (__ieee754_atan2,
HAS_ARCH_FEATURE (FMA4_Usable) ? __ieee754_atan2_fma4 HAS_ARCH_FEATURE (FMA4_Usable) ? __ieee754_atan2_fma4
@ -19,8 +18,7 @@ libm_ifunc (__ieee754_atan2,
? __ieee754_atan2_avx : __ieee754_atan2_sse2)); ? __ieee754_atan2_avx : __ieee754_atan2_sse2));
strong_alias (__ieee754_atan2, __atan2_finite) strong_alias (__ieee754_atan2, __atan2_finite)
# define __ieee754_atan2 __ieee754_atan2_sse2 #define __ieee754_atan2 __ieee754_atan2_sse2
#endif
#include <sysdeps/ieee754/dbl-64/e_atan2.c> #include <sysdeps/ieee754/dbl-64/e_atan2.c>

22
sysdeps/x86_64/fpu/multiarch/e_exp.c
View File

@ -1,17 +1,16 @@
#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT #include <init-arch.h>
# include <init-arch.h> #include <math.h>
# include <math.h> #include <math_private.h>
# include <math_private.h>
extern double __ieee754_exp_sse2 (double); extern double __ieee754_exp_sse2 (double);
extern double __ieee754_exp_avx (double); extern double __ieee754_exp_avx (double);
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_exp_fma4 (double); extern double __ieee754_exp_fma4 (double);
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __ieee754_exp_fma4 ((void *) 0) # define __ieee754_exp_fma4 ((void *) 0)
# endif #endif
libm_ifunc (__ieee754_exp, libm_ifunc (__ieee754_exp,
HAS_ARCH_FEATURE (FMA4_Usable) ? __ieee754_exp_fma4 HAS_ARCH_FEATURE (FMA4_Usable) ? __ieee754_exp_fma4
@ -19,8 +18,7 @@ libm_ifunc (__ieee754_exp,
? __ieee754_exp_avx : __ieee754_exp_sse2)); ? __ieee754_exp_avx : __ieee754_exp_sse2));
strong_alias (__ieee754_exp, __exp_finite) strong_alias (__ieee754_exp, __exp_finite)
# define __ieee754_exp __ieee754_exp_sse2 #define __ieee754_exp __ieee754_exp_sse2
#endif
#include <sysdeps/ieee754/dbl-64/e_exp.c> #include <sysdeps/ieee754/dbl-64/e_exp.c>

22
sysdeps/x86_64/fpu/multiarch/e_log.c
View File

@ -1,17 +1,16 @@
#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT #include <init-arch.h>
# include <init-arch.h> #include <math.h>
# include <math.h> #include <math_private.h>
# include <math_private.h>
extern double __ieee754_log_sse2 (double); extern double __ieee754_log_sse2 (double);
extern double __ieee754_log_avx (double); extern double __ieee754_log_avx (double);
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_log_fma4 (double); extern double __ieee754_log_fma4 (double);
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __ieee754_log_fma4 ((void *) 0) # define __ieee754_log_fma4 ((void *) 0)
# endif #endif
libm_ifunc (__ieee754_log, libm_ifunc (__ieee754_log,
HAS_ARCH_FEATURE (FMA4_Usable) ? __ieee754_log_fma4 HAS_ARCH_FEATURE (FMA4_Usable) ? __ieee754_log_fma4
@ -19,8 +18,7 @@ libm_ifunc (__ieee754_log,
? __ieee754_log_avx : __ieee754_log_sse2)); ? __ieee754_log_avx : __ieee754_log_sse2));
strong_alias (__ieee754_log, __log_finite) strong_alias (__ieee754_log, __log_finite)
# define __ieee754_log __ieee754_log_sse2 #define __ieee754_log __ieee754_log_sse2
#endif
#include <sysdeps/ieee754/dbl-64/e_log.c> #include <sysdeps/ieee754/dbl-64/e_log.c>

20
sysdeps/x86_64/fpu/multiarch/s_atan.c
View File

@ -1,23 +1,21 @@
#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT #include <init-arch.h>
# include <init-arch.h> #include <math.h>
# include <math.h>
extern double __atan_sse2 (double); extern double __atan_sse2 (double);
extern double __atan_avx (double); extern double __atan_avx (double);
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
extern double __atan_fma4 (double); extern double __atan_fma4 (double);
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __atan_fma4 ((void *) 0) # define __atan_fma4 ((void *) 0)
# endif #endif
libm_ifunc (atan, (HAS_ARCH_FEATURE (FMA4_Usable) ? __atan_fma4 : libm_ifunc (atan, (HAS_ARCH_FEATURE (FMA4_Usable) ? __atan_fma4 :
HAS_ARCH_FEATURE (AVX_Usable) HAS_ARCH_FEATURE (AVX_Usable)
? __atan_avx : __atan_sse2)); ? __atan_avx : __atan_sse2));
# define atan __atan_sse2 #define atan __atan_sse2
#endif
#include <sysdeps/ieee754/dbl-64/s_atan.c> #include <sysdeps/ieee754/dbl-64/s_atan.c>

17
sysdeps/x86_64/fpu/multiarch/s_fma.c
View File

@ -21,8 +21,6 @@
#include <math.h> #include <math.h>
#include <init-arch.h> #include <init-arch.h>
#ifdef HAVE_AVX_SUPPORT
extern double __fma_sse2 (double x, double y, double z) attribute_hidden; extern double __fma_sse2 (double x, double y, double z) attribute_hidden;
@ -34,18 +32,18 @@ __fma_fma3 (double x, double y, double z)
} }
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
static double static double
__fma_fma4 (double x, double y, double z) __fma_fma4 (double x, double y, double z)
{ {
asm ("vfmaddsd %3, %2, %1, %0" : "=x" (x) : "x" (x), "x" (y), "x" (z)); asm ("vfmaddsd %3, %2, %1, %0" : "=x" (x) : "x" (x), "x" (y), "x" (z));
return x; return x;
} }
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __fma_fma4 ((void *) 0) # define __fma_fma4 ((void *) 0)
# endif #endif
libm_ifunc (__fma, HAS_ARCH_FEATURE (FMA_Usable) libm_ifunc (__fma, HAS_ARCH_FEATURE (FMA_Usable)
@ -53,7 +51,6 @@ libm_ifunc (__fma, HAS_ARCH_FEATURE (FMA_Usable)
? __fma_fma4 : __fma_sse2)); ? __fma_fma4 : __fma_sse2));
weak_alias (__fma, fma) weak_alias (__fma, fma)
# define __fma __fma_sse2 #define __fma __fma_sse2
#endif
#include <sysdeps/ieee754/dbl-64/s_fma.c> #include <sysdeps/ieee754/dbl-64/s_fma.c>

17
sysdeps/x86_64/fpu/multiarch/s_fmaf.c
View File

@ -20,8 +20,6 @@
#include <math.h> #include <math.h>
#include <init-arch.h> #include <init-arch.h>
#ifdef HAVE_AVX_SUPPORT
extern float __fmaf_sse2 (float x, float y, float z) attribute_hidden; extern float __fmaf_sse2 (float x, float y, float z) attribute_hidden;
@ -33,18 +31,18 @@ __fmaf_fma3 (float x, float y, float z)
} }
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
static float static float
__fmaf_fma4 (float x, float y, float z) __fmaf_fma4 (float x, float y, float z)
{ {
asm ("vfmaddss %3, %2, %1, %0" : "=x" (x) : "x" (x), "x" (y), "x" (z)); asm ("vfmaddss %3, %2, %1, %0" : "=x" (x) : "x" (x), "x" (y), "x" (z));
return x; return x;
} }
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __fmaf_fma4 ((void *) 0) # define __fmaf_fma4 ((void *) 0)
# endif #endif
libm_ifunc (__fmaf, HAS_ARCH_FEATURE (FMA_Usable) libm_ifunc (__fmaf, HAS_ARCH_FEATURE (FMA_Usable)
@ -52,7 +50,6 @@ libm_ifunc (__fmaf, HAS_ARCH_FEATURE (FMA_Usable)
? __fmaf_fma4 : __fmaf_sse2)); ? __fmaf_fma4 : __fmaf_sse2));
weak_alias (__fmaf, fmaf) weak_alias (__fmaf, fmaf)
# define __fmaf __fmaf_sse2 #define __fmaf __fmaf_sse2
#endif
#include <sysdeps/ieee754/dbl-64/s_fmaf.c> #include <sysdeps/ieee754/dbl-64/s_fmaf.c>

26
sysdeps/x86_64/fpu/multiarch/s_sin.c
View File

@ -1,21 +1,20 @@
#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT #include <init-arch.h>
# include <init-arch.h> #include <math.h>
# include <math.h> #undef NAN
# undef NAN
extern double __cos_sse2 (double); extern double __cos_sse2 (double);
extern double __sin_sse2 (double); extern double __sin_sse2 (double);
extern double __cos_avx (double); extern double __cos_avx (double);
extern double __sin_avx (double); extern double __sin_avx (double);
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
extern double __cos_fma4 (double); extern double __cos_fma4 (double);
extern double __sin_fma4 (double); extern double __sin_fma4 (double);
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __cos_fma4 ((void *) 0) # define __cos_fma4 ((void *) 0)
# define __sin_fma4 ((void *) 0) # define __sin_fma4 ((void *) 0)
# endif #endif
libm_ifunc (__cos, (HAS_ARCH_FEATURE (FMA4_Usable) ? __cos_fma4 : libm_ifunc (__cos, (HAS_ARCH_FEATURE (FMA4_Usable) ? __cos_fma4 :
HAS_ARCH_FEATURE (AVX_Usable) HAS_ARCH_FEATURE (AVX_Usable)
@ -27,9 +26,8 @@ libm_ifunc (__sin, (HAS_ARCH_FEATURE (FMA4_Usable) ? __sin_fma4 :
? __sin_avx : __sin_sse2)); ? __sin_avx : __sin_sse2));
weak_alias (__sin, sin) weak_alias (__sin, sin)
# define __cos __cos_sse2 #define __cos __cos_sse2
# define __sin __sin_sse2 #define __sin __sin_sse2
#endif
#include <sysdeps/ieee754/dbl-64/s_sin.c> #include <sysdeps/ieee754/dbl-64/s_sin.c>

20
sysdeps/x86_64/fpu/multiarch/s_tan.c
View File

@ -1,23 +1,21 @@
#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT #include <init-arch.h>
# include <init-arch.h> #include <math.h>
# include <math.h>
extern double __tan_sse2 (double); extern double __tan_sse2 (double);
extern double __tan_avx (double); extern double __tan_avx (double);
# ifdef HAVE_FMA4_SUPPORT #ifdef HAVE_FMA4_SUPPORT
extern double __tan_fma4 (double); extern double __tan_fma4 (double);
# else #else
# undef HAS_ARCH_FEATURE # undef HAS_ARCH_FEATURE
# define HAS_ARCH_FEATURE(feature) 0 # define HAS_ARCH_FEATURE(feature) 0
# define __tan_fma4 ((void *) 0) # define __tan_fma4 ((void *) 0)
# endif #endif
libm_ifunc (tan, (HAS_ARCH_FEATURE (FMA4_Usable) ? __tan_fma4 : libm_ifunc (tan, (HAS_ARCH_FEATURE (FMA4_Usable) ? __tan_fma4 :
HAS_ARCH_FEATURE (AVX_Usable) HAS_ARCH_FEATURE (AVX_Usable)
? __tan_avx : __tan_sse2)); ? __tan_avx : __tan_sse2));
# define tan __tan_sse2 #define tan __tan_sse2
#endif
#include <sysdeps/ieee754/dbl-64/s_tan.c> #include <sysdeps/ieee754/dbl-64/s_tan.c>

22
sysdeps/x86_64/multiarch/strcmp.S
View File

@ -107,11 +107,9 @@ END(STRCMP)
ENTRY(__strcasecmp) ENTRY(__strcasecmp)
.type __strcasecmp, @gnu_indirect_function .type __strcasecmp, @gnu_indirect_function
LOAD_RTLD_GLOBAL_RO_RDX LOAD_RTLD_GLOBAL_RO_RDX
# ifdef HAVE_AVX_SUPPORT
leaq __strcasecmp_avx(%rip), %rax leaq __strcasecmp_avx(%rip), %rax
HAS_ARCH_FEATURE (AVX_Usable) HAS_ARCH_FEATURE (AVX_Usable)
jnz 3f jnz 3f
# endif
HAS_ARCH_FEATURE (Slow_SSE4_2) HAS_ARCH_FEATURE (Slow_SSE4_2)
jnz 2f jnz 2f
leaq __strcasecmp_sse42(%rip), %rax leaq __strcasecmp_sse42(%rip), %rax
@ -129,11 +127,9 @@ weak_alias (__strcasecmp, strcasecmp)
ENTRY(__strncasecmp) ENTRY(__strncasecmp)
.type __strncasecmp, @gnu_indirect_function .type __strncasecmp, @gnu_indirect_function
LOAD_RTLD_GLOBAL_RO_RDX LOAD_RTLD_GLOBAL_RO_RDX
# ifdef HAVE_AVX_SUPPORT
leaq __strncasecmp_avx(%rip), %rax leaq __strncasecmp_avx(%rip), %rax
HAS_ARCH_FEATURE (AVX_Usable) HAS_ARCH_FEATURE (AVX_Usable)
jnz 3f jnz 3f
# endif
HAS_ARCH_FEATURE (Slow_SSE4_2) HAS_ARCH_FEATURE (Slow_SSE4_2)
jnz 2f jnz 2f
leaq __strncasecmp_sse42(%rip), %rax leaq __strncasecmp_sse42(%rip), %rax
@ -155,16 +151,14 @@ weak_alias (__strncasecmp, strncasecmp)
# include "strcmp-sse42.S" # include "strcmp-sse42.S"
# ifdef HAVE_AVX_SUPPORT # if defined USE_AS_STRCASECMP_L || defined USE_AS_STRNCASECMP_L
# if defined USE_AS_STRCASECMP_L || defined USE_AS_STRNCASECMP_L # define LABEL(l) .L##l##_avx
# define LABEL(l) .L##l##_avx # define GLABEL(l) l##_avx
# define GLABEL(l) l##_avx # define USE_AVX 1
# define USE_AVX 1 # undef STRCMP_SSE42
# undef STRCMP_SSE42 # define STRCMP_SSE42 STRCMP_AVX
# define STRCMP_SSE42 STRCMP_AVX # define SECTION avx
# define SECTION avx # include "strcmp-sse42.S"
# include "strcmp-sse42.S"
# endif
# endif # endif