aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/x86_64
diff options
context:
space:
mode:
authorUlrich Drepper <drepper@gmail.com>2011-10-25 21:34:55 -0400
committerUlrich Drepper <drepper@gmail.com>2011-10-25 21:34:55 -0400
commite0016b11d6186a7003d7204cb100acab9bbcd940 (patch)
tree8c702a330664d113edbdaff9ba4ec67e266bbb80 /sysdeps/x86_64
parentffb124cc518c182e238c20abb0ff8bdeaefefd66 (diff)
downloadglibc-e0016b11d6186a7003d7204cb100acab9bbcd940.tar
glibc-e0016b11d6186a7003d7204cb100acab9bbcd940.tar.gz
glibc-e0016b11d6186a7003d7204cb100acab9bbcd940.tar.bz2
glibc-e0016b11d6186a7003d7204cb100acab9bbcd940.zip
Add AVX optimized versions for some x86-64 math functions
Diffstat (limited to 'sysdeps/x86_64')
-rw-r--r--sysdeps/x86_64/fpu/multiarch/Makefile33
-rw-r--r--sysdeps/x86_64/fpu/multiarch/brandred-avx.c4
-rw-r--r--sysdeps/x86_64/fpu/multiarch/doasin-avx.c4
-rw-r--r--sysdeps/x86_64/fpu/multiarch/dosincos-avx.c6
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_asin-avx.c11
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_asin.c19
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_atan2-avx.c10
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_atan2.c12
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_exp-avx.c6
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_exp.c13
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_log-avx.c8
-rw-r--r--sysdeps/x86_64/fpu/multiarch/e_log.c14
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mpa-avx.c12
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mpatan-avx.c10
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mpatan2-avx.c9
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mpexp-avx.c9
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mplog-avx.c8
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mpsqrt-avx.c8
-rw-r--r--sysdeps/x86_64/fpu/multiarch/mptan-avx.c7
-rw-r--r--sysdeps/x86_64/fpu/multiarch/s_atan-avx.c9
-rw-r--r--sysdeps/x86_64/fpu/multiarch/s_atan.c11
-rw-r--r--sysdeps/x86_64/fpu/multiarch/s_sin-avx.c12
-rw-r--r--sysdeps/x86_64/fpu/multiarch/s_sin.c17
-rw-r--r--sysdeps/x86_64/fpu/multiarch/s_tan-avx.c9
-rw-r--r--sysdeps/x86_64/fpu/multiarch/s_tan.c11
-rw-r--r--sysdeps/x86_64/fpu/multiarch/sincos32-avx.c15
-rw-r--r--sysdeps/x86_64/fpu/multiarch/slowexp-avx.c9
-rw-r--r--sysdeps/x86_64/multiarch/init-arch.h32
28 files changed, 294 insertions, 34 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
index 70cb740aac..f97ce1699d 100644
--- a/sysdeps/x86_64/fpu/multiarch/Makefile
+++ b/sysdeps/x86_64/fpu/multiarch/Makefile
@@ -30,7 +30,36 @@ CFLAGS-s_atan-fma4.c = -mfma4
CFLAGS-sincos32-fma4.c = -mfma4
CFLAGS-slowexp-fma4.c = -mfma4
CFLAGS-slowpow-fma4.c = -mfma4
-CLFAGS-s_sin-fma4.c = -mfma4
-CLFAGS-s_tan-fma4.c = -mfma4
+CFLAGS-s_sin-fma4.c = -mfma4
+CFLAGS-s_tan-fma4.c = -mfma4
+endif
+
+ifeq ($(config-cflags-avx),yes)
+libm-sysdep_routines += e_exp-avx e_log-avx s_atan-avx \
+ e_asin-avx e_atan2-avx s_sin-avx s_tan-avx \
+ mplog-avx mpa-avx slowexp-avx \
+ sincos32-avx doasin-avx dosincos-avx \
+ brandred-avx mpexp-avx \
+ mpatan2-avx mpatan-avx mpsqrt-avx mptan-avx
+
+CFLAGS-brandred-avx.c = -mavx
+CFLAGS-doasin-avx.c = -mavx
+CFLAGS-dosincos-avx.c = -mavx
+CFLAGS-e_asin-avx.c = -mavx
+CFLAGS-e_atan2-avx.c = -mavx
+CFLAGS-e_exp-avx.c = -mavx
+CFLAGS-e_log-avx.c = -mavx
+CFLAGS-mpa-avx.c = -mavx
+CFLAGS-mpatan-avx.c = -mavx
+CFLAGS-mpatan2-avx.c = -mavx
+CFLAGS-mpexp-avx.c = -mavx
+CFLAGS-mplog-avx.c = -mavx
+CFLAGS-mpsqrt-avx.c = -mavx
+CFLAGS-mptan-avx.c = -mavx
+CFLAGS-s_atan-avx.c = -mavx
+CFLAGS-s_sin-avx.c = -mavx
+CFLAGS-sincos32-avx.c = -mavx
+CFLAGS-slowexp-avx.c = -mavx
+CFLAGS-s_tan-avx.c = -mavx
endif
endif
diff --git a/sysdeps/x86_64/fpu/multiarch/brandred-avx.c b/sysdeps/x86_64/fpu/multiarch/brandred-avx.c
new file mode 100644
index 0000000000..2f6edffadd
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/brandred-avx.c
@@ -0,0 +1,4 @@
+#define __branred __branred_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/branred.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/doasin-avx.c b/sysdeps/x86_64/fpu/multiarch/doasin-avx.c
new file mode 100644
index 0000000000..86781c1562
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/doasin-avx.c
@@ -0,0 +1,4 @@
+#define __doasin __doasin_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/doasin.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/dosincos-avx.c b/sysdeps/x86_64/fpu/multiarch/dosincos-avx.c
new file mode 100644
index 0000000000..ffa18743df
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/dosincos-avx.c
@@ -0,0 +1,6 @@
+#define __docos __docos_avx
+#define __dubcos __dubcos_avx
+#define __dubsin __dubsin_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/dosincos.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_asin-avx.c b/sysdeps/x86_64/fpu/multiarch/e_asin-avx.c
new file mode 100644
index 0000000000..7f89e4f14e
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_asin-avx.c
@@ -0,0 +1,11 @@
+#define __ieee754_acos __ieee754_acos_avx
+#define __ieee754_asin __ieee754_asin_avx
+#define __cos32 __cos32_avx
+#define __doasin __doasin_avx
+#define __docos __docos_avx
+#define __dubcos __dubcos_avx
+#define __dubsin __dubsin_avx
+#define __sin32 __sin32_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/e_asin.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_asin.c b/sysdeps/x86_64/fpu/multiarch/e_asin.c
index 8882cead9d..bb711b1ffc 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_asin.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_asin.c
@@ -1,18 +1,29 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math_private.h>
extern double __ieee754_acos_sse2 (double);
-extern double __ieee754_acos_fma4 (double);
extern double __ieee754_asin_sse2 (double);
+extern double __ieee754_acos_avx (double);
+extern double __ieee754_asin_avx (double);
+# ifdef HAVE_FMA4_SUPPORT
+extern double __ieee754_acos_fma4 (double);
extern double __ieee754_asin_fma4 (double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __ieee754_acos_fma4 ((void *) 0)
+# define __ieee754_asin_fma4 ((void *) 0)
+# endif
libm_ifunc (__ieee754_acos,
- HAS_FMA4 ? __ieee754_acos_fma4 : __ieee754_acos_sse2);
+ HAS_FMA4 ? __ieee754_acos_fma4
+ : (HAS_AVX ? __ieee754_acos_avx : __ieee754_acos_sse2));
strong_alias (__ieee754_acos, __acos_finite)
libm_ifunc (__ieee754_asin,
- HAS_FMA4 ? __ieee754_asin_fma4 : __ieee754_asin_sse2);
+ HAS_FMA4 ? __ieee754_asin_fma4
+ : (HAS_AVX ? __ieee754_asin_avx : __ieee754_asin_sse2));
strong_alias (__ieee754_asin, __asin_finite)
# define __ieee754_acos __ieee754_acos_sse2
diff --git a/sysdeps/x86_64/fpu/multiarch/e_atan2-avx.c b/sysdeps/x86_64/fpu/multiarch/e_atan2-avx.c
new file mode 100644
index 0000000000..13155c87dc
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_atan2-avx.c
@@ -0,0 +1,10 @@
+#define __ieee754_atan2 __ieee754_atan2_avx
+#define __add __add_avx
+#define __dbl_mp __dbl_mp_avx
+#define __dvd __dvd_avx
+#define __mpatan2 __mpatan2_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/e_atan2.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_atan2.c b/sysdeps/x86_64/fpu/multiarch/e_atan2.c
index 12fc929068..6867c6e64e 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_atan2.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_atan2.c
@@ -1,12 +1,20 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math_private.h>
extern double __ieee754_atan2_sse2 (double, double);
+extern double __ieee754_atan2_avx (double, double);
+# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_atan2_fma4 (double, double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __ieee754_atan2_fma4 ((void *) 0)
+# endif
libm_ifunc (__ieee754_atan2,
- HAS_FMA4 ? __ieee754_atan2_fma4 : __ieee754_atan2_sse2);
+ HAS_FMA4 ? __ieee754_atan2_fma4
+ : (HAS_AVX ? __ieee754_atan2_avx : __ieee754_atan2_sse2));
strong_alias (__ieee754_atan2, __atan2_finite)
# define __ieee754_atan2 __ieee754_atan2_sse2
diff --git a/sysdeps/x86_64/fpu/multiarch/e_exp-avx.c b/sysdeps/x86_64/fpu/multiarch/e_exp-avx.c
new file mode 100644
index 0000000000..ee5dd6d2dc
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_exp-avx.c
@@ -0,0 +1,6 @@
+#define __ieee754_exp __ieee754_exp_avx
+#define __exp1 __exp1_avx
+#define __slowexp __slowexp_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/e_exp.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_exp.c b/sysdeps/x86_64/fpu/multiarch/e_exp.c
index fc1096b54c..3c650287c5 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_exp.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_exp.c
@@ -1,11 +1,20 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math_private.h>
extern double __ieee754_exp_sse2 (double);
+extern double __ieee754_exp_avx (double);
+# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_exp_fma4 (double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __ieee754_exp_fma4 ((void *) 0)
+# endif
-libm_ifunc (__ieee754_exp, HAS_FMA4 ? __ieee754_exp_fma4 : __ieee754_exp_sse2);
+libm_ifunc (__ieee754_exp,
+ HAS_FMA4 ? __ieee754_exp_fma4
+ : (HAS_AVX ? __ieee754_exp_avx : __ieee754_exp_sse2));
strong_alias (__ieee754_exp, __exp_finite)
# define __ieee754_exp __ieee754_exp_sse2
diff --git a/sysdeps/x86_64/fpu/multiarch/e_log-avx.c b/sysdeps/x86_64/fpu/multiarch/e_log-avx.c
new file mode 100644
index 0000000000..c669019bc2
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/e_log-avx.c
@@ -0,0 +1,8 @@
+#define __ieee754_log __ieee754_log_avx
+#define __mplog __mplog_avx
+#define __add __add_avx
+#define __dbl_mp __dbl_mp_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/e_log.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/e_log.c b/sysdeps/x86_64/fpu/multiarch/e_log.c
index c54264609c..3b468d0f46 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_log.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_log.c
@@ -1,11 +1,21 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math_private.h>
extern double __ieee754_log_sse2 (double);
+extern double __ieee754_log_avx (double);
+# ifdef HAVE_FMA4_SUPPORT
extern double __ieee754_log_fma4 (double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __ieee754_log_fma4 ((void *) 0)
+# endif
-libm_ifunc (__ieee754_log, HAS_FMA4 ? __ieee754_log_fma4 : __ieee754_log_sse2);
+libm_ifunc (__ieee754_log,
+ HAS_FMA4 ? __ieee754_log_fma4
+ : (HAS_AVX ? __ieee754_log_avx
+ : __ieee754_log_sse2));
strong_alias (__ieee754_log, __log_finite)
# define __ieee754_log __ieee754_log_sse2
diff --git a/sysdeps/x86_64/fpu/multiarch/mpa-avx.c b/sysdeps/x86_64/fpu/multiarch/mpa-avx.c
new file mode 100644
index 0000000000..a92dbedc05
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpa-avx.c
@@ -0,0 +1,12 @@
+#define __add __add_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define __dbl_mp __dbl_mp_avx
+#define __dvd __dvd_avx
+
+#define NO___CPY 1
+#define NO___MP_DBL 1
+#define NO___ACR 1
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mpa.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpatan-avx.c b/sysdeps/x86_64/fpu/multiarch/mpatan-avx.c
new file mode 100644
index 0000000000..e26e1fa3e9
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpatan-avx.c
@@ -0,0 +1,10 @@
+#define __mpatan __mpatan_avx
+#define __add __add_avx
+#define __dvd __dvd_avx
+#define __mpsqrt __mpsqrt_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define AVOID_MPATAN_H 1
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mpatan.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpatan2-avx.c b/sysdeps/x86_64/fpu/multiarch/mpatan2-avx.c
new file mode 100644
index 0000000000..fa824d69bf
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpatan2-avx.c
@@ -0,0 +1,9 @@
+#define __mpatan2 __mpatan2_avx
+#define __add __add_avx
+#define __dvd __dvd_avx
+#define __mpatan __mpatan_avx
+#define __mpsqrt __mpsqrt_avx
+#define __mul __mul_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mpatan2.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpexp-avx.c b/sysdeps/x86_64/fpu/multiarch/mpexp-avx.c
new file mode 100644
index 0000000000..87f29c96c9
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpexp-avx.c
@@ -0,0 +1,9 @@
+#define __mpexp __mpexp_avx
+#define __add __add_avx
+#define __dbl_mp __dbl_mp_avx
+#define __dvd __dvd_avx
+#define __mul __mul_avx
+#define AVOID_MPEXP_H 1
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mpexp.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mplog-avx.c b/sysdeps/x86_64/fpu/multiarch/mplog-avx.c
new file mode 100644
index 0000000000..fd783d9a67
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mplog-avx.c
@@ -0,0 +1,8 @@
+#define __mplog __mplog_avx
+#define __add __add_avx
+#define __mpexp __mpexp_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mplog.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mpsqrt-avx.c b/sysdeps/x86_64/fpu/multiarch/mpsqrt-avx.c
new file mode 100644
index 0000000000..26f2fb30fb
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mpsqrt-avx.c
@@ -0,0 +1,8 @@
+#define __mpsqrt __mpsqrt_avx
+#define __dbl_mp __dbl_mp_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define AVOID_MPSQRT_H 1
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mpsqrt.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/mptan-avx.c b/sysdeps/x86_64/fpu/multiarch/mptan-avx.c
new file mode 100644
index 0000000000..02514b7839
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/mptan-avx.c
@@ -0,0 +1,7 @@
+#define __mptan __mptan_avx
+#define __c32 __c32_avx
+#define __dvd __dvd_avx
+#define __mpranred __mpranred_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/mptan.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_atan-avx.c b/sysdeps/x86_64/fpu/multiarch/s_atan-avx.c
new file mode 100644
index 0000000000..8e7022f51f
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_atan-avx.c
@@ -0,0 +1,9 @@
+#define atan __atan_avx
+#define __add __add_avx
+#define __dbl_mp __dbl_mp_avx
+#define __mpatan __mpatan_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/s_atan.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_atan.c b/sysdeps/x86_64/fpu/multiarch/s_atan.c
index ffc4a56fa8..3160201c17 100644
--- a/sysdeps/x86_64/fpu/multiarch/s_atan.c
+++ b/sysdeps/x86_64/fpu/multiarch/s_atan.c
@@ -1,11 +1,18 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math.h>
extern double __atan_sse2 (double);
+extern double __atan_avx (double);
+# ifdef HAVE_FMA4_SUPPORT
extern double __atan_fma4 (double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __atan_fma4 ((void *) 0)
+# endif
-libm_ifunc (atan, HAS_FMA4 ? __atan_fma4 : __atan_sse2);
+libm_ifunc (atan, HAS_FMA4 ? __atan_fma4 : HAS_AVX ? __atan_avx : __atan_sse2);
# define atan __atan_sse2
#endif
diff --git a/sysdeps/x86_64/fpu/multiarch/s_sin-avx.c b/sysdeps/x86_64/fpu/multiarch/s_sin-avx.c
new file mode 100644
index 0000000000..c0ccf4be77
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_sin-avx.c
@@ -0,0 +1,12 @@
+#define __cos __cos_avx
+#define __sin __sin_avx
+#define __branred __branred_avx
+#define __docos __docos_avx
+#define __dubsin __dubsin_avx
+#define __mpcos __mpcos_avx
+#define __mpcos1 __mpcos1_avx
+#define __mpsin __mpsin_avx
+#define __mpsin1 __mpsin1_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/s_sin.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_sin.c b/sysdeps/x86_64/fpu/multiarch/s_sin.c
index a7c35dc858..1ba9dbc50c 100644
--- a/sysdeps/x86_64/fpu/multiarch/s_sin.c
+++ b/sysdeps/x86_64/fpu/multiarch/s_sin.c
@@ -1,17 +1,26 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math.h>
# undef NAN
extern double __cos_sse2 (double);
-extern double __cos_fma4 (double);
extern double __sin_sse2 (double);
+extern double __cos_avx (double);
+extern double __sin_avx (double);
+# ifdef HAVE_FMA4_SUPPORT
+extern double __cos_fma4 (double);
extern double __sin_fma4 (double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __cos_fma4 ((void *) 0)
+# define __sin_fma4 ((void *) 0)
+# endif
-libm_ifunc (__cos, HAS_FMA4 ? __cos_fma4 : __cos_sse2);
+libm_ifunc (__cos, HAS_FMA4 ? __cos_fma4 : HAS_AVX ? __cos_avx : __cos_sse2);
weak_alias (__cos, cos)
-libm_ifunc (__sin, HAS_FMA4 ? __sin_fma4 : __sin_sse2);
+libm_ifunc (__sin, HAS_FMA4 ? __sin_fma4 : HAS_AVX ? __sin_avx : __sin_sse2);
weak_alias (__sin, sin)
# define __cos __cos_sse2
diff --git a/sysdeps/x86_64/fpu/multiarch/s_tan-avx.c b/sysdeps/x86_64/fpu/multiarch/s_tan-avx.c
new file mode 100644
index 0000000000..f31ff65b6c
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_tan-avx.c
@@ -0,0 +1,9 @@
+#define tan __tan_avx
+#define __branred __branred_avx
+#define __dbl_mp __dbl_mp_avx
+#define __mpranred __mpranred_avx
+#define __mptan __mptan_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/s_tan.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_tan.c b/sysdeps/x86_64/fpu/multiarch/s_tan.c
index cca02b54de..8f6601e17c 100644
--- a/sysdeps/x86_64/fpu/multiarch/s_tan.c
+++ b/sysdeps/x86_64/fpu/multiarch/s_tan.c
@@ -1,11 +1,18 @@
-#ifdef HAVE_FMA4_SUPPORT
+#if defined HAVE_FMA4_SUPPORT || defined HAVE_AVX_SUPPORT
# include <init-arch.h>
# include <math.h>
extern double __tan_sse2 (double);
+extern double __tan_avx (double);
+# ifdef HAVE_FMA4_SUPPORT
extern double __tan_fma4 (double);
+# else
+# undef HAS_FMA4
+# define HAS_FMA4 0
+# define __tan_fma4 ((void *) 0)
+# endif
-libm_ifunc (tan, HAS_FMA4 ? __tan_fma4 : __tan_sse2);
+libm_ifunc (tan, HAS_FMA4 ? __tan_fma4 : HAS_AVX ? __tan_avx : __tan_sse2);
# define tan __tan_sse2
#endif
diff --git a/sysdeps/x86_64/fpu/multiarch/sincos32-avx.c b/sysdeps/x86_64/fpu/multiarch/sincos32-avx.c
new file mode 100644
index 0000000000..5af51ab38d
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/sincos32-avx.c
@@ -0,0 +1,15 @@
+#define __cos32 __cos32_avx
+#define __sin32 __sin32_avx
+#define __c32 __c32_avx
+#define __mpsin __mpsin_avx
+#define __mpsin1 __mpsin1_avx
+#define __mpcos __mpcos_avx
+#define __mpcos1 __mpcos1_avx
+#define __mpranred __mpranred_avx
+#define __add __add_avx
+#define __dbl_mp __dbl_mp_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/sincos32.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/slowexp-avx.c b/sysdeps/x86_64/fpu/multiarch/slowexp-avx.c
new file mode 100644
index 0000000000..d01c6d71a4
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/slowexp-avx.c
@@ -0,0 +1,9 @@
+#define __slowexp __slowexp_avx
+#define __add __add_avx
+#define __dbl_mp __dbl_mp_avx
+#define __mpexp __mpexp_avx
+#define __mul __mul_avx
+#define __sub __sub_avx
+#define SECTION __attribute__ ((section (".text.avx")))
+
+#include <sysdeps/ieee754/dbl-64/slowexp.c>
diff --git a/sysdeps/x86_64/multiarch/init-arch.h b/sysdeps/x86_64/multiarch/init-arch.h
index 2fb6f75b66..2a1df393c4 100644
--- a/sysdeps/x86_64/multiarch/init-arch.h
+++ b/sysdeps/x86_64/multiarch/init-arch.h
@@ -23,16 +23,19 @@
#define bit_Fast_Unaligned_Load (1 << 4)
#define bit_Prefer_PMINUB_for_stringop (1 << 5)
+#define bit_SSE2 (1 << 26)
+#define bit_SSSE3 (1 << 9)
+#define bit_SSE4_1 (1 << 19)
+#define bit_SSE4_2 (1 << 20)
+#define bit_AVX (1 << 28)
+#define bit_POPCOUNT (1 << 23)
+#define bit_FMA (1 << 12)
+#define bit_FMA4 (1 << 16)
+
#ifdef __ASSEMBLER__
# include <ifunc-defines.h>
-# define bit_SSE2 (1 << 26)
-# define bit_SSSE3 (1 << 9)
-# define bit_SSE4_1 (1 << 19)
-# define bit_SSE4_2 (1 << 20)
-# define bit_AVX (1 << 28)
-
# define index_SSE2 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
# define index_SSSE3 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
# define index_SSE4_1 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
@@ -104,17 +107,18 @@ extern const struct cpu_features *__get_cpu_features (void)
# endif
# define HAS_CPU_FEATURE(idx, reg, bit) \
- ((__get_cpu_features ()->cpuid[idx].reg & (1 << (bit))) != 0)
+ ((__get_cpu_features ()->cpuid[idx].reg & (bit)) != 0)
/* Following are the feature tests used throughout libc. */
-# define HAS_SSE2 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, edx, 26)
-# define HAS_POPCOUNT HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 23)
-# define HAS_SSSE3 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 9)
-# define HAS_SSE4_1 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 19)
-# define HAS_SSE4_2 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 20)
-# define HAS_FMA HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 12)
-# define HAS_FMA4 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_80000001, ecx, 16)
+# define HAS_SSE2 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, edx, bit_SSE2)
+# define HAS_POPCOUNT HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_POPCOUNT)
+# define HAS_SSSE3 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSSE3)
+# define HAS_SSE4_1 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSE4_1)
+# define HAS_SSE4_2 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSE4_2)
+# define HAS_FMA HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_FMA)
+# define HAS_AVX HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_AVX)
+# define HAS_FMA4 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_80000001, ecx, bit_FMA4)
# define index_Fast_Rep_String FEATURE_INDEX_1
# define index_Fast_Copy_Backward FEATURE_INDEX_1